Imputation Methods for Surveys: A Demonstration of the Impute Procedure in Sudaan

• Activity Number: 444
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Kimberly Ault*+
• Companies: RTI International
• Keywords: Imputation ; weighted sequential hot deck ; cell mean ; linear regression ; logistic regression ; SUDAAN

Abstract

Survey researchers commonly face issues with missing data during analysis. Ad-hoc missing data methods, such as complete-case analysis, are easy to implement but they have well-known disadvantages of potentially yielding biased results and having reduced power due to deleting observations with missing values. The impact of missing data on survey estimates depends on the pattern of missing data, percent of missing data, and parameters to be estimated. Since most surveys experience some missing data, survey data analysis should account for missing data. Weighting adjustments may compensate for non-coverage and total nonresponse, but imputation methods that assign values for missing responses are more commonly used to compensate for item nonresponse. The IMPUTE Procedure in SUDAAN v11 performs the following imputation methods: weighted sequential hot deck imputation, cell mean imputation, linear regression for continuous variables, and logistic regression for binary variables. Data from public use files for the 2004 National Health Interview Survey are used to provide examples of each imputation method and the advantages and disadvantages of each.