Online Program

Combining Administrative and Survey Data to Estimate County-Level Crime Rates: A Comparison of Two