Visit TrendStats to create custom bar charts, stacked bar charts, pie charts, and line graphs from your TrendStats analysis.
It is easy. On the TrendStats homepage, click the Create Chart tab and choose your analysis type.
Now available in PowerStats: the 1988 National Study of Postsecondary Faculty (NSOPF:88). Learn more about the National Study of Postsecondary Faculty (NSOPF).
Now available in PowerStats: the 1999 National Study of Postsecondary Faculty (NSOPF:99). Learn more about the National Study of Postsecondary Faculty (NSOPF).
Now available in PowerStats: the 1993 National Study of Postsecondary Faculty (NSOPF:93). Learn more about the National Study of Postsecondary Faculty (NSOPF).
Tables for One Year After a Bachelor’s Degree: A Profile of 2015-16 Graduates are now available in the DataLab Tables Library.
Tables for First-Time Subbaccalaureate Students: An Overview of Their Institutions, Programs, Completion, and Labor Market Outcomes After 3 Years are now available in the DataLab Tables Library.
Now available in PowerStats: the 1992-93 National Postsecondary Student Aid Study (NPSAS:93). Learn more about the Postsecondary Student Aid Study (NPSAS).
Now available in PowerStats: the 1986-87 National Postsecondary Student Aid Study (NPSAS:87). Learn more about the Postsecondary Student Aid Study (NPSAS).
Now available in PowerStats: the 1989-90 National Postsecondary Student Aid Study (NPSAS:90). Learn more about the Postsecondary Student Aid Study (NPSAS).
Tables for What Is the Price of College? Total, Net, and Out-of-Pocket Prices in 2015-16 are now available in the DataLab Tables Library.
Tables for High School Longitudinal Study of 2009 (HSLS:09): A First Look at the Postsecondary Transcripts and Student Financial Aid Records of Fall 2009 Ninth-Graders are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: the 1999-2000 School Survey on Crime and Safety (SSOCS). Learn more about SSOCS.
Download the Creating a TrendStats Chart (PDF, 3.04 MB) tutorial to learn more about creating a custom chart in TrendStats.
Now available in PowerStats and QuickStats: 2016/2017 Baccalaureate and Beyond (B&B:16/17). Learn more about Baccalaureate and Beyond.
Now available in PowerStats and QuickStats: the 2003-04 School Survey on Crime and Safety (SSOCS). Learn more about SSOCS.
Tables for Trends in Graduate Student Financing: Selected Years, 2003–04 to 2015–16 are now available in the DataLab Tables Library.
Tables for Trends in Undergraduate Nonfederal Grant and Scholarship Aid by Demographic and Enrollment Characteristics: Selected Years, 2003–04 to 2015–16 are now available in the DataLab Tables Library.
Tables for Trends in Pell Grant Receipt and the Characteristics of Pell Grant Recipients: Selected Years, 2003–04 to 2015–16 are now available in the DataLab Tables Library.
Tables for Trends in Ratio of Pell Grant to Total Price of Attendance and Federal Loan Receipt are now available in the DataLab Tables Library.
Newly released data are now available in PowerStats and QuickStats. The 2011-12 Beginning Postsecondary Students Longitudinal Study (BPS:12/17) now includes data from the second follow-up in 2017. Learn more about the 2011-12 Beginning Postsecondary Students Longitudinal Study (BPS:12/17).
Tables for Baccalaureate and Beyond (B&B:16/17): A First Look at the Employment and Educational Experiences of College Graduates, 1 Year Later are now available in the DataLab Tables Library.
Tables for Profile of Very Low- and Low-Income Undergraduates in 2015–16 are now available in the DataLab Tables Library.
Tables for Considerations for Using the School Courses for the Exchange of Data (SCED) Classification System in High School Transcript Studies: Applications for Converting Course Codes from the Classification of Secondary School Courses (CSSC) are now available in the DataLab Tables Library.
Tables for Changes in Undergraduate Program Completers’ Borrowing Rates and Loan Amounts by Age: 1995–96 Through 2015–16 are now available in the DataLab Tables Library.
Tables for Advanced Placement, International Baccalaureate, and Dual-Enrollment Courses: Availability, Participation, and Related Outcomes for 2009 Ninth-Graders: 2013 are now available in the DataLab Tables Library.
TrendStas ChartsVisit TrendStats to create custom bar charts, stacked bar charts, pie charts, and line graphs from your TrendStats analysis.
Tables for Going Back to College: Undergraduates Who Already Held a Postsecondary Credential are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: the High School and Beyond Longitudinal Study and the National Education Longitudinal Study of 1988. Learn more about HS&B and NELS.
Tables for Student Financing of Undergraduate Education in 2015–16: Students’ Net Price, Expected Family Contribution, and Financial Need are now available in the DataLab Tables Library
Tables for Student Financing of Undergraduate Education in 2015–16: Income, Tuition, and Total Price are now available in the DataLab Tables Library
Tables for Student Financing of Undergraduate Education in 2015–16: Financial Aid by Type and Source are now available in the DataLab Tables Library.
Tables for Profile and Financial Aid Estimates of Graduate Students: 2015–16 are now available in the DataLab Tables Library.
Tables for Profile of Undergraduate Students: Attendance, Distance and Remedial Education, Degree Program and Field of Study, Demographics, Financial Aid, Financial Literacy, Employment, and Military Status: 2015–16 are now available in the DataLab Tables Library.
Tables for Persistence, Retention, and Attainment of 2011–12 First-Time Beginning Postsecondary Students as of Spring 2017 are now available in the DataLab Tables Library.
Tables for Military Service and Educational Attainment of High School Sophomores After 9/11: Experiences of 2002 High School Sophomores as of 2012 are now available in the DataLab Tables Library.
New variables released in Baccalaureate and Beyond: 2008/2012 PowerStats. See a list of variables or learn more.
B&B 2012 Students New variables now available in B&B 2012 PowerStats. The following variables have been added: Variable Name Variable Label B2DISTINSTE Distance between primary job in 2012 and bachelor's degree institution B2DISTINSTR Distance between residence in 2012 and bachelor's degree institution B2SMSTE Primary job in 2012 is in same state as bachelor's degree institution state B2SMSTER Primary job and residence in 2012 are in same state as bachelor's degree institution state B2SMSTR Residence in 2012 is in same state as bachelor's degree institution state B2STCDE State of primary job: 2012 B2STCDR State of residence: 2012
New variables now available in B&B 2012 PowerStats. The following variables have been added:
Now available in PowerStats, QuickStats, and TrendStats: the National Household Education Survey (NHES) for 2012 and 2016 Parent and Family Involvement in Education (PFI). Learn more about PFI.
Tables for What High Schoolers and Their Parents Know About Public 4-Year Tuition and Fees in Their State are now available in the DataLab Tables Library.
New IPEDS Tables are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: the National Household Education Survey (NHES) for 2016 Adult Training and Education (ATES). Learn more about ATES.
Now available in PowerStats, QuickStats, and TrendStats: the National Household Education Survey (NHES) for 2012 and 2016 Early Childhood Program Participation (ECPP). Learn more about ECPP.
QuickStats ChartsVisit QuickStats to view the latest enhancements to charting options, including new chart types, color switching, and more.
Introducing the DataLab Tables LibraryThe College & Career Tables Library is now the DataLab Tables Library. Visit the DataLab Tables Library.
Tables for Trends in Free Application for Federal Student Aid (FAFSA) Submissions are now available in the DataLab Tables Library.
Tables for Four Years Later: 2007–08 College Graduates' Employment, Debt, and Enrollment in 2012 (2018435) are now available in the DataLab Tables Library.
Tables for Working Before, During, and After Beginning at a Public 2-Year Institution: Labor Market Experiences of Community College Students (2018428) are now available in the DataLab Tables Library.
Tables for First-Generation Students: College Access, Persistence, and Postbachelor’s Outcomes (2018401) are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: the 2015-16 School Survey on Crime and Safety (SSOCS). Learn more about SSOCS.
Now available in PowerStats and QuickStats: the High School Longitudinal Study of 2009 (HSLS:09) Second Follow-Up. Learn more about HSLS.
Tables for High School Longitudinal Study of 2009 (HSLS:09) Second Follow Up: A First Look at Fall 2009 Ninth–Graders in 2016 (2018139) are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: the 2015-16 National Teacher and Principal Survey (NTPS). Learn more about NTPS.
Now available in PowerStats and QuickStats: the 2015-16 National Postsecondary Student Aid Study, Undergraduate Students (NPSAS:2016 UG). Learn more about NPSAS.
Now available in PowerStats and QuickStats: the 2015-16 National Postsecondary Student Aid Study, Graduate Students (NPSAS:2016 GR). Learn more about NPSAS.
New variables released in NPSAS 1995-96, 1999-2000, 2003-04, 2007-08, and 2011-12 Undergraduate PowerStats. See a list of variables or learn more.
NPSAS Undergraduate Students New variables now available in NPSAS Undergraduates PowerStats. The following variables have been added: Variable Name Variable Label INGRTAMT2Institutional grants total (updated) INSTAMT2Institutional aid total (updated) INSTNEED2Institutional need-based grants (updated) STATNEED2State need-based grants (updated) STGTAMT2State grants total (updated) STATEAMT2State aid total (updated)
New variables now available in NPSAS Undergraduates PowerStats. The following variables have been added:
Tables for National Postsecondary Student Aid Study (NPSAS:16): Student Financial Aid Estimates for 2015–16 First Look (2018466) are now available in the DataLab Tables Library.
Tables for First-Generation Students: College Access, Persistence, and Postbachelor's Outcomes (2018421) are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: the 1999-00, 2003-04, and 2007-08 Schools and Staffing District Survey (SASS). Learn more about SASS.
Now available in PowerStats and QuickStats: the 1999-00, 2003-04, and 2007-08 Schools and Staffing Library Media Center Survey (SASS). Learn more about SASS.
Tables for Characteristics and Outcomes of Undergraduates with Disabilities (2018432) are now available in the DataLab Tables Library.
Tables for Science, Technology, Engineering, and Mathematics (STEM) Majors: Where Are They 4 Years After Receiving a Bachelor's Degree? (2018423) are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: the 1999-00, 2003-04, and 2007-08 Schools and Staffing School Survey (SASS). Analyze these datasets by public, private, or combined public and private schools. Learn more about SASS.
Tables for Beginning College Students Who Change Their Majors Within 3 Years of Enrollment (2018434) are now available in the DataLab Tables Library.
2015 Federal Student Aid Supplements added to BPS:1996/2001 and BPS:2004/2009 PowerStats. Additional variables related to borrowing, repayment, and student characteristics are now available.
Tables for Repayment of Student Loans as of 2015 Among 1995-96 and 2003-04 First-Time Beginning Students (2018410) are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: the 1999-00, 2003-04, and 2007-08 Schools and Staffing Principal Survey (SASS). Analyze these datasets by public, private, or combined public and private school principals. Learn more about SASS.
Now available in PowerStats and QuickStats: the 1999-00, 2003-04, and 2007-08 Schools and Staffing Teacher Survey (SASS). Analyze these datasets by public, private, or combined public and private school teachers. Learn more about SASS.
Tables for Four Years After a Bachelor’s Degree: Employment, Enrollment, and Debt Among College Graduates (2017438) are now available in the DataLab Tables Library.
NULL
Tables for The Debt Burden of Bachelor’s Degree Recipients (2017436) are now available in the DataLab Tables Library.
Tables for A Profile of the Enrollment Patterns and Demographic Characteristics of Undergraduates at For-Profit Institutions (2017416) are now available in the DataLab Tables Library.
Interested in trends related to crime and safety in US public schools? Identify trends over time in school crime with the newest addition to TrendStats, the School Survey on Crime and Safety.
Now available in PowerStats and QuickStats: the 2011-12 Private School Universe Survey (PSS). Learn more about PSS.
IPEDS Fall 2013 State/Compendium Tables are now available in the DataLab Tables Library.
IPEDS Spring 2013 State/Compendium Tables are now available in the DataLab Tables Library.
Tables for New American Undergraduates Enrollment Trends and Age at Arrival of Immigrant and Second-Generation Students (NCES 2017414) are now available in the DataLab Tables Library.
Tables for Use of Private Loans by Postsecondary Students: Selected Years 2003-04 Through 2011-12 (2017420) tables are now available in the DataLab Tables Library.
Tables for Employment and Enrollment Status of Baccalaureate Degree Recipients 1 Year After Graduation: 1994, 2001, and 2009 (2017407) are now available in the DataLab Tables Library.
Tables for First-time Postsecondary Students in 2011-12: Three-year Withdrawal, Stopout, and Transfer Rates (2016139) are now available in the DataLab Tables Library.
Tables for First-time Postsecondary Students in 2011-12: Three-year Persistence and Attainment at Any Institution (2016138) are now available in the DataLab Tables Library.
Tables for First-time Postsecondary Students in 2011-12: Three-year Retention and Attainment at First Institution (2016137) are now available in the DataLab Tables Library.
Tables for First-time Postsecondary Students in 2011-12: A Profile (2016136) are now available in the DataLab Tables Library.
Tables for Reaching the Limit: Undergraduates Who Borrow the Maximum Amount in Federal Direct Loans: 2011-12 (2016408) are now available in the DataLab Tables Library.
Tables for Bachelor's degree recipients 1 year after graduation: employment and enrollment in 1994, 2001, and 2009 (2016435) are now available in the DataLab Tables Library.
Tables for Changes in Pell Grant Participation and Median Income of Recipients (2016407) are now available in the DataLab Tables Library.
Tables for A Profile of Military Undergraduates: 2011-12 (2016415) are now available in the DataLab Tables Library.
Tables for Undergraduates Who Do Not Apply for Financial Aid (2016406) are now available in the DataLab Tables Library.
Additional 2011-12 National Postsecondary Student Aid Study (NPSAS:12) variables released. See list.
New Variables in NPSAS:2011-12 PowerStats The following NPSAS Undergraduate variables are now available in PowerStats. Name Label AIDAPP2Applied for any aid (including nonfederal only) AIDCST3Ratio of aid (excluding private loans and Direct PLUS loans to parents) to student budget
The following NPSAS Undergraduate variables are now available in PowerStats.
New variables added to High School Longitudinal Study of 2009 PowerStats. See list.
New Variables added to High School Longitudinal Study of 2009 PowerStats The following HSLS:09 variables are now available in PowerStats. Name Label P1DISABP1 Doctor/school has told parent 9th grader has a disability P1EAREYEP1 D03D Doctor/school has told parent 9th grader has hearing/vision problem P1JOINTP1 D03E Doctor/school has told parent 9th grader has bone/joint/muscle problem S3CLGSECTORU13 Postsecondary institutional sector
The following HSLS:09 variables are now available in PowerStats.
Newly released data are now available in PowerStats and QuickStats. The 2011-12 Beginning Postsecondary Students Longitudinal Study, First Follow-up (BPS:12/14) now includes over 250 new variables focusing on students’ borrowing while enrolled and their labor market outcomes 3 years after beginning postsecondary education. Learn more about the 2011-12 Beginning Postsecondary Students Longitudinal Study, First Follow-up (BPS:12/14).
Additional 2008-12 Baccalaureate and Beyond (B&B:08/12) variables released. See list.
New Variables in 2008-12 Baccalaureate and Beyond (B&B:08/12) PowerStats The following B&B:08/12 variables are now available in PowerStats. Name Label B2CMRJSTPrimary job: Employed in primary job in 2012 B2EMPTYPPrimary job: Employer type, 2012 BA_ENRMonths between bachelor's degree award date and first post-bachelor's enrollment BA_JOB1Months between bachelor's degree award date and first job JOB1GT3Held first job longer than 3 months
The following B&B:08/12 variables are now available in PowerStats.
Tables from K-12 Teaching Experience Among 2007-08 College Graduates: 2012 (2016641) are now available in the DataLab Tables Library.
Now available in PowerStats: the School Survey on Crime and Safety 2007-08 (SSOCS:2008) and School Survey on Crime and Safety 2005-06 (SSOCS:2006).
Now available in PowerStats and QuickStats: High School Longitudinal Study of 2009 (HSLS:2009).
Additional NPSAS Undergraduate variables added to TrendStats. See list.
NPSAS Undergraduate Variables Added to TrendStats The following NPSAS Undergraduate variables are now available in TrendStats. Name Description AIDCST3Total aid received excluding private loans and Direct PLUS loans to parents as a percentage of the total student budget. AIDSNEEDThe amount of financial aid exceeding federal need. AIDSRCAid package by source of aid received. ATTEND2Student's attendance status during the fall term. ATTENDMRThe student's main reason for enrolling at NPSAS. ATTNPTStudent's attendance intensity at all institutions in the given survey year. CALSYSThe NPSAS institution's academic calendar system. CLOCKDenotes if the NPSAS institution is on a clock or credit hour system. CNTLAFFIThe NPSAS institution's control or affiliation. CRNUMCRDThe number of credit cards a student has in his or her own name. CUMLNTP1Indicates the source of the student's loans during undergraduate education. DEGPRIndicates whether the student has earned degrees or certificates since high school. DEGPRAAIndicates whether the student has already earned an associate's degree since high school. DEGPRCRTIndicates whether the student has already earned an undergraduate certificate/diploma since high school. DEGPRMSIndicates whether the student has already earned a master's degree since high school. DEGPRPTBIndicates whether the student has already earned a post-baccalaureate certificate since high school. DEGPRPTMIndicates whether the student has already earned a post-master's certificate since high school. DEPEND4Student's dependency status including dependents and marital status. DEPNUMNumber of student's dependents (children and others). DEPOTHERIndicates whether the student had dependents other than children. DEPTYPEType of student's dependents. DEPYNGThe age of the student's youngest child. DSTUINCDependent student's own income in the year two years prior to the survey year, excluding the income of the parents. EFCAIDThe total amount of financial aid that is subject to federal EFC limitations if the student received need-based federal aid. EFFORT18Net price after all aid except work-study as a percentage of total income. EFFORT20Net price after grants and loans as a percentage of total income. EMPLWAIVTuition waivers for staff and families of staff at the institution. EMPLYAM3Indicates the amount of tuition aid received from the student's or the parents' employers. EMPLYAMTTotal amount of aid received from employers. ENLENNumber of months enrolled. ENR01Monthly enrollment status for July of the academic year. ENR02Monthly enrollment status for August of the academic year. ENR03Monthly enrollment status for September of the academic year. ENR04Monthly enrollment status for October of the academic year. ENR05Monthly enrollment status for November of the academic year. ENR06Monthly enrollment status for December of the academic year. ENR07Monthly enrollment status for January of the academic year. ENR08Monthly enrollment status for February of the academic year. ENR09Monthly enrollment status for March of the academic year. ENR10Monthly enrollment status for April of the academic year. ENR11Monthly enrollment status for May of the academic year. ENR12Monthly enrollment status for June of the academic year. ENRFALLStudent was enrolled during the July through December term. ENRSPRStudent was enrolled during the January through June term. ENRSTATStudent's enrollment pattern during the academic year. ESUBMX2Indicates whether undergraduates who took out subsidized Stafford loans took out their individual maximum subsidized amount, that is, the maximum they could borrow that was allowed for the program (based on their class level) and also taking into account their financial need after all aid except for any subsidized Stafford loans. ETOTMX2Indicates whether undergraduates who took out Stafford loans in took out their individual maximum total amount (sum of subsidized and unsubsidized), that is, the maximum they could borrow that was allowed for the program (based on their class level and dependency) and also taking into account their total price of attendance reduced by all financial aid except for any loans. EVER2PUBThe student has ever taken classes for credit offered through a community college. EVER4YRThe student has ever attended a 4-year college. FEDBENThis variable indicates whether any member of the student's household received any of the following federal benefits during the academic year: - Food Stamps Benefits - Free/Reduced Price School Lunch Benefits - Supplemental Security Income Benefits - TANF Benefits - WIC Benefits. FEDBENAThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: Food Stamp Benefit. FEDBENBThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: Free or Reduced Price School Lunch Benefits. FEDBENCThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: Supplemental Security Income Benefits. FEDBENDThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: TANF Benefits. FEDBENEThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: WIC Benefits. FEDGRPCTRatio of total federal grants to total aid received during the academic year. FEDLNPAKFederal loan package by type of loan received during the academic year. FEDNEEDTotal amount of federal need-based aid received during the academic year. FEDPCTRatio of total federal aid to total aid received during the academic year. FLNPCT6Ratio of federal loans to federal aid (excluding PLUS loans and Veterans' benefit) received during the academic year. GRNTSRCGrant package by source of grant received during the academic year. GRTCSTTotal grants received as a percentage of the total student budget. GRTLOANRatio of total grants to total loans received during the academic year. GRTPCTTNTotal grants received as a percentage of tuition and fees at NPSAS. GRTRATIORatio of total grants to total grants and loans received during the academic year. GRTSNEEDGrant amount exceeding federal need. HCHONORSCount of high school subject areas (English, math, foreign languages, science, and social studies) in which the student has taken an advanced placement, accelerated or honors course, according to self-report on standardized test questionnaire. HCMATHHIHighest level of math completed or planned to take, according to self-report on standardized test questionnaire and the student interview. HCSCINUMNumber of science courses the student took or planned to take, according to self-report on standardized test questionnaire. HCTKBIOLStudent took or planned to take to take Biology, according to self-report on standardized test questionnaire. HCTKCHEMStudent took or planned to take to take Chemistry, according to self-report on standardized test questionnaire. HCTKPHYSStudent took or planned to take to take Physics, according to self-report on standardized test questionnaire. HCYSENGLYears of high school coursework completed or planned in English, according to self-report on standardized test questionnaire. HCYSLANGYears of high school coursework completed or planned in foreign languages, according to self-report on standardized test questionnaire. HCYSMATHYears of high school coursework completed or planned in math, according to self-report on standardized test questionnaire and the student interview. HCYSSCIEYears of high school coursework completed or planned in science, according to self-report on standardized test questionnaire. HCYSSOCIYears of high school coursework completed or planned in social studies, according to self-report on standardized test questionnaire. HIGHLVEXThe highest level of education that the student ever expects to complete. HLOFFERHighest level of postsecondary degree or award offered at the NPSAS sample institution attended during the academic year. HOMESTUDThe student owns a home or pays a mortgage on a home. HSCRDANYIndicates whether or not student earned any college credits in high school. College credits can be college credits earned at a college or Advanced Placement credits earned in high school. HSCRDAPIndicates whether or not student earned Advanced Placement credits in high school. HSCRDCOLIndicates whether or not student earned college credits at a college during high school. HSGPAHigh school grade point average on the standardized test date, according to self-report on test questionnaire. HSGRADYYThe calendar year the student graduated from high school. HSIZEStudent's family size during the academic year. INCPCT2The tuition charged at the NPSAS institution as a percentage of total income. INJURISIndicates whether the tuition charged at public NPSAS institutions was in or out of jurisdiction. INSTCATInstitutional category was derived using the level of offerings reported on the Institutional Characteristics (IC) component and the number and level of awards that were reported on the Completions (C) component. INSTGPCTRatio of total institutional grants to total aid received during the academic year. INSTPACKAid package with institutional aid received during the academic year. For years 1996 and 2000 the categories for "No institutional" and "No aid received" are combined to be comparable to "No Institutional Aid" in subsequent years. INSTPCTRatio of total institutional aid to total aid received during the academic year. INSWAIVTotal amount of institutional tuition waivers received during the academic year. JOBEARNThe student's total amount earned from work (excluding work-study, assistantship, and traineeship) during the academic year. JOBONOFFThe location of the student's job. JOBROLEThe student's primary role while enrolled at NPSAS and also working. JOBTYPE2Indicates the student has a work-study/assistantship job, a regular job, or both. LNREPAYThe student expects help with repaying their student loans, from an individual other than their spouse. LOANCSTTotal student loans (excluding Parent PLUS) received as a percentage of the total student budget. LOANCST2Total student loans (including Parent PLUS loans) received as a percentage of the total student budget. LOANPCTRatio of total loans (excluding Parent PLUS loans) to total aid received during the academic year. LOANPCT2Ratio of total loans (including Parent PLUS loans) to total aid received during the academic year. LOCALEThe degree of urbanization in which the NPSAS institution is located. LOCALESTThe degree of urbanization in which the students home is located. MAJCHGFQThe frequency with which the student formally changed his or her major. MAJORS23The student's undergraduate major or field of study during the academic year. MAJORS2YThe student's undergraduate major or field of study during the academic year. MAJORS4YThe student's undergraduate major or field of study during the academic year. MFTNumber of months enrolled full-time between July and June of the academic year. MHTNumber of months enrolled half-time between July and June of the academic year. MLTNumber of months enrolled less than half-time between July and June of the academic year. MNTRENTThe average monthly rent or mortgage the student paid during the academic year. MPTNumber of months enrolled part time between July and June of the academic year. NEEDAID1Indicates the amount of need-based aid received. NETCST10Tuition and fees minus federal grants for the academic year. NETCST12Tuition and fees minus state grants for the academic year. NETCST13Tuition and fees minus institutional grants from the NPSAS institution for the academic year. NETCST14Tuition and fees minus non-federal grants for the academic year. NETCST15Tuition and fees minus all state and institutional grants for the academic year. NETCST16Total net price after all federal and state grants for the academic year. NETCST17Total net price after all grants and loans for the academic year. NETCST18Total net price after all financial aid except work-study for the academic year. NETCST4Total net price after all grants and one-half of all loans for the academic year. NETCST41Net total price after all financial aid except private loans for the academic year. OTHFDGRTTotal amount of grants from various small federal programs. OTHRSCRTotal aid from outside sources received. OTHTYPE2Total amount of aid received during the academic year that was not classified by type as grants and loans to students, but included work-study. PARBORNStudent's parent(s) were born in the United States. PDADEDFather's highest level of education. PELLRAT1Ratio of Pell grants to total aid received during the academic year. PELLRAT2Ratio of Pell grants to total grants received during the academic year. PELLYRSNumber of years that a Pell grant was received. PERKCUM1Indicates the cumulative amount of Perkins loans ever borrowed for undergraduate education through July 1 of the survey year. PFAMNUMFamily size of dependent student during the academic year. PFEDTAXThe amount of federal income tax paid by the parents of dependent student in the year two years prior to the survey year. PINCOLNumber of parent's dependent children in college during the academic year. PLUSPCTRatio of Parent PLUS loans to total aid received during the academic year. PMARITALParent's marital status during the academic year. PMOMEDMother's highest level of education. PRIVAMTTotal private source grants and loans for the academic year. PRIVCSTPrivate loans received as a percentage of the total student budget. PRIVLRATRatio of private loans to total loans received during the academic year. PRIVPCTRatio of private loans to total aid received during the academic year. PTAXFILEWhether or not parents of dependent student filed federal income tax in the year two years prior to the survey year. REANOAPAThe student did not apply for financial aid because he/she did not want to take on debt. REANOAPBThe student did not apply for financial aid because the application forms were too much work or too time consuming. REANOAPCThe student did not apply for financial aid he/she did not have enough information about how to apply for financial aid. REANOAPDThe student did not apply for financial aid because he/she did not need financial aid. REANOAPEThe student did not apply for financial aid because he/she thought he/she would be ineligible. SECTOR1Sector of the NPSAS sample institution attended during the academic year. SFAMNUMIndicates the number of persons in the independent student's own family (including the student). SFEDTAXThe amount of federal income tax the independent student paid for the year two years prior to the survey. SIBINCOLIndicates whether the student has siblings who attended college or graduate school during the academic year. SINCOLNumber of persons in the independent student's household who attended college during the academic year. SJHOURSThe average number of hours the student worked per week during the academic year. SJMAJORThe student's work-study job was related to his/her major or field of study. SJONOFFThe location of the student's work-study job. SJSCHOOLThe student's work study job was for NPSAS or for another institution or organization. SNEED3The remaining need after all federal grants, calculated as the total student budget minus the Expected Family Contribution and minus federal grants. SNEED4The remaining need after all grants and federal need-based aid, calculated as the total student budget minus the Expected Family Contribution and minus aid subject to the federal EFC limitation. SNEED7The remaining need after federal and state grants, calculated as the total student budget minus the Expected Family Contribution and minus federal and state grants. SNEED8The remaining need after all federal, state, and other grants, calculated as the total student budget minus the Expected Family Contribution and minus federal state, and outside grants. SPINCOLThe student's spouse attended college or graduate school during the academic year. SPSINCSpouse's earned income in the year two years prior to the survey year, if married. STAFFRATRatio of Stafford loans to total loans received during the academic year. STAFFSTThis variable indicates the first year that a Direct Subsidized or Unsubsidized Loan (also known as subsidized and unsubsidized Stafford Loan) was received. STAFLSTThis variable indicates the last year that a Direct Subsidized or Unsubsidized Loan (also known as subsidized and unsubsidized Stafford Loan) was received. STAFYRSThe number of years that the student received Direct Subsidized Loans or Direct Unsubsidized Loans (also known as subsidized and unsubsidized Stafford Loans). STAPCTRatio of total state aid to total aid received during the academic year. STAXFILEWhether or not independent student filed federal income tax for the year two years prior to the survey year. STGRPCTRatio of total state grants to total aid received during the academic year. STNOND1Indicates the amount of state grants received in the academic year that were based neither on need nor merit. STSBCUM1Cumulative subsidized Stafford loan amounts borrowed for undergraduate education through July 1 of the survey year. STSUBMXIndicates whether undergraduates who took out subsidized Stafford loans took out the maximum subsidized amount that was allowed for the program, based on their class level. STTOTMXIndicates whether undergraduates who took out Stafford loans in took out the maximum total amount (sum of subsidized and unsubsidized) that was allowed for the program, based on their class level and dependency status. SUBCUM1Cumulative federal subsidized loan amounts borrowed for undergraduate education through July 1 of the survey year. T4LNAMT2Total amount of federal Title IV loans (including Parent PLUS loans) received during the academic year. TEACTDERACT composite score, derived from either a reported ACT score or the SAT I combined score converted to an estimated ACT composite score. TESATCP1The average percentile rank of reported SAT verbal and math scores, among all test takers. TESATCREThe sum of reported SAT verbal and math scores. TESATMDESAT I math score, derived as either the actual SAT I math score or the ACT math score converted to an estimated SAT I math score. TESATMP1Percentile rank of reported SAT math score (recentered), among all test takers. TESATMREThe reported SAT math score (recentered). TESATNP1Percentile rank of derived SAT math score, among all test takers (nationwide). TESATVDESAT I verbal score derived as either the actual SAT I verbal score or the ACT English and reading score converted to an estimated SAT I verbal score. TESATVP1Percentile rank of reported SAT verbal score (recentered), among all test takers. TESATVREThe reported SAT verbal score (recentered). TETOOKIndicates whether the student took the SAT I or ACT college entrance exam. TFEDAID2Total amount of federal aid (including Veterans' benefit & DOD) received during the academic year. TFEDAID6Total amount of federal aid excluding Parent PLUS loans and Veterans' benefit received during the academic year. TNFEDAIDTotal amount of non-federal financial aid received during the academic year. TOTAID2Total amount of federal Title IV, state, and institutional aid received during the academic year. TOTAID4Total amount of all financial aid received except Parent PLUS loans during the academic year. TOTAID6Total amount of all financial aid received except Parent PLUS loans and federal Veterans' benefit during the academic year. TOTAID7Total amount of all financial aid received during the academic year except for federal Veterans' benefit. TOTAID8Total amount of all financial aid received during the academic year except for private loans. TOTGRT4Total amount of state and institutional grants received in the academic year. TOTNOND3Indicates non-need based aid received from both institutional non-need aid and state non-need aid programs during the academic year. UNSBLOANTotal amount of all unsubsidized loans from any source received during the academic year. UNTAXBFCIndicates if the student received worker's compensation during the academic year. USBORNThe student was born in the United States. VETBENIndicates the total amount of federal veterans' benefit received in the academic year. VOCHELPTotal amount of vocational rehabilitation and job training grants received during the academic year. WORKPCTRatio of total work-study to total aid received during the academic year.
