Appendix B: Documentation for the American Community Survey-Comparable Wage Index (ACS-CWI): 2013–15

The ACS-CWI was developed and produced by the U.S. Census Bureau in collaboration with Lori L. Taylor, Bush School of Government and Public Service, Texas A&M University. Stephen Q. Cornman, National Center for Education Statistics; Laura C. Nixon and Matthew J. Spence, Education Demographic, Geographic, and Economic Statistics (EDGE) Branch, Economic Reimbursable Surveys Division (ERD) provided direct assistance in this collaborative effort. 

Introduction

The Comparable Wage Index (CWI) was initially created by the National Center for Education Statistics (NCES) to facilitate comparison of educational expenditures across locales, principally school districts or local education agencies (LEAs) and states or state education agencies (SEAs).¹ The CWI is a measure of the systematic regional variations in the wages and salaries of college graduates who are not prekindergarten through grade 12 educators (in this context, those with occupations or employers in elementary or secondary education). It can be used by researchers to adjust district-level finance data at different levels and ultimately make better comparisons across geographic areas.

This documentation describes the creation of a CWI based primarily on the American Community Survey (ACS). The ACS, an ongoing survey conducted by the U.S. Census Bureau, has replaced the decennial census as the primary source of detailed demographic information about the U.S. population. It provides information about the earnings, age, occupation, industry, and other demographic characteristics of millions of U.S. workers. The American Community Survey-Comparable Wage Index (ACS-CWI) measures wage and salary differences of college graduates using an analysis that is modeled on the baseline analysis used to construct the original CWI released by NCES in 2006.

The remainder of this documentation includes background information, detailed information about the ACS-CWI, a user guide, and a glossary of terms.

Top

Background

Geographic cost data for states, metropolitan areas, and school districts are frequently and widely requested by policymakers, practitioners, the school finance research community, and the public. In response, NCES has engaged in a long tradition of publishing research and analysis on geographic cost indexes.² This report documents the newly developed American Community Survey-Comparable Wage Index (ACS-CWI).

The goal of any geographic cost index is to measure uncontrollable differences in the purchasing power of school districts so that comparisons among districts or across time can be based on real educational resources. Where costs are high, districts are unable to purchase as many real resources for each dollar of expenditure; where costs are low, districts have greater purchasing power and are able to purchase more real resources. In other words, districts in high-cost environments must spend more than districts in low-cost environments to provide the same level of educational services. A geographic cost index describes how much more. The cost of labor, particularly the wages paid to teachers, is one of the primary costs for districts. For this reason, NCES has focused on measuring the variation in labor costs by geographic location.

The ACS-CWI is designed to identify geographic variation in wages of college-educated workers outside the education field after controlling for job-related and demographic characteristics.³ The basic premise of any CWI is that all types of workers demand higher wages in areas where the cost of living is high or desirable local amenities (such as good climate, low crime rates, or access to beaches, museums, and restaurants) are lacking. As a result, it should be possible to measure most of the geographic variation in the cost of hiring teachers and other prekindergarten (preK) through grade 12 educators by observing systematic regional variations in the wages of comparable workers who are not preK–12 educators.⁴

In theory, if accountants, nurses, and computer programmers, for example, all earn 5 percent more than the national average for their professions in Houston, then it is reasonable to expect that the cost of hiring teachers in Houston would also be 5 percent more than the national average for teachers.

The ACS-CWI has been developed as a special tabulation of restricted-use data from the three most recent years of the ACS. The ACS-CWI measures local differences in the prevailing wage for college graduates in all jobs, except education.
The ACS-CWI updates and improves on the baseline analysis used to estimate the initial CWI developed by NCES (Taylor and Fowler 2006). The initial CWI was based on public-use data from the 2000 Census. The initial CWI based labor market definitions on Public Use Microdata Areas (PUMAs), which are “special non-overlapping areas that partition each state into congruous geographic units containing no fewer than 100,000 people each.”⁵ In constructing the ACS-CWI, U.S. Census Bureau researchers have access to the restricted-use files and are therefore able to base the labor market definitions on counties, which are the units of analysis most commonly used by the U.S. Bureau of Labor Statistics to define labor markets.⁶ As a result, in stark contrast to the initial CWI released by NCES that provided labor cost estimates for 800 labor market areas, the ACS-CWI provides labor cost estimates for 1,570 labor market areas.

The ACS-CWI incorporates the recommendations of a panel of experts on the CWI, which was convened by NCES in January 2012. The panel recommended that NCES annually produce and release geographic adjustment factors for educational expenditures.⁷ The panel recommended that the factors be based on 1 year of restricted-access data produced by the ACS but also recognized that it may be desirable to base the estimation on multiple years. The ACS-CWI uses 3 years of restricted-access data to contain sufficient sample sizes in optimally sized labor markets for high data quality. Going forward, a rolling sample of the 3 most recent years of the ACS will be utilized to update the ACS-CWI each year.

Top

Strengths and weaknesses of the Comparable Wage Index

A CWI offers many advantages over other geographic cost adjustment methodologies.⁸ A CWI can be estimated from existing data, making it more cost effective to estimate and update than other approaches.

