Characteristics of Public School Districts in the United States:
NCES 2009-320
June 2009

Appendix B. Methodology and Technical Notes—Data Processing and Imputation

The Census Bureau used both central processing and headquarters staff to check returned questionnaires, capture data, and implement quality control procedures. Questionnaires that had a preliminary classification of a complete interview were submitted to a series of computer edits consisting of a range check, a consistency edit, and a blanking edit.6 After these edits were run and reviewed by analysts, the records were put through another edit to make a final determination as to whether the case was eligible for the survey and whether sufficient data had been collected for the case to be classified as a complete interview.

After the final edits were run, cases with "not-answered" values for items remained. Values were imputed using a two stage process. In the first stage, items were imputed with a valid response using data either from the sample frame, other items in the same SASS questionnaire, or another questionnaire associated with the same school or school district. In addition, data were ratio adjusted in some circumstances so that items were consistent with one another. In the second stage, donor-respondent methods, such as hotdeck imputation, were used. If no suitable donor case could be matched, the few remaining items were imputed with a mean or mode from groups of similar cases. After each stage of imputation, computer edits were run again to verify that the imputed data were consistent with the existing questionnaire data. If that was not the case, an imputed value was blanked out by one of these computer edits due to inconsistency with other data within the same questionnaire or because it was out of the range of acceptable values. In these situations, Census Bureau analysts looked at the items and tried to determine an appropriate value. Imputation flags, indicating which imputation method was used, were assigned to each imputed survey variable. For further information, see the section on data processing and imputation in the Documentation for the 2007–08 Schools and Staffing Survey (Tourkin et al. forthcoming).

6 Blanking edits delete answers to questions that should not have been filled in (e.g., if a respondent followed a wrong skip pattern).