The following NPSAS Undergraduate variables are now available in TrendStats.
Additional NPSAS Graduate variables added to TrendStats. See list.
NPSAS Graduate Variables Added to TrendStats The following NPSAS Graduate variables are now available in TrendStats. Name Description AIDCSTTotal aid received as a percentage of the total student budget. AIDSNEEDThe amount of financial aid exceeding federal need. ATTEND2Student's attendance status during the fall term. ATTNPTStudent's attendance intensity at all institutions attended in the academic year. CALSYSThe NPSAS institution's academic calendar system. CLOCKDenotes if the NPSAS institution is on a clock or credit hour system. CNTLAFFIThe NPSAS institution's control or affiliation. CUMLNTP1Indicates the source of the student's loans during undergraduate education. DEGPRThe student has earned degrees or certificates since high school. DEGPRAAThe student has already earned an associate's degree since high school. DEGPRCRTThe student has already earned an undergraduate certificate/diploma since high school. DEGPRMSThe student has already earned a master's degree since high school. DEGPRPTBThe student has already earned a post-baccalaureate certificate since high school. DEGPRPTMThe student has already earned a post-master's certificate since high school. DEPCAREThe student has dependent children in daycare during the academic year. DEPCOSTThe student's monthly daycare costs for dependent children during the academic year. DEPEND5AStudent's dependency status including dependents and marital status during the academic year. DEPNUMNumber of student's dependents (children and others) during the academic year. DEPOTHERIndicates whether the student had dependents other than children. DEPTYPEIndicates whether the student has dependents who are their children, not their children, or both. DEPYNGThe age of the student's youngest child during the academic year. EFCAIDThe total amount of financial aid that is subject to federal EFC limitations if the student received need-based federal aid. EFFORT18Net price after all aid except work-study as a percentage of total income. EFFORT3Net price after grants as a percentage of total income in the year two years prior to the survey year. EFFORT9Net tuition after all grants as a percentage of total income in the year two years prior to the survey year. EMPLWAIVTuition waivers for staff and families of staff at the institution attended during the academic year. EMPLYAM3Indicates the amount of tuition aid received from the student's or the parents' employers in the academic year. ENLENNumber of months enrolled between July and June of the academic year. ENR01Monthly enrollment status for July of the academic year. ENR02Monthly enrollment status for August of the academic year. ENR03Monthly enrollment status for September of the academic year. ENR04Monthly enrollment status for October of the academic year. ENR05Monthly enrollment status for November of the academic year. ENR06Monthly enrollment status for December of the academic year. ENR07Monthly enrollment status for January of the academic year. ENR08Monthly enrollment status for February of the academic year. ENR09Monthly enrollment status for March of the academic year. ENR10Monthly enrollment status for April of the academic year. ENR11Monthly enrollment status for May of the academic year. ENR12Monthly enrollment status for June of the academic year. ENRFALLStudent was enrolled during the July through December term. ENRLSIZEFall enrollment. ENRSPRStudent was enrolled during the January through June term. ENRSTATStudent's enrollment pattern during the academic year. FEDBENThis variable indicates whether any member of the student's household received any of the following federal benefits during the academic year: food Stamps Benefits, free/Reduced Price School Lunch Benefits, supplemental Security Income Benefits, TANF Benefits, or WIC Benefits FEDBENAThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: Food Stamp Benefit. FEDBENBThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: Free or Reduced Price School Lunch Benefits. FEDBENCThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: Supplemental Security Income Benefits. FEDBENDThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: TANF Benefits. FEDBENEThis variable indicates whether any member of the student's household received the following federal benefit during the academic year: WIC Benefits. FEDCUM1Cumulative federal loan amounts borrowed for undergraduate education through July 1 of the survey year. FEDOWE1Indicates total amount owed on all federal loans for undergraduate education as of late in the survey year. FGRTLNTotal amount of federal student loans and federal grants received during the academic year. FLNPCT6Ratio of federal loans to federal aid (excluding veterans benefits) received during the academic year. GRNTSRCGrant package by source of grant received during the academic year. GRTCSTTotal grants received as a percentage of the total student budget. GRTLOANRatio of total grants to total loans received during the academic year. GRTRATIORatio of total grants to total grants and loans received during the academic year. GRTSNEEDGrant amount exceeding federal need. HIGHLVEXThe highest level of education that the student ever expects to complete. HISPANICStudent is of Hispanic or Latino origin. HISPTYPEType of Hispanic or Latino origin. HLOFFERHighest level of postsecondary degree or award offered at the NPSAS sample institution attended during the academic year. HOMESTUDThe student owns a home or pays a mortgage on a home. IMMIGENNumber of generations the student's family has been in the U.S. IMMIGRAStudent's immigrant status. INCPCT1The adjusted cost of attendance at the NPSAS institution as a percentage of total income in the year two years prior to the survey year. INCPCT2The tuition charged at the NPSAS institution as a percentage of total income. INJURISIndicates whether the tuition charged at public NPSAS institutions was in or out of jurisdiction. JOBEARNThe student's total amount earned from work (excluding work-study, assistantship, and traineeship) during the academic year. JOBENRThe student's intensity of work (excluding work-study/assistantship/traineeship) while enrolled during the academic year. JOBHOURThe average number of hours worked per week during the academic year (excluding work-study, fellowships, assistantships, and traineeships). JOBONOFFThe location of the student's job. JOBROLEThe student's primary role while enrolled at NPSAS and also working. JOBTYPE2Indicates the student has a work-study/assistantship job, a regular job, or both. LNREPAYThe student expects help with repaying their student loans, from an individual other than their spouse. LOANCSTTotal student loans (excluding Parent PLUS) received as a percentage of the total student budget. LOCALEThe degree of Urbanization in which the NPSAS institution is located. LOCALESTInput to this derived variable was the best-known current address after data collection. MFTNumber of months enrolled full time between July and June of the academic year. MHTNumber of months enrolled half-time between July and June of the academic year. MLTNumber of months enrolled less than half-time between July and June of the academic year. MNTRENTThe average monthly rent or mortgage the student paid during the academic year. MPTNumber of months enrolled part-time between July and June of the academic year. NETCST10Tuition and fees minus federal grants for the academic year. NETCST12Tuition and fees minus state grants for the academic year. NETCST14Tuition and fees minus non-federal grants for the academic year. NETCST15Tuition and fees minus all state and institutional grants for the academic year. NETCST16Net total price after all federal and state grants for the academic year. NETCST17Net total price after all grants and loans for the academic year. NETCST18Net total price after all financial aid except work-study for the academic year. NETCST2Net total price after all federal grants for the academic year. NETCST4Net total price after all grants and one-half of all loans for the academic year. NETCST41Net total price after all financial aid except private loans for the academic year four years prior to the survey year. NETCST9Tuition and fees minus all grants for the academic year. NFEDCUM1Cumulative non-federal loan amounts borrowed for undergraduate education through July 1 of the survey year. OBEREGRegion where NPSAS sample institution is located. OTHTYPE2Total amount of aid received during the academic year that was not classified by type as grants and loans to students, but included work-study. PCTPOVIndicates total income as a percentage of the federal poverty level thresholds for the year two years prior to the survey year. PDADEDFather's highest level of education. PELLFSTThe first year that a Pell grant was received during the years covered by the survey. PELLLSTThe last year that a Pell grant was received during the years covered by the survey. PELLYRSNumber of years that a Pell grant was received. PERKCUM1Indicates the cumulative amount of Perkins loans ever borrowed for undergraduate education through July 1 of the survey year. PMOMEDMother's highest level of education. PRIVCSTPrivate loans received as a percentage of the total student budget. PRIVLRATRatio of private loans to total loans received during the academic year. PRIVPACKLoan package by whether the loan received was private (alternative) or not during the academic year. PRIVPCTRatio of private loans to total aid received during the academic year. REANOAPAThe student did not apply for financial aid because he/she did not want to take on debt. REANOAPBThe student did not apply for financial aid because the application forms were too much work or too time consuming. REANOAPCThe student did not apply for financial aid he/she did not have enough information about how to apply for financial aid. REANOAPDThe student did not apply for financial aid because he/she did not need financial aid. REANOAPEThe student did not apply for financial aid because he/she thought he/she would be ineligible. SECTOR1Sector of the NPSAS sample institution attended during the academic year. SECTOR4Sector of the NPSAS sample institution attended during the academic year, for students who attended only one institution. SFEDTAXThe amount of federal income tax paid by the independent student in the year two years prior to the survey year. SINCOLNumber of persons in the independent student's household who attended college during the academic year. SJEARNThe total amount the student earned from his or her assistantship/fellowship/traineeship/work-study job during the academic year. SJHOURSThe average number of hours the student worked per week during the academic year. SNEED1The student's total need for need-based financial aid. SNEED2The remaining need after all financial aid (need-based and non-need-based) received. SNEED3The remaining need after all federal grants. Equal to the total student budget minus Expected Family Contribution, and minus federal grants. SNEED4The remaining need after all grants and federal need-based aid. Equal to the total student budget minus Expected Family Contribution, and minus aid subject to the federal EFC limitation. SNEED7The remaining need after federal and state grant aid. Equal to the total student budget minus expected family contribution, and minus federal and state grants. SNEED8The remaining need after all federal, state, and other grants. Equal to the total student budget minus the Expected Family Contribution, minus federal, state, and outside grants. SNEED9The remaining need after all financial aid received except private loans. SPINCOLThe student's spouse attended college or graduate school during the academic year. STAFFRATRatio of Stafford loans to total loans received during the academic year. STAFFSTThis variable indicates the first year that a Direct Subsidized or Unsubsidized Loan (also known as subsidized and unsubsidized Stafford Loan) was received. STAFLSTThis variable indicates the last year that a Direct Subsidized or Unsubsidized Loan (also known as subsidized and unsubsidized Stafford Loan) was received. STAFTYPEThis variable indicates the combination of subsidized and unsubsidized Stafford loans received at all institutions attended during the academic year. STAFYRSThe number of years that the student received Direct Subsidized Loans or Direct Unsubsidized Loans (also known as subsidized and unsubsidized Stafford Loans). STAPCTRatio of total state aid to total aid received during the academic year. STAXFILEWhether or not independent student filed federal income tax for the year two years prior to the survey year. STEMMAJThis variable indicates the student's major field of study during the academic year with a focus on science, technology, engineering, and mathematics (STEM) fields. STGRPCTRatio of total state grants to total aid received during the academic year. STSBCUM1Cumulative subsidized Stafford loan amounts borrowed for undergraduate education through July 1 of the survey year. STUDMULTNumber of institutions attended during the academic year. STYPELSTStudent type at the NPSAS sample institution during the academic year. SUBCUM1Cumulative federal subsidized loan amounts borrowed for undergraduate education through July 1 of the survey year. SUBLOANTotal amount of federal Title IV subsidized loans received during the academic year. TFEDGRT2Total amount of all federal grants, veterans benefits and Department of Defense aid received during the academic year. TGRTLNTotal amount of loans and grants received during the academic year. TITIVAMTTotal amount of federal Title IV financial aid received during the academic year. TNFEDAIDTotal amount of non-federal financial aid received during the academic year. TNFEDGRTTotal amount of non-federal grants received during the academic year. TNFEDLNTotal amount of non-federal loans received during the academic year. TOTAID5Total amount of all financial aid received except for work-study during the academic year. TOTAID7Total amount of all financial aid received during the academic year except for federal veterans benefits. TOTGRT2Total amount of all grants, veteran's benefits and Department of Defense aid received during the academic year. TOTGRT4Total amount of state and institutional grants received in the survey year. TOTLOAN3Total amount of all loans excluding private loans received during the academic year. UNSBLOANTotal amount of all unsubsidized loans from any source received during the academic year. USBORNThe student was born in the United States. VETBENIndicates the total amount of federal veterans' benefit received in the academic year. WORKPCTRatio of total work-study to total aid received during the academic year.
The following NPSAS Graduate variables are now available in TrendStats.
Tables from Persistence and Attainment of 2011–12 First-Time Postsecondary Students After 3 Years (BPS:12/14 First Look) are now available in the DataLab Tables Library.
Newly released study is now available in PowerStats and QuickStats. Learn more about the 2011-12 Beginning Postsecondary Students Longitudinal Study, First Follow-up (BPS:12/14).
Important update to financial aid variable documentation. Learn More.
Financial Aid Variables Unless otherwise specified, variables based on NSLDS data were derived with a filter that removed loans borrowed prior to July 1995. This loan-level filter was applied to balance the goal of including as many loans as possible while minimizing the use of partial loan information. However, since consolidated loans contain multiple loans, each with potentially different dates of origination, pre-1995 loans may be included in the total consolidated amount. See codebook entries for NSLDS derived variables for more information. Respondents who had taken out a loan before July 1995 can be identified with variable LOANBF071995.
Unless otherwise specified, variables based on NSLDS data were derived with a filter that removed loans borrowed prior to July 1995. This loan-level filter was applied to balance the goal of including as many loans as possible while minimizing the use of partial loan information. However, since consolidated loans contain multiple loans, each with potentially different dates of origination, pre-1995 loans may be included in the total consolidated amount. See codebook entries for NSLDS derived variables for more information. Respondents who had taken out a loan before July 1995 can be identified with variable LOANBF071995.
2008-12 Baccalaureate and Beyond (B&B:08/12) New variables now available in the 2008-12 Baccalaureate and Beyond PowerStats. The following variables have been added: Variable Name Variable Label B2DFROCRECONNumber of separate deferments granted for economic difficulty as of 2012 B2DFROCRENRNumber of separate deferments granted for student enrollment as of 2012 B2DFROCRFAMNumber of separate deferments granted for family or disability as of 2012 B2DFROCRGOVNumber of separate deferments granted for government program (Action, Peace Corps, Head Start, NOAA) as of 2012 B2DFROCRMILNumber of separate deferments granted for military or law enforcement as of 2012 B2DFROCRNUMTotal number of separate deferment incidents as of 2012 B2DFROCRREASReason for most frequently-granted deferment as of 2012 B2DFROCRTEANumber of separate deferments granted for teacher, medical, or non-profit as of 2012 LOANBF071995Borrowed federal loans before July 1995
New variables now available in the 2008-12 Baccalaureate and Beyond PowerStats. The following variables have been added:
2008-12 Baccalaureate and Beyond (B&B:08/12) variables revised. Learn More.
Variables With Revised Data in B&B:2012 Newly revised variables now available in the 2008-12 Baccalaureate and Beyond PowerStats and QuickStats. The following variables have been revised: Name Label How variable was revised # of cases changed B2FDDUE1Cumulative federal amount owed (principal and interest) for undergraduate as of 2012Cumulative amount was revised due to re-allocation of consolidated amounts to graduate level (B2FDCUM2)1,320 B2FDDUE2Cumulative federal amount owed (principal and interest) for graduate as of 2012Cumulative amount was revised due to re-allocation of consolidated amounts to graduate level (B2FDCUM2)1,310 B2DFR_AVGAverage number of deferments per loan as of 2012Consolidated and cancelled loan amounts were removed from the denominator50 B2DLQ_AVGAverage number of delinquencies per federal loan as of 2012Consolidated and cancelled loan amounts were removed from the denominator30 B2FBPERLNAverage number of forbearances per loan as of 2012Consolidated and cancelled loan amounts were removed from the denominator40 B2CNSCUMAmount of federal loans consolidated as of 2012Parent Plus loans and duplicate loans were removed from the total consolidated amount600 The values for the following variables have changed as a result of the changes to the variables above: Name Label # of cases changed B2BORATCumulative amount borrowed for education as of 201270 B2DATFBLatest forbearance date for borrower as of 201210 B2DEBTRTRatio of federal loans to annualized salary as of 201220 B2DEFEREver had a deferment on a loan as of 201210 B2DFR_ECONNumber of deferments for economic difficulty for all loans as of 201210 B2DFR_ENRNumber of deferments for student enrollment for all loans as of 201220 B2DFR_FAMNumber of deferments for family or disability for all loans as of 2012<10 B2DFR_GOVNumber of deferments for government program (Action, Peace Corps, Head Start, NOAA) for all loans as of 2012<10 B2DFR_MILNumber of deferments for military or law enforcement for all loans as of 2012<10 B2DFR_NUMTotal number of deferments for all loans as of 201220 B2DFR_REASMost common deferment reason for borrower for all loans as of 201210 B2DFR_TEANumber of deferments for teacher, medical, or non-profit for all loans as of 2012<10 B2DLQ_NOWCurrently in delinquent status - has a federal loan in delinquency in the 2011-12 academic year<10 B2EVERDAFBEver had loans in deferment or forbearance as of 201210 B2EVERPIFEver had a loan paid in full as of 2012<10 B2FDDUE3Cumulative federal amount owed (principal and interest) for all education as of 201210 B2FDOWE1Cumulative federal amount owed (principal) for undergraduate as of 2012 1,320 B2FEDCUM1Cumulative amount borrowed in federal loans as of 2012 - undergraduate level70 B2FEDCUM3Cumulative total amount borrowed in federal loans as of 201270 B2FEDFYEARFirst year borrowed federal loans as of 2012<10 B2FEDLYEARLast year borrowed federal loans as of 2012<10 B2FORBAREver had any loans in forbearance as of 201210 B2GP_USEBorrower consolidated loans as of 2012<10 B2LASTLEVGrade level when last federal loan was received as of 2012<10 B2LASTSTDTDate of status of latest federal loan as of 201220 B2LNSTATStatus of latest federal loan as of 201210 B2LOANPAIDAll federal loans were paid in full as of 2012<10 B2OWELRPTotal amount owed at time last entered repayment40 B2OWEPNLRPOutstanding principle amount at date last entered repayment40 B2OWEPRINLatest federal amount owed - principal as of 201250 B2PAYSTATRepayment status for any loans in 2012 (federal and private)<10 B2REPLNRepayment plan of latest federal loan in 201210 B2T4XDUECumulative Stafford and Perkins loan amount owed (principal and interest) as of 2012<10 B2TOTDUE3Cumulative amount owed for education loans as of 2012 (federal and private, principal and interest)10
Newly revised variables now available in the 2008-12 Baccalaureate and Beyond PowerStats and QuickStats. The following variables have been revised:
The values for the following variables have changed as a result of the changes to the variables above:
Now available in the 2008-12 Baccalaureate and Beyond PowerStats New variables now available in the 2008-12 Baccalaureate and Beyond PowerStats. The following variables have been added: Name Label B2DFROCRECONNumber of separate deferments granted for economic difficulty as of 2012 B2DFROCRENRNumber of separate deferments granted for student enrollment as of 2012 B2DFROCRFAMNumber of separate deferments granted for family or disability as of 2012 B2DFROCRGOVNumber of separate deferments granted for government program (Action, Peace Corps, Head Start, NOAA) as of 2012 B2DFROCRMILNumber of separate deferments granted for military or law enforcement as of 2012 B2DFROCRNUMTotal number of separate deferment incidents as of 2012 B2DFROCRREASReason for most frequently-granted deferment as of 2012 B2DFROCRTEANumber of separate deferments granted for teacher, medical, or non-profit as of 2012 LOANBF071995Borrowed federal loans before July 1995
Important update to financial aid variable documentation Unless otherwise specified, variables based on NSLDS data were derived with a filter that removed loans borrowed prior to July 1995. This loan-level filter was applied to balance the goal of including as many loans as possible while minimizing the use of partial loan information. However, since consolidated loans contain multiple loans, each with potentially different dates of origination, pre-1995 loans may be included in the total consolidated amount. See codebook entries for NSLDS derived variables for more information. Respondents who had taken out a loan before July 1995 can be identified with variable LOANBF071995.
Tables from Trends in Undergraduate Nonfederal Grant and Scholarship Aid by Demographic and Enrollment Characteristics, Selected Years: 1999-2000 to 2011-12 (NPSAS:96, NPSAS:2000, NPSAS:04, NPSAS:08 and NPSAS:12) are now available in the DataLab Tables Library.
Tables from Demographic and Enrollment Characteristics of Nontraditional Undergraduates: 2011-12 (NPSAS:12) are now available in the DataLab Tables Library.
Tables from Trends in Pell Grant Receipt and the Characteristics of Pell Grant Recipients: Selected Years, 1999-2000 to 2011-12 (NPSAS:2000, NPSAS:04, NPSAS:08 and NPSAS:12) are now available in the DataLab Tables Library.
Now available in PowerStats: the School Survey on Crime and Safety 2009-10 (SSOCS:2010).
Now available in PowerStats: Education Longitudinal Study of 2002 (ELS:2002).
Now available in DataLab: TrendStats.TrendStats allows users to compare estimates for like variables across multiple dataset years within a single table. Users can create percentage distribution tables or average, median, and percent greater than tables for both Undergraduate and Graduate respondents of the 1996, 2000, 2004, 2008, and 2012 National Postsecondary Student Aid Surveys (NPSAS).Visit the Learning Center to learn more.
2008-2012 Baccalaureate and Beyond (B&B:08/12) PowerStats updated. This update adds many new variables to the 2008-2012 Baccalaureate and Beyond PowerStats, including approximately 70 variables on teaching.
Tables from High School Dropouts and Stopouts: Demographic Backgrounds, Academic Experiences, Engagement, and School Characteristics (HSLS:09) are now available in the DataLab Tables Library.
Tables for Trends in Graduate Student Financing: Selected Years, 1995–96 to 2011–12 (NPSAS:96, NPSAS:2000, NPSAS:04, NPSAS:08 and NPSAS:12) are now available in the DataLab Tables Library.
Tables from What Is the Price of College? Total, Net, and Out-of-Pocket Prices by Type of Institution in 2011-12 (NPSAS:12) are now available in the DataLab Tables Library.
Tables from Baccalaureate Degree Recipients' Early Labor Market and Education Outcomes: 1994, 2001, and 2009 (B&B:93/94, B&B:2000/01, and B&B:08/09) are now available in the DataLab Tables Library.
Tables from Student Financing of Undergraduate Education Web Table: 2011-12 (NPSAS:12) are now available in the DataLab Tables Library.
New variables added to 2011-12 National Postsecondary Student Aid Study, Undergraduate Students and Graduate Students (NPSAS:12) PowerStats: DECMAJ, BUDGETBK, and MAJORCTE added to Undergraduate Students and BUDGETBK added to Graduate Students.
Tables for Profile and Financial Aid Estimates of Graduate Students: 2011-12 (NPSAS:12) are now available in the DataLab Tables Library.
Tables for Profile of Undergraduate Students: 2011-12 (Web Tables) (NPSAS:12) are now available in the DataLab Tables Library.
Tables for Transferability of Postsecondary Credit Following Student Transfer or Coenrollment (BPS:04/09) are now available in the DataLab Tables Library.
New variables added to 2007-08 National Postsecondary Student Aid Study, Undergraduate Students (NPSAS:2007-08). INSTCAT, DISTLOC2, TCHPLN, and TCHCRS are now available on PowerStats.
Now available in PowerStats: 2011-12 Schools and Staffing Survey (SASS). Data from the following questionnaires are included: Public and Private Schools; Public and Private School Teachers; Public and Private School Principals; Public School Districts; and Public School Library Media Centers.
Tables for Baccalaureate and Beyond: A First Look at the Employment Experiences and Lives of College Graduates, 4 Years On (B&B:08/12) are now available in the DataLab Tables Library.
Now available in PowerStats and QuickStats: 2008-2012 Baccalaureate and Beyond (B&B:08/12).