A CWI clearly measures costs that are beyond the control of school district administrators. Unlike cost adjustments that are based on analyses of school district expenditures (as in Chambers 1998 and Taylor, Chambers, and Robinson 2004) there is no risk that a CWI confuses high-spending school districts with high-cost school districts and no need to rely on statistical techniques and researcher judgment to separate controllable from uncontrollable costs.

A CWI is also appropriate regardless of the competitiveness of teacher labor markets. If a lack of competition in the teacher market distorts teacher compensation patterns, then cost indexes based on teacher compensation will be biased, but a CWI will not (Goldhaber 1999; Hanushek 1999).

A CWI reflects differences in amenities as well as the cost of living. As such, it is a more complete price index than the cost of living indexes used for regional cost adjustments in the Colorado and Wyoming school funding formulas (Taylor 2015). Cost of living indexes, like the Wyoming Cost of Living Index, have been criticized for overestimating labor costs in locations where attractive amenities make those locations desirable places to live and work (Rothstein and Smith 1997; Stoddard 2005; Taylor 2015).

Another advantage of a CWI is its general applicability. Because the resulting cost index is based on systematic differences in the general wage level, it can be used to measure labor costs not only for public elementary and secondary education but also for private schools, job training programs, and postsecondary institutions.

There are also several limitations to using a CWI to measure variations in the cost of education. First, a CWI is a labor cost index, and labor cost is only part of the total cost of education—albeit a very large part. It could be problematic to apply a labor cost index, such as NCES’s initial CWI or the ACS-CWI, to school district expenditures that are not affected by labor cost differentials, such as energy costs (Smith et al. 2003).

Second, the labor cost model underlying any CWI presumes that workers are mobile. If moving costs or other barriers to moving slow worker migration, then “labor cost may temporarily diverge from what would be expected given local amenities and the local cost of living. Employers in fast-growing industries and school districts in fast-growing areas may need to pay a temporary premium to attract workers. [A] CWI cannot capture this effect” (Taylor 2006, p. 352).

Third, a CWI is constructed with the assumption that educators and the noneducator population under analysis are comparable with respect to their tastes for amenities and the cost of living. If comparability breaks down, then a CWI becomes a poor proxy for the cost of educator labor. Another aspect of this limitation exists when there are teacher preferences for teaching in certain types of schools rather than others, and local schools offer higher wage rates for specific types of schools or in certain subjects. For example, in some areas, teachers are offered higher salaries as an incentive to teach in high-poverty schools. These relatively higher salaries for teachers may not be reflected in the ACS-CWI model.

Fourth, a CWI is an estimate from a sample survey and is subject to the usual criticisms of sample-based research, including sampling error.⁹ As a result, data users will need to account for this variability when making claims about differences between estimated means. The ACS-CWI estimates are reported along with standard errors to facilitate this review.

Finally, a CWI is based on labor markets, not school districts. It is not designed to capture variations in cost across school districts within a single labor market, such as those cost differences that might be attributable to working conditions in specific school districts. It is also not designed to map perfectly onto school district boundaries. When school districts operate in multiple labor markets (as may be the case when districts cross county lines), researchers must develop strategies for matching index values to school districts. Such strategies may introduce measurement error.

Top

The ACS-CWI

The ACS-CWI is derived from a regression analysis of individual wage data. The data for the analysis come from the 2013, 2014, and 2015 survey years of the ACS because a 3-year span yields a much larger sample and more precise estimates of wages by location than could be generated using a single year of data. The larger sample also permits a much finer geographic breakdown than would be possible in an analysis based on a single year of data.

The ACS asks respondents about employment characteristics, including location of workplace. Geography contributes to and is involved in ACS sampling, data collection, weighting, and data tabulation activities. The place of work geographies are derived from the respondents’ answers to the survey and are not based on where the surveys are sent, which helps reduce the possibility of disclosure. The place of work geographies for this tabulation are counties.

The ACS collects respondents’ total wages and not wages by job. Respondents with more than one job are identified by their primary occupation and industry, but their total wages and hours worked may be based on more than one job. If hourly earnings differ between a respondent’s primary and secondary job, this introduces a possible source of measurement error because the ACS-CWI regression model attributes all wages to the primary occupation and industry.¹⁰ The estimated coefficients for specific occupations or industries in which multiple job holding is more common, such as firefighters, emergency medical technicians, and dental hygienists, may be particularly affected. (Teachers are another occupation with relatively high rates of multiple job holding, but they are excluded from the estimation of the ACS-CWI.)

Top

The sample

The estimation sample has been constructed to ensure that the noneducator population is comparable to teachers with respect to their sensitivity to housing costs and local amenities.

