Small Area Estimation Alternatives for the National Crime Victimization Survey

The National Crime Victimization Survey (NCVS) has provided annual estimates of the number of victimizations for several types of crime since 1972, with an almost exclusive focus on national rates. Most of the programs to prevent or reduce crime are implemented locally, however. To respond to a resulting interest in subnational statistics, the Bureau of Justice Statistics (BJS) has been recently supporting research on a variety of approaches to produce subnational estimates. In this paper, we report on the potential application of model-based small area estimation methods based on the NCVS and auxiliary data, particularly the FBI’s Uniform Crime Reports, using empirical best linear unbiased estimation (EBLUP). We compare a time-series model introduced by Rao and Yu to a new variant, termed here the dynamic model. We will also indicate how the small area approach might be integrated with other approaches that BJS is currently considering, including possible expansion of the NCVS sample size to augment the survey’s capacity to produce direct estimates for some or all states.