Among college graduates, 13% go into teaching within four years of graduation (Source: B&B:08/12)
Among first-generation college students, 40% were no longer enrolled 3 years after beginning their postsecondary education (Source: BPS:12/14)
As of 2012, the highest levels of education completed by 2002 High School Sophomores were: 33% bachelor’s degree or higher; 9% associate’s degree; 10% undergraduate certificate; 32% postsecondary attendance but no postsecondary credential; 13% high school diploma or equivalent; and 3% less than high school completion (Source: ELS:2002)
On average, fall 2009 ninth-graders had earned 3.6 credits in math and 3.3 credits in science by 2013. On average, students had earned 7.6 credits in STEM courses (Source: HSLS:09)
From 1996 to 2012, the percentage of undergraduates that applied for federal financial aid rose from 44.5% to 70.1% (Source: NPSAS:1995-96 & NPSAS:2011-12)
Part-time instructional faculty and staff averaged $44,800 in outside income other than consulting income in 2003 compared with $12,400 earned by full-time faculty (Source: NSOPF:2004)
Preschoolers with disabilities were disproportionately male, 70 percent versus 30 percent female (Source: PEELS)
The majority of private elementary schools in 2011-12 (13,459 of 30,861 total schools) enrolled less than 50 students (Source: PSS:2011-12)
More private school students in 2013–14 were enrolled in kindergarten (463,067) than in any other grade level (Source: PSS:2011-12)
Among Public School Teachers surveyed in 2011-12, 39.9% reported having a Bachelor’s Degree, 47.7% a Master’s Degree, and 8.7% reported having higher than a Master’s degree (Source: SASS:2011-12)
Early Childhood Program ParticipationNational Postsecondary Student Aid Study, UndergraduateNational Postsecondary Student Aid Study, GraduateParent and Family Involvement in EducationSchool Survey on Crime and Safety Pre-Elementary Education Longitudinal StudyPEELSPre-elementary students who received preschool special education services, as they progressed through the early elementary years Preschool special education, Programs and services received, Transitions between preschool and elementary school, Function and performance in preschool, kindergarten, and elementary schoolhttps://ies.ed.gov/ncser/projects/peels482003/2008qsOnpsOntsOff3,000ImputationImputation was conducted for selected items on the teacher questionnaire and parent interview data. In general, the item missing rate was low. The risk of imputation-related bias was judged to be minimal. The variance inflation due to imputation was also low due to the low imputation rate of 10 percent. Imputation for the supplemental sample increased the amount of data usable for analysis, offsetting the potential risk of bias.The methods of imputation included: hot-deck imputation, regression, external data source, and a derivation method, based on the internal consistency of inter-related variables.View methodology reportpeels_subject.pdf6.71 MBpeels_varname.pdf6.63 MB00148Private School Universe SurveyPSSPrivate schoolsSchool Affiliation/Associations, Enrollment, Grades Taught, Staffing, General Informationhttps://nces.ed.gov/surveys/pss/752011-2012qsOnpsOntsOff26,983WeightingThe final weights are needed to have the estimates reflect the population of private schools when analyzing the data. The data from the area frame component were weighted to reflect the sampling rates (probability of selection) of the PSUs. Survey data from both the list and area frame components were adjusted for school nonresponse. The final weight for PSS data items is the product of the Base Weight and the Nonresponse Adjustment Factor, where:Base Weight is the inverse of the probability of selection of the school. The base weight is equal to one for all list-frame schools. For area-frame schools, the base weight is equal to the inverse of the probability of selecting the PSU in which the school resides.Nonresponse Adjustment Factor is an adjustment that accounts for school nonresponse. It is the weighted (base weight) ratio of the total eligible in-scope schools (interviewed schools plus noninterviewed schools) to the total responding in-scope schools (interviewed schools) within cells. Noninterviewed and out-of-scope cases are assigned a nonresponse adjustment factor of zero.Because we have more information for list-frame schools, the cells used to compute the nonresponse adjustment were defined differently for list-frame and area-frame schools. For schools in the list frame, the cells were defined by affiliation (17 categories), locale type (4 categories), grade level (4 categories), Census region (4 categories), and enrollment (3 categories). The nonresponse adjustment cells for area frame schools were defined by three-level typology (3 categories) and grade level (4 categories). If the number of schools in a cell was fewer than 15 or the nonresponse adjustment factor was greater than 1.5, then that cell was collapsed into a similar cell. The variables used to collapse the cells and the collapse order varied according to whether the school was from the list or area frame and whether a school was a traditional or k-terminal school. The cells for traditional schools from the list frame were collapsed within enrollment category, locale type, grade level, and Census region. Cells for k-terminal schools from the list frame were collapsed within enrollment category, locale type, Census region, and affiliation. Cells for traditional schools from the area frame were collapsed within grade level and then within three-level typology. Cells for k-terminal schools from the area frame were collapsed within three level typology.ImputationAfter the data edit processing was complete, there were missing values within some records classified as interviews. These were cases where the respondent had not answered some applicable questionnaire items (and data for these items were not added in the pre-edit, consistency, or logic edit) or the response had been deleted during editing. Values were imputed to the missing data during imputation. Two types of imputation were employed: donor and analyst imputation.Donor ImputationIn donor imputation, values were created by extracting data from the record for a sample case (donor) with similar characteristics, using a procedure known as the “sequential nearest neighbor hot deck” (Kalton and Kasprzyk 1982, 1986; Kalton 1983; Little and Rubin 1987; Madow, Olkin, and Rubin 1983). In order to match incomplete records to those with complete data, “imputation” variables that identify certain characteristics of the school that were deemed to be important to the reporting of the data in each item (e.g., religious affiliation, enrollment, school level of instruction) were used. Items were grouped according to the perceived relevance of the imputation variables to the data collected by the item. For example, school level of instruction was used for matching incomplete records and donors to fill item 16 (length of school year) but was not used for item 7 (students by race).Analyst ImputationAfter the donor imputation was completed, there were records that still had missing values for 64 items. These were cases where the imputation failed to create a value because there was no suitable record to use as a donor, or the value imputed was deleted because it was outside the acceptable range for the item or was inconsistent with other data on the same record, or the religious orientation or purpose, or the religious orientation or affiliation, was not reported (items 14a and 14c) and no previous PSS information was available.For these cases, values were imputed by analysts to the items with missing data. That is, staff reviewed the data record, sample file record, and the questionnaire and identified a value consistent with the information from these sources for imputation.pss2012_subject.pdf377 KBpss2012_varname.pdf368 KB00260Schools and Staffing Survey, TeachersSASSPublic and private school teachersClass Organization, Education and Training, Certification, Professional Development, Working Conditions, School Climate and Teacher Attitudes, Employment and Background Informationhttps://nces.ed.gov/surveys/sass622011-2012qsOnpsOntsOff42,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12teachpub_subject.pdf5.60 MBsass12teachpub_varname.pdf5.50 MB1332632011-2012qsOnpsOntsOff42,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12teachpriv_subject.pdf4.90 MBsass12teachpriv_varname.pdf4.90 MB2332642011-2012qsOnpsOntsOff42,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12teachcombined_subject.pdf5.60 MBsass12teachcombined_varname.pdf5.55 MB3332902007-2008qsOnpsOntsOff38,200PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass08teachpub_subject.pdf4.39 MBsass08teachpub_varname.pdf7.01 MB1337912007-2008qsOnpsOntsOff6,000PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass08teachpriv_subject.pdf3.27 MBsass08teachpriv_varname.pdf5.11 MB2337922007-2008qsOnpsOntsOff44,200PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass08teachcombined_subject.pdf3.25 MBsass08teachcombined_varname.pdf1.09 MB3337872003-2004qsOnpsOntsOff43,200PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass04teachpub_subject.pdf4.39 MBsass04teachpub_varname.pdf4.40 MB13312882003-2004qsOnpsOntsOff8,000PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass04teachpriv_subject.pdf6.73 MBsass04teachpriv_varname.pdf3.98 MB23312892003-2004qsOnpsOntsOff51,200PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass04teachcombined_subject.pdf1.15 MBsass04teachcombined_varname.pdf1.17 MB33312931999-2000qsOffpsOntsOff52,000PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass00teachpub_subject.pdf7.91 MBsass00teachpub_varname.pdf1.39 MB13316941999-2000qsOffpsOntsOff52,000PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass00teachpriv_subject.pdf6.34 MBsass00teachpriv_varname.pdf1.38 MB23316951999-2000qsOffpsOntsOff52,000PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass00teachcombined_subject.pdf6.60 MBsass00teachcombined_varname.pdf1.91 MB33316Schools and Staffing Survey, PrincipalsSASSPublic and private school principalsExperience, Training, Education, and Professional Development, Goals and Decision Making, Teacher and Aide Professional Development, School Climate and Safety, Instructional Time, Working Conditions and Principal Perceptions, Teacher and School Performancehttps://nces.ed.gov/surveys/sass652011-2012qsOnpsOntsOff9,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12prinpub_subject.pdf1.99 MBsass12prinpub_varname.pdf1.97 MB1343662011-2012qsOnpsOntsOff9,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12prinpriv_subject.pdf1.98 MBsass12prinpriv_varname.pdf1.90 MB2343672011-2012qsOnpsOntsOff9,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12princombined_subject.pdf1.92 MBsass12princombined_varname.pdf2.05 MB33431022007-2008qsOnpsOntsOff7,500PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass08prinpub_subject.pdf2.00 MBsass08prinpub_varname.pdf1.83 MB13481032007-2008qsOnpsOntsOff1,900PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass08prinpriv_subject.pdf1.80 MBsass08prinpriv_varname.pdf1.61 MB23481042007-2008qsOnpsOntsOff9,400PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass08princombined_subject.pdf1.86 MBsass08princombined_varname.pdf1.61 MB3348992003-2004qsOnpsOntsOff8,100PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass04prinpub_subject.pdf553 KBsass04prinpub_varname.pdf2.20 MB134131002003-2004qsOnpsOntsOff2,400PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass04prinpriv_subject.pdf466 KBsass04prinpriv_varname.pdf445 KB234131012003-2004qsOnpsOntsOff10,500PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass04princombined_subject.pdf449 KBsass04princombined_varname.pdf433 KB33413961999-2000qsOffpsOntsOff12,000PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass00prinpub_subject.pdf1.90 MBsass00prinpub_varname.pdf1.61 MB13417971999-2000qsOffpsOntsOff12,000PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass00prinpriv_subject.pdf1.58 MBsass00prinpriv_varname.pdf1.25 MB23417981999-2000qsOffpsOntsOff12,000PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass00princombined_subject.pdf1.48 MBsass00princombined_varname.pdf1.26 MB33417Schools and Staffing Survey, SchoolsSASSPublic and private schoolsTeacher demand, teacher and principal characteristics, general conditions in schools, principals' and teachers' perceptions of school climate and problems in their schools, teacher compensation, district hiring and retention practices, basic characteristics of the student populationhttps://nces.ed.gov/surveys/sass592011-2012qsOnpsOntsOff9,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12schoolpub_subject.pdf520 KBsass12schoolpub_varname.pdf530 KB1351602011-2012qsOnpsOntsOff9,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12schoolpriv_subject.pdf720 KBsass12schoolpriv_varname.pdf675 KB2351612011-2012qsOnpsOntsOff9,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12schoolcombined_subject.pdf1.60 MBsass12schoolcombined_varname.pdf1.55 MB33511172007-2008qsOnpsOntsOff7,600Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass08schoolspub_subject.pdf2.23 MBsass08schoolspub_varname.pdf2.52 MB135201182007-2008qsOnpsOntsOff2,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass08schoolspriv_subject.pdf2.51 MBsass08schoolspriv_varname.pdf3.07 MB235201192007-2008qsOnpsOntsOff9,500Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass08schoolscombined_subject.pdf1.92 MBsass08schoolscombined_varname.pdf2.27 MB335201142003-2004qsOnpsOntsOff8,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass04schoolspublic_subject.pdf2.27 MBsass04schoolspublic_varname.pdf2.30 MB135191152003-2004qsOnpsOntsOff2,500Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass04schoolsprivate_subject.pdf3.26 MBsass04schoolsprivate_varname.pdf1.55 MB235191162003-2004qsOnpsOntsOff10,400Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass04schoolscombined_subject.pdf1.90 MBsass04schoolscombined_varname.pdf2.00 MB335191111999-2000qsOffpsOntsOff9,300Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass00schoolspublic_subject.pdf2.00 MBsass00schoolspublic_varname.pdf2.07 MB135181121999-2000qsOffpsOntsOff2,600Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass00schoolsprivate_subject.pdf2.89 MBsass00schoolsprivate_varname.pdf3.18 MB235181131999-2000qsOffpsOntsOff11,900Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass00schoolscombined_subject.pdf2.20 MBsass00schoolscombined_varname.pdf2.50 MB33518Schools and Staffing Survey, DistrictsSASSPublic school districtsRecruitment and Hiring of Staff, Principal and Teacher Compensation, Student Assignment, Graduation Requirements, Migrant Education, District Performancehttps://nces.ed.gov/surveys/sass582011-2012qsOnpsOntsOff4,500Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12district_subject.pdf1.15 MBsass12district_varname.pdf1.10 MB31641102007-2008qsOnpsOntsOff4,600PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass08district_subject.pdf0.51 MBsass08district_varname.pdf0.53 MB316261092003-2004qsOnpsOntsOff4,400PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass04district_subject.pdf0.88 MBsass04district_varname.pdf0.93 MB316251081999-2000qsOffpsOntsOff4,700PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass00district_subject.pdf1.10 MBsass00district_varname.pdf0.68 MB31624Schools and Staffing Survey, Library Media CentersSASSLibrary media centersSchool information, Facilities, services, and policies, Staffing information, Technology and information literacy, Collections and expenditureshttps://nces.ed.gov/surveys/sass572011-2012qsOnpsOntsOff7,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics. View methodology informationsass12LMC_subject.pdf675 KBsass12LMC_varname.pdf695 KB31751072007-2008qsOnpsOntsOff7,300PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass08LMC_subject.pdf0.59 MBsass08LMC_varname.pdf0.61 MB317231062003-2004qsOnpsOntsOff7,200PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass04LMC_subject.pdf0.80 MBsass04LMC_varname.pdf0.81 MB317221051999-2000qsOffpsOntsOff7,700PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.View methodology informationsass00LMC_subject.pdf1.16 MBsass00LMC_varname.pdf1.18 MB31721School Survey on Crime and SafetySSOCSElementary and secondary schoolsSchool Practices and Programs, Parent and Community Involvement at School, School Security, Staff Training, Limitations on Crime Prevention, Frequency of Crime and Violence, Frequency of hate and gang-related crimes, Disciplinary problems and actionshttps://nces.ed.gov/surveys/ssocs12832015-2016qsOnpsOntsOn3,500ImputationCompleted SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond: In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical. WeightingData are weighted to compensate for differential probabilities of selection and to adjust for the effects of nonresponse.Sample weights allow inferences to be made about the population from which the sample units are drawn. Because of the complex nature of the SSOCS sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Due to nonresponse, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias due to nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The same predictor variables from the SSOCS:2004 CHAID analysis were used for SSOCS:2006: instructional level, region, enrollment size, percent minority, student-to-FTE teaching staff ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time equivalent (FTE) teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted so that the weighted distribution of the responding schools resembled the initial distribution of the total sample. The nonresponse-adjusted weights were then poststratified to calibrate the sample to known population totals. Two dimension margins were set up for the poststratification—(1) instructional level and school enrollment size; and (2) instructional level and locale—and an iterative process known as the raking ratio adjustment brought the weights into agreement with known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. All three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2004). Miller, A.K. (2004). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.ssocs2016_subject.pdf375 KBssocs2016_varname.pdf375 KB008677032009-2010qsOnpsOntsOn2,600ImputationCompleted SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond: In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical. WeightingData are weighted to compensate for differential probabilities of selection and to adjust for the effects of nonresponse.Sample weights allow inferences to be made about the population from which the sample units are drawn. Because of the complex nature of the SSOCS sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Due to nonresponse, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias due to nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The same predictor variables from the SSOCS:2004 CHAID analysis were used for SSOCS:2006: instructional level, region, enrollment size, percent minority, student-to-FTE teaching staff ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time equivalent (FTE) teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted so that the weighted distribution of the responding schools resembled the initial distribution of the total sample. The nonresponse-adjusted weights were then poststratified to calibrate the sample to known population totals. Two dimension margins were set up for the poststratification—(1) instructional level and school enrollment size; and (2) instructional level and locale—and an iterative process known as the raking ratio adjustment brought the weights into agreement with known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. All three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2004). Miller, A.K. (2004). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.ssocs2010_subject.pdf565 KBssocs2010_varname.pdf365 KB008577432007-2008qsOnpsOntsOn2,560ImputationCompleted SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond: In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical. WeightingData are weighted to compensate for differential probabilities of selection and to adjust for the effects of nonresponse.Sample weights allow inferences to be made about the population from which the sample units are drawn. Because of the complex nature of the SSOCS sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Due to nonresponse, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias due to nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The same predictor variables from the SSOCS:2004 CHAID analysis were used for SSOCS:2006: instructional level, region, enrollment size, percent minority, student-to-FTE teaching staff ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time equivalent (FTE) teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted so that the weighted distribution of the responding schools resembled the initial distribution of the total sample. The nonresponse-adjusted weights were then poststratified to calibrate the sample to known population totals. Two dimension margins were set up for the poststratification—(1) instructional level and school enrollment size; and (2) instructional level and locale—and an iterative process known as the raking ratio adjustment brought the weights into agreement with known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. All three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2004). Miller, A.K. (2004). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.ssocs2008_subject.pdf1.96 MBssocs2008_varname.pdf912 KB008587332005-2006qsOnpsOntsOn2,720ImputationCompleted SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond: In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical. WeightingData are weighted to compensate for differential probabilities of selection and to adjust for the effects of nonresponse.Sample weights allow inferences to be made about the population from which the sample units are drawn. Because of the complex nature of the SSOCS sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Due to nonresponse, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias due to nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The same predictor variables from the SSOCS:2004 CHAID analysis were used for SSOCS:2006: instructional level, region, enrollment size, percent minority, student-to-FTE teaching staff ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time equivalent (FTE) teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted so that the weighted distribution of the responding schools resembled the initial distribution of the total sample. The nonresponse-adjusted weights were then poststratified to calibrate the sample to known population totals. Two dimension margins were set up for the poststratification—(1) instructional level and school enrollment size; and (2) instructional level and locale—and an iterative process known as the raking ratio adjustment brought the weights into agreement with known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. All three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2004). Miller, A.K. (2004). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.ssocs2006_subject.pdf8.82 MBssocs2006_varname.pdf3.58 MB008591382003-2004qsOnpsOntsOff2,800ImputationCompleted SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond. In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical.WeightingSample weights allow inferences to be made about the population from which the sample units were drawn. Because of the complex nature of the SSOCS:2004 sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. The procedures used to create the SSOCS sampling weights are described below.An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Because some schools refused to participate, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias from nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (i.e., chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The predictor variables for the analysis were instructional level, region, enrollment size, percent minority, student-to-teacher ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time-equivalent teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted by dividing the base weight by the response rate in each class, so that the weighted distribution of the responding schools resembled the initial distribution of the total sample.The non-response-adjusted weights were then poststratified to calibrate the sample to known population totals. For SSOCS:2004, two dimension margins were set up for the poststratification: (1) instructional level and school enrollment size, and (2) instructional level and locale. An iterative process known as the raking ratio adjustment brought the weights into agreement with the known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. Similar to SSOCS:2000, all three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2003).Miller, A.K. (2003). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.View methodology reportssocs2004_subject.pdf1.2 MBssocs2004_varname.pdf2.1 MB008741391999-2000qsOnpsOntsOff2,300ImputationAll key data items with missing values were imputed using well-known procedures. Depending on the type of data to be imputed and the extent of missing values, logical imputation, poststratum means, or "hot-deck" imputation methods were employed. For three data items, imputation was done using information from the 1998-99 CCD file. Logical imputation is the assignment of data values based on other information in the data record. In the poststratum means method, a record with missing data was assigned the mean value of those cases in the same "poststratum" for which information on the item was available. The poststrata or "imputation classes" were defined on the basis of variables that were correlated with the item being imputed. Preliminary exploratory analyses (e.g., using chi-square tests of association, correlation analysis, and regression analysis) were carried out to identify the relevant classification variables. The strength of association of the variables in combination with subjective assessment was used to prioritize the importance of the variables in forming the imputation classes. In the "hot-deck" technique, cases with missing items were assigned the corresponding value of a "similar" respondent in the same "poststratum". Similar to the poststratum means approach, preliminary exploratory analyses were carried out to identify the relevant classification variables to be used to define the poststrata. The classification variables were separated into two groups -- "hard" and "soft" boundary variables. The hard boundary variables were considered to be so important that the imputation classes were always formed within those boundaries. The boundaries formed by the soft boundary variables were crossed, if necessary, to form the imputation class.WeightingA stratified random sample design was used to select schools for the SSOCS:2000. Over 3,000 schools were selected at rates that varied by sampling stratum; i.e., the classes formed by crossing instructional level (elementary, middle, secondary, combined), type of locale (city, urban fringe, town, rural), and enrollment size class (less than 300, 300-499, 500-999, 1,000+). Since the schools were selected with unequal probabilities, sampling weights are required for analysis to inflate the survey responses to population levels. Weighting is also used to reduce the potential bias resulting from nonresponse and possible undercoverage of the sampling frame.One method of computing sampling errors to reflect various aspects of the sample design and estimation procedures is the replication method. Under replication methods, a specified number of subsamples of the full sample (called "replicates") are created. The survey estimates can then be computed for each of the replicates by creating replicate weights that mimic the actual sample design and estimation procedures used in the full sample. The variability of the estimates computed from the replicate weights is then used to estimate the sampling errors of the estimates from the full sample. An important advantage of the replication methods is that they preclude the need to specify cumbersome variance formulas that are typically needed for complex sample designs (McCarthy, 1966).1 Another advantage is that they can readily be adapted to reflect the variance resulting from nonresponse (and other weight) adjustment procedures. The two most prevalent replication methods are balanced repeated replication (BRR) and jackknife replication. The two methods differ in the manner in which the replicates are constructed. For the SSOCS:2000, a variant of jackknife replication was used to develop replicate weights for variance estimation because the jackknife method is believed to perform somewhat better than BRR for estimates of moderately rare events (e.g., number of schools in which a serious crime was committed). Under the jackknife method, the replicates are formed by deleting specified subsets of units from the full sample. The jackknife method provides a relatively simple way of creating the replicates for variance estimation and has been used extensively in NCES surveys.1. McCarthy, P. (1966). Replication: An Approach to the Analysis of Data from Complex Surveys. Vital and Health Statistics, Series 2, No. 14. Washington, DC: U.S. Department of Health, Education and Welfare.View methodology reportssocs2000_subject.pdfX KBssocs2000_varname.pdfX KB00875Education Longitudinal StudyELSStudents who were high school sophomores in 2001-02 or high school seniors in 2003-04Student and Family Background, School and Classroom Characteristics, High School Completion and Dropout Status, Postsecondary Education Choice and Enrollment, Postsecondary Attainment, Employment, Transition to Adult Roleshttps://nces.ed.gov/surveys/els2002682002qsOnpsOntsOff14,000 to 16,000ImputationStochastic methods were used to impute the missing values for the ELS:2002 third follow-up data. Specifically, a weighted sequential hot-deck (WSHD) statistical imputation procedure (Cox 1980; Iannacchione 1982) using the final analysis weight (F3QWT) was applied to the missing values for the variables in table 12 in the order in which they are listed. The WSHD procedure replaces missing data with valid data from a donor record within an imputation class. In general, variables with lower item nonresponse rates were imputed earlier in the process.View methodology reportels2002sophomores_subject.pdf7.58 MBels2002sophomores_varname.pdf7.49 MB32954692002qsOffpsOntsOff14,000 to 16,000ImputationStochastic methods were used to impute the missing values for the ELS:2002 third follow-up data. Specifically, a weighted sequential hot-deck (WSHD) statistical imputation procedure (Cox 1980; Iannacchione 1982) using the final analysis weight (F3QWT) was applied to the missing values for the variables in table 12 in the order in which they are listed. The WSHD procedure replaces missing data with valid data from a donor record within an imputation class. In general, variables with lower item nonresponse rates were imputed earlier in the process.View methodology reportels2002seniors_subject.pdf6.22 MBels2002seniors_varname.pdf6.16 MB42954High School Longitudinal StudyHSLSStudents who were high school freshmen in the fall of 2009Student Background, Math and Science Education, Classroom Characteristics, The Changing Environment of High School, Postsecondary Education Choice and Enrollment, Transition to Adult Roleshttps://nces.ed.gov/surveys/hsls09722009qsOnpsOntsOff23,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, HSLS:09 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. ImputationStochastic methods were used to impute the missing values. Specifically, a weighted sequential hot-deck (WSHD; statistical) imputation procedure (Cox 1980; Iannacchione 1982) using the final student analysis weight (W2STUDENT) was applied to the missing values for variables. The WSHD procedure replaces missing data with valid data from a donor record (i.e., first follow-up student [item] respondent) within an imputation class. In general, variables with lower item nonresponse rates were imputed earlier in the process. Skips and Missing Values The HSLS:09 data were edited using procedures developed and implemented for previous studies sponsored by NCES Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in HSLS:09. Please consult the methodology report (coming soon) for more information. Description of missing data codes Missing data code Description -1 Don't know -4 Item not administered: abbreviated interview -5 Suppressed -6 Component not applicable -7 Item legitimate skip/NA -8 Unit nonresponse -9 Missing hsls2009_subject.pdf5.34 MBhsls2009_varname.pdf8.91 MB001056Baccalaureate and BeyondB&BBachelor degree recipients who were surveyed at the time of graduation, one year after graduation, four years after graduation, and ten years after graduationOutcomes for bachelor's degree recipients, Graduate and professional program access, Labor market experiences, Rates of return on investment in education, Post-baccalaureate education, Teacher preparation, Certifications and licenses, Enrollment while employedhttps://nces.ed.gov/surveys/b&b1342016/2017qsOnpsOntsOff29,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values During and following data collection, the data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a number of explanations for missing data; for example, the item may not have been applicable to certain respondents or a respondent may not have known the answer to the question. With the exception of the not applicable codes, missing data were stochastically imputed. Moreover, for hierarchical analyses and developing survey estimates for faculty members corresponding to sample institutions that provided faculty lists and responded to the institution survey, contextual weights were produced for such subsets of the responding faculty members. The table below shows codes for missing values used. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -3 Legitimate skip -7 Not reached -9 Missing 1Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable imputed and observed will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction. Methodology report coming soonbb2017_subject.pdf608 KBbb2017_varname.pdf420 KB001173542008/2012qsOnpsOntsOff15,500Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, B&B:08/12 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. B&B:08/12 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. Imputation Variables with missing data were imputed for graduates who were respondents in a study wave . The imputation procedures employed a two-step process. The first step is a logical imputation . If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation. This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The B&B: 08/12 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:08. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in B&B:08/12. Please consult the First Look for more information. Description of missing value codes Missing data codeDescription -1Don’t know -2Independent student -3Skipped -9Missing 1In other words, if a graduate was a respondent in B&B:09, he or she will have no missing data for variables created as part of the B&B:09 wave. Similarly, if a graduate was a respondent in B&B:12, he or she will have no missing data for variables created as part of the B&B:12 wave, but may have missing data for variables created as part of the B&B:09 wave if he or she was not a respondent in B&B:09. 2Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 3Sequential hot deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction. View methodology reportbb12_subject.pdf26.6 MBbb12_varname.pdf15.6 MB001149311993/2003qsOnpsOntsOff11,200Imputation Variables used in cross-sectional estimates in the Baccalaureate and Beyond descriptive reports were imputed. The variables identified for imputation were used in the two B&B:93/03 descriptive reports (Bradburn, Nevill, and Forrest Cataldi 2006; Alt and Henke 2007). The imputations were performed in three steps. First, the interview variables were imputed using the sequential hot deck imputation method.1 This imputation procedure involves identifying a relatively homogenous group of observations, and within the group selecting a random donor’s value to impute a value for the recipient. Second, using the interview variables, including the newly imputed variable values, derived variables were constructed. Skips and Missing Values Both during and upon completion of data collection, edit checks were performed on the B&B:93/03 data file to confirm that the intended skip patterns were implemented during the interview. At the conclusion of data collection, special codes were added as needed to indicate the reason for missing data. Missing data within individual data elements can occur for a variety of reasons. The Table below lists each missing value code and its associated meaning in the B&B:93/03 interview. For more information, see the Baccalaureate and Beyond Longitudinal Study (B&B:93/03) methodology report. Description of missing data codes Missing data code Description -1 Missing -2 Not applicable -3 Skipped -4 B&B:97 nonrespondent not sampled -6 Uncodeable, out of range -7 Not reached -8 Item was not reached due to an error -9 Missing, blank 1Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. Under this methodology, while each respondent record has a chance to be selected for use as a hot-deck donor, the number of times a respondent record can be used for imputation will be controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictor) for each item being imputed were defined. Imputation classes were developed by using a Chi-squared Automatic Interaction.View methodology reportbb03_subject.pdf4.56 MBbb03_varname.pdf3.98 MB001151202000/2001qsOffpsOntsOff10,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, B&B:01 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Both during and upon completion of data collection, edit checks were performed on the B&B:00/01 data file to confirm that the intended skip patterns were implemented during the interview. Following data collection, the information collected in CATI was subjected to various checks and examinations. These checks were intended to confirm that the database reflected appropriate skip-pattern relationships and different types of missing data by inserting special codes. The Table below lists each missing value code and its associated meaning in the B&B:00/01 interview. For more information, see the Baccalaureate and Beyond Longitudinal Study (B&B:00/01) methodology report . Description of missing data codes Missing data code Description -1 Don’t know (CATI variables), Data not available (CADE variables) -2 Refused (CATI variables only) -3 Not applicable (CADE and CATI variables only) -4 B&B:97 nonrespondent not sampled -6 Bad data, out of range -7 Item was not reached (abbreviated and partial CATI interviews) -8 Item was not reached due to a CATI error -9 Data missing, reason unknown (CATI variables) View methodology reportbb01_subject.pdf3.44 MBbb01_varname.pdf3.38 MB001132Baccalaureate and Beyond, Graduate StudentsB&B:GRBachelor degree recipients who were surveyed at the time of graduation, one year after graduation, four years after graduation, and ten years after graduationOutcomes for bachelor's degree recipients, Graduate and professional program access, Labor market experiences, Rates of return on investment in education, Post-baccalaureate education, Teacher preparation, Certifications and licenses, Enrollment while employedhttps://nces.ed.gov/surveys/b&b561993/2003qsOffpsOntsOff4,000Imputation Variables used in cross-sectional estimates in the Baccalaureate and Beyond descriptive reports were imputed. The variables identified for imputation were used in the two B&B:93/03 descriptive reports (Bradburn, Nevill, and Forrest Cataldi 2006; Alt and Henke 2007). The imputations were performed in three steps. First, the interview variables were imputed using the sequential hot deck imputation method.1 This imputation procedure involves identifying a relatively homogenous group of observations, and within the group selecting a random donor’s value to impute a value for the recipient. Second, using the interview variables, including the newly imputed variable values, derived variables were constructed. Skips and Missing Values Both during and upon completion of data collection, edit checks were performed on the B&B:93/03 data file to confirm that the intended skip patterns were implemented during the interview. At the conclusion of data collection, special codes were added as needed to indicate the reason for missing data. Missing data within individual data elements can occur for a variety of reasons. The Table below lists each missing value code and its associated meaning in the B&B:93/03 interview. For more information, see the Baccalaureate and Beyond Longitudinal Study (B&B:93/03) methodology report. Description of missing data codes Missing data code Description -1 Missing -2 Not applicable -3 Skipped -4 B&B:97 nonrespondent not sampled -6 Uncodeable, out of range -7 Not reached -8 Item was not reached due to an error -9 Missing, blank 1Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. Under this methodology, while each respondent record has a chance to be selected for use as a hot-deck donor, the number of times a respondent record can be used for imputation will be controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictor) for each item being imputed were defined. Imputation classes were developed by using a Chi-squared Automatic Interaction. View methodology reportbb03_subject_students.pdf9.75 MBbb03_varname_students.pdf8.81 MB001252Beginning Postsecondary StudentsBPSBeginning students who were surveyed at the end of their first year, and then three and six years after first starting in postsecondary education. Students’ persistence, progress and attainment of a degree, Labor force experienceshttps://nces.ed.gov/surveys/bps/712012/2017qsOnpsOntsOff22,500Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, BPS:12/17 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:12/17 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The BPS:12/17 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:12. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in BPS:12/17. Please consult the methodology report (coming soon) for more information. Description of missing data codes Missing data code Description -1 Not classified -2 Not applicable -3 Skipped -8 Double non-respondent -9 Data missing 1Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction. bps2017_subject.html9.99 MBbps2017_varname.html9.99 MB001353532004/2009qsOnpsOntsOff16,500Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, BPS:04/09 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:04/09 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The BPS:04/09 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:04. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in BPS:04/09. Please consult the methodology report (coming soon) for more information. Description of missing data codes Missing data code Description -2 Independent student -3 Skipped -9 Data missing 1Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction. View methodology reportbps2009_subject.pdf7.50 MBbps2009_varname.pdf6.20 MB00133311996/2001qsOnpsOntsOff12,000Imputation Logical imputations were performed where items were missing but their values could be implicitly determined. Skips and Missing Values During and following data collection, the CATI/CAPI data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a variety of explanations for missing data within individual data elements. The table below shows codes for missing values used in BPS:01. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -1 Don’t know -2 Refused -3 Legitimate skip (item was intentionally not collected because variable was not applicable to this student) -6 Bad data, out of range, uncodeable userexit string -7 Not reached -8 Missing, CATI error -9 Missing View methodology reportbps2001_subject.pdf9.20 MBbps2001_varname.pdf7.10 MB001334321990/1994qsOnpsOntsOff6,600Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, BPS:94 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:94 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The BPS:94 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data.A variety of explanations are possible for missing data.The table below shows codes for missing values used in BPS:94. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -2 Independent student -3 Skipped -9 Data missing View methodology reportbps1994_subject.pdf4.34 MBbps1994_varname.pdf4.17 MB001335National Postsecondary Student Aid Study, UndergraduateNPSAS:UGStudents who were undergraduates at the time of interviewGeneral demographics, Types of aid and amounts received, Cost of attending college, Combinations of work, study, and borrowing, Enrollment patternshttps://nces.ed.gov/surveys/npsas12112016qsOnpsOntsOn89,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 2016. Please consult the data file documentation report for more information. Description of missing data codes Missing data code Description -3 Skipped -9 Missing 1Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variableimputed and observedwill resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction. View methodology reportnpsas2016ug_subject.pdf8.7 MBnpsas2016ug_varname.pdf6.7 MB0014628212012qsOnpsOntsOn95,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Missing Values and Imputation Following data collection, the data are subjected to various consistency and quality control checks before release. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Except for data that were missing for cases to which they did not apply (e.g., whether a spouse is enrolled in college for unmarried students) and in a small number of items describing institutional characteristics, missing data were imputed using a two-step process. The first step is a logical imputation.1 If a value could be calculated from the logical relationships with other variables, then that information was used to impute the value for the observation with a missing value. The second step is weighted hot deck imputation.2 This procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor's value to impute a value for the observation with a missing value. The table below shows the set of missing value codes for missing values that were not imputed in NPSAS:12. More information is available from the NPSAS:12 Data File Documentation (http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2014182). Description of missing value codes Missing data codeDescription -1Not classified -2Not applicable -3Skipped -9Missing 1Logical imputation is a process that aims to infer or deduce the missing values from values for other items. 2Sequential hot deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent's answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using the chi-square automatic interaction detection algorithm. View methodology reportnpsas2012ug_subject.pdf6.90 MBnpsas2012ug_varname.pdf5.45 MB0014365112008qsOnpsOntsOn113,500Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 2008. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -1 Not classified -2 Not applicable -6 Out of range -8 Item was not reached due to an error -9 Missing 1Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction. View methodology reportnpsas2008ug_subject.pdf8.10 MBnpsas2008ug_varname.pdf6.40 MB0014372412004qsOnpsOntsOn79,900 Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation The imputation procedures employed a two-step process. In the first step, the matching criteria and imputation classes that were used to stratify the dataset were identified such that all imputation was processed independently within each class. In the second step, the weighted sequential hot deck process1 was implemented, whereby missing data were replaced with valid data from donor records that match the recipients with respect to the matching criteria. Variables requiring imputation were not imputed simultaneously. However, some variables that were related substantively were grouped together into blocks, and the variables within a block were imputed simultaneously. Basic demographic variables were imputed first using variables with full information to determine the matching criteria. The order in which variables were imputed was also determined to some extent by the substantive nature of the variables. For example, basic demographics (such as age) were imputed first and these were used to process education variables (such as student level and enrollment intensity) which in turn were used to impute the financial aid variables (such as aid receipt and loan amounts). Skips and Missing Values Edit checks were performed on the NPSAS:04 student interview data and CADE data, both during and upon completion of data collection, to confirm that the intended skip patterns were implemented in both instruments. At the conclusion of data collection, special codes were added as needed to indicate the reason for missing data. Missing data within individual data elements can occur for a variety of reasons. The table below shows the set of reserve codes for missing values used in NPSAS 2004. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -1 Not classified -3 Legitimate skip -9 Missing 1Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction.View methodology reportnpsas2004ug_subject.pdf7.75 MBnpsas2004ug_varname.pdf6.00 MB0014383512000qsOffpsOntsOn50,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, NPSAS:00 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing ValuesThe NPSAS:00 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in NPSAS:00 Please consult the methodology report for more information. Description of missing data codes Missing data code Description -2 Independent student -3 Skipped -9 Data missing View methodology reportnpsas2000ug_subject.pdf8.68 MBnpsas2000ug_varname.pdf7.25 MB0014393611996qsOffpsOntsOn41,500Imputation Values for 22 analysis variables were imputed. The variables were imputed using a weighted hot deck procedure, with the exception of estimated family contribution (EFC), which was imputed through a multiple regression approach.The weighed hot deck imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing ValuesThe NPSAS:96 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in NPSAS:96 Please consult the methodology report for more information. Description of missing data codes Missing data code Description -1 Don't know -2 Refused -3 Skipped -8 Data source not available -9 Data missing View methodology reportnpsas1996ug_subject.pdf3.47 MBnpsas1996ug_varname.pdf3.09 MB001440221993qsOffpsOntsOff52,700Derived Variables and Imputed Values Approximately 800 variables have been constructed based on data collected in the NPSAS:93. As a general rule, the constructions of derive variables that concern financial aid and other financial descriptors depend first on record abstract data from the CADE system. These data are supplemented in many cases with information collected in the telephone interviews with parents and students. As between parent and student data, precedence was generally given to parent data for variables concerning family income and assets. Imputations were performed on seven variables that contained missing values. Skips and Missing Values Both the student and parent CATI programs were designed to accommodate responses of "refusal" and "don't know" to any single question. Typically, refusal responses are given for items considered too sensitive by the respondent. "Don't know" responses may be given for any one of several reasons: (1) the respondent misunderstands the question wording, and is not offered subsequent explanation by the interviewer; (2) the respondent is hesitant to provide "best guess" responses, with insufficient prompting from the interviewer; (3) the respondent truly does not know the answer; or (4) the respondent chooses to respond with "don't know" as an implicit refusal to answer the question. Whenever they occur, indeterminate responses in the data set must be resolved by imputation or otherwise dealt with during analysis. The table below shows the set of reserve codes for missing values used in NPSAS 1993. Please consult the data file documentation report for more information. Description of missing data codes Missingdata code Description -1 Legitimate skip -7 Missing, refused -8 Missing, don't know -9 Missing, blank View methodology reportnpsas93ug_subject.pdf500 KBnpsas93ug_varname.pdf500 KB001440141990qsOffpsOntsOff46,800Imputation Variables with more than 5 percent missing cases were imputed. After using information from all appropriate secondary sources, there remained eight variables which required some statistical imputation. Two methods of statistical imputation were used, regression-based or hot deck. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 1990. Please consult the data file documentation report for more information. Description of missing data codes Missingdata codeDescription-1 Legitimate skip -9 Missing, blankView methodology reportnpsas90ug_subject.pdf500 KBnpsas90ug_varname.pdf500 KB001440171987qsOffpsOntsOff34,500Derived Variables and Imputed Values Approximately 800 variables have been constructed based on data collected in the NPSAS:87. As a general rule, the constructions of derive variables that concern financial aid and other financial descriptors depend first on record abstract data from the CADE system. These data are supplemented in many cases with information collected in the telephone interviews with parents and students. As between parent and student data, precedence was generally given to parent data for variables concerning family income and assets. Imputations were performed on seven variables that contained missing values. Skips and Missing Values Both the student and parent CATI programs were designed to accommodate responses of "refusal" and "don't know" to any single question. Typically, refusal responses are given for items considered too sensitive by the respondent. "Don't know" responses may be given for any one of several reasons: (1) the respondent misunderstands the question wording, and is not offered subsequent explanation by the interviewer; (2) the respondent is hesitant to provide "best guess" responses, with insufficient prompting from the interviewer; (3) the respondent truly does not know the answer; or (4) the respondent chooses to respond with "don't know" as an implicit refusal to answer the question. Whenever they occur, indeterminate responses in the data set must be resolved by imputation or otherwise dealt with during analysis. The table below shows the set of reserve codes for missing values used in NPSAS 1993. Please consult the data file documentation report for more information. Description of missing data codes Missingdata code Description -1 Legitimate skip -7 Missing, refused -8 Missing, don't know -9 Missing, blank View methodology reportnpsas87ug_subject.pdf500 KBnpsas87ug_varname.pdf500 KB001440National Postsecondary Student Aid Study, GraduateNPSAS:GRStudents who were graduate and first-professional students at the time of interview General demographics, Types of aid and amounts received, Cost of attending college, Combinations of work, study, and borrowing, Enrollment patternshttps://nces.ed.gov/surveys/npsas12222016qsOnpsOntsOn24,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 2016. Please consult the data file documentation report for more information. Description of missing data codes Missing data code Description -3 Skipped -9 Missing 1Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variableimputed and observedwill resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction. View methodology reportnpsas2016gr_subject.pdf6.6 MBnpsas2016gr_varname.pdf5.4 MB0015638322012qsOnpsOntsOn16,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Missing Values and Imputation Following data collection, the data are subjected to various consistency and quality control checks before release. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Except for data that were missing for cases to which they did not apply (e.g., whether a spouse is enrolled in college for unmarried students) and in a small number of items describing institutional characteristics, missing data were imputed using a two-step process. The first step is a logical imputation.1 If a value could be calculated from the logical relationships with other variables, then that information was used to impute the value for the observation with a missing value. The second step is weighted hot deck imputation.2 This procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor's value to impute a value for the observation with a missing value. The table below shows the set of missing value codes for missing values that were not imputed in NPSAS:12. More information is available from the NPSAS:12 Data File Documentation (http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2014182). Description of missing value codes Missing data codeDescription -1Not classified -2Not applicable -3Skipped -9Missing 1Logical imputation is a process that aims to infer or deduce the missing values from values for other items. 2Sequential hot deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent's answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using the chi-square automatic interaction detection algorithm. View methodology reportnpsas2012gr_subject.pdf1.47 MBnpsas2012gr_varname.pdf4.20 MB0015415222008qsOnpsOntsOn14,200Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation.1 If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 2008. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -1 Not classified -3 Not applicable -6 Out of range -8 Item was not reached due to an error -9 Missing 1Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction.View methodology reportnpsas2008gr_subject.pdf1.02 MBnpsas2008gr_varname.pdf748 KB0015421222004qsOnpsOntsOn10,900 Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation The imputation procedures employed a two-step process. In the first step, the matching criteria and imputation classes that were used to stratify the dataset were identified such that all imputation was processed independently within each class. In the second step, the weighted sequential hot deck process1 was implemented, whereby missing data were replaced with valid data from donor records that match the recipients with respect to the matching criteria. Variables requiring imputation were not imputed simultaneously. However, some variables that were related substantively were grouped together into blocks, and the variables within a block were imputed simultaneously. Basic demographic variables were imputed first using variables with full information to determine the matching criteria. The order in which variables were imputed was also determined to some extent by the substantive nature of the variables. For example, basic demographics (such as age) were imputed first and these were used to process education variables (such as student level and enrollment intensity) which in turn were used to impute the financial aid variables (such as aid receipt and loan amounts). Skips and Missing Values Edit checks were performed on the NPSAS:04 student interview data and CADE data, both during and upon completion of data collection, to confirm that the intended skip patterns were implemented in both instruments. At the conclusion of data collection, special codes were added as needed to indicate the reason for missing data. Missing data within individual data elements can occur for a variety of reasons. The table below shows the set of reserve codes for missing values used in NPSAS 2004. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -1 Not classified -3 Legitimate skip -9 Missing 1Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction.View methodology reportnpsas2004gr_subject.pdf1.06 MBnpsas2004gr_varname.pdf787 KB0015433722000qsOffpsOntsOn12,000Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, NPSAS:00 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The NPSAS:00 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in NPSAS:00 Please consult the methodology report for more information. Description of missing data codes Missing data code Description -2 Independent student -3 Skipped -9 Data missing View methodology reportnpsas2000gr_subject.pdf1.71 MBnpsas2000gr_varname.pdf1.43 MB0015443821996qsOffpsOntsOn7,000Imputation Values for 22 analysis variables were imputed. The variables were imputed using a weighted hot deck procedure, with the exception of estimated family contribution (EFC), which was imputed through a multiple regression approach.The weighed hot deck imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing ValuesThe NPSAS:96 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in NPSAS:96 Please consult the methodology report for more information. Description of missing data codes Missing data code Description -1 Don''t know -2 Refused -3 Skipped -8 Data source not available -9 Data missing View methodology reportnpsas1996gr_subject.pdf2.53 MBnpsas1996gr_varname.pdf2.13 MB001545131993qsOffpsOntsOff13,400Derived Variables and Imputed Values Approximately 800 variables have been constructed based on data collected in the NPSAS:93. As a general rule, the constructions of derive variables that concern financial aid and other financial descriptors depend first on record abstract data from the CADE system. These data are supplemented in many cases with information collected in the telephone interviews with parents and students. As between parent and student data, precedence was generally given to parent data for variables concerning family income and assets. Imputations were performed on seven variables that contained missing values. Skips and Missing Values Both the student and parent CATI programs were designed to accommodate responses of "refusal" and "don't know" to any single question. Typically, refusal responses are given for items considered too sensitive by the respondent. "Don't know" responses may be given for any one of several reasons: (1) the respondent misunderstands the question wording, and is not offered subsequent explanation by the interviewer; (2) the respondent is hesitant to provide "best guess" responses, with insufficient prompting from the interviewer; (3) the respondent truly does not know the answer; or (4) the respondent chooses to respond with "don't know" as an implicit refusal to answer the question. Whenever they occur, indeterminate responses in the data set must be resolved by imputation or otherwise dealt with during analysis. The table below shows the set of reserve codes for missing values used in NPSAS 1993. Please consult the data file documentation report for more information. Description of missing data codes Missingdata code Description -1 Legitimate skip -7 Missing, refused -8 Missing, don't know -9 Missing, blank View methodology reportnpsas93gr_subject.pdf500 KBnpsas93gr_varname.pdf500 KB001540181987qsOffpsOntsOff8,600Derived Variables and Imputed Values Approximately 800 variables have been constructed based on data collected in the NPSAS:93. As a general rule, the constructions of derive variables that concern financial aid and other financial descriptors depend first on record abstract data from the CADE system. These data are supplemented in many cases with information collected in the telephone interviews with parents and students. As between parent and student data, precedence was generally given to parent data for variables concerning family income and assets. Imputations were performed on seven variables that contained missing values. Skips and Missing Values Both the student and parent CATI programs were designed to accommodate responses of "refusal" and "don't know" to any single question. Typically, refusal responses are given for items considered too sensitive by the respondent. "Don't know" responses may be given for any one of several reasons: (1) the respondent misunderstands the question wording, and is not offered subsequent explanation by the interviewer; (2) the respondent is hesitant to provide "best guess" responses, with insufficient prompting from the interviewer; (3) the respondent truly does not know the answer; or (4) the respondent chooses to respond with "don't know" as an implicit refusal to answer the question. Whenever they occur, indeterminate responses in the data set must be resolved by imputation or otherwise dealt with during analysis. The table below shows the set of reserve codes for missing values used in NPSAS 1993. Please consult the data file documentation report for more information. Description of missing data codes Missingdata code Description -1 Legitimate skip -7 Missing, refused -8 Missing, don't know -9 Missing, blank View methodology reportnpsas87gr_subject.pdf500 KBnpsas87gr_varname.pdf500 KB001540161990qsOffpsOntsOff14,300Imputation Variables with more than 5 percent missing cases were imputed. After using information from all appropriate secondary sources, there remained eight variables which required some statistical imputation. Two methods of statistical imputation were used, regression-based or hot deck. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 1990. Please consult the data file documentation report for more information. Description of missing data codes Missingdata codeDescription-1 Legitimate skip -9 Missing, blankView methodology reportnpsas90gr_subject.pdf500 KBnpsas90gr_varname.pdf500 KB001540National Study of Postsecondary FacultyNSOPFPostsecondary facultyWorkload, Equity issues, Involvement in undergraduate teaching, Relationship between teaching and researchhttps://nces.ed.gov/surveys/nsopf282004qsOnpsOntsOff26,100Perturbation A restricted faculty-level data file was created for release to individuals who apply for and meet standards for such data releases. While this file does not include personally identifying information (i.e., name and Social Security number), other data (i.e., institution, Integrated Postsecondary Education Data System [IPEDS] ID, demographic information, and salary data) may be manipulated in such a way to seem to identify data records corresponding to a particular faculty member. To protect further against such situations, some of the variable values were swapped between faculty respondents. This procedure perturbed and added additional uncertainty to the data. Thus, associations made among variable values to identify a faculty respondent may be based on the original or edited, imputed and/or swapped data. For the same reasons, the data from the institution questionnaire were also swapped to avoid data disclosure. Imputation Item imputation for the faculty questionnaire was performed in several steps. In the first step, the missing values of gender, race, and ethnicity were filled—using cold-deck imputation1— based on the sampling frame information or institution record data. These three key demographic variables were imputed prior to any other variables since they were used as key predictors for all other variables on the data file. After all logical2 and cold-deck imputation procedures were performed, the remaining variables were imputed using the weighted sequential hot-deck method.3 Initially, variables were separated into two groups: unconditional and conditional variables. The first group (unconditional) consisted of variables that applied to all respondents, while the second group (conditional) consisted of variables that applied to only a subset of the respondents. That is, conditional variables were subject to “gate” questions. After this initial grouping, these groups were divided into finer subgroups. After all variables were imputed, consistency checks were applied to the entire faculty data file to ensure that the imputed values did not conflict with other questionnaire items, observed or imputed. This process involved reviewing all of the logical imputation and editing rules as well. Skips and Missing Values During and following data collection, the data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a number of explanations for missing data; for example, the item may not have been applicable to certain respondents or a respondent may not have known the answer to the question. With the exception of the not applicable codes, missing data were stochastically imputed. Moreover, for hierarchical analyses and developing survey estimates for faculty members corresponding to sample institutions that provided faculty lists and responded to the institution survey, contextual weights were produced for such subsets of the responding faculty members. The table below shows codes for missing values used in NSOPF:04. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -3 Legitimate skip -7 Not reached -9 Missing 1Cold-deck imputation involves replacing the missing values with data from sources such as data used for sampling frame construction. While resource intensive, these methods often obtain the actual value that is missing. Stochastic imputation methods, such as sequential hot-deck imputation, rely on the observed data to provide replacing values (donors) for records with missing values. 2Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions. 3Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. Under this methodology, while each respondent record has a chance to be selected for use as a hot-deck donor, the number of times a respondent record can be used for imputation will be controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictor) for each item being imputed were defined. Imputation classes were developed by using a Chi-squared Automatic Interaction.View methodology reportnsopf04_subject.pdf1.16 MBnsopf04_varname.pdf926 KB001646261999qsOffpsOntsOff18,000Both the faculty and institution questionnaire data were edited using seven principles designed to improve data quality and consistency.Menu items. For many questions there were several sub-items listed where the respondent was asked to give a response for each sub-item. These questions were cleaned with several procedures. First if the main question had an “NA” (Not Applicable) check box and that box was checked, all of the sub-items were set to a value of “no” or “zero” depending on the wording of the question. Second, if the respondent had filled out one or more of the sub-items with a “yes” response or a positive number but had left other sub-items blank, the missing sub-items were set to “no,” “zero,” or “don’t know” depending on the question wording. If all sub-items were missing and there was no “NA” box, or the “NA” box was not checked, the case was flagged and the data values were imputed for that question. Examples of these types of questions are Question 21 in the institution questionnaire and Question 29 in the faculty questionnaire.Inter-item consistency checks. Many types of inter-item consistency checks were performed on the data. One procedure was to check groups of related items for internal consistency and to make adjustments to make them consistent. For example, in questions that asked about a spouse in the faculty questionnaire (Questions 66i, Q76i, and 77a) if respondents indicated that they did not have a spouse in one or more of the questions, the other questions were checked for consistency and corrected as necessary. Another procedure checked “NA” boxes. If the respondent had checked the “NA” box for a question but had filled in any of the sub-items for that question the “NA” box was set to blank. For example, this procedure was used with Question 21 in the institution questionnaire and Question 16 in the faculty questionnaire. A third procedure was to check filter items for which more detail was sought in a follow-up open-ended or closed-ended question. If detail was provided, then the filter question was checked to make sure the appropriate response was recorded. For example, this procedure was used with Question 11 in the institution questionnaire and Question 12E in the faculty questionnaire.Percent items. All items where respondents were asked to give a percentage were checked to make sure they summed to 100 percent. The editing program also looked for any numbers between 0 and 1 to make sure that respondents did not fill in the question with a decimal rather than a percentage. All fractions of a percent were rounded to the nearest whole percent. An example of this type of item is Question 31 in the faculty questionnaire.Data imputation for the faculty questionnaire was performed in four steps. The imputation method for each variable is specified in the labels for the imputation flags in the faculty dataset.Logical imputation. The logical imputation was conducted during the data cleaning steps as explained in the immediately preceding section. Cold deck. Missing responses were filled in with data from the sample frame whenever the relevant data were available. Examples include gender, race, and employment status.Hot deck. This procedure selected non-missing values from “sequential nearest neighbors” within the imputation class. All questions that were categorical and had more than 16 categories were imputed with this method. An example is Question Q14 – principal field of teaching. The imputation class for this question was created using faculty stratum and instructional duty status (Q1). Regression type. This procedure employed SAS PROC IMPUTE21. All items that were still missing after the logical, cold deck, and hot deck imputation procedures were imputed with this method. Project staff selected the independent variables by first looking through the questionnaire for logically related items and then by conducting a correlation analysis of the questions against each other to find the top correlates for each item.View methodology reportnsopf99_subject.pdf500 KBnsopf99_varname.pdf500 KB001646251993qsOffpsOntsOff31,000Depending on the scale of the variable being imputed, one of two methods were used:1) Regression imputation was used for continuous and dichotomous variables; and2) Hotdeck imputation was used for unordered polytomous variables.The regression method incorporated in NCES’s PROC IMPUTE was used to impute missing values for approximately 90 percent of the 395 items on the faculty questionnaire.Of the total of 395 items, 353 were imputed using the regression-based imputation procedures only.View methodology reportnsopf93_subject.pdf500 KBnsopf93_varname.pdf500 KB001646231988qsOffpsOntsOff25,000NSOPF:88 was conducted with a sample of 480 institutions (including 2-year, 4-year, doctoral-granting, and other colleges and universities), some 11,010 faculty, and more than 3,000 department chairpersons. Institutions were sampled from the 1987 IPEDS universe and were stratified by modified Carnegie Classifications and size (faculty counts). These strata were (1) public, research; (2) private, research; (3) public, other Ph.D. institution (not defined in any other stratum); (4) private, other Ph.D. institution (not defined in any other stratum); (5) public, comprehensive; (6) private, comprehensive; (7) liberal arts; (8) public, 2-year; (9) private, 2-year; (10) religious; (11) medical; and (12) “other” schools (not defined in any other stratum). Within each stratum, institutions were randomly selected. Of the 480 institutions selected, 450 (94 percent) agreed to participate and provided lists of their faculty and department chairpersons. Within 4-year institutions, faculty and department chairpersons were stratified by program area and randomly sampled within each stratum; within 2-year institutions, simple random samples of faculty and department chairpersons were selected; and within specialized institutions (religious, medical, etc.), faculty samples were randomly selected (department