The sample consists of people who

1. Are employed in private for-profit, private nonprofit, or government industries (excludes unemployed and self-employed or unpaid family workers).
2. Are between the ages of 18 and 80.
3. Work at least 20 but fewer than 90 hours per week.
4. Worked between 27 and 52 weeks in the past 12 months.
5. Have at least a bachelor’s degree.
6. Have annual wage and salary earnings above $5,000.
7. Work in one of the 50 states or Washington, D.C.
8. Do not work in the elementary or secondary education industry and are not education administrators, teachers, librarians, teaching assistants, or miscellaneous other education workers (see Taylor and Fowler 2006).

Individuals who are self-employed are excluded because their reported wage and salary earnings may not represent the market value of their time. Individuals who report working less than half time or for more than 90 hours a week are excluded, as are workers under the age of 18 and over the age of 80 and workers without a bachelor’s degree, because they are unlikely to be comparable to teachers. Individuals who report earning less than $5,000 in the past year (despite working at least half time) are excluded because their responses are improbable, at least in the context of fully compensated work. Workers for whom the Census Bureau has to allocate key attributes of their job (e.g., wages, occupation, industry, hours worked) from donor sample cases are excluded for statistical reasons. Finally, individuals employed outside the United States are excluded because their wages may represent compensation for foreign travel or other working conditions not faced by domestic workers.

The estimation sample does not include anyone who has a teaching or education administration occupation or who is employed in the elementary or secondary education industry.¹¹ Such persons are excluded from the analysis because it is conceptually important that the wages and salaries reflected in the ACS-CWI are outside of school district control (i.e., are independent of school district hiring practices or the influences of unionization).

All other occupations and industries have been included in the analysis. Retaining all noneducator occupations and industries greatly increases the sample size and reduces the noise in the estimates of local wage levels. Furthermore, as discussed in Taylor and Fowler (2006), a CWI is not influenced by differences in pay levels or job characteristics from one occupation or industry to another because it is based on demographically adjusted pay differentials within each occupation or industry. Without evidence that differences in job descriptions imply differences in tastes for housing or local amenities, there would be no gain from restricting the sample to a subset of occupations or industries.

Top

The variables

The dependent variable is the log of reported wage and salary earnings in the past year. Ideally, the dependent variables would reflect total compensation and include not only wages and salaries but also fringe benefits. Unfortunately, survey respondents are not asked about the value of their fringe benefits¹² (if any) so more complete data on worker compensation are not available.¹³

The independent variables describe the workers and the jobs they held. The worker characteristics include continuous variables for age, age squared, and the number of hours worked per week; a categorical variable for weeks worked per year; and indicator variables for gender, race, English-speaking ability, educational attainment, and undergraduate degree field.¹⁴ The model includes the interaction between sex and age to allow for the possibility that men and women have different career paths and, therefore, different age-earnings profiles.¹⁵ The job characteristics include indicator variables for occupation and industry for each year. This specification allows wages to rise (or fall) more slowly in some occupations or industries than they do in others. Such flexibility is particularly important because the analysis period includes the period immediately after the “Great Recession,” and some industries and occupations are recovering more slowly than others.

Finally, the regression includes indicator variables for each labor market area. The labor market indicators¹⁶ capture the effect on wages of all market-specific characteristics, including the price of housing, the crime rate, and the climate.¹⁷

The regression model is produced using ACS person data collected from 2013 through 2015. The models are produced for “labor market” geographic areas. Data are produced for 1,570 labor markets in the United States. The 1,570 labor market areas are based on counties or county equivalents and an individual’s reported place of work, not place of residence. As such, it is possible for an individual to live in one county but work in another. Each individual’s compensation contributes to the estimate of the prevailing wage in his or her place of work, regardless of his or her place of residence.

NCES requested a minimum number of sample cases to help improve data quality and prediction accuracy. Each labor market must contain at least 100 unweighted universe cases per county based on data collected from 2013 through 2015. Those that do not meet the minimum are successively combined with the neighboring county within the same state that has the fewest cases until every labor market has at least 100 unweighted universe cases. The neighboring counties are determined by a county adjacency file that was created from the U.S. Census Bureau’s 2015 TIGER geographic shapefiles. Counties must share a least one mile of border to be considered “neighboring.”

Top

The estimation

Table 1.B presents selected coefficients from generalized least-squares estimation of the ACS-CWI wage model. The estimation sample contains 1,391,896 survey respondents, and the regression is weighted using the person weights provided by the Census Bureau. Replicate weights are used to incorporate known sampling error into adjusted standard errors for the coefficients.¹⁸ ACS implements a replication method for variance estimation. An advantage of this method is that the variance estimates can be computed without consideration of the form of the statistics or the complexity of the sampling or weighting procedures, such as those being used by the ACS.¹⁹ The ACS replicate weights were applied to the CWI model using SAS PROC SURVEYREG to help account for known sampling error.²⁰

As the table illustrates, the estimated model is consistent with reasonable expectations about labor markets. Wages and salaries increase with the amount of time worked per week and the number of weeks worked per year. Wages and salaries also rise as workers get older, but the increase is more rapid for men than for women (perhaps because age is not as good an indicator of experience for women as it is for men). Workers with advanced degrees systematically earn more than workers with bachelor’s degrees. Non-Hispanic whites systematically earn more than comparable individuals from other racial or ethnic groups. Workers who do not speak English earn substantially less than other college-educated workers, all other things being equal.

