Imputing NCVS Income

• NCJ Number: 248563
• Author(s): Marcus Berzofsky Dr.PH; Darryl Creel M.S.; Andrew Moore M.S.; Hope Smiley-McDonald Ph.D.; Chris Krebs Ph.D.
• Date Published: January 2014
• Length: 88 pages
• Grant Number(s): 2011-NV-CX-K068
• Sponsoring Agency: Bureau of Justice Statistics (BJS)
• Corporate Author: RTI International
• Publication Format: Document; Document (Online)
• Language: English
• Country: United States of America
• Note: This document was prepared using Federal funds provided by the U.S. Department of Justice under Cooperative Agreement number 2011-NV-CX-K068. Any opinions and conclusions expressed herein are those of the author(s) and do not necessarily represent the views of the Bureau of Justice Statistics and the U.S. Department of Justice. The BJS Project Managers were Lynn Langton, BJS Statistician, and Michael Planty, BJS Victimization Unit Chief.

Annotation

This report describes the problem of missing income data in the National Crime Victimization Survey (NCVS), details and assesses several potential approaches to imputing the missing data, and makes recommendations on the use of the single imputation hot deck method for imputation of NCVS household income data, or any other variables, when there is item nonresponse.

Abstract

The National Crime Victimization Survey (NCVS) is an important source of information on criminal victimization in the United States. Each year, data are obtained from a nationally representative sample of about 40,000 households comprising nearly 75,000 persons on the frequency, characteristics, and consequences of criminal victimization. The survey enables the Bureau of Justice Statistics (BJS) to estimate the likelihood of experiencing rape, sexual assault, robbery, assault, theft, household burglary, and motor vehicle theft victimization for the population as a whole, as well as for segments of the population. Virtually all data collection efforts experience the challenge of missing data, and the NCVS is no exception. The NCVS is a panel data collection effort with three possible types of nonresponse: 1) unit, 2) wave, and 3) item. BJS uses weight adjustments to account for unit and wave nonresponse; however nothing is done to address item nonresponse. This missing data may result in a potentially biased estimate. This is particularly problematic when it comes to the measurement of household income, one of the NCVS items that has historically suffered from high rates of nonresponse. As BJS works to continually improve the utility of the NCVS for understanding changes in crime and its correlates over time, the high levels of missing income data has to be addressed to accurately establish associations between victimization and socioeconomic factors, including income. Continuing to omit the nonresponse income data may lead to false interpretations of the data collected and biased survey results. It is recommended that imputation methods be used to impute income data for the longitudinal data. Although alternative methods for addressing missing data during analysis exist, imputation is a common and trusted approach.