chairpersons were not sampled). At all institutions, faculty were also stratified on the basis of employment status—full-time and part-time. Note that teaching assistants and teaching fellows were excluded in NSOPF:88.Although NSOPF:88 consisted of three questionnaires, imputations were only performed for faculty item nonresponse. The within-cell random imputation method was used to fill in most Faculty Questionnaire items that had missing data.nsopf88_subject.pdf500 KBnsopf88_varname.pdf500 KB001646National Study of Postsecondary Faculty, InstitutionsNSOPFPostsecondary institutionsFaculty tenure policies, Union representation, and Faculty attritionhttps://nces.ed.gov/surveys/nsopf292004qsOnpsOntsOff900 Imputation The imputation process for the missing data from the institution questionnaire involved similar steps to those used for imputation of the faculty data. The missing data for variables were imputed using the weighted sequential hot-deck method.1 Analogous to the imputation process for the faculty data, the variables were partitioned into conditional and unconditional groups. The unconditional variables were sorted by percent missing and then imputed in the order from the lowest percent missing to the highest. The conditional group was partitioned into three subgroups based on the level of conditionality for each variable, and then imputed in that order. The imputation class for both unconditional and conditional variables consisted of the institution sampling stratum, and the sorting variables included the number of full-time and part-time faculty members. Skips and Missing Values During and following data collection, the data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a number of explanations for missing data; for example, the item may not have been applicable to certain respondents or a respondent may not have known the answer to the question. With the exception of the not applicable codes, missing data were stochastically imputed. Moreover, for hierarchical analyses and developing survey estimates for faculty members corresponding to sample institutions that provided faculty lists and responded to the institution survey, contextual weights were produced for such subsets of the responding faculty members. The table below shows codes for missing values used in NSOPF:04. Please consult the methodology report for more information. Description of missing data codes Missing data code Description -3 Legitimate skip -7 Not reached -9 Missing 1Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. Under this methodology, while each respondent record has a chance to be selected for use as a hot-deck donor, the number of times a respondent record can be used for imputation will be controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictor) for each item being imputed were defined. Imputation classes were developed by using a Chi-squared Automatic Interaction. View methodology reportnsopf04inst_subject.pdf543 KBnsopf04inst_varname.pdf471 KB001747National Teacher and Principal Survey, Public School PrincipalsNTPSPublic school principalsExperience, Training, Education, and Professional Development, Goals and Decision Making, Teacher and Aide Professional Development, School Climate and Safety, Instructional Time, Working Conditions and Principal Perceptions, Teacher and School Performancehttps://nces.ed.gov/surveys/ntps1262015-2016qsOnpsOntsOff8,300ImputationThe NTPS used two main approaches to impute data. First, donor respondent methods, such as hot-deck imputation, were used. Second, if no suitable donor case could be matched, the few remaining items were imputed using mean or mode from groups of similar cases to impute a value to the item with missing data. Finally, in rare cases for which imputed values were inconsistent with existing questionnaire data or out of the range of acceptable values, Census Bureau analysts looked at the items and tried to determine an appropriate value.WeightingWeighting of the sample units was carried out to produce national estimates for public schools, principals, and teachers. The weighting procedures used in NTPS had three purposes: to take into account the school's selection probability; to reduce biases that may result from unit nonresponse; and to make use of available information from external sources to improve the precision of sample estimates.ntps2016principals_subject.pdf1.53 MBntps2016principals_varname.pdf1.70 MB001965National Teacher and Principal Survey, Public SchoolsNTPSPublic schoolsTeacher demand, teacher and principal characteristics, general conditions in schools, principals' and teachers' perceptions of school climate and problems in their schools, teacher compensation, district hiring and retention practices, basic characteristics of the student populationhttps://nces.ed.gov/surveys/ntps1272015-2016qsOnpsOntsOff8,300ImputationThe NTPS used two main approaches to impute data. First, donor respondent methods, such as hot-deck imputation, were used. Second, if no suitable donor case could be matched, the few remaining items were imputed using mean or mode from groups of similar cases to impute a value to the item with missing data. Finally, in rare cases for which imputed values were inconsistent with existing questionnaire data or out of the range of acceptable values, Census Bureau analysts looked at the items and tried to determine an appropriate value.WeightingWeighting of the sample units was carried out to produce national estimates for public schools, principals, and teachers. The weighting procedures used in NTPS had three purposes: to take into account the school's selection probability; to reduce biases that may result from unit nonresponse; and to make use of available information from external sources to improve the precision of sample estimates.ntps2016schools_subject.pdf2.59 MBntps2016schools_varname.pdf3.35 MB002066Early Childhood Program ParticipationECPPChildren who were enrolled in some type of childcare programChildren's participation, Relative care, Nonrelative care, Center-based care, Head Start and Early Head start programs, time spent in care, number of children and care providershttps://nces.ed.gov/nhes13062016qsOnpsOntsOn5,800ImputationFour approaches to imputation were used in the NHES:2016: logic-based imputation, which was used whenever possible; unweighted sequential hot deck imputation, which was used for the majority of the missing data (i.e., for all variables that were not boundary and sort variables—described below); weighted random imputation, which was used for a small number of variables including boundary and sort variables; and manual imputation, which was used in a very small number of cases for a small number of variables.For more information about these approaches, please see the NHES: 2016 Data File User's Manual. ecpp2016_subject.pdfecpp2016_varname.pdf00216912962012qsOnpsOntsOn7,900ImputationThree approaches to imputation were used in the NHES:2012: unweighted sequential hot deck imputation, which was used for the majority of the missing data, that is, for all variables that were not required for Interview Status Recode (ISR) classification, as described in chapter 4; weighted random imputation, which was used for a small number of variables; and manual imputation, which was used in a very small number of cases for most variables.For more information about these approaches, please see the NHES: 2012 Data File User's Manual. ecpp2012_subject.pdfecpp2012_varname.pdf002168Adult Training and Education SurveyATESAdults who were enrolled in a training or literacy programEducation, Certifications and Licenses, Certificates, Work Experience Programs, Employment, Backgroundhttps://nces.ed.gov/nhes1332016qsOnpsOntsOff47,700ImputationFour approaches to imputation were used in the NHES:2016: logic-based imputation, which was used whenever possible; unweighted sequential hot deck imputation, which was used for the majority of the missing data (i.e., for all variables that were not boundary and sort variables—described below); weighted random imputation, which was used for a small number of variables including boundary and sort variables; and manual imputation, which was used in a very small number of cases for a small number of variables.For more information about these approaches, please see the NHES: 2016 Data File User's Manual. ates2016_subject.pdf2.84 MBates2016_varname.pdf2.90 MB002270Parent and Family Involvement in EducationPFIParents and families who were involved in their child's educationChildren's schooling, Families and schools, Homework, Family activities, Health, Background, Householdhttps://nces.ed.gov/nhes13272016qsOnpsOntsOn13,500ImputationFour approaches to imputation were used in the NHES:2016: logic-based imputation, which was used whenever possible; unweighted sequential hot deck imputation, which was used for the majority of the missing data (i.e., for all variables that were not boundary and sort variables—described below); weighted random imputation, which was used for a small number of variables including boundary and sort variables; and manual imputation, which was used in a very small number of cases for a small number of variables.For more information about these approaches, please see the NHES: 2016 Data File User's Manual. pfi2016_subject.pdf2.5 MBpfi2016_varname.pdf2.1 MB00237213172012qsOnpsOntsOn17,200ImputationThree approaches to imputation were used in the NHES:2012: unweighted sequential hot deck imputation, which was used for the majority of the missing data, that is, for all variables that were not required for Interview Status Recode (ISR) classification, as described in chapter 4; weighted random imputation, which was used for a small number of variables; and manual imputation, which was used in a very small number of cases for most variables.For more information about these approaches, please see the NHES: 2012 Data File User's Manual. pfi2012_subject.pdfpfi2012_varname.pdf002371High School and BeyondHSBStudents who were high school sophomores in 1980Social background, Test battery and school record, Home educational support system, Postsecondary education choice and enrollment, Employment, Outcomeshttps://nces.ed.gov/surveys/hsb/index.asp1351980qsOnpsOntsOff14,800Nonresponse Nonresponse inevitably introduces some degree of error into survey results. In examining the impact of nonresponse, it is useful to think of the survey population as including two strata--a respondent stratum that consists of all units that would have provided data had they been selected for the survey, and a nonrespondent stratum that consists of all units that would not have provided data had they been selected. The actual sample of respondents necessarily consists entirely of units from the respondent stratum. Thus, sample statistics can serve as unbiased estimates only for the respondent stratum; as estimates for the entire population, the sample statistics will be biased to the extent that the characteristics of the respondents differ from those of the entire population.In the High School and Beyond study, there were two stages of sample selection and therefore two stages of nonresponse. During the base year survey, sample schools were asked to permit the selection of individual sophomores and seniors from school rosters and to designate "survey days" for the collection of student questionnaire and test data. Schools that refused to cooperate in either of these activities were dropped from the sample. Individual students at cooperating schools could also fail to take part in the base year survey. Unlike "refusal" schools, nonparticipating students were not dropped from the sample; they remained eligible for selection into the follow-up samples.Estimates based on student data from the base year surveys include two components of nonresponse bias: bias introduced by nonresponse at the school level, and bias introduced by nonresponse on the part of students attending cooperating schools. Each component of the overall bias depends on two factors--the level of nonresponse and the difference between respondents and nonrespondents: Bias = P1(Y1R - Y1NR) + P2(Y2R - Y2NR)in which P1 = the proportion of the population of students attending schools that would have been nonrespondents,YlNR = the parameter describing the population of students attending nonrespondent schools, P2 = the proportion of students attending respondent schools who would have been nonrespondents, and Y2NR = the parameter describing this group of students.Nonresponse bias will be small if the nonrespondent strata constitute only a small portion of the survey population or if the differences between respondents and nonrespondents are small. The proportions P1 and P2 can generally be estimated from survey data using appropriately weighted nonresponse rates. The implications of the equation can be easily seen in terms of a particular base year estimate. On the average, sophomores got 10.9 items right on a standardized vocabulary test. This figure is an estimate of Y2R, the population mean for all participating students at cooperating schools. Now, suppose that sophomores at cooperating schools average two more correct than sophomores attending refusal schools (Y1R - Y1NR = 2), and suppose further that among sophomores attending cooperating schools, student respondents average one more correct answer than student nonrespondents (Y2R - Y2NR = 1). Noting that the base year school nonresponse rate was about .30 and the student nonresponse rate for sophomores was about .12, we can use these figures as estimates of P1 and P2 and we can use this equation to calculate the bias as: Bias = .30(2) + .12(1) = .72 That is, the sample estimate is biased by about .7 of a test score point.This example assumes knowledge of the relevant population means; in practice, of course, they are not known and, although Pl and P2 can generally be estimated from the nonresponse rates, the lack of survey data for nonrespondents prevents the estimation of the nonresponse bias. The High School and Beyond study is an exception to this general rule: during the first follow-up, school questionnaire data were obtained from most of the base year refusal schools, and student data were obtained from most of the base year student nonrespondents selected for the first follow-up sample. These data provide a basis for assessing the magnitude of nonresponse bias in base year estimates.The bias introduced by base year school-level refusals is of particular concern since it carries over into successive rounds of the survey. Students attending refusal schools were not sampled during the base year and have no chance for selection into subsequent rounds of observation. To the extent that these students differ from students from cooperating schools during later waves of the study, the bias introduced by base year school nonresponse will persist. Student nonresponse is not carried over in this way since student nonrespondents remain eligible for sampling in later waves of the study.The results of three types of analyses concerning nonresponse are described in an earlier report. Based on school questionnaire data, schools that participated during the base year were compared with all eligible schools. Based on the first follow-up student data, base year student respondents were compared with nonrespondents. Finally, student nonresponse during the first follow-up survey was analyzed. Taken together, these earlier analyses indicated that nonresponse had little effect on base year and first follow-up estimates. The results presented there suggest that the school-level component of the bias affected base year estimates by 2 percent or less and that the student-level component had even less impact.hsb1980_subject.pdf17.9 MBhsb1980_varname.pdf13.4 MB002556National Education Longitudinal Study of 1988NELS:88Students who were eighth graders in 1988School, work, home experiences, educational resources and support, the role in education of parents and peers, neighborhood characteristics, educational and occupational aspirations, other student perceptionshttps://nces.ed.gov/surveys/nels88/1361988qsOnpsOntsOff15,000NonresponseSchool-level nonresponse is a serious concern because it carries over into successive rounds of NELS:88. Students attending schools that did not cooperate in the base year were not sampled and had little or no chance of selection into the follow-up samples. To the extent that students at noncooperating schools differ from students at cooperating schools, the student-level bias introduced by base-year school noncooperation persists during subsequent waves. Nonresponse adjustments to weights are an attempt to compensate for bias in the estimate for a particular subgroup; they do not adjust for nonresponse bias within subgroups.In the base year, nonresponding schools were asked to supply information about key school questionnaire variables, and virtually all did so. Based on these data, analysis of school-level nonresponse suggests that, to the extent that schools can be characterized by size, control, organizational structure, student composition, and other characteristics, the impact of nonresponding schools on school level estimates is small.25 Readers interested in more information about the analyses of school nonresponse rates and bias for the NELS:88 base year should refer to the NELS:88 Base-Year Sample Design Report (Spencer et al. 1990). School nonresponse was not assessed in the first or second follow-ups for two reasons. First, there was practically no school-level nonresponse; institutional cooperation levels approached 99 percent in both rounds. Second, the first and second follow-up samples were student-driven, unlike the two-stage initial sample design in the base year. Hence, even if a school refused in either the first or second follow-ups, the individual student was pursued outside of school.25. The use of school questionnaire variables to assess bias in estimates concerning characteristics of the student population is not entirely straightforward. Still, to the extent that school characteristics are closely related to the characteristics of the students attending them, estimates based on school questionnaire data can serve as reasonable proxies for more direct estimates of student-level unit nonresponse bias.nels1988_subject.pdf6.9 MBnels1988_varname.pdf15.1 MB002657Adult Training and Education Survey: 20162016Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017Adult EducationqsOnpsOntsOffBeginning Postsecondary Students: 1990/19941990/1994PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 1996/20011996/2001PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 2004/20092004/2009PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 2012/20172012/2017PostsecondaryqsOnpsOntsOffEarly Childhood Program Participation: 20122012P-12qsOnpsOntsOn6Early Childhood Program Participation: 20162016P-12qsOnpsOntsOn6Education Longitudinal Study of 20022002PostsecondaryqsOnpsOntsOffEducation Longitudinal Study of 20022002P-12qsOnpsOntsOffHigh School and Beyond1980PostsecondaryqsOnpsOntsOffHigh School and Beyond1980P-12qsOnpsOntsOffHigh School Longitudinal Study of 20092009PostsecondaryqsOnpsOntsOffHigh School Longitudinal Study of 20092009P-12qsOnpsOntsOffNational Education Longitudinal Study of 19881988PostsecondaryqsOnpsOntsOffNational Education Longitudinal Study of 19881988P-12qsOnpsOntsOffNational Postsecondary Student Aid Study: 1987 Graduate Students1987PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Undergraduates1987PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Graduate Students1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Undergraduates1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Graduate Students1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Undergraduates1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1996 Graduate Students1996PostsecondaryqsOffpsOntsOn2National Postsecondary Student Aid Study: 1996 Undergraduates1996PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 2000 Graduate Students2000PostsecondaryqsOffpsOntsOn2National Postsecondary Student Aid Study: 2000 Undergraduates2000PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 2004 Graduate Students2004PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2004 Undergraduates2004PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2008 Graduate Students2008PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2008 Undergraduates2008PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2012 Graduate Students2012PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2012 Undergraduates2012PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2016 Graduate Students2016PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2016 Undergraduates2016PostsecondaryqsOnpsOntsOn1National Study of Postsecondary Faculty: 1988 Faculty1988PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1993 Faculty1993PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1999 Faculty1999PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 2004 Faculty2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Institution2004PostsecondaryqsOnpsOntsOffNational Teacher and Principal Survey, 2015-16 Public School Principals2015-2016P-12qsOnpsOntsOffNational Teacher and Principal Survey, 2015-16 Public Schools2015-2016P-12qsOnpsOntsOffParent and Family Involvement in Education: 20122012P-12qsOnpsOntsOn7Parent and Family Involvement in Education: 20162016P-12qsOnpsOntsOn7Pre-Elementary Education Longitudinal Study, Waves 1-52003/2008P-12qsOnpsOntsOffPrivate School Universe Survey: 2011-122011-2012P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 1999-20001999-2000P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 2003-042003-2004P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 2005-062005-2006P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2007-082007-2008P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2009-102009-2010P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2015-162015-2016P-12qsOnpsOntsOn3Schools and Staffing Survey, Districts: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Districts: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Library Media Centers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Library Media Centers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Districts: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 1999-001999-2000P-12qsOffpsOntsOffSchool Survey on Crime and Safety: 2015-162015-2016P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2009-102009-2010P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2007-082007-2008P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2005-062005-2006P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2003-042003-2004P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 1999-20001999-2000P-12qsOnpsOntsOffPrivate School Universe Survey: 2011-122011-2012P-12qsOnpsOntsOffPre-Elementary Education Longitudinal Study, Waves 1-52003/2008P-12qsOnpsOntsOffParent and Family Involvement in Education: 20162016P-12qsOnpsOntsOn7Parent and Family Involvement in Education: 20122012P-12qsOnpsOntsOn7National Teacher and Principal Survey, 2015-16 Public Schools2015-2016P-12qsOnpsOntsOffNational Teacher and Principal Survey, 2015-16 Public School Principals2015-2016P-12qsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Institution2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Faculty2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 1999 Faculty1999PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1993 Faculty1993PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1988 Faculty1988PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 2016 Undergraduates2016PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2016 Graduate Students2016PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2012 Undergraduates2012PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2012 Graduate Students2012PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2008 Undergraduates2008PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2008 Graduate Students2008PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2004 Undergraduates2004PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2004 Graduate Students2004PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2000 Undergraduates2000PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 2000 Graduate Students2000PostsecondaryqsOffpsOntsOn2National Postsecondary Student Aid Study: 1996 Undergraduates1996PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 1996 Graduate Students1996PostsecondaryqsOffpsOntsOn2National Postsecondary Student Aid Study: 1993 Undergraduates1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Graduate Students1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Undergraduates1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Graduate Students1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Undergraduates1987PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Graduate Students1987PostsecondaryqsOffpsOntsOffNational Education Longitudinal Study of 19881988PostsecondaryqsOnpsOntsOffNational Education Longitudinal Study of 19881988P-12qsOnpsOntsOffHigh School Longitudinal Study of 20092009PostsecondaryqsOnpsOntsOffHigh School Longitudinal Study of 20092009P-12qsOnpsOntsOffHigh School and Beyond1980PostsecondaryqsOnpsOntsOffHigh School and Beyond1980P-12qsOnpsOntsOffEducation Longitudinal Study of 20022002PostsecondaryqsOnpsOntsOffEducation Longitudinal Study of 20022002P-12qsOnpsOntsOffEarly Childhood Program Participation: 20162016P-12qsOnpsOntsOn6Early Childhood Program Participation: 20122012P-12qsOnpsOntsOn6Beginning Postsecondary Students: 2012/20172012/2017PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 2004/20092004/2009PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 1996/20011996/2001PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 1990/19941990/1994PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003Adult EducationqsOnpsOntsOffAdult Training and Education Survey: 20162016Adult EducationqsOnpsOntsOffHigh School and Beyond1980PostsecondaryqsOnpsOntsOffHigh School and Beyond1980P-12qsOnpsOntsOffNational Postsecondary Student Aid Study: 1987 Undergraduates1987PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Graduate Students1987PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1988 Faculty1988PostsecondaryqsOffpsOntsOffNational Education Longitudinal Study of 19881988PostsecondaryqsOnpsOntsOffNational Education Longitudinal Study of 19881988P-12qsOnpsOntsOffNational Postsecondary Student Aid Study: 1990 Undergraduates1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Graduate Students1990PostsecondaryqsOffpsOntsOffBeginning Postsecondary Students: 1990/19941990/1994PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 1993 Faculty1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Undergraduates1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Graduate Students1993PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003Adult EducationqsOnpsOntsOffNational Postsecondary Student Aid Study: 1996 Undergraduates1996PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 1996 Graduate Students1996PostsecondaryqsOffpsOntsOn2Beginning Postsecondary Students: 1996/20011996/2001PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 1999 Faculty1999PostsecondaryqsOffpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private Schools: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Library Media Centers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Districts: 1999-001999-2000P-12qsOffpsOntsOffSchool Survey on Crime and Safety: 1999-20001999-2000P-12qsOnpsOntsOffNational Postsecondary Student Aid Study: 2000 Undergraduates2000PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 2000 Graduate Students2000PostsecondaryqsOffpsOntsOn2Baccalaureate and Beyond: 2000/20012000/2001PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001Adult EducationqsOffpsOntsOffEducation Longitudinal Study of 20022002PostsecondaryqsOnpsOntsOffEducation Longitudinal Study of 20022002P-12qsOnpsOntsOffPre-Elementary Education Longitudinal Study, Waves 1-52003/2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2003-042003-2004P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 2003-042003-2004P-12qsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Institution2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Faculty2004PostsecondaryqsOnpsOntsOffNational Postsecondary Student Aid Study: 2004 Undergraduates2004PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2004 Graduate Students2004PostsecondaryqsOnpsOntsOn2Beginning Postsecondary Students: 2004/20092004/2009PostsecondaryqsOnpsOntsOffSchool Survey on Crime and Safety: 2005-062005-2006P-12qsOnpsOntsOn3Schools and Staffing Survey, Public and Private Teachers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2007-082007-2008P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 2007-082007-2008P-12qsOnpsOntsOn3National Postsecondary Student Aid Study: 2008 Undergraduates2008PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2008 Graduate Students2008PostsecondaryqsOnpsOntsOn2Baccalaureate and Beyond: 2008/20122008/2012PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012Adult EducationqsOnpsOntsOffHigh School Longitudinal Study of 20092009PostsecondaryqsOnpsOntsOffHigh School Longitudinal Study of 20092009P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 2009-102009-2010P-12qsOnpsOntsOn3Schools and Staffing Survey, Public and Private Teachers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2011-122011-2012P-12qsOnpsOntsOffPrivate School Universe Survey: 2011-122011-2012P-12qsOnpsOntsOffParent and Family Involvement in Education: 20122012P-12qsOnpsOntsOn7National Postsecondary Student Aid Study: 2012 Undergraduates2012PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2012 Graduate Students2012PostsecondaryqsOnpsOntsOn2Early Childhood Program Participation: 20122012P-12qsOnpsOntsOn6Beginning Postsecondary Students: 2012/20172012/2017PostsecondaryqsOnpsOntsOffSchool Survey on Crime and Safety: 2015-162015-2016P-12qsOnpsOntsOn3National Teacher and Principal Survey, 2015-16 Public Schools2015-2016P-12qsOnpsOntsOffNational Teacher and Principal Survey, 2015-16 Public School Principals2015-2016P-12qsOnpsOntsOffParent and Family Involvement in Education: 20162016P-12qsOnpsOntsOn7National Postsecondary Student Aid Study: 2016 Undergraduates2016PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2016 Graduate Students2016PostsecondaryqsOnpsOntsOn2Early Childhood Program Participation: 20162016P-12qsOnpsOntsOn6Adult Training and Education Survey: 20162016Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017Adult EducationqsOnpsOntsOffParent and Family Involvement in Education: 20162016P-12qsOnpsOntsOn7National Postsecondary Student Aid Study: 2016 Undergraduates2016PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2016 Graduate Students2016PostsecondaryqsOnpsOntsOn2Early Childhood Program Participation: 20162016P-12qsOnpsOntsOn6Adult Training and Education Survey: 20162016Adult EducationqsOnpsOntsOffSchool Survey on Crime and Safety: 2015-162015-2016P-12qsOnpsOntsOn3National Teacher and Principal Survey, 2015-16 Public Schools2015-2016P-12qsOnpsOntsOffNational Teacher and Principal Survey, 2015-16 Public School Principals2015-2016P-12qsOnpsOntsOffBeginning Postsecondary Students: 2012/20172012/2017PostsecondaryqsOnpsOntsOffParent and Family Involvement in Education: 20122012P-12qsOnpsOntsOn7National Postsecondary Student Aid Study: 2012 Undergraduates2012PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2012 Graduate Students2012PostsecondaryqsOnpsOntsOn2Early Childhood Program Participation: 20122012P-12qsOnpsOntsOn6Schools and Staffing Survey, Public and Private Teachers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2011-122011-2012P-12qsOnpsOntsOffPrivate School Universe Survey: 2011-122011-2012P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 2009-102009-2010P-12qsOnpsOntsOn3High School Longitudinal Study of 20092009PostsecondaryqsOnpsOntsOffHigh School Longitudinal Study of 20092009P-12qsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012Adult EducationqsOnpsOntsOffNational Postsecondary Student Aid Study: 2008 Undergraduates2008PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2008 Graduate Students2008PostsecondaryqsOnpsOntsOn2Schools and Staffing Survey, Public and Private Teachers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2007-082007-2008P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 2007-082007-2008P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2005-062005-2006P-12qsOnpsOntsOn3Beginning Postsecondary Students: 2004/20092004/2009PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Institution2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Faculty2004PostsecondaryqsOnpsOntsOffNational Postsecondary Student Aid Study: 2004 Undergraduates2004PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2004 Graduate Students2004PostsecondaryqsOnpsOntsOn2Schools and Staffing Survey, Public and Private Teachers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2003-042003-2004P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 2003-042003-2004P-12qsOnpsOntsOffPre-Elementary Education Longitudinal Study, Waves 1-52003/2008P-12qsOnpsOntsOffEducation Longitudinal Study of 20022002PostsecondaryqsOnpsOntsOffEducation Longitudinal Study of 20022002P-12qsOnpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001Adult EducationqsOffpsOntsOffNational Postsecondary Student Aid Study: 2000 Undergraduates2000PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 2000 Graduate Students2000PostsecondaryqsOffpsOntsOn2Schools and Staffing Survey, Public and Private Teachers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private Schools: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Library Media Centers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Districts: 1999-001999-2000P-12qsOffpsOntsOffSchool Survey on Crime and Safety: 1999-20001999-2000P-12qsOnpsOntsOffNational Study of Postsecondary Faculty: 1999 Faculty1999PostsecondaryqsOffpsOntsOffBeginning Postsecondary Students: 1996/20011996/2001PostsecondaryqsOnpsOntsOffNational Postsecondary Student Aid Study: 1996 Undergraduates1996PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 1996 Graduate Students1996PostsecondaryqsOffpsOntsOn2Baccalaureate and Beyond: 1993/2003 Graduate students1993/2003PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003Adult EducationqsOnpsOntsOffNational Study of Postsecondary Faculty: 1993 Faculty1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Undergraduates1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Graduate Students1993PostsecondaryqsOffpsOntsOffBeginning Postsecondary Students: 1990/19941990/1994PostsecondaryqsOnpsOntsOffNational Postsecondary Student Aid Study: 1990 Undergraduates1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Graduate Students1990PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1988 Faculty1988PostsecondaryqsOffpsOntsOffNational Education Longitudinal Study of 19881988PostsecondaryqsOnpsOntsOffNational Education Longitudinal Study of 19881988P-12qsOnpsOntsOffNational Postsecondary Student Aid Study: 1987 Undergraduates1987PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Graduate Students1987PostsecondaryqsOffpsOntsOffHigh School and Beyond1980PostsecondaryqsOnpsOntsOffHigh School and Beyond1980P-12qsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003Adult EducationqsOnpsOntsOffAdult Training and Education Survey: 20162016Adult EducationqsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Library Media Centers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Districts: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 1999-001999-2000P-12qsOffpsOntsOffSchool Survey on Crime and Safety: 2015-162015-2016P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2009-102009-2010P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2007-082007-2008P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2005-062005-2006P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2003-042003-2004P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 1999-20001999-2000P-12qsOnpsOntsOffPrivate School Universe Survey: 2011-122011-2012P-12qsOnpsOntsOffPre-Elementary Education Longitudinal Study, Waves 1-52003/2008P-12qsOnpsOntsOffParent and Family Involvement in Education: 20162016P-12qsOnpsOntsOn7Parent and Family Involvement in Education: 20122012P-12qsOnpsOntsOn7National Teacher and Principal Survey, 2015-16 Public Schools2015-2016P-12qsOnpsOntsOffNational Teacher and Principal Survey, 2015-16 Public School Principals2015-2016P-12qsOnpsOntsOffNational Education Longitudinal Study of 19881988P-12qsOnpsOntsOffHigh School Longitudinal Study of 20092009P-12qsOnpsOntsOffHigh School and Beyond1980P-12qsOnpsOntsOffEducation Longitudinal Study of 20022002P-12qsOnpsOntsOffEarly Childhood Program Participation: 20162016P-12qsOnpsOntsOn6Early Childhood Program Participation: 20122012P-12qsOnpsOntsOn6National Study of Postsecondary Faculty: 2004 Institution2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Faculty2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 1999 Faculty1999PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1993 Faculty1993PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1988 Faculty1988PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 2016 Undergraduates2016PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2016 Graduate Students2016PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2012 Undergraduates2012PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2012 Graduate Students2012PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2008 Undergraduates2008PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2008 Graduate Students2008PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2004 Undergraduates2004PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2004 Graduate Students2004PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2000 Undergraduates2000PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 