Top

The construction of market-level index values

The predicted wage level in each labor market area captures systematic variations in labor earnings while controlling for demographics, industry and occupation mix, and amount of time worked. Using the coefficient estimates from the regression analysis, the researchers predicted the log wage and salary that a person with average characteristics would earn in each location.²¹ Using those local predictions, they also predicted the log wage and salary for each state and for the nation as a whole.²² The predicted wage level for each location is the exponent of the corresponding predicted log wage and salary. In turn, the ACS-CWI for each location is the predicted wage level for that location divided by $62,655, which was the national average predicted wage. The ACS-CWI ranges from 0.649 in rural Montana to 1.377 in New York County (Manhattan), New York.

When predicting the log wage and salary for each local labor market, the researchers also calculated the standard error of the prediction, incorporating both model and survey error. The standard error for the ACS-CWI in each local labor market area is calculated by dividing one standard error of the predicted wage by the national average predicted wage.²³ Among the 1,570 local labor market areas in the United States, the standard error for the ACS-CWI ranges from 0.004 in Los Angeles, California, to 0.160 in rural Colorado.

Top

The construction of LEA-level index values

As a general rule, the ACS-CWI for a school district is the ACS-CWI for the corresponding county. However, some LEAs span multiple counties. In those cases, the ACS-CWI for the LEA is a population-weighted average of the ACS-CWIs for each county in the LEA. The weights reflect the shares of school-age children in each LEA who live in each county.²⁴ For example, Abernathy Independent School District (ISD) straddles the border between Hale County, Texas, and Lubbock County, Texas. The U.S. Census Bureau estimates that 71 percent of Abernathy ISD’s students live in Hale County and the remaining 29 percent live in Lubbock County. Thus, because the ACS-CWI for Hale County is 0.813 and the ACS-CWI for Lubbock County is 0.866, the ACS-CWI for Abernathy ISD is 0.828 (0.71*0.813 + 0.29*0.866).

---

Table B.1. Selected coefficient estimates from the ACS model of log annual wage and salary income: 2013, 2014, and 2015

Variable	Estimate	Standard error
Usual hours worked per week (in logs)	0.9166	0.0030	**
50 to 52 weeks worked per year	0.5545	0.0045	**
48 to 49 weeks worked per year	0.4486	0.0063	**
40 to 47 weeks worked per year	0.3045	0.0058	**
27 to 39 weeks worked per year	0.0000	0.0000
Female	0.3105	0.0137	**
Male	0.0000	0.0000
Age	0.0847	0.0005	**
Age squared	-0.7861	0.0056	**
Female* age	-0.0161	0.0007	**
Female* age squared	0.1275	0.0074	**
Not an English speaker	-0.5075	0.0242	**
Bachelor’s degree	-0.2696	0.0028	**
Master’s degree	-0.1561	0.0027	**
Professional degree	-0.0519	0.0038	**
Doctoral degree	0.0000	0.0000
Hispanic	-0.1077	0.0022	**
White	0.0617	0.0039	**
Black or African American	-0.0734	0.0043	**
American Indian/Alaska Native	-0.0195	0.0089	*
Asian	-0.0390	0.0041	**
Pacific Islander	-0.0395	0.0209
Some other race	-0.0111	0.0079
Two or More Races	0.0000	0.0000
Undergraduate degree field indicators?	Yes
Industry* year indicators?	Yes
Occupation* year indicators?	Yes
Labor market indicators?	Yes
Number of observations	1,391,896

---

** Indicates that the coefficient is significantly different from zero at the 1 percent level.
* Indicates a coefficient that is significantly different from zero at the 5 percent level.
NOTE: Due to Office of Management and Budget guidelines, respondents are asked separately about race and Hispanic origin. Respondents who   identify as Hispanic will also have a race identified. For this data, 73 percent of Hispanic respondents identified as White, 15 percent identified as Some Other Race, 5 percent identified as Two or More Races, 3 percent identified as Asian, 3 percent identified as Black or African American, and the remainder identified as American Indian/Alaska Native or Pacific Islander.
SOURCE: U.S. Department of Commerce, U.S. Census Bureau, special tabulation.

---

Top

Changes from the initial CWI

Although the ACS-CWI is modeled after the baseline specification used to estimate the initial CWI, there are key differences beyond simple updating. Some of the differences arise from differences between the decennial census and the ACS. Others arise from the differences between restricted-use and public-use data files. Still other differences arise from enhancements in the modeling technique.

The most obvious difference is the geography: the ACS-CWI provides labor cost estimates for 1,570 labor market areas based on counties and clusters of counties. In contrast, the initial CWI was based on 800 census-defined place of work areas. The increased geographic detail in the ACS-CWI—which is only possible with the restricted-use data—provides better representations of local labor market conditions than were possible with the initial CWI.

