605 – Imputation of Missing Data
Methods to Impute Household Income in the National Crime Victimization Survey
Andrew Moore
RTI International
Marcus E. Berzofsky
RTI International
Darryl Creel
RTI International
Tommy Holder
RTI International
This paper presents an analysis to determine the best approach to impute household income in a large, national survey. The National Crime Victimization Survey (NCVS) is a multi-stage, rotating panel design of households sponsored by the Bureau of Justice Statistics. The NCVS is designed to allow estimation of annual counts and rates of criminal victimization for both the population as a whole as well as subgroups of interest. Due to the strong relationship between socioeconomic status and criminal victimization, household income is a key characteristic often chosen to partition the population. Like many other surveys, the NCVS suffers from a high rate of missing data on household income and, while weighting is used to adjust for unit nonresponse, nothing is currently done to address item nonresponse. Failure to properly account for this missing data can lead to a loss of power and potentially biased estimates. We evaluate several potential approaches to imputing missing income data in the NCVS and assess each option on several criteria including consistency, variability, and usability. Final imputed results are also compared to the ACS for external validation of the chosen method.