2000 Graduate Students2000PostsecondaryqsOffpsOntsOn2National Postsecondary Student Aid Study: 1996 Undergraduates1996PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 1996 Graduate Students1996PostsecondaryqsOffpsOntsOn2National Postsecondary Student Aid Study: 1993 Undergraduates1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Graduate Students1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Undergraduates1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Graduate Students1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Undergraduates1987PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Graduate Students1987PostsecondaryqsOffpsOntsOffNational Education Longitudinal Study of 19881988PostsecondaryqsOnpsOntsOffHigh School Longitudinal Study of 20092009PostsecondaryqsOnpsOntsOffHigh School and Beyond1980PostsecondaryqsOnpsOntsOffEducation Longitudinal Study of 20022002PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 2012/20172012/2017PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 2004/20092004/2009PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 1996/20011996/2001PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 1990/19941990/1994PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Institution2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 2004 Faculty2004PostsecondaryqsOnpsOntsOffNational Study of Postsecondary Faculty: 1999 Faculty1999PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1993 Faculty1993PostsecondaryqsOffpsOntsOffNational Study of Postsecondary Faculty: 1988 Faculty1988PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 2016 Undergraduates2016PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2016 Graduate Students2016PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2012 Undergraduates2012PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2012 Graduate Students2012PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2008 Undergraduates2008PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2008 Graduate Students2008PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2004 Undergraduates2004PostsecondaryqsOnpsOntsOn1National Postsecondary Student Aid Study: 2004 Graduate Students2004PostsecondaryqsOnpsOntsOn2National Postsecondary Student Aid Study: 2000 Undergraduates2000PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 2000 Graduate Students2000PostsecondaryqsOffpsOntsOn2National Postsecondary Student Aid Study: 1996 Undergraduates1996PostsecondaryqsOffpsOntsOn1National Postsecondary Student Aid Study: 1996 Graduate Students1996PostsecondaryqsOffpsOntsOn2National Postsecondary Student Aid Study: 1993 Undergraduates1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1993 Graduate Students1993PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Undergraduates1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1990 Graduate Students1990PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Undergraduates1987PostsecondaryqsOffpsOntsOffNational Postsecondary Student Aid Study: 1987 Graduate Students1987PostsecondaryqsOffpsOntsOffNational Education Longitudinal Study of 19881988PostsecondaryqsOnpsOntsOffHigh School Longitudinal Study of 20092009PostsecondaryqsOnpsOntsOffHigh School and Beyond1980PostsecondaryqsOnpsOntsOffEducation Longitudinal Study of 20022002PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 2012/20172012/2017PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 2004/20092004/2009PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 1996/20011996/2001PostsecondaryqsOnpsOntsOffBeginning Postsecondary Students: 1990/19941990/1994PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2016/20172016/2017PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012PostsecondaryqsOnpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003PostsecondaryqsOffpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003PostsecondaryqsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Teachers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private Schools: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Public and Private School Principals: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Library Media Centers: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Library Media Centers: 1999-001999-2000P-12qsOffpsOntsOffSchools and Staffing Survey, Districts: 2011-122011-2012P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2007-082007-2008P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 2003-042003-2004P-12qsOnpsOntsOffSchools and Staffing Survey, Districts: 1999-001999-2000P-12qsOffpsOntsOffSchool Survey on Crime and Safety: 2015-162015-2016P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2009-102009-2010P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2007-082007-2008P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2005-062005-2006P-12qsOnpsOntsOn3School Survey on Crime and Safety: 2003-042003-2004P-12qsOnpsOntsOffSchool Survey on Crime and Safety: 1999-20001999-2000P-12qsOnpsOntsOffPrivate School Universe Survey: 2011-122011-2012P-12qsOnpsOntsOffPre-Elementary Education Longitudinal Study, Waves 1-52003/2008P-12qsOnpsOntsOffParent and Family Involvement in Education: 20162016P-12qsOnpsOntsOn7Parent and Family Involvement in Education: 20122012P-12qsOnpsOntsOn7National Teacher and Principal Survey, 2015-16 Public Schools2015-2016P-12qsOnpsOntsOffNational Teacher and Principal Survey, 2015-16 Public School Principals2015-2016P-12qsOnpsOntsOffNational Education Longitudinal Study of 19881988P-12qsOnpsOntsOffHigh School Longitudinal Study of 20092009P-12qsOnpsOntsOffHigh School and Beyond1980P-12qsOnpsOntsOffEducation Longitudinal Study of 20022002P-12qsOnpsOntsOffEarly Childhood Program Participation: 20162016P-12qsOnpsOntsOn6Early Childhood Program Participation: 20122012P-12qsOnpsOntsOn6Baccalaureate and Beyond: 2016/20172016/2017Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2008/20122008/2012Adult EducationqsOnpsOntsOffBaccalaureate and Beyond: 2000/20012000/2001Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 1993/2003 Graduate students1993/2003Adult EducationqsOffpsOntsOffBaccalaureate and Beyond: 1993/20031993/2003Adult EducationqsOnpsOntsOffAdult Training and Education Survey: 20162016Adult EducationqsOnpsOntsOff1 Percentage distribution of 1995–96 beginning postsecondary students' highest degree attained by 2001, by work status Highest degree completed as of June 2001 Certificate(%) Associate(%) Bachelor(%) Never attained(%) Total Estimates Total 11.7 9.8 29.8 48.6 100% Job 1995–96: hours worked per week while enrolled Did not work while enrolled 14.0 9.8 38.5 37.8 100% Worked part time 8.9 11.6 35.5 44.0 100% Worked full time 14.5 7.2 8.3 69.9 100% NOTE: Rows may not add up to 100% due to rounding.SOURCE: U.S. Department of Education, National Center for Education Statistics, 1995–96 Beginning Postsecondary Students Longitudinal Study, Second Follow-up (BPS:96/01). Computation by NCES QuickStats on 6/22/2009 ckeakb72 Percentage distribution of 1995–96 beginning postsecondary students' highest degree attained by 2001, by number of advanced placement tests taken Persistence and completion at any institution as of 2000-01 Never attained(%) Certificate(%) Associate(%) Bachelor(%) Total Estimates Total 48.6 11.7 9.9 29.8 100% Number of Advanced Placement tests taken 0 51.1 7.7 12.1 29.1 100% 1 38.1 2.6 6.0 53.4 100% 2 33.6 0.4 3.4 62.6 100% Three or more 13.8 0.1 1.4 84.8 100% NOTE: Rows may not add up to 100% due to rounding.SOURCE: U.S. Department of Education, National Center for Education Statistics, 1995–96 Beginning Postsecondary Students Longitudinal Study, Second Follow-up (BPS:96/01). Computation by NCES QuickStats on 6/22/2009 ckeak193 Percentage of beginning postsecondary students who received Pell grants, by race/ethnicity: 1995–96 Pell Grant amount 1995-96(%>0) Estimates Total 26.4 Race/ethnicity White, non-Hispanic 19.0 Black, non-Hispanic 49.3 Hispanic 42.4 Asian/Pacific Islander 35.5 American Indian/Alaska Native 33.2 Other ‡ ‡ Reporting standards not met. Source: U.S. Department of Education, National Center for Education Statistics, 1995–96 Beginning Postsecondary Students Longitudinal Study, Second Follow-up (BPS:96/01). Computation by NCES QuickStats on 3/10/2009 cgfak7e4 Percentage distribution of 1995–96 beginning postsecondary students' grade point average (GPA) through 2001, by income percentile rank Cumulative Grade Point Average (GPA) as of 2001 Mostly A’s (%) A’s and B’s (%) Mostly B’s (%) B’s and C’s (%) Mostly C’s (%) C’s and D’s (%) Mostly D’s or below (%) Total Estimates Total 13.3 31.8 35.3 14.4 4.4 0.7 0.1 100% Income percentile rank 1994 1-25 13.1 28.2 37.8 14.7 4.7 1.4 0.2 100% 26-50 13.5 30.2 37.3 12.8 5.8 0.3 0.2 100% 51-75 12.9 36.1 33.1 14.0 3.4 0.4 0.. 100% More than 75 13.7 32.7 33.0 16.3 3.7 0.7 0.0 100% NOTE: Rows may not add up to 100% due to rounding.SOURCE: U.S. Department of Education, National Center for Education Statistics, 1995–96 Beginning Postsecondary Students Longitudinal Study, Second Follow-up (BPS:96/01). Computation by NCES QuickStats on 6/22/2009 ckeak03 5 Percentage distribution of 1995–96 beginning postsecondary students' persistence at any institution through 2001, by gender Persistence at any institution through 2001 Attained, still enrolled(%) Attained, not enrolled(%) Never attained, still enrolled(%) Never attained, not enrolled(%) Total Estimates Total 5.9 45.5 14.9 33.7 100% Gender Male 5.9 41.8 15.8 36.5 100% Female 5.8 48.5 14.2 31.5 100% NOTE: Rows may not add up to 100% due to rounding.SOURCE: U.S. Department of Education, National Center for Education Statistics, 1995–96 Beginning Postsecondary Students Longitudinal Study, Second Follow-up (BPS:96/01). Computation by NCES QuickStats on 6/22/2009 cgeakd41 Percent of graduate students who borrowed, by type of graduate program: 2003–04 Loans: total student loans all sources(%>0) Estimates Total 40.0 Graduate study: program Business administration (MBA) 39.1 Education (any master's) 34.8 Other master of arts (MA) 41.3 Other master of science (MS) 31.8 Other master's degree 49.3 PhD except in education 19.9 Education (any doctorate) 27.1 Other doctoral degree 49.5 Medicine (MD) 77.3 Other health science degree 81.7 Law (LLB or JD) 81.0 Theology (MDiv, MHL, BD) 30.0 Post-baccalaureate certificate 30.1 Not in a degree program 28.0 SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 bbfak2a2 Percentage of graduate students with assistantships, by graduate field of study: 2003–04 Assistantships(%>0) Estimates Total 15.3 Graduate study: major field Humanities 20.8 Social/behavioral sciences 31.7 Life sciences 47.4 Math/Engineering/Computer science 37.9 Education 7.6 Business/management 7.9 Health 10.3 Law 5.8 Others 23.8 Undeclared or not in a degree program 5.4 SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 ckeak393 Percentage distribution of graduates students' student/employee role, by graduate field of study: 2003–04 Work: primarily student or employee A student working to meet expenses(%) An employee enrolled in school(%) No job(%) Total Estimates Total 35.8 45.1 19.1 100% Graduate study: major field Humanities 44.9 35.9 19.2 100% Social/behavioral sciences 58.9 24.6 16.5 100% Life sciences 61.0 20.7 18.3 100% Math/Engineering/Computer science 47.4 38.3 14.3 100% Education 26.3 63.3 10.4 100% Business/management 24.8 61.8 13.3 100% Health 39.4 19.0 41.6 100% Law 39.6 11.6 48.8 100% Others 47.0 38.5 14.5 100% Undeclared or not in a degree program 20.5 67.3 12.2 100% NOTE: Rows may not add up to 100% due to rounding. SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 ckeakce4 Percentage of graduate students who have ever borrowed loans, by institution type: 2003–04 Total loan debt (cumulative)(%>0) Estimates Total 65.2 Type of 4-year institution Public 4-year nondoctorate 61.4 Public 4-year doctorate 60.6 Private not-for-profit 4-yr nondoctorate 61.6 Private not-for-profit 4-year doctorate 71.3 Private for-profit 4-year 85.9 Attended more than one institution 68.9 SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 ckeak5f5 Average loan amount for graduate students, by parents' education, 2003–04 Loans: total student loans all sources(Mean[0]) Estimates Total 6,302.0 Parent's highest education Do not know parent's education level 7,677.5 High school diploma or less 5,878.7 Some college 6,016.3 Bachelor's degree 5,794.3 Master's degree or higher 7,185.9 SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 ckeakef1 Percentage of undergraduate students who applied for aid, by parents' income: 2003–04 Aid: applied for federal aid Yes(%) No(%) Total Estimates Total 58.3 41.7 100% Income: dependent student household income Less than $32,000 78.7 21.3 100% $32,000-59,999 66.6 33.4 100% $60,000-91,999 56.9 43.1 100% $92,000 or more 47.1 52.9 100% NOTE: Rows may not add up to 100% due to rounding. SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 cgeak8c2 Percentage distribution of undergraduates' cumulative grade point average (GPA) categories, by major field of study: 2003–04 Cumulative Grade Point Average (GPA) as of 2003-04 Less than 2.75(%) 2.75 - 3.74(%) 3.75 or higher(%) Total Estimates Total 34.4 49.0 16.7 100% College study: major Humanities 35.9 50.4 13.6 100% Social/behavioral sciences 35.0 52.1 12.8 100% Life sciences 34.9 52.7 12.4 100% Physical sciences 31.5 54.3 14.2 100% Math 29.1 55.3 15.6 100% Computer/information science 34.0 48.1 17.9 100% Engineering 37.4 48.1 14.5 100% Education 31.9 52.6 15.5 100% Business/management 35.6 49.3 15.1 100% Health 32.2 50.7 17.0 100% Vocational/technical 33.3 47.1 19.6 100% Other technical/professional 36.7 49.9 13.4 100% Undeclared or not in a degree program 33.2 44.1 22.8 100% NOTE: Rows may not add up to 100% due to rounding. SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 cgeake93 Mean net price of attendance for undergraduate students, by type of institution: 2003–04 Net price after all aid(Mean[0]) Estimates Total 6,656.0 Institution: type Public less-than-2-year 5,616.5 Public 2-year 4,716.3 Public 4-year nondoctorate 6,253.5 Public 4-year doctorate 7,564.1 Private not-for-profit less-than-4-year 7,382.3 Private not-for-profit 4-yr nondoctorate 9,208.7 Private not-for-profit 4-year doctorate 14,812.2 Private for-profit less-than-2-year 7,842.9 Private for-profit 2 years or more 6,737.6 Attended more than one institution ‡ ‡ Reporting standards not met. SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 bcfak0c4 Percentage distribution of undergraduates' parents' highest level of education, by type of institution: 2003–04 Parent's highest education High school or less(%) Some college(%) Bachelor's degree or higher(%) Total Estimates Total 37.1 21.6 41.3 100% Institution: type Public less-than-2-year 54.2 17.4 28.4 100% Public 2-year 43.3 23.9 32.7 100% Public 4-year nondoctorate 28.7 20.5 50.8 100% Public 4-year doctorate 46.9 18.8 34.2 100% Private not-for-profit less than 4-year 29.6 18.1 52.3 100% Private not-for-profit 4-year nondoctorate 55.6 17.4 27.0 100% Private not-for-profit 4-year doctorate 53.8 20.2 25.9 100% NOTE: Rows may not add up to 100% due to rounding.SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 10/14/2009 cgeakd55 Average amount of Pell grants received by undergraduates, by income and dependency status: 2003–04 Grants: Pell Grants(Avg>0) Estimates Total 2,449.7 Income: categories by dependency status Dependent: Less than $10,000 3,242.2 Dependent: $10,000-$19,999 3,176.1 Dependent: $20,000-$29,999 2,715.0 Dependent: $30,000-$39,999 1,958.3 Dependent: $40,000-$49,999 1,508.6 Dependent: $50,000-$59,999 1,309.0 Dependent: $60,000-$69,999 1,241.7 Dependent: $70,000-$79,999 1,404.4 Dependent: $80,000-$99,999 ‡ Dependent: $100,000 or more ‡ Independent: Less than $5,000 2,860.3 Independent: $5,000-$9,999 2,642.9 Independent: $10,000-$19,999 2,291.7 Independent: $20,000-$29,999 2,328.3 Independent: $30,000-$49,999 1,561.9 Independent: $50,000 or more 1,124.3 ‡ Reporting standards not met. SOURCE: U.S. Department of Education, National Center for Education Statistics, 2003–04 National Postsecondary Student Aid Study (NPSAS:04). NOTICE OF REVISIONS: The NPSAS:04 weights were revised in June 2009. The revised weights will produce 2003-04 estimates that differ somewhat from those in any tables and publications produced before June 2009. See the description for the total Stafford loan variable (STAFFAMT) for details. Computation by NCES QuickStats on 8/25/2009 cgeak381 Percentage distribution of instructional faculty and staff's employment status, by institution type, Fall 2003 Employment status at this job Full time(%) Part time(%) Total Estimates Total 56.3 43.7 100% Institution: type and control Public doctoral 77.8 22.2 100% Private not-for-profit doctoral 68.7 31.3 100% Public master's 63.3 36.7 100% Private not-for-profit master's 45.0 55.0 100% Private not-for-profit baccalaureate 63.2 36.8 100% Public associates 33.3 66.7 100% Other 49.2 50.8 100% NOTE: Rows may not add up to 100% due to rounding. SOURCE: U.S. Department of Education, National Center for Education Statistics, 2004 National Study of Postsecondary Faculty (NSOPF:04). Computation by NCES QuickStats on 6/19/2009 ckeak012 Percentage distribution of full-time instructional faculty and staff, by race/ethnicity, institution type: Fall 2003 Race/ethnicity White, non-Hispanic(%) Black, non-Hispanic(%) Asian/Pacific Islander(%) Hispanic(%) Other(%) Estimates Total 80.3 5.9 8.6 3.4 1.2 Institution: type and control Public doctoral 79.4 4.5 12.0 3.0 1.0 Private not-for-profit doctoral 79.1 5.3 11.9 2.9 0.8 Public master’s 78.3 8.9 7.6 3.6 1.6 Private not-for-profit master’s 85.4 5.1 5.7 2.5 1.3 Private not-for-profit baccalaureate 85.8 6.8 4.2 2.2 1.1 Public associates 81.2 7.2 4.4 5.5 1.7 Other 86.9 4.6 5.8 1.7 1.0 NOTE: Rows may not add up to 100% due to rounding.SOURCE: U.S. Department of Education, National Center for Education Statistics, 2004 National Study of Postsecondary Faculty (NSOPF:04). Computation by NCES QuickStats on 6/19/2009 ckeak043 Percentage distribution of full-time instructional faculty and staff, by tenure status, institution type: Fall 2003 Tenure status Tenured(%) On tenure track but not tenured(%) Not on tenure track(%) Not tenured-no tenure system(%) Total Estimates Total 49.3 21.3 20.9 8.5 100% Institution: type and control Public doctoral 53.0 20.4 25.9 0.7 100% Private not-for-profit doctoral 47.1 19.6 28.8 4.5 100% Public master’s 53.7 28.3 16.9 1.0 100% Private not-for-profit master’s 41.9 28.1 21.5 8.6 100% Private not-for-profit baccalaureate 42.9 25.1 21.6 10.4 100% Public associates 49.1 15.6 9.3 26.0 100% Other 39.4 17.3 18.7 24.6 100% NOTE: Rows may not add up to 100% due to rounding.SOURCE: U.S. Department of Education, National Center for Education Statistics, 2004 National Study of Postsecondary Faculty (NSOPF:04). Computation by NCES QuickStats on 6/19/2009 ckeak984 Percentage distribution of part-time instructional faculty and staff, by academic rank, institution type: Fall 2003 Academic rank Professor(%) Associate professor(%) Assistant professor(%) Instructor or lecturer(%) Other ranks/not applicable(%) Estimates Total 4.6 2.9 3.4 42.2 46.9 Institution: type and control Public doctoral 6.3 4.5 8.1 45.0 36.0 Private not-for-profit doctoral 5.6 4.9 9.1 31.9 48.5 Public master’s 6.4 2.3 2.0 40.7 48.7 Private not-for-profit master’s 2.7 3.4 2.6 30.3 60.9 Private not-for-profit baccalaureate 4.6 4.2 5.4 32.5 53.3 Public associates 3.4 1.5 1.0 49.5 44.6 Other 7.1 4.9 5.2 33.3 49.4 NOTE: Rows may not add up to 100% due to rounding.SOURCE: U.S. Department of Education, National Center for Education Statistics, 2004 National Study of Postsecondary Faculty (NSOPF:04). Computation by NCES QuickStats on 6/19/2009 ckeakff5 Average hours worked per week among full-time instructional faculty and staff, by tenure status: Fall 2003 Hours worked per week(Mean>0) Estimates Total 47.4 Tenure status Tenured 53.3 On tenure track but not tenured 53.7 Not on tenure track 43.0 Not tenured-no tenure system 45.4 SOURCE: U.S. Department of Education, National Center for Education Statistics, 2004 National Study of Postsecondary Faculty (NSOPF:04). Computation by NCES QuickStats on 6/19/2009 bkfakf31 Percentage of institutions with full- or part-time faculty represented by a union, by institution type: Fall 2003 Faculty represented by a union Not represented by a union(%) Represented by a union(%) Total Estimates Total 68.1 31.9 100% Institution: type and control
ImputationImputation was conducted for selected items on the teacher questionnaire and parent interview data. In general, the item missing rate was low. The risk of imputation-related bias was judged to be minimal. The variance inflation due to imputation was also low due to the low imputation rate of 10 percent. Imputation for the supplemental sample increased the amount of data usable for analysis, offsetting the potential risk of bias.The methods of imputation included: hot-deck imputation, regression, external data source, and a derivation method, based on the internal consistency of inter-related variables.
Imputation
Imputation was conducted for selected items on the teacher questionnaire and parent interview data. In general, the item missing rate was low. The risk of imputation-related bias was judged to be minimal. The variance inflation due to imputation was also low due to the low imputation rate of 10 percent. Imputation for the supplemental sample increased the amount of data usable for analysis, offsetting the potential risk of bias.
The methods of imputation included: hot-deck imputation, regression, external data source, and a derivation method, based on the internal consistency of inter-related variables.
WeightingThe final weights are needed to have the estimates reflect the population of private schools when analyzing the data. The data from the area frame component were weighted to reflect the sampling rates (probability of selection) of the PSUs. Survey data from both the list and area frame components were adjusted for school nonresponse. The final weight for PSS data items is the product of the Base Weight and the Nonresponse Adjustment Factor, where:Base Weight is the inverse of the probability of selection of the school. The base weight is equal to one for all list-frame schools. For area-frame schools, the base weight is equal to the inverse of the probability of selecting the PSU in which the school resides.Nonresponse Adjustment Factor is an adjustment that accounts for school nonresponse. It is the weighted (base weight) ratio of the total eligible in-scope schools (interviewed schools plus noninterviewed schools) to the total responding in-scope schools (interviewed schools) within cells. Noninterviewed and out-of-scope cases are assigned a nonresponse adjustment factor of zero.Because we have more information for list-frame schools, the cells used to compute the nonresponse adjustment were defined differently for list-frame and area-frame schools. For schools in the list frame, the cells were defined by affiliation (17 categories), locale type (4 categories), grade level (4 categories), Census region (4 categories), and enrollment (3 categories). The nonresponse adjustment cells for area frame schools were defined by three-level typology (3 categories) and grade level (4 categories). If the number of schools in a cell was fewer than 15 or the nonresponse adjustment factor was greater than 1.5, then that cell was collapsed into a similar cell. The variables used to collapse the cells and the collapse order varied according to whether the school was from the list or area frame and whether a school was a traditional or k-terminal school. The cells for traditional schools from the list frame were collapsed within enrollment category, locale type, grade level, and Census region. Cells for k-terminal schools from the list frame were collapsed within enrollment category, locale type, Census region, and affiliation. Cells for traditional schools from the area frame were collapsed within grade level and then within three-level typology. Cells for k-terminal schools from the area frame were collapsed within three level typology.ImputationAfter the data edit processing was complete, there were missing values within some records classified as interviews. These were cases where the respondent had not answered some applicable questionnaire items (and data for these items were not added in the pre-edit, consistency, or logic edit) or the response had been deleted during editing. Values were imputed to the missing data during imputation. Two types of imputation were employed: donor and analyst imputation.Donor ImputationIn donor imputation, values were created by extracting data from the record for a sample case (donor) with similar characteristics, using a procedure known as the “sequential nearest neighbor hot deck” (Kalton and Kasprzyk 1982, 1986; Kalton 1983; Little and Rubin 1987; Madow, Olkin, and Rubin 1983). In order to match incomplete records to those with complete data, “imputation” variables that identify certain characteristics of the school that were deemed to be important to the reporting of the data in each item (e.g., religious affiliation, enrollment, school level of instruction) were used. Items were grouped according to the perceived relevance of the imputation variables to the data collected by the item. For example, school level of instruction was used for matching incomplete records and donors to fill item 16 (length of school year) but was not used for item 7 (students by race).Analyst ImputationAfter the donor imputation was completed, there were records that still had missing values for 64 items. These were cases where the imputation failed to create a value because there was no suitable record to use as a donor, or the value imputed was deleted because it was outside the acceptable range for the item or was inconsistent with other data on the same record, or the religious orientation or purpose, or the religious orientation or affiliation, was not reported (items 14a and 14c) and no previous PSS information was available.For these cases, values were imputed by analysts to the items with missing data. That is, staff reviewed the data record, sample file record, and the questionnaire and identified a value consistent with the information from these sources for imputation.
Weighting
Because we have more information for list-frame schools, the cells used to compute the nonresponse adjustment were defined differently for list-frame and area-frame schools. For schools in the list frame, the cells were defined by affiliation (17 categories), locale type (4 categories), grade level (4 categories), Census region (4 categories), and enrollment (3 categories). The nonresponse adjustment cells for area frame schools were defined by three-level typology (3 categories) and grade level (4 categories). If the number of schools in a cell was fewer than 15 or the nonresponse adjustment factor was greater than 1.5, then that cell was collapsed into a similar cell. The variables used to collapse the cells and the collapse order varied according to whether the school was from the list or area frame and whether a school was a traditional or k-terminal school. The cells for traditional schools from the list frame were collapsed within enrollment category, locale type, grade level, and Census region. Cells for k-terminal schools from the list frame were collapsed within enrollment category, locale type, Census region, and affiliation. Cells for traditional schools from the area frame were collapsed within grade level and then within three-level typology. Cells for k-terminal schools from the area frame were collapsed within three level typology.
After the data edit processing was complete, there were missing values within some records classified as interviews. These were cases where the respondent had not answered some applicable questionnaire items (and data for these items were not added in the pre-edit, consistency, or logic edit) or the response had been deleted during editing. Values were imputed to the missing data during imputation. Two types of imputation were employed: donor and analyst imputation.
Donor Imputation
In donor imputation, values were created by extracting data from the record for a sample case (donor) with similar characteristics, using a procedure known as the “sequential nearest neighbor hot deck” (Kalton and Kasprzyk 1982, 1986; Kalton 1983; Little and Rubin 1987; Madow, Olkin, and Rubin 1983). In order to match incomplete records to those with complete data, “imputation” variables that identify certain characteristics of the school that were deemed to be important to the reporting of the data in each item (e.g., religious affiliation, enrollment, school level of instruction) were used. Items were grouped according to the perceived relevance of the imputation variables to the data collected by the item. For example, school level of instruction was used for matching incomplete records and donors to fill item 16 (length of school year) but was not used for item 7 (students by race).
Analyst Imputation
After the donor imputation was completed, there were records that still had missing values for 64 items. These were cases where the imputation failed to create a value because there was no suitable record to use as a donor, or the value imputed was deleted because it was outside the acceptable range for the item or was inconsistent with other data on the same record, or the religious orientation or purpose, or the religious orientation or affiliation, was not reported (items 14a and 14c) and no previous PSS information was available.
For these cases, values were imputed by analysts to the items with missing data. That is, staff reviewed the data record, sample file record, and the questionnaire and identified a value consistent with the information from these sources for imputation.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables. 2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.
Perturbation
To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.
Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible. After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation.
Skips and Missing Values
Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information.
1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.
2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.
PerturbationTo protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.ImputationThree types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation. Skips and Missing ValuesFollowing data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Please consult the survey methodology for more information. 1Donors were selected based on the type of data the donor would supply to the record undergoing imputation. Matching variables were selected based on their close relationship to the item requiring imputation, and a pool of donors was selected based on their answers to these matching variables.2Goldring, R., Taie, S., Rizzo, L., Colby, D., and Fraser, A. (2013). User’s Manual for the 2011–12 Schools and Staffing Survey, Volume 1: Overview (NCES 2013-330). U.S. Department of Education. Washington, DC: National Center for Education Statistics.
Three types of edits were performed on the SASS data: blanking, consistency, and logic edits. Blanking edits delete extraneous entries that result from respondents failing to follow skip patterns correctly and assign “missing” codes to items that respondents should have answered and didn’t. Consistency edits ensured that responses to related items were consistent and did not contradict other survey data. Finally, logic edits were performed, using information collected from the same questionnaire, associated questionnaires in the same school or district, or information from the sampling frame to fill missing items, where possible.After blanking, consistency, and logic edits were completed, any missing items that remained were filled using imputation. Data were imputed from items found on questionnaires of the same type that had certain characteristics in common or from the aggregated answers of similar questionnaires. These records are called “donor records1,” and the method of imputation that involves imputing data from donor records is called “hot-deck2” imputation.
ImputationCompleted SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond: In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical. WeightingData are weighted to compensate for differential probabilities of selection and to adjust for the effects of nonresponse.Sample weights allow inferences to be made about the population from which the sample units are drawn. Because of the complex nature of the SSOCS sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Due to nonresponse, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias due to nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The same predictor variables from the SSOCS:2004 CHAID analysis were used for SSOCS:2006: instructional level, region, enrollment size, percent minority, student-to-FTE teaching staff ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time equivalent (FTE) teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted so that the weighted distribution of the responding schools resembled the initial distribution of the total sample. The nonresponse-adjusted weights were then poststratified to calibrate the sample to known population totals. Two dimension margins were set up for the poststratification—(1) instructional level and school enrollment size; and (2) instructional level and locale—and an iterative process known as the raking ratio adjustment brought the weights into agreement with known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. All three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2004). Miller, A.K. (2004). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.
Completed SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond: In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical.
Data are weighted to compensate for differential probabilities of selection and to adjust for the effects of nonresponse.Sample weights allow inferences to be made about the population from which the sample units are drawn. Because of the complex nature of the SSOCS sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Due to nonresponse, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias due to nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The same predictor variables from the SSOCS:2004 CHAID analysis were used for SSOCS:2006: instructional level, region, enrollment size, percent minority, student-to-FTE teaching staff ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time equivalent (FTE) teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted so that the weighted distribution of the responding schools resembled the initial distribution of the total sample. The nonresponse-adjusted weights were then poststratified to calibrate the sample to known population totals. Two dimension margins were set up for the poststratification—(1) instructional level and school enrollment size; and (2) instructional level and locale—and an iterative process known as the raking ratio adjustment brought the weights into agreement with known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. All three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2004). Miller, A.K. (2004). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.
ImputationCompleted SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond. In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical.WeightingSample weights allow inferences to be made about the population from which the sample units were drawn. Because of the complex nature of the SSOCS:2004 sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. The procedures used to create the SSOCS sampling weights are described below.An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Because some schools refused to participate, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias from nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (i.e., chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The predictor variables for the analysis were instructional level, region, enrollment size, percent minority, student-to-teacher ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time-equivalent teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted by dividing the base weight by the response rate in each class, so that the weighted distribution of the responding schools resembled the initial distribution of the total sample.The non-response-adjusted weights were then poststratified to calibrate the sample to known population totals. For SSOCS:2004, two dimension margins were set up for the poststratification: (1) instructional level and school enrollment size, and (2) instructional level and locale. An iterative process known as the raking ratio adjustment brought the weights into agreement with the known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. Similar to SSOCS:2000, all three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2003).Miller, A.K. (2003). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.
Completed SSOCS surveys contain some level of item nonresponse after the conclusion of the data collection phase. Imputation procedures were used to impute missing values of key items in SSOCS:2000 and missing values of all items in each subsequent SSOCS. All imputed values are flagged as such.SSOCS:2004 and Beyond. In subsequent collections, imputation procedures were used to create values for all questionnaire items with missing data. This procedural change from SSOCS:2000 was implemented because the analysis of incomplete datasets may cause different users to arrive at different conclusions, depending on how the missing data are treated. The imputation methods used in SSOCS:2004 and later surveys were tailored to the nature of each survey item. Four methods were used: aggregate proportions, logical, best match, and clerical.