The improved geographic detail of the ACS-CWI also facilitates the construction of more finely grained index values for LEAs. With the initial CWI, school districts were matched to the labor market areas according to the county of record. Thus, an LEA that spanned more than one county typically had the CWI of the county where the head office was located. In contrast, the ACS-CWI for an LEA is a population-weighted average of the ACS-CWIs for each county in the LEA. Because LEAs may cross county lines, this change means that it is no longer necessarily the case that all the LEAs in a metropolitan area have the same index values. In addition, changes from one year to the next in the ACS-CWI for a specific LEA could now arise from changes in the population weights as well as changes in the wage levels.

Differences in the survey questions between the decennial census and the ACS have led to changes in the specification of the hedonic wage model. The ACS measure of weeks worked per year is categorical rather than continuous (as was the case with the decennial census) so the wage model changed accordingly. The ACS also contains data on the undergraduate degree field that were not available with the decennial census. As was recommended by the expert panel convened by NCES to review the initial CWI, indicators for degree fields have been incorporated into the ACS hedonic wage model. Finally, the ACS collects data on occupations and industries that are based on more recent coding schemes than those used in the 2000 Census; those updated codes are used in the estimation of the ACS hedonic wage model.

Additional changes in the specification represent enhancements in modeling technique. The revised model includes the interaction between sex and age to allow for age-earnings profile differences between men and women. It also includes indicators for whether or not the worker is Hispanic or speaks English. Because the ACS model incorporates data from multiple years, it also incorporates the interaction between year indicators and the occupation or industry fixed effects. Whereas the initial hedonic wage model included random effects for states in the estimation but did not include those random effects in the construction of the wage predictions, the ACS model does not include random effects at either stage (estimation or prediction), making the wage predictions more consistent with the underlying model. Unlike the initial model, the ACS model also incorporates replicate weights, which is consistent with the recommendations of the expert review panel for the initial CWI.

Top

User Guide

CWI estimates for three geographic levels

• The school district ACS-CWIs are created for each local education agency (LEA) in the FY15 Title I Database. The ACS-CWI for each LEA is either the ACS-CWI for the corresponding county or a school-age child population weighted average of the ACS-CWIs for the corresponding counties when the LEA straddles county lines.
• The county ACS-CWIs are created for each of the 3,143 counties or county equivalents in the United States. The 1,570 labor market areas used to construct the ACS-CWI are based on counties or county equivalents.²⁵ The ACS-CWI ratio is created by dividing the exponent of the log wage of the labor market area by the national average wage ($62,6455.
• The state ACS-CWIs are based on the state’s average predicted log wage for each state (including Washington, D.C.). The state average predicted log wage is a weighted average of the county-level predicted log wages, where the weights are the local employment shares among the college graduates in the regression sample adjusted for differences in sampling weights. The state’s ACS-CWI ratio is the exponent of the state’s average predicted log wage divided by the national average wage ($62,655).

Top

Using the index to make geographic adjustments

One important reason for the development of the ACS-CWI is to enable more meaningful comparisons across school districts. To normalize dollar amounts and make them comparable, divide the dollar amounts by the district-level ACS-CWI, which are already normalized to the national average wage. For example, suppose one wished to make an adjustment to current expenditure data from the Elementary and Secondary Information (ELSI) system for the 2013–14 school year. The ACS-CWI for the Los Angeles Unified School District (LAUSD) is 1.129. So, the $6,137 total current expenditures on salary per pupil in LAUSD in 2013–14,²⁶ when normalized, are equal to $5,436 ($6,137 / 1.129). In comparison, the ACS-CWI for Palm Beach County (Florida) School District (PBCSD) is 0.957, and the 2013–14 total current expenditures on salary per pupil were $5,433. Normalized to reflect the lower cost of hiring in this area, they are the equivalent of $5,677 ($5,433 / 0.957). In other words, even though LAUSD spent more than PBCSD in nominal terms, once the two dollar figures were adjusted for the difference in purchasing power between the two districts, PBCSD effectively spent $241 more per pupil than did LAUSD

Top

Geographic adjustments applied to state aid

Since one of the great advantages of the ACS-CWI is that it is outside of school district control, another application of the ACS-CWI is to adjust state aid to a school district for differences in wages. For example, consider a program intended to provide an additional $100 per pupil, adjusted for geographical variations in the cost of education. The ACS-CWI for New Rochelle, New York, in 2015 is 1.16, or 16 percent higher than the national average; the ACS-CWI for Buffalo, New York, is 0.902, or 10 percent lower than the national average. Therefore, to receive the same increase in purchasing power as a $100 increase in Buffalo City School District, New Rochelle City School District would need to receive $128.94 ($100*(1.163 / 0.902)).

Top

Standard errors

The standard error of each predicted wage level indicates the precision with which it was measured. Dividing one standard error of each predicted wage by the national average wage ($62,655) yields the standard error of the ACS-CWI, which ranges from 0.004 in Los Angeles, California, to 0.160 in rural Colorado.

Top

Glossary

American Community Survey (ACS)
An ongoing survey conducted by the U.S. Census Bureau. It has replaced the decennial census as the primary source of detailed demographic information about the U.S. population.

