Combined Methods for Imputing School Variables in Principal Data Files

• Activity Number: 427
• Type: Contributed
• Date/Time: Tuesday, August 2, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Yan Wang*+ and Matthew Doyle
• Companies: American Institutes for Research and American Institutes for Research
• Keywords: School and Staffing Survey ; cold-deck imputation ; regression impuation ; school nonresponse ; data utility ; principal data files

Abstract

Due to school nonresponse, three school variables-school level, school enrollment, and urbanicity-have about 10 percent missing values in the 1987-88, 1990-91, and 1993-94 School and Staffing Survey principal data files. To create fully imputed files to meeting reporting standards on key variables, a combination of imputation procedures were used for these variables based on the availability of auxiliary information. The first imputation method was to use existing sampling variables - the school stratum codes for public and private schools to extract school level values. The second method was a cold-deck imputation using universe data (Common Core Data school files and Private School Universe Survey data) to fill in school enrollment values when possible. Thirdly, as a last resort, a set of survey variables was selected to predict imputed variables and regression imputation was used to fill the rest of the missing values on school enrollment and urbanicity. The utility of the imputed data was evaluated by comparing the original data files with the imputed files on the weighted cell counts, standard errors, and pairwise correlations and multivariate associations.