Sample weights allow inferences to be made about the population from which the sample units were drawn. Because of the complex nature of the SSOCS:2004 sample design, these weights are necessary to obtain population-based estimates, to minimize bias arising from differences between responding and nonresponding schools, and to calibrate the data to known population characteristics in a way that reduces sampling error. The procedures used to create the SSOCS sampling weights are described below.An initial (base) weight was first determined within each stratum by calculating the ratio of the number of schools available in the sampling frame to the number of schools selected. Because some schools refused to participate, the responding schools did not necessarily constitute a random sample from the schools in the stratum. In order to reduce the potential of bias from nonresponse, weighting classes were determined by using a statistical algorithm similar to CHAID (i.e., chi-square automatic interaction detector) to partition the sample such that schools within a weighting class were homogenous with respect to their probability of responding. The predictor variables for the analysis were instructional level, region, enrollment size, percent minority, student-to-teacher ratio, percentage of students eligible for free or reduced-price lunch, and number of full-time-equivalent teachers. When the number of responding schools in a class was sufficiently small, the weighting class was combined with another to avoid the possibility of large weights. After combining the necessary classes, the base weights were adjusted by dividing the base weight by the response rate in each class, so that the weighted distribution of the responding schools resembled the initial distribution of the total sample.The non-response-adjusted weights were then poststratified to calibrate the sample to known population totals. For SSOCS:2004, two dimension margins were set up for the poststratification: (1) instructional level and school enrollment size, and (2) instructional level and locale. An iterative process known as the raking ratio adjustment brought the weights into agreement with the known control totals. Poststratification works well when the population not covered by the survey is similar to the covered population within each poststratum. Thus, to be effective, the variables that define the poststrata must be correlated with the variables of interest, they must be well measured in the survey, and control totals must be available for the population as a whole. Similar to SSOCS:2000, all three requirements were satisfied by the aforementioned poststratification margins. Instructional level, school enrollment, and locale have been shown to be correlated with crime (Miller 2003).Miller, A.K. (2003). Violence in U.S. Public Schools: 2000 School Survey on Crime and Safety (NCES 2004-314R). National Center for Education Statistics, Institute of Education Sciences, U.S. Department of Education. Washington, DC.
ImputationAll key data items with missing values were imputed using well-known procedures. Depending on the type of data to be imputed and the extent of missing values, logical imputation, poststratum means, or "hot-deck" imputation methods were employed. For three data items, imputation was done using information from the 1998-99 CCD file. Logical imputation is the assignment of data values based on other information in the data record. In the poststratum means method, a record with missing data was assigned the mean value of those cases in the same "poststratum" for which information on the item was available. The poststrata or "imputation classes" were defined on the basis of variables that were correlated with the item being imputed. Preliminary exploratory analyses (e.g., using chi-square tests of association, correlation analysis, and regression analysis) were carried out to identify the relevant classification variables. The strength of association of the variables in combination with subjective assessment was used to prioritize the importance of the variables in forming the imputation classes. In the "hot-deck" technique, cases with missing items were assigned the corresponding value of a "similar" respondent in the same "poststratum". Similar to the poststratum means approach, preliminary exploratory analyses were carried out to identify the relevant classification variables to be used to define the poststrata. The classification variables were separated into two groups -- "hard" and "soft" boundary variables. The hard boundary variables were considered to be so important that the imputation classes were always formed within those boundaries. The boundaries formed by the soft boundary variables were crossed, if necessary, to form the imputation class.WeightingA stratified random sample design was used to select schools for the SSOCS:2000. Over 3,000 schools were selected at rates that varied by sampling stratum; i.e., the classes formed by crossing instructional level (elementary, middle, secondary, combined), type of locale (city, urban fringe, town, rural), and enrollment size class (less than 300, 300-499, 500-999, 1,000+). Since the schools were selected with unequal probabilities, sampling weights are required for analysis to inflate the survey responses to population levels. Weighting is also used to reduce the potential bias resulting from nonresponse and possible undercoverage of the sampling frame.One method of computing sampling errors to reflect various aspects of the sample design and estimation procedures is the replication method. Under replication methods, a specified number of subsamples of the full sample (called "replicates") are created. The survey estimates can then be computed for each of the replicates by creating replicate weights that mimic the actual sample design and estimation procedures used in the full sample. The variability of the estimates computed from the replicate weights is then used to estimate the sampling errors of the estimates from the full sample. An important advantage of the replication methods is that they preclude the need to specify cumbersome variance formulas that are typically needed for complex sample designs (McCarthy, 1966).1 Another advantage is that they can readily be adapted to reflect the variance resulting from nonresponse (and other weight) adjustment procedures. The two most prevalent replication methods are balanced repeated replication (BRR) and jackknife replication. The two methods differ in the manner in which the replicates are constructed. For the SSOCS:2000, a variant of jackknife replication was used to develop replicate weights for variance estimation because the jackknife method is believed to perform somewhat better than BRR for estimates of moderately rare events (e.g., number of schools in which a serious crime was committed). Under the jackknife method, the replicates are formed by deleting specified subsets of units from the full sample. The jackknife method provides a relatively simple way of creating the replicates for variance estimation and has been used extensively in NCES surveys.1. McCarthy, P. (1966). Replication: An Approach to the Analysis of Data from Complex Surveys. Vital and Health Statistics, Series 2, No. 14. Washington, DC: U.S. Department of Health, Education and Welfare.
All key data items with missing values were imputed using well-known procedures. Depending on the type of data to be imputed and the extent of missing values, logical imputation, poststratum means, or "hot-deck" imputation methods were employed. For three data items, imputation was done using information from the 1998-99 CCD file. Logical imputation is the assignment of data values based on other information in the data record. In the poststratum means method, a record with missing data was assigned the mean value of those cases in the same "poststratum" for which information on the item was available. The poststrata or "imputation classes" were defined on the basis of variables that were correlated with the item being imputed. Preliminary exploratory analyses (e.g., using chi-square tests of association, correlation analysis, and regression analysis) were carried out to identify the relevant classification variables. The strength of association of the variables in combination with subjective assessment was used to prioritize the importance of the variables in forming the imputation classes. In the "hot-deck" technique, cases with missing items were assigned the corresponding value of a "similar" respondent in the same "poststratum". Similar to the poststratum means approach, preliminary exploratory analyses were carried out to identify the relevant classification variables to be used to define the poststrata. The classification variables were separated into two groups -- "hard" and "soft" boundary variables. The hard boundary variables were considered to be so important that the imputation classes were always formed within those boundaries. The boundaries formed by the soft boundary variables were crossed, if necessary, to form the imputation class.
A stratified random sample design was used to select schools for the SSOCS:2000. Over 3,000 schools were selected at rates that varied by sampling stratum; i.e., the classes formed by crossing instructional level (elementary, middle, secondary, combined), type of locale (city, urban fringe, town, rural), and enrollment size class (less than 300, 300-499, 500-999, 1,000+). Since the schools were selected with unequal probabilities, sampling weights are required for analysis to inflate the survey responses to population levels. Weighting is also used to reduce the potential bias resulting from nonresponse and possible undercoverage of the sampling frame.One method of computing sampling errors to reflect various aspects of the sample design and estimation procedures is the replication method. Under replication methods, a specified number of subsamples of the full sample (called "replicates") are created. The survey estimates can then be computed for each of the replicates by creating replicate weights that mimic the actual sample design and estimation procedures used in the full sample. The variability of the estimates computed from the replicate weights is then used to estimate the sampling errors of the estimates from the full sample. An important advantage of the replication methods is that they preclude the need to specify cumbersome variance formulas that are typically needed for complex sample designs (McCarthy, 1966).1 Another advantage is that they can readily be adapted to reflect the variance resulting from nonresponse (and other weight) adjustment procedures. The two most prevalent replication methods are balanced repeated replication (BRR) and jackknife replication. The two methods differ in the manner in which the replicates are constructed. For the SSOCS:2000, a variant of jackknife replication was used to develop replicate weights for variance estimation because the jackknife method is believed to perform somewhat better than BRR for estimates of moderately rare events (e.g., number of schools in which a serious crime was committed). Under the jackknife method, the replicates are formed by deleting specified subsets of units from the full sample. The jackknife method provides a relatively simple way of creating the replicates for variance estimation and has been used extensively in NCES surveys.
1. McCarthy, P. (1966). Replication: An Approach to the Analysis of Data from Complex Surveys. Vital and Health Statistics, Series 2, No. 14. Washington, DC: U.S. Department of Health, Education and Welfare.
ImputationStochastic methods were used to impute the missing values for the ELS:2002 third follow-up data. Specifically, a weighted sequential hot-deck (WSHD) statistical imputation procedure (Cox 1980; Iannacchione 1982) using the final analysis weight (F3QWT) was applied to the missing values for the variables in table 12 in the order in which they are listed. The WSHD procedure replaces missing data with valid data from a donor record within an imputation class. In general, variables with lower item nonresponse rates were imputed earlier in the process.
Stochastic methods were used to impute the missing values for the ELS:2002 third follow-up data. Specifically, a weighted sequential hot-deck (WSHD) statistical imputation procedure (Cox 1980; Iannacchione 1982) using the final analysis weight (F3QWT) was applied to the missing values for the variables in table 12 in the order in which they are listed. The WSHD procedure replaces missing data with valid data from a donor record within an imputation class. In general, variables with lower item nonresponse rates were imputed earlier in the process.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, HSLS:09 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. ImputationStochastic methods were used to impute the missing values. Specifically, a weighted sequential hot-deck (WSHD; statistical) imputation procedure (Cox 1980; Iannacchione 1982) using the final student analysis weight (W2STUDENT) was applied to the missing values for variables. The WSHD procedure replaces missing data with valid data from a donor record (i.e., first follow-up student [item] respondent) within an imputation class. In general, variables with lower item nonresponse rates were imputed earlier in the process. Skips and Missing Values The HSLS:09 data were edited using procedures developed and implemented for previous studies sponsored by NCES Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in HSLS:09. Please consult the methodology report (coming soon) for more information. Description of missing data codes
To protect the confidentiality of NCES data that contain information about specific individuals, HSLS:09 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies.
Stochastic methods were used to impute the missing values. Specifically, a weighted sequential hot-deck (WSHD; statistical) imputation procedure (Cox 1980; Iannacchione 1982) using the final student analysis weight (W2STUDENT) was applied to the missing values for variables. The WSHD procedure replaces missing data with valid data from a donor record (i.e., first follow-up student [item] respondent) within an imputation class. In general, variables with lower item nonresponse rates were imputed earlier in the process.
The HSLS:09 data were edited using procedures developed and implemented for previous studies sponsored by NCES Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data.
The table below shows codes for missing values used in HSLS:09. Please consult the methodology report (coming soon) for more information.
Description of missing data codes
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values During and following data collection, the data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a number of explanations for missing data; for example, the item may not have been applicable to certain respondents or a respondent may not have known the answer to the question. With the exception of the not applicable codes, missing data were stochastically imputed. Moreover, for hierarchical analyses and developing survey estimates for faculty members corresponding to sample institutions that provided faculty lists and responded to the institution survey, contextual weights were produced for such subsets of the responding faculty members. The table below shows codes for missing values used. Please consult the methodology report for more information. Description of missing data codes
All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient.
During and following data collection, the data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a number of explanations for missing data; for example, the item may not have been applicable to certain respondents or a respondent may not have known the answer to the question. With the exception of the not applicable codes, missing data were stochastically imputed. Moreover, for hierarchical analyses and developing survey estimates for faculty members corresponding to sample institutions that provided faculty lists and responded to the institution survey, contextual weights were produced for such subsets of the responding faculty members.
The table below shows codes for missing values used. Please consult the methodology report for more information.
1Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions.
2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable imputed and observed will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, B&B:08/12 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. B&B:08/12 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. Imputation Variables with missing data were imputed for graduates who were respondents in a study wave . The imputation procedures employed a two-step process. The first step is a logical imputation . If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation. This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The B&B: 08/12 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:08. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in B&B:08/12. Please consult the First Look for more information. Description of missing value codes
To protect the confidentiality of NCES data that contain information about specific individuals, B&B:08/12 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. B&B:08/12 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies.
Variables with missing data were imputed for graduates who were respondents in a study wave . The imputation procedures employed a two-step process. The first step is a logical imputation . If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation. This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient.
The B&B: 08/12 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:08. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data.
The table below shows codes for missing values used in B&B:08/12. Please consult the First Look for more information.
Description of missing value codes
1In other words, if a graduate was a respondent in B&B:09, he or she will have no missing data for variables created as part of the B&B:09 wave. Similarly, if a graduate was a respondent in B&B:12, he or she will have no missing data for variables created as part of the B&B:12 wave, but may have missing data for variables created as part of the B&B:09 wave if he or she was not a respondent in B&B:09.
2Logical imputation is a process that aims to infer or deduce the missing values from answers to other questions.
3Sequential hot deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction.
Imputation Variables used in cross-sectional estimates in the Baccalaureate and Beyond descriptive reports were imputed. The variables identified for imputation were used in the two B&B:93/03 descriptive reports (Bradburn, Nevill, and Forrest Cataldi 2006; Alt and Henke 2007). The imputations were performed in three steps. First, the interview variables were imputed using the sequential hot deck imputation method.1 This imputation procedure involves identifying a relatively homogenous group of observations, and within the group selecting a random donor’s value to impute a value for the recipient. Second, using the interview variables, including the newly imputed variable values, derived variables were constructed. Skips and Missing Values Both during and upon completion of data collection, edit checks were performed on the B&B:93/03 data file to confirm that the intended skip patterns were implemented during the interview. At the conclusion of data collection, special codes were added as needed to indicate the reason for missing data. Missing data within individual data elements can occur for a variety of reasons. The Table below lists each missing value code and its associated meaning in the B&B:93/03 interview. For more information, see the Baccalaureate and Beyond Longitudinal Study (B&B:93/03) methodology report. Description of missing data codes
Variables used in cross-sectional estimates in the Baccalaureate and Beyond descriptive reports were imputed. The variables identified for imputation were used in the two B&B:93/03 descriptive reports (Bradburn, Nevill, and Forrest Cataldi 2006; Alt and Henke 2007). The imputations were performed in three steps. First, the interview variables were imputed using the sequential hot deck imputation method.1 This imputation procedure involves identifying a relatively homogenous group of observations, and within the group selecting a random donor’s value to impute a value for the recipient. Second, using the interview variables, including the newly imputed variable values, derived variables were constructed.
Both during and upon completion of data collection, edit checks were performed on the B&B:93/03 data file to confirm that the intended skip patterns were implemented during the interview. At the conclusion of data collection, special codes were added as needed to indicate the reason for missing data. Missing data within individual data elements can occur for a variety of reasons.
The Table below lists each missing value code and its associated meaning in the B&B:93/03 interview. For more information, see the Baccalaureate and Beyond Longitudinal Study (B&B:93/03) methodology report.
1Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. Under this methodology, while each respondent record has a chance to be selected for use as a hot-deck donor, the number of times a respondent record can be used for imputation will be controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictor) for each item being imputed were defined. Imputation classes were developed by using a Chi-squared Automatic Interaction.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, B&B:01 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Both during and upon completion of data collection, edit checks were performed on the B&B:00/01 data file to confirm that the intended skip patterns were implemented during the interview. Following data collection, the information collected in CATI was subjected to various checks and examinations. These checks were intended to confirm that the database reflected appropriate skip-pattern relationships and different types of missing data by inserting special codes. The Table below lists each missing value code and its associated meaning in the B&B:00/01 interview. For more information, see the Baccalaureate and Beyond Longitudinal Study (B&B:00/01) methodology report . Description of missing data codes
To protect the confidentiality of NCES data that contain information about specific individuals, B&B:01 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors.
Both during and upon completion of data collection, edit checks were performed on the B&B:00/01 data file to confirm that the intended skip patterns were implemented during the interview. Following data collection, the information collected in CATI was subjected to various checks and examinations. These checks were intended to confirm that the database reflected appropriate skip-pattern relationships and different types of missing data by inserting special codes.
The Table below lists each missing value code and its associated meaning in the B&B:00/01 interview. For more information, see the Baccalaureate and Beyond Longitudinal Study (B&B:00/01) methodology report .
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, BPS:12/17 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:12/17 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The BPS:12/17 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:12. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in BPS:12/17. Please consult the methodology report (coming soon) for more information. Description of missing data codes
To protect the confidentiality of NCES data that contain information about specific individuals, BPS:12/17 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:12/17 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies.
The BPS:12/17 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:12. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data.
The table below shows codes for missing values used in BPS:12/17. Please consult the methodology report (coming soon) for more information.
2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, BPS:04/09 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:04/09 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The BPS:04/09 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:04. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in BPS:04/09. Please consult the methodology report (coming soon) for more information. Description of missing data codes
To protect the confidentiality of NCES data that contain information about specific individuals, BPS:04/09 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:04/09 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies.
The BPS:04/09 data were edited using procedures developed and implemented for previous studies sponsored by NCES, including the base-year study, NPSAS:04. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data.
The table below shows codes for missing values used in BPS:04/09. Please consult the methodology report (coming soon) for more information.
Imputation Logical imputations were performed where items were missing but their values could be implicitly determined. Skips and Missing Values During and following data collection, the CATI/CAPI data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a variety of explanations for missing data within individual data elements. The table below shows codes for missing values used in BPS:01. Please consult the methodology report for more information. Description of missing data codes
Logical imputations were performed where items were missing but their values could be implicitly determined.
During and following data collection, the CATI/CAPI data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a variety of explanations for missing data within individual data elements.
The table below shows codes for missing values used in BPS:01. Please consult the methodology report for more information.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, BPS:94 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:94 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The BPS:94 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data.A variety of explanations are possible for missing data.The table below shows codes for missing values used in BPS:94. Please consult the methodology report for more information. Description of missing data codes
To protect the confidentiality of NCES data that contain information about specific individuals, BPS:94 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. BPS:94 has multiple sources of data for some variables (CPS, NLSDS, student interview, etc.), and reporting differences can occur in each. Data swapping and other forms of perturbation, implemented to protect respondent confidentiality, can also lead to inconsistencies.
The BPS:94 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data.
The table below shows codes for missing values used in BPS:94. Please consult the methodology report for more information.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 2016. Please consult the data file documentation report for more information. Description of missing data codes
Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely.
The table below shows the set of reserve codes for missing values used in NPSAS 2016. Please consult the data file documentation report for more information.
2Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variableimputed and observedwill resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Missing Values and Imputation Following data collection, the data are subjected to various consistency and quality control checks before release. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. Except for data that were missing for cases to which they did not apply (e.g., whether a spouse is enrolled in college for unmarried students) and in a small number of items describing institutional characteristics, missing data were imputed using a two-step process. The first step is a logical imputation.1 If a value could be calculated from the logical relationships with other variables, then that information was used to impute the value for the observation with a missing value. The second step is weighted hot deck imputation.2 This procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor's value to impute a value for the observation with a missing value. The table below shows the set of missing value codes for missing values that were not imputed in NPSAS:12. More information is available from the NPSAS:12 Data File Documentation (http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2014182). Description of missing value codes
To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Data swapping and other forms of perturbation can lead to inconsistencies.
Missing Values and Imputation
Following data collection, the data are subjected to various consistency and quality control checks before release. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely.
Except for data that were missing for cases to which they did not apply (e.g., whether a spouse is enrolled in college for unmarried students) and in a small number of items describing institutional characteristics, missing data were imputed using a two-step process. The first step is a logical imputation.1 If a value could be calculated from the logical relationships with other variables, then that information was used to impute the value for the observation with a missing value. The second step is weighted hot deck imputation.2 This procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor's value to impute a value for the observation with a missing value.
The table below shows the set of missing value codes for missing values that were not imputed in NPSAS:12. More information is available from the NPSAS:12 Data File Documentation (http://nces.ed.gov/pubsearch/pubsinfo.asp?pubid=2014182).
1Logical imputation is a process that aims to infer or deduce the missing values from values for other items.
2Sequential hot deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent's answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using the chi-square automatic interaction detection algorithm.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 2008. Please consult the methodology report for more information. Description of missing data codes
The table below shows the set of reserve codes for missing values used in NPSAS 2008. Please consult the methodology report for more information.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation The imputation procedures employed a two-step process. In the first step, the matching criteria and imputation classes that were used to stratify the dataset were identified such that all imputation was processed independently within each class. In the second step, the weighted sequential hot deck process1 was implemented, whereby missing data were replaced with valid data from donor records that match the recipients with respect to the matching criteria. Variables requiring imputation were not imputed simultaneously. However, some variables that were related substantively were grouped together into blocks, and the variables within a block were imputed simultaneously. Basic demographic variables were imputed first using variables with full information to determine the matching criteria. The order in which variables were imputed was also determined to some extent by the substantive nature of the variables. For example, basic demographics (such as age) were imputed first and these were used to process education variables (such as student level and enrollment intensity) which in turn were used to impute the financial aid variables (such as aid receipt and loan amounts). Skips and Missing Values Edit checks were performed on the NPSAS:04 student interview data and CADE data, both during and upon completion of data collection, to confirm that the intended skip patterns were implemented in both instruments. At the conclusion of data collection, special codes were added as needed to indicate the reason for missing data. Missing data within individual data elements can occur for a variety of reasons. The table below shows the set of reserve codes for missing values used in NPSAS 2004. Please consult the methodology report for more information. Description of missing data codes
The imputation procedures employed a two-step process. In the first step, the matching criteria and imputation classes that were used to stratify the dataset were identified such that all imputation was processed independently within each class. In the second step, the weighted sequential hot deck process1 was implemented, whereby missing data were replaced with valid data from donor records that match the recipients with respect to the matching criteria. Variables requiring imputation were not imputed simultaneously. However, some variables that were related substantively were grouped together into blocks, and the variables within a block were imputed simultaneously. Basic demographic variables were imputed first using variables with full information to determine the matching criteria. The order in which variables were imputed was also determined to some extent by the substantive nature of the variables. For example, basic demographics (such as age) were imputed first and these were used to process education variables (such as student level and enrollment intensity) which in turn were used to impute the financial aid variables (such as aid receipt and loan amounts).
Edit checks were performed on the NPSAS:04 student interview data and CADE data, both during and upon completion of data collection, to confirm that the intended skip patterns were implemented in both instruments. At the conclusion of data collection, special codes were added as needed to indicate the reason for missing data. Missing data within individual data elements can occur for a variety of reasons.
The table below shows the set of reserve codes for missing values used in NPSAS 2004. Please consult the methodology report for more information.
1Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. While each respondent record may be selected for use as a hot-deck donor, the number of times a respondent record is used for imputation is controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictors) for each item being imputed are defined. Imputation classes are developed by using a Chi-squared Automatic Interaction.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, NPSAS:00 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing ValuesThe NPSAS:00 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in NPSAS:00 Please consult the methodology report for more information. Description of missing data codes
To protect the confidentiality of NCES data that contain information about specific individuals, NPSAS:00 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors.
The table below shows codes for missing values used in NPSAS:00 Please consult the methodology report for more information.
Imputation Values for 22 analysis variables were imputed. The variables were imputed using a weighted hot deck procedure, with the exception of estimated family contribution (EFC), which was imputed through a multiple regression approach.The weighed hot deck imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing ValuesThe NPSAS:96 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in NPSAS:96 Please consult the methodology report for more information. Description of missing data codes
Values for 22 analysis variables were imputed. The variables were imputed using a weighted hot deck procedure, with the exception of estimated family contribution (EFC), which was imputed through a multiple regression approach.The weighed hot deck imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient.
The table below shows codes for missing values used in NPSAS:96 Please consult the methodology report for more information.
Derived Variables and Imputed Values Approximately 800 variables have been constructed based on data collected in the NPSAS:93. As a general rule, the constructions of derive variables that concern financial aid and other financial descriptors depend first on record abstract data from the CADE system. These data are supplemented in many cases with information collected in the telephone interviews with parents and students. As between parent and student data, precedence was generally given to parent data for variables concerning family income and assets. Imputations were performed on seven variables that contained missing values. Skips and Missing Values Both the student and parent CATI programs were designed to accommodate responses of "refusal" and "don't know" to any single question. Typically, refusal responses are given for items considered too sensitive by the respondent. "Don't know" responses may be given for any one of several reasons: (1) the respondent misunderstands the question wording, and is not offered subsequent explanation by the interviewer; (2) the respondent is hesitant to provide "best guess" responses, with insufficient prompting from the interviewer; (3) the respondent truly does not know the answer; or (4) the respondent chooses to respond with "don't know" as an implicit refusal to answer the question. Whenever they occur, indeterminate responses in the data set must be resolved by imputation or otherwise dealt with during analysis. The table below shows the set of reserve codes for missing values used in NPSAS 1993. Please consult the data file documentation report for more information. Description of missing data codes
Derived Variables and Imputed Values
Approximately 800 variables have been constructed based on data collected in the NPSAS:93. As a general rule, the constructions of derive variables that concern financial aid and other financial descriptors depend first on record abstract data from the CADE system. These data are supplemented in many cases with information collected in the telephone interviews with parents and students. As between parent and student data, precedence was generally given to parent data for variables concerning family income and assets. Imputations were performed on seven variables that contained missing values.
Both the student and parent CATI programs were designed to accommodate responses of "refusal" and "don't know" to any single question. Typically, refusal responses are given for items considered too sensitive by the respondent. "Don't know" responses may be given for any one of several reasons: (1) the respondent misunderstands the question wording, and is not offered subsequent explanation by the interviewer; (2) the respondent is hesitant to provide "best guess" responses, with insufficient prompting from the interviewer; (3) the respondent truly does not know the answer; or (4) the respondent chooses to respond with "don't know" as an implicit refusal to answer the question. Whenever they occur, indeterminate responses in the data set must be resolved by imputation or otherwise dealt with during analysis.
The table below shows the set of reserve codes for missing values used in NPSAS 1993. Please consult the data file documentation report for more information.
Imputation Variables with more than 5 percent missing cases were imputed. After using information from all appropriate secondary sources, there remained eight variables which required some statistical imputation. Two methods of statistical imputation were used, regression-based or hot deck. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 1990. Please consult the data file documentation report for more information. Description of missing data codes
Variables with more than 5 percent missing cases were imputed. After using information from all appropriate secondary sources, there remained eight variables which required some statistical imputation. Two methods of statistical imputation were used, regression-based or hot deck.
The table below shows the set of reserve codes for missing values used in NPSAS 1990. Please consult the data file documentation report for more information.
Derived Variables and Imputed Values Approximately 800 variables have been constructed based on data collected in the NPSAS:87. As a general rule, the constructions of derive variables that concern financial aid and other financial descriptors depend first on record abstract data from the CADE system. These data are supplemented in many cases with information collected in the telephone interviews with parents and students. As between parent and student data, precedence was generally given to parent data for variables concerning family income and assets. Imputations were performed on seven variables that contained missing values. Skips and Missing Values Both the student and parent CATI programs were designed to accommodate responses of "refusal" and "don't know" to any single question. Typically, refusal responses are given for items considered too sensitive by the respondent. "Don't know" responses may be given for any one of several reasons: (1) the respondent misunderstands the question wording, and is not offered subsequent explanation by the interviewer; (2) the respondent is hesitant to provide "best guess" responses, with insufficient prompting from the interviewer; (3) the respondent truly does not know the answer; or (4) the respondent chooses to respond with "don't know" as an implicit refusal to answer the question. Whenever they occur, indeterminate responses in the data set must be resolved by imputation or otherwise dealt with during analysis. The table below shows the set of reserve codes for missing values used in NPSAS 1993. Please consult the data file documentation report for more information. Description of missing data codes
Approximately 800 variables have been constructed based on data collected in the NPSAS:87. As a general rule, the constructions of derive variables that concern financial aid and other financial descriptors depend first on record abstract data from the CADE system. These data are supplemented in many cases with information collected in the telephone interviews with parents and students. As between parent and student data, precedence was generally given to parent data for variables concerning family income and assets. Imputations were performed on seven variables that contained missing values.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in non-sampling errors. Data swapping and other forms of perturbation can lead to inconsistencies. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation.1 If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values Following data collection, the data are subjected to various consistency and quality control checks before release for use by analysts. One important check is examining all variables with missing data and substituting specific values to indicate the reason for the missing data. For example, an item may not have been applicable to some groups of respondents, a respondent may not have known the answer to a question, or a respondent may have skipped the item entirely. The table below shows the set of reserve codes for missing values used in NPSAS 2008. Please consult the methodology report for more information. Description of missing data codes
All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation.1 If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient.