Elementary/Secondary Education
Programs providing instruction, or assisting in providing instruction, for students in grades preK–12 and ungraded programs.

Fiscal Year (FY)
The 12-month period to which the annual operating budget applies. At the end of the fiscal year, the agency determines its financial condition and the results of its operations.

Labor Market
An economically integrated area within which individuals can reside and find employment within a reasonable distance or can readily change jobs without changing their place of residence (as defined by the U.S. Bureau of Labor Statistics: https://www.bls.gov/lau/laufaq.htm#Q06). Labor markets are the units of analysis for the Comparable Wage Index study. They are geographic regions (either individual counties or groupings of neighboring counties) that have the same value for a comparable wage index.

Local Education Agency (LEA)
Often called a school district; primary responsibility is to operate public schools or to contract for public school services.

Top

---

References

Alexander, C.D., Gronberg, T., Jansen, D., Keller, H., Taylor, L.L., and Treisman, P.U. (2000). A Study of Uncontrollable Variations in the Costs of Texas Public Education (summary report prepared for the 77th Texas Legislature). Austin, TX: Charles A. Dana Center, University of Texas at Austin.

Barrow, S. (1994). Cost of Education Differentials Across the States (NCES 94-05). U.S. Department of Education. Washington, DC: National Center for Education Statistics Working Paper.

Brazer, H.E., and Anderson, A.P. (1975). A Cost Adjustment Index for Michigan School Districts. In E. Tron (Ed.), Selected Papers in School Finance, 1975 (pp. 23–81). Washington, DC: U.S. Office of Education.

Cohen, C., and Johnson, F. (2004). Revenues and Expenditures for Public Elementary and Secondary Education: School Year 2001-02 (NCES 2004-341). U.S. Department of Education. Washington DC: National Center for Education Statistics.

Chambers, J.G. (1997). Measuring Inflation in Public Schools Costs (NCES 97-43). U.S. Department of Education. Washington, DC: National Center for Education Statistics Working Paper.

Chambers, J.G. (1998). Geographic Variations in Public Schools Costs (NCES 98-04). U.S. Department of Education. Washington, DC: National Center for Education Statistics Working Paper.

Duncombe, W., Lukemeyer, A., and Yinger, J. (2003). Financing an Adequate Education: A Case Study of New York. In W.J. Fowler Jr. (Ed.), Developments in School Finance: 2001–02 (NCES 2003-403) (pp. 127–153). U.S. Department of Education, National Center for Education Statistics. Washington, DC: U.S. Government Printing Office.

Fowler, W.J., Jr., and Monk, D.M. (2001). A Primer for Making Cost Adjustments in Education (NCES 2001-323). U.S. Department of Education. Washington, DC: National Center for Education Statistics.

Geverdt, D. (2015). Education Demographic and Geographic Estimates Program (EDGE): Locale Boundaries User’s Manual (NCES 2016-012). U.S. Department of Education. Washington, DC: National Center for Education Statistics. Retrieved July 5, 2017 from https://nces.ed.gov/pubsearch.

Goldhaber, D. (1999). An Alternative Measure of Inflation in Teacher Salaries. In W.J. Fowler Jr. (Ed.), Selected Papers in School Finance, 1997–99 (NCES 1999-334) (pp. 29–54). U.S. Department of Education. Washington, DC: National Center for Education Statistics.

Guthrie, J.W, and Rothstein, R. (1999). Enabling “Adequacy” to Achieve Reality: Translating Adequacy into State School Finance Distribution Arrangements. In H.F. Ladd, R. Chalk, and J.S. Hansen (Eds.), Equity and Adequacy in Education Finance (pp. 209–259). Washington, DC: National Academy Press.

Hanushek, E.A. (1999). Adjusting for Differences in the Costs of Educational Inputs. In W.J. Fowler Jr. (Ed.), Selected Papers in School Finance, 1997–99 (NCES 1999-334) (pp. 13–27). U.S. Department of Education. Washington, DC: National Center for Education Statistics.

Hirsch, B.T., Husain, M.M., and Winters, J.V. (2016). Multiple Job Holding, Local Labor Markets, and the Business Cycle. IZA Journal of Labor Economics, 5: 4.

Rothstein, R., and Smith, J.R. (1997). Adjusting Oregon Education Expenditures for Regional Cost Differences: A Feasibility Study (submitted to the Confederation of Oregon School Administrators). Sacramento, CA: Management Analysis & Planning Associates, LLC.

Smith, T., Porch, R., Farris, E., and Fowler, W. (2003). Effects of Energy Needs and Expenditures on U.S. Public Schools (NCES 2003-108). U.S. Department of Education. Washington DC: National Center for Education Statistics.

Stinebricker, T.R. (2002). An Analysis of Occupational Change and Departure from the Labor Force: Evidence of the Reasons Teachers Leave. Journal of Human Resources, 37(1): 192–216.

Stoddard, C. (2005). Adjusting Teacher Salaries for the Cost of Living: The Effect on Salary Comparisons and Policy Conclusions. Economics of Education Review, 24(3): 323–339.

Taylor, L.L. (2006). Comparable Wages, Inflation, and School Finance Equity. Education Finance and Policy, 1(3): 349–371.

Taylor, L.L. (2015). When Equality is not Equity: Regional Cost Differences and the Real Allocation of Educational Resources. In A.H Normore, P.A.L Ehrensal, P.F. First, and M.S. Torres (Eds.), Legal Frontiers in Education: Complex Law Issues for Leaders, Policymakers and Policy Implementers (pp. 247–266). Bingley, UK: Emerald Group Publishing Limited.

Taylor, L.L., Alexander, C.D., Gronberg, T.J., Jansen, D.W., and Keller, H. (2002). Updating the Texas Cost of Education Index. The Journal of Education Finance, 28(2): 261–284.

Taylor, L.L., and Fowler, W.J., Jr. (2006). A Comparable Wage Approach to Geographic Cost Adjustment (NCES 2006-321). U.S. Department of Education. Washington, DC: National Center for Education Statistics.
 
Taylor, L.L., and Keller H. (2003). Competing Perspectives on the Cost of Education. In William J. Fowler Jr. (Ed.), Developments in School Finance: 2001–02 (NCES 2003-403) (pp. 111–126). U.S. Department of Education. Washington, DC: National Center for Education Statistics.

Taylor, L.L., Chambers, J., and Robinson, J.P. (2004). A New Geographic Cost of Education Index for Alaska: Old Approaches with Some New Twists. The Journal of Education Finance, 30(1): 51–78.