Perturbation To protect the confidentiality of NCES data that contain information about specific individuals, NPSAS:00 data were subject to perturbation procedures to minimize disclosure risk. Perturbation procedures, which have been approved by the NCES Disclosure Review Board, preserve the central tendency estimates but may result in slight increases in nonsampling errors. Imputation All variables with missing data were imputed. The imputation procedures employed a two-step process. The first step is a logical imputation1. If the imputed value could be deduced from the logical relationships with other variables, then that information was used to impute the value for the recipient. The second step is weighted hot-deck imputation.2 This imputation procedure involves identifying a relatively homogenous group of observations, and, from within the group, selecting a random donor’s value to impute a value for the recipient. Skips and Missing Values The NPSAS:00 data were edited using procedures developed and implemented for previous studies sponsored by NCES. Following data collection, the information collected in the student instrument was subjected to various quality control checks and examinations. These checks were to confirm that the collected data reflected appropriate skip patterns. Another evaluation examined all variables with missing data and substituted specific values to indicate the reason for the missing data. A variety of explanations are possible for missing data. The table below shows codes for missing values used in NPSAS:00 Please consult the methodology report for more information. Description of missing data codes
Perturbation A restricted faculty-level data file was created for release to individuals who apply for and meet standards for such data releases. While this file does not include personally identifying information (i.e., name and Social Security number), other data (i.e., institution, Integrated Postsecondary Education Data System [IPEDS] ID, demographic information, and salary data) may be manipulated in such a way to seem to identify data records corresponding to a particular faculty member. To protect further against such situations, some of the variable values were swapped between faculty respondents. This procedure perturbed and added additional uncertainty to the data. Thus, associations made among variable values to identify a faculty respondent may be based on the original or edited, imputed and/or swapped data. For the same reasons, the data from the institution questionnaire were also swapped to avoid data disclosure. Imputation Item imputation for the faculty questionnaire was performed in several steps. In the first step, the missing values of gender, race, and ethnicity were filled—using cold-deck imputation1— based on the sampling frame information or institution record data. These three key demographic variables were imputed prior to any other variables since they were used as key predictors for all other variables on the data file. After all logical2 and cold-deck imputation procedures were performed, the remaining variables were imputed using the weighted sequential hot-deck method.3 Initially, variables were separated into two groups: unconditional and conditional variables. The first group (unconditional) consisted of variables that applied to all respondents, while the second group (conditional) consisted of variables that applied to only a subset of the respondents. That is, conditional variables were subject to “gate” questions. After this initial grouping, these groups were divided into finer subgroups. After all variables were imputed, consistency checks were applied to the entire faculty data file to ensure that the imputed values did not conflict with other questionnaire items, observed or imputed. This process involved reviewing all of the logical imputation and editing rules as well. Skips and Missing Values During and following data collection, the data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a number of explanations for missing data; for example, the item may not have been applicable to certain respondents or a respondent may not have known the answer to the question. With the exception of the not applicable codes, missing data were stochastically imputed. Moreover, for hierarchical analyses and developing survey estimates for faculty members corresponding to sample institutions that provided faculty lists and responded to the institution survey, contextual weights were produced for such subsets of the responding faculty members. The table below shows codes for missing values used in NSOPF:04. Please consult the methodology report for more information. Description of missing data codes
A restricted faculty-level data file was created for release to individuals who apply for and meet standards for such data releases. While this file does not include personally identifying information (i.e., name and Social Security number), other data (i.e., institution, Integrated Postsecondary Education Data System [IPEDS] ID, demographic information, and salary data) may be manipulated in such a way to seem to identify data records corresponding to a particular faculty member. To protect further against such situations, some of the variable values were swapped between faculty respondents. This procedure perturbed and added additional uncertainty to the data. Thus, associations made among variable values to identify a faculty respondent may be based on the original or edited, imputed and/or swapped data. For the same reasons, the data from the institution questionnaire were also swapped to avoid data disclosure.
Item imputation for the faculty questionnaire was performed in several steps. In the first step, the missing values of gender, race, and ethnicity were filled—using cold-deck imputation1— based on the sampling frame information or institution record data. These three key demographic variables were imputed prior to any other variables since they were used as key predictors for all other variables on the data file. After all logical2 and cold-deck imputation procedures were performed, the remaining variables were imputed using the weighted sequential hot-deck method.3 Initially, variables were separated into two groups: unconditional and conditional variables. The first group (unconditional) consisted of variables that applied to all respondents, while the second group (conditional) consisted of variables that applied to only a subset of the respondents. That is, conditional variables were subject to “gate” questions. After this initial grouping, these groups were divided into finer subgroups. After all variables were imputed, consistency checks were applied to the entire faculty data file to ensure that the imputed values did not conflict with other questionnaire items, observed or imputed. This process involved reviewing all of the logical imputation and editing rules as well.
The table below shows codes for missing values used in NSOPF:04. Please consult the methodology report for more information.
1Cold-deck imputation involves replacing the missing values with data from sources such as data used for sampling frame construction. While resource intensive, these methods often obtain the actual value that is missing. Stochastic imputation methods, such as sequential hot-deck imputation, rely on the observed data to provide replacing values (donors) for records with missing values.
3Sequential hot-deck imputation involves defining imputation classes, which generally consist of a cross-classification of covariates, and then replacing missing values sequentially from a single pass through the survey data within the imputation classes. When this form of imputation is performed using the sampling weights, the procedure is called weighted sequential hot-deck imputation. This procedure takes into account the unequal probabilities of selection in the original sample to specify the expected number of times a particular respondent’s answer will be used as a donor. These expected selection frequencies are specified so that, over repeated applications of the algorithm, the weighted distribution of all values for that variable—imputed and observed—will resemble that of the target universe. Under this methodology, while each respondent record has a chance to be selected for use as a hot-deck donor, the number of times a respondent record can be used for imputation will be controlled. To implement the weighted sequential hot-deck procedure, imputation classes and sorting variables that are relevant (strong predictor) for each item being imputed were defined. Imputation classes were developed by using a Chi-squared Automatic Interaction.
Both the faculty and institution questionnaire data were edited using seven principles designed to improve data quality and consistency.Menu items. For many questions there were several sub-items listed where the respondent was asked to give a response for each sub-item. These questions were cleaned with several procedures. First if the main question had an “NA” (Not Applicable) check box and that box was checked, all of the sub-items were set to a value of “no” or “zero” depending on the wording of the question. Second, if the respondent had filled out one or more of the sub-items with a “yes” response or a positive number but had left other sub-items blank, the missing sub-items were set to “no,” “zero,” or “don’t know” depending on the question wording. If all sub-items were missing and there was no “NA” box, or the “NA” box was not checked, the case was flagged and the data values were imputed for that question. Examples of these types of questions are Question 21 in the institution questionnaire and Question 29 in the faculty questionnaire.Inter-item consistency checks. Many types of inter-item consistency checks were performed on the data. One procedure was to check groups of related items for internal consistency and to make adjustments to make them consistent. For example, in questions that asked about a spouse in the faculty questionnaire (Questions 66i, Q76i, and 77a) if respondents indicated that they did not have a spouse in one or more of the questions, the other questions were checked for consistency and corrected as necessary. Another procedure checked “NA” boxes. If the respondent had checked the “NA” box for a question but had filled in any of the sub-items for that question the “NA” box was set to blank. For example, this procedure was used with Question 21 in the institution questionnaire and Question 16 in the faculty questionnaire. A third procedure was to check filter items for which more detail was sought in a follow-up open-ended or closed-ended question. If detail was provided, then the filter question was checked to make sure the appropriate response was recorded. For example, this procedure was used with Question 11 in the institution questionnaire and Question 12E in the faculty questionnaire.Percent items. All items where respondents were asked to give a percentage were checked to make sure they summed to 100 percent. The editing program also looked for any numbers between 0 and 1 to make sure that respondents did not fill in the question with a decimal rather than a percentage. All fractions of a percent were rounded to the nearest whole percent. An example of this type of item is Question 31 in the faculty questionnaire.Data imputation for the faculty questionnaire was performed in four steps. The imputation method for each variable is specified in the labels for the imputation flags in the faculty dataset.Logical imputation. The logical imputation was conducted during the data cleaning steps as explained in the immediately preceding section. Cold deck. Missing responses were filled in with data from the sample frame whenever the relevant data were available. Examples include gender, race, and employment status.Hot deck. This procedure selected non-missing values from “sequential nearest neighbors” within the imputation class. All questions that were categorical and had more than 16 categories were imputed with this method. An example is Question Q14 – principal field of teaching. The imputation class for this question was created using faculty stratum and instructional duty status (Q1). Regression type. This procedure employed SAS PROC IMPUTE21. All items that were still missing after the logical, cold deck, and hot deck imputation procedures were imputed with this method. Project staff selected the independent variables by first looking through the questionnaire for logically related items and then by conducting a correlation analysis of the questions against each other to find the top correlates for each item.
Menu items. For many questions there were several sub-items listed where the respondent was asked to give a response for each sub-item. These questions were cleaned with several procedures. First if the main question had an “NA” (Not Applicable) check box and that box was checked, all of the sub-items were set to a value of “no” or “zero” depending on the wording of the question. Second, if the respondent had filled out one or more of the sub-items with a “yes” response or a positive number but had left other sub-items blank, the missing sub-items were set to “no,” “zero,” or “don’t know” depending on the question wording. If all sub-items were missing and there was no “NA” box, or the “NA” box was not checked, the case was flagged and the data values were imputed for that question. Examples of these types of questions are Question 21 in the institution questionnaire and Question 29 in the faculty questionnaire.
Inter-item consistency checks. Many types of inter-item consistency checks were performed on the data. One procedure was to check groups of related items for internal consistency and to make adjustments to make them consistent. For example, in questions that asked about a spouse in the faculty questionnaire (Questions 66i, Q76i, and 77a) if respondents indicated that they did not have a spouse in one or more of the questions, the other questions were checked for consistency and corrected as necessary. Another procedure checked “NA” boxes. If the respondent had checked the “NA” box for a question but had filled in any of the sub-items for that question the “NA” box was set to blank. For example, this procedure was used with Question 21 in the institution questionnaire and Question 16 in the faculty questionnaire. A third procedure was to check filter items for which more detail was sought in a follow-up open-ended or closed-ended question. If detail was provided, then the filter question was checked to make sure the appropriate response was recorded. For example, this procedure was used with Question 11 in the institution questionnaire and Question 12E in the faculty questionnaire.
Percent items. All items where respondents were asked to give a percentage were checked to make sure they summed to 100 percent. The editing program also looked for any numbers between 0 and 1 to make sure that respondents did not fill in the question with a decimal rather than a percentage. All fractions of a percent were rounded to the nearest whole percent. An example of this type of item is Question 31 in the faculty questionnaire.
Logical imputation. The logical imputation was conducted during the data cleaning steps as explained in the immediately preceding section.
Cold deck. Missing responses were filled in with data from the sample frame whenever the relevant data were available. Examples include gender, race, and employment status.
Hot deck. This procedure selected non-missing values from “sequential nearest neighbors” within the imputation class. All questions that were categorical and had more than 16 categories were imputed with this method. An example is Question Q14 – principal field of teaching. The imputation class for this question was created using faculty stratum and instructional duty status (Q1).
Regression type. This procedure employed SAS PROC IMPUTE21. All items that were still missing after the logical, cold deck, and hot deck imputation procedures were imputed with this method. Project staff selected the independent variables by first looking through the questionnaire for logically related items and then by conducting a correlation analysis of the questions against each other to find the top correlates for each item.
Depending on the scale of the variable being imputed, one of two methods were used:1) Regression imputation was used for continuous and dichotomous variables; and2) Hotdeck imputation was used for unordered polytomous variables.The regression method incorporated in NCES’s PROC IMPUTE was used to impute missing values for approximately 90 percent of the 395 items on the faculty questionnaire.Of the total of 395 items, 353 were imputed using the regression-based imputation procedures only.
Depending on the scale of the variable being imputed, one of two methods were used:1) Regression imputation was used for continuous and dichotomous variables; and2) Hotdeck imputation was used for unordered polytomous variables.The regression method incorporated in NCES’s PROC IMPUTE was used to impute missing values for approximately 90 percent of the 395 items on the faculty questionnaire.
Of the total of 395 items, 353 were imputed using the regression-based imputation procedures only.
NSOPF:88 was conducted with a sample of 480 institutions (including 2-year, 4-year, doctoral-granting, and other colleges and universities), some 11,010 faculty, and more than 3,000 department chairpersons. Institutions were sampled from the 1987 IPEDS universe and were stratified by modified Carnegie Classifications and size (faculty counts). These strata were (1) public, research; (2) private, research; (3) public, other Ph.D. institution (not defined in any other stratum); (4) private, other Ph.D. institution (not defined in any other stratum); (5) public, comprehensive; (6) private, comprehensive; (7) liberal arts; (8) public, 2-year; (9) private, 2-year; (10) religious; (11) medical; and (12) “other” schools (not defined in any other stratum). Within each stratum, institutions were randomly selected. Of the 480 institutions selected, 450 (94 percent) agreed to participate and provided lists of their faculty and department chairpersons. Within 4-year institutions, faculty and department chairpersons were stratified by program area and randomly sampled within each stratum; within 2-year institutions, simple random samples of faculty and department chairpersons were selected; and within specialized institutions (religious, medical, etc.), faculty samples were randomly selected (department chairpersons were not sampled). At all institutions, faculty were also stratified on the basis of employment status—full-time and part-time. Note that teaching assistants and teaching fellows were excluded in NSOPF:88.Although NSOPF:88 consisted of three questionnaires, imputations were only performed for faculty item nonresponse. The within-cell random imputation method was used to fill in most Faculty Questionnaire items that had missing data.
NSOPF:88 was conducted with a sample of 480 institutions (including 2-year, 4-year, doctoral-granting, and other colleges and universities), some 11,010 faculty, and more than 3,000 department chairpersons. Institutions were sampled from the 1987 IPEDS universe and were stratified by modified Carnegie Classifications and size (faculty counts). These strata were (1) public, research; (2) private, research; (3) public, other Ph.D. institution (not defined in any other stratum); (4) private, other Ph.D. institution (not defined in any other stratum); (5) public, comprehensive; (6) private, comprehensive; (7) liberal arts; (8) public, 2-year; (9) private, 2-year; (10) religious; (11) medical; and (12) “other” schools (not defined in any other stratum). Within each stratum, institutions were randomly selected. Of the 480 institutions selected, 450 (94 percent) agreed to participate and provided lists of their faculty and department chairpersons. Within 4-year institutions, faculty and department chairpersons were stratified by program area and randomly sampled within each stratum; within 2-year institutions, simple random samples of faculty and department chairpersons were selected; and within specialized institutions (religious, medical, etc.), faculty samples were randomly selected (department chairpersons were not sampled). At all institutions, faculty were also stratified on the basis of employment status—full-time and part-time. Note that teaching assistants and teaching fellows were excluded in NSOPF:88.
Although NSOPF:88 consisted of three questionnaires, imputations were only performed for faculty item nonresponse. The within-cell random imputation method was used to fill in most Faculty Questionnaire items that had missing data.
Imputation The imputation process for the missing data from the institution questionnaire involved similar steps to those used for imputation of the faculty data. The missing data for variables were imputed using the weighted sequential hot-deck method.1 Analogous to the imputation process for the faculty data, the variables were partitioned into conditional and unconditional groups. The unconditional variables were sorted by percent missing and then imputed in the order from the lowest percent missing to the highest. The conditional group was partitioned into three subgroups based on the level of conditionality for each variable, and then imputed in that order. The imputation class for both unconditional and conditional variables consisted of the institution sampling stratum, and the sorting variables included the number of full-time and part-time faculty members. Skips and Missing Values During and following data collection, the data were reviewed to confirm that the data collected reflected the intended skip-pattern relationships. At the conclusion of data collection, special codes were inserted in the database to reflect the different types of missing data. There are a number of explanations for missing data; for example, the item may not have been applicable to certain respondents or a respondent may not have known the answer to the question. With the exception of the not applicable codes, missing data were stochastically imputed. Moreover, for hierarchical analyses and developing survey estimates for faculty members corresponding to sample institutions that provided faculty lists and responded to the institution survey, contextual weights were produced for such subsets of the responding faculty members. The table below shows codes for missing values used in NSOPF:04. Please consult the methodology report for more information. Description of missing data codes
The imputation process for the missing data from the institution questionnaire involved similar steps to those used for imputation of the faculty data. The missing data for variables were imputed using the weighted sequential hot-deck method.1 Analogous to the imputation process for the faculty data, the variables were partitioned into conditional and unconditional groups. The unconditional variables were sorted by percent missing and then imputed in the order from the lowest percent missing to the highest. The conditional group was partitioned into three subgroups based on the level of conditionality for each variable, and then imputed in that order. The imputation class for both unconditional and conditional variables consisted of the institution sampling stratum, and the sorting variables included the number of full-time and part-time faculty members.
ImputationThe NTPS used two main approaches to impute data. First, donor respondent methods, such as hot-deck imputation, were used. Second, if no suitable donor case could be matched, the few remaining items were imputed using mean or mode from groups of similar cases to impute a value to the item with missing data. Finally, in rare cases for which imputed values were inconsistent with existing questionnaire data or out of the range of acceptable values, Census Bureau analysts looked at the items and tried to determine an appropriate value.WeightingWeighting of the sample units was carried out to produce national estimates for public schools, principals, and teachers. The weighting procedures used in NTPS had three purposes: to take into account the school's selection probability; to reduce biases that may result from unit nonresponse; and to make use of available information from external sources to improve the precision of sample estimates.
The NTPS used two main approaches to impute data. First, donor respondent methods, such as hot-deck imputation, were used. Second, if no suitable donor case could be matched, the few remaining items were imputed using mean or mode from groups of similar cases to impute a value to the item with missing data. Finally, in rare cases for which imputed values were inconsistent with existing questionnaire data or out of the range of acceptable values, Census Bureau analysts looked at the items and tried to determine an appropriate value.
Weighting of the sample units was carried out to produce national estimates for public schools, principals, and teachers. The weighting procedures used in NTPS had three purposes: to take into account the school's selection probability; to reduce biases that may result from unit nonresponse; and to make use of available information from external sources to improve the precision of sample estimates.
ImputationFour approaches to imputation were used in the NHES:2016: logic-based imputation, which was used whenever possible; unweighted sequential hot deck imputation, which was used for the majority of the missing data (i.e., for all variables that were not boundary and sort variables—described below); weighted random imputation, which was used for a small number of variables including boundary and sort variables; and manual imputation, which was used in a very small number of cases for a small number of variables.For more information about these approaches, please see the NHES: 2016 Data File User's Manual.
Four approaches to imputation were used in the NHES:2016: logic-based imputation, which was used whenever possible; unweighted sequential hot deck imputation, which was used for the majority of the missing data (i.e., for all variables that were not boundary and sort variables—described below); weighted random imputation, which was used for a small number of variables including boundary and sort variables; and manual imputation, which was used in a very small number of cases for a small number of variables.For more information about these approaches, please see the NHES: 2016 Data File User's Manual.
ImputationThree approaches to imputation were used in the NHES:2012: unweighted sequential hot deck imputation, which was used for the majority of the missing data, that is, for all variables that were not required for Interview Status Recode (ISR) classification, as described in chapter 4; weighted random imputation, which was used for a small number of variables; and manual imputation, which was used in a very small number of cases for most variables.For more information about these approaches, please see the NHES: 2012 Data File User's Manual.
Three approaches to imputation were used in the NHES:2012: unweighted sequential hot deck imputation, which was used for the majority of the missing data, that is, for all variables that were not required for Interview Status Recode (ISR) classification, as described in chapter 4; weighted random imputation, which was used for a small number of variables; and manual imputation, which was used in a very small number of cases for most variables.For more information about these approaches, please see the NHES: 2012 Data File User's Manual.
Nonresponse Nonresponse inevitably introduces some degree of error into survey results. In examining the impact of nonresponse, it is useful to think of the survey population as including two strata--a respondent stratum that consists of all units that would have provided data had they been selected for the survey, and a nonrespondent stratum that consists of all units that would not have provided data had they been selected. The actual sample of respondents necessarily consists entirely of units from the respondent stratum. Thus, sample statistics can serve as unbiased estimates only for the respondent stratum; as estimates for the entire population, the sample statistics will be biased to the extent that the characteristics of the respondents differ from those of the entire population.In the High School and Beyond study, there were two stages of sample selection and therefore two stages of nonresponse. During the base year survey, sample schools were asked to permit the selection of individual sophomores and seniors from school rosters and to designate "survey days" for the collection of student questionnaire and test data. Schools that refused to cooperate in either of these activities were dropped from the sample. Individual students at cooperating schools could also fail to take part in the base year survey. Unlike "refusal" schools, nonparticipating students were not dropped from the sample; they remained eligible for selection into the follow-up samples.Estimates based on student data from the base year surveys include two components of nonresponse bias: bias introduced by nonresponse at the school level, and bias introduced by nonresponse on the part of students attending cooperating schools. Each component of the overall bias depends on two factors--the level of nonresponse and the difference between respondents and nonrespondents: Bias = P1(Y1R - Y1NR) + P2(Y2R - Y2NR)in which P1 = the proportion of the population of students attending schools that would have been nonrespondents,YlNR = the parameter describing the population of students attending nonrespondent schools, P2 = the proportion of students attending respondent schools who would have been nonrespondents, and Y2NR = the parameter describing this group of students.Nonresponse bias will be small if the nonrespondent strata constitute only a small portion of the survey population or if the differences between respondents and nonrespondents are small. The proportions P1 and P2 can generally be estimated from survey data using appropriately weighted nonresponse rates. The implications of the equation can be easily seen in terms of a particular base year estimate. On the average, sophomores got 10.9 items right on a standardized vocabulary test. This figure is an estimate of Y2R, the population mean for all participating students at cooperating schools. Now, suppose that sophomores at cooperating schools average two more correct than sophomores attending refusal schools (Y1R - Y1NR = 2), and suppose further that among sophomores attending cooperating schools, student respondents average one more correct answer than student nonrespondents (Y2R - Y2NR = 1). Noting that the base year school nonresponse rate was about .30 and the student nonresponse rate for sophomores was about .12, we can use these figures as estimates of P1 and P2 and we can use this equation to calculate the bias as: Bias = .30(2) + .12(1) = .72 That is, the sample estimate is biased by about .7 of a test score point.This example assumes knowledge of the relevant population means; in practice, of course, they are not known and, although Pl and P2 can generally be estimated from the nonresponse rates, the lack of survey data for nonrespondents prevents the estimation of the nonresponse bias. The High School and Beyond study is an exception to this general rule: during the first follow-up, school questionnaire data were obtained from most of the base year refusal schools, and student data were obtained from most of the base year student nonrespondents selected for the first follow-up sample. These data provide a basis for assessing the magnitude of nonresponse bias in base year estimates.The bias introduced by base year school-level refusals is of particular concern since it carries over into successive rounds of the survey. Students attending refusal schools were not sampled during the base year and have no chance for selection into subsequent rounds of observation. To the extent that these students differ from students from cooperating schools during later waves of the study, the bias introduced by base year school nonresponse will persist. Student nonresponse is not carried over in this way since student nonrespondents remain eligible for sampling in later waves of the study.The results of three types of analyses concerning nonresponse are described in an earlier report. Based on school questionnaire data, schools that participated during the base year were compared with all eligible schools. Based on the first follow-up student data, base year student respondents were compared with nonrespondents. Finally, student nonresponse during the first follow-up survey was analyzed. Taken together, these earlier analyses indicated that nonresponse had little effect on base year and first follow-up estimates. The results presented there suggest that the school-level component of the bias affected base year estimates by 2 percent or less and that the student-level component had even less impact.
Nonresponse
Nonresponse inevitably introduces some degree of error into survey results. In examining the impact of nonresponse, it is useful to think of the survey population as including two strata--a respondent stratum that consists of all units that would have provided data had they been selected for the survey, and a nonrespondent stratum that consists of all units that would not have provided data had they been selected. The actual sample of respondents necessarily consists entirely of units from the respondent stratum. Thus, sample statistics can serve as unbiased estimates only for the respondent stratum; as estimates for the entire population, the sample statistics will be biased to the extent that the characteristics of the respondents differ from those of the entire population.
In the High School and Beyond study, there were two stages of sample selection and therefore two stages of nonresponse. During the base year survey, sample schools were asked to permit the selection of individual sophomores and seniors from school rosters and to designate "survey days" for the collection of student questionnaire and test data. Schools that refused to cooperate in either of these activities were dropped from the sample. Individual students at cooperating schools could also fail to take part in the base year survey. Unlike "refusal" schools, nonparticipating students were not dropped from the sample; they remained eligible for selection into the follow-up samples.
Estimates based on student data from the base year surveys include two components of nonresponse bias: bias introduced by nonresponse at the school level, and bias introduced by nonresponse on the part of students attending cooperating schools. Each component of the overall bias depends on two factors--the level of nonresponse and the difference between respondents and nonrespondents:
Bias = P1(Y1R - Y1NR) + P2(Y2R - Y2NR)
in which
P1 = the proportion of the population of students attending schools that would have been nonrespondents,
YlNR = the parameter describing the population of students attending nonrespondent schools,
P2 = the proportion of students attending respondent schools who would have been nonrespondents, and
Y2NR = the parameter describing this group of students.
Nonresponse bias will be small if the nonrespondent strata constitute only a small portion of the survey population or if the differences between respondents and nonrespondents are small. The proportions P1 and P2 can generally be estimated from survey data using appropriately weighted nonresponse rates.
The implications of the equation can be easily seen in terms of a particular base year estimate. On the average, sophomores got 10.9 items right on a standardized vocabulary test. This figure is an estimate of Y2R, the population mean for all participating students at cooperating schools. Now, suppose that sophomores at cooperating schools average two more correct than sophomores attending refusal schools (Y1R - Y1NR = 2), and suppose further that among sophomores attending cooperating schools, student respondents average one more correct answer than student nonrespondents (Y2R - Y2NR = 1). Noting that the base year school nonresponse rate was about .30 and the student nonresponse rate for sophomores was about .12, we can use these figures as estimates of P1 and P2 and we can use this equation to calculate the bias as:
Bias = .30(2) + .12(1) = .72
That is, the sample estimate is biased by about .7 of a test score point.
This example assumes knowledge of the relevant population means; in practice, of course, they are not known and, although Pl and P2 can generally be estimated from the nonresponse rates, the lack of survey data for nonrespondents prevents the estimation of the nonresponse bias. The High School and Beyond study is an exception to this general rule: during the first follow-up, school questionnaire data were obtained from most of the base year refusal schools, and student data were obtained from most of the base year student nonrespondents selected for the first follow-up sample. These data provide a basis for assessing the magnitude of nonresponse bias in base year estimates.
The bias introduced by base year school-level refusals is of particular concern since it carries over into successive rounds of the survey. Students attending refusal schools were not sampled during the base year and have no chance for selection into subsequent rounds of observation. To the extent that these students differ from students from cooperating schools during later waves of the study, the bias introduced by base year school nonresponse will persist. Student nonresponse is not carried over in this way since student nonrespondents remain eligible for sampling in later waves of the study.
The results of three types of analyses concerning nonresponse are described in an earlier report. Based on school questionnaire data, schools that participated during the base year were compared with all eligible schools. Based on the first follow-up student data, base year student respondents were compared with nonrespondents. Finally, student nonresponse during the first follow-up survey was analyzed. Taken together, these earlier analyses indicated that nonresponse had little effect on base year and first follow-up estimates. The results presented there suggest that the school-level component of the bias affected base year estimates by 2 percent or less and that the student-level component had even less impact.
NonresponseSchool-level nonresponse is a serious concern because it carries over into successive rounds of NELS:88. Students attending schools that did not cooperate in the base year were not sampled and had little or no chance of selection into the follow-up samples. To the extent that students at noncooperating schools differ from students at cooperating schools, the student-level bias introduced by base-year school noncooperation persists during subsequent waves. Nonresponse adjustments to weights are an attempt to compensate for bias in the estimate for a particular subgroup; they do not adjust for nonresponse bias within subgroups.In the base year, nonresponding schools were asked to supply information about key school questionnaire variables, and virtually all did so. Based on these data, analysis of school-level nonresponse suggests that, to the extent that schools can be characterized by size, control, organizational structure, student composition, and other characteristics, the impact of nonresponding schools on school level estimates is small.25 Readers interested in more information about the analyses of school nonresponse rates and bias for the NELS:88 base year should refer to the NELS:88 Base-Year Sample Design Report (Spencer et al. 1990). School nonresponse was not assessed in the first or second follow-ups for two reasons. First, there was practically no school-level nonresponse; institutional cooperation levels approached 99 percent in both rounds. Second, the first and second follow-up samples were student-driven, unlike the two-stage initial sample design in the base year. Hence, even if a school refused in either the first or second follow-ups, the individual student was pursued outside of school.25. The use of school questionnaire variables to assess bias in estimates concerning characteristics of the student population is not entirely straightforward. Still, to the extent that school characteristics are closely related to the characteristics of the students attending them, estimates based on school questionnaire data can serve as reasonable proxies for more direct estimates of student-level unit nonresponse bias.
School-level nonresponse is a serious concern because it carries over into successive rounds of NELS:88. Students attending schools that did not cooperate in the base year were not sampled and had little or no chance of selection into the follow-up samples. To the extent that students at noncooperating schools differ from students at cooperating schools, the student-level bias introduced by base-year school noncooperation persists during subsequent waves. Nonresponse adjustments to weights are an attempt to compensate for bias in the estimate for a particular subgroup; they do not adjust for nonresponse bias within subgroups.
In the base year, nonresponding schools were asked to supply information about key school questionnaire variables, and virtually all did so. Based on these data, analysis of school-level nonresponse suggests that, to the extent that schools can be characterized by size, control, organizational structure, student composition, and other characteristics, the impact of nonresponding schools on school level estimates is small.25 Readers interested in more information about the analyses of school nonresponse rates and bias for the NELS:88 base year should refer to the NELS:88 Base-Year Sample Design Report (Spencer et al. 1990). School nonresponse was not assessed in the first or second follow-ups for two reasons. First, there was practically no school-level nonresponse; institutional cooperation levels approached 99 percent in both rounds. Second, the first and second follow-up samples were student-driven, unlike the two-stage initial sample design in the base year. Hence, even if a school refused in either the first or second follow-ups, the individual student was pursued outside of school.