Top

---

¹ The CWI was initially developed by Lori L. Taylor at the Bush School of Government and Public Service, Texas A&M University, and William J. Fowler Jr. at NCES. Taylor’s research was supported by a contract with NCES. The complete description of the research is provided in the NCES Research and Development report A Comparable Wage Approach to Geographic Cost Adjustment (NCES 2006-321).
² For example, see Barrow (1994), Brazer and Anderson (1983), Chambers (1998), Fowler and Monk (2001), Goldhaber (1999), Taylor and Fowler (2006), and Taylor and Keller (2003).
³ The ACS asks respondents questions related to income in the past 12 months. If respondents report receiving income from “[w]ages, salary, commissions, bonuses, or tips from all jobs,”they are asked to “[r]eport [the] amount before deductions for taxes, bonds, dues, or other items” for the 12 months prior to the response date. Any future reference to “wage(s)” or “wage(s) and salary(ies)” in this documentation includes all of the income items contained in the questions.
⁴ For example, see Alexander et al. (2000), Goldhaber (1999), Guthrie and Rothstein (1999), Rothstein and Smith (1997), Stoddard (2005), Taylor (2006), Taylor (2015), and Taylor et al. (2002).
⁵ For the full PUMA definition, go to https://www.census.gov/programs-surveys/acs/technical-documentation/pums/about.html.
⁶ The U.S. Bureau of Labor Statistics provides wage and employment data for counties, metropolitan areas, and nonmetropolitan areas. Metropolitan areas “consist of one or more counties (or towns and cities in New England) and contain a core area with a substantial population that has a high degree of economic and social integration with the surrounding areas.” For more information, go to https://stats.bls.gov/bls/blswage.htm.
⁷ The panel made the following recommendations:
          • NCES should annually produce and release geographic adjustment factors for educational expenditures (GAFEEs).
          • GAFEEs should (a) support both cross-sectional and temporal comparisons and (b) be accompanied by detailed           documentation of the data sources, methodology, and statistical uncertainties in their values.
          • GAFEEs should be based on 1 year of restricted-access data produced by ACS and be reported as rapidly as           possible once data become available.
          • GAFEEs should be calculated using a modification of the current CWI base-year methodology that (a) accounts           properly for the state-level random effects and estimates these effects correctly, (b) does not exclude industry and           occupation classifications related to education, (c) includes the ACS degree field variable as a predictor, and (d)           properly includes weights in the mixed model as well as in mean calculations across geographic localities.
          • GAFEEs should be reported to no more than two decimal places (x.yy).
⁸ For a more detailed discussion of the advantages and disadvantages of the CWI approach, see Taylor (2015) and Taylor and Fowler (2006).
⁹ The ACS-CWI is also subject to nonsampling error such as nonresponse error, coverage error, measurement error, and processing error. For more information on ACS methodology, go to https://www.census.gov/programs-surveys/acs/.
¹⁰ Hirsch, Husain and Winters (2016) “guestimate” that the difference in wages between primary and secondary jobs is approximately 6 percent. However, their estimates include all levels of educational attainment and include teachers, who are among the most common types of workers to hold secondary jobs. We have no data on the extent of wage differentials among college graduates who are not educators (the CWI population).
¹¹ The expert panel on the initial CWI released by NCES recommended that the geographic cost adjustment factors should not exclude industry and occupation classifications related to education, arguing that the original CWI was not very sensitive to the occupation and industry exclusions and that including the education sector would increase the sample size. However, the fact that the CWI reflects wage differences outside of education is crucially important conceptually and one of the major reasons why this approach appeals to researchers. Including educators in the estimation sample would fundamentally change the nature of the wage index.
¹² The only question about fringe benefits included in the ACS was a yes/no question that did not differentiate between health insurance tied to the respondent’s current job and health insurance tied to a family member’s job or to the respondent’s previous job. The question asks, “Is this person covered by any of the following types of health insurance or health coverage plans: Insurance through a current or former employer or union (of this person or another family member?).”
¹³ To the extent that fringe benefits differ systematically across industries or occupations, they will be captured by regression fixed effects and have no impact on the ACS-CWI. However, as discussed in Taylor and Fowler (2006), systematic differences in benefits across states—such as those that might arise if workers take more of their compensation in the form of benefits in states with income tax than they do in states without income tax—could bias the ACS-CWI.
¹⁴ The degree fields are aggregated to the two-digit level and include the following: agricultural sciences; environmental sciences; architecture; area ethnic and civilization studies; communications; communication technologies; computer and information sciences; cosmetology and culinary arts; education; engineering; engineering technologies; languages; family and consumer sciences; prelaw and legal studies; literature; liberal arts and humanities; library science; biological sciences; mathematics and statistics; military technologies; multidisciplinary studies; physical fitness, parks, recreation, and leisure; philosophy and religious studies; theology; physical and related sciences; applied biotechnology; psychology; criminal justice and fire protection; public administration, public policy, and social work; social science; construction services; electrical and mechanical repairs and technologies; precision production; transportation sciences and technologies; visual and performing arts; healthcare; business; and history.
¹⁵ This is a change from Taylor and Fowler (2006) and represents an enhancement in the modeling. The estimation suggests that the age-earnings profiles of men and women are different in statistically and analytically meaningful ways.
¹⁶ The labor market indicators, which are also known as labor market fixed effects, capture both measurable and unmeasurable characteristics of labor markets.
¹⁷ In contrast to the current ACS-CWI, the baseline model for the original CWI also included random effects for states. Although the expert panel recommended that the predicted wages used to generate the CWI incorporate the average state-level random effects, this would have been particularly consequential for index values in metropolitan areas that straddle state lines (such as Kansas City or New York City). The ACS-CWI labor markets do not cross state lines, which removes the need for any state-level random effects.
¹⁸ Replicate weights were used to adjust the standard errors of the CWI for survey error in addition to model error. The use of replicate weights has no effect on the CWI values themselves. For more information on ACS design and methodology, go to https://www.census.gov/programs-surveys/acs/methodology/design-and-methodology.html. Information on replicate weights and variance estimation can be found in chapters 11 and 12. For more information on ACS variance estimation, see chapter 12 of American Community Survey Design and Methodology (https://www2.census.gov/programs-surveys/acs/methodology/design_and_methodology/acs_design_methodology_ch12_2014.pdf).
¹⁹ Since the start of the survey, the ACS has used the Successive Differences Replication (SDR) method to calculate estimates of variance.
²⁰ The model incorporated ACS replicate weights to estimate sampling errors of the estimators using SAS PROC SURVEYREG. For more information on the SAS SURVEYREG procedure, go to https://support.sas.com/documentation/cdl/en/statug/63347/HTML/default/viewer.htm#statug_surveyreg_sect001.htm.
²¹ Formally, the predicted log wage level in each labor market area (i.e., the least-squares mean or population marginal mean) is the mean wage level that would be expected from a balanced design holding all continuous variables at their means and all indicator variables at their population frequencies.
²² At the state and national levels, the predicted log wage is a weighted average of the local predicted log wages, where the weights are the local employment shares among the college graduates in the regression sample, adjusted for differences in sampling weights.
²³ In other words, the dollar value of one standard error of the predicted wage divided by $62,655 is the ACS-CWI standard error. The dollar value of one standard error of the predicted wage is calculated by adding one standard error of the predicted log wage to the log wage, taking the exponent, and then subtracting the predicted wage.
²⁴ Data on the population ages 5–17 come from the U.S. Census Bureau’s Small Area Income and Poverty Estimates (SAIPE) for school districts for income year 2013. Of the estimates available for county pieces of school districts, the shares of the population ages 5–17 are most correlated with the shares of teachers.
²⁵ Contiguous counties in sparsely populated areas have been aggregated into labor market areas containing at least 100 survey respondents that meet the estimation sample criteria.
²⁶ U.S. Department of Education, National Center for Education Statistics, Common Core of Data (CCD), "Local Education Agency (School District) Universe Survey Directory Data," 2014–15 v.1a; "School District Finance Survey (F-33)," 2013–14 (FY 2014) v.1a.
²⁷ Current expenditures for salary comprise expenditures for the day-to-day operation of schools and school districts for public elementary and secondary education. General administration expenditures and school administration expenditures are included in current expenditures. Expenditures associated with repaying debts and capital outlays (e.g., purchases of land, school construction, and equipment) are excluded from current expenditures. Programs outside the scope of public preK–12 education, such as community services and adult education, are excluded from current expenditures. Payments to private schools and charter schools outside of the school district are also excluded from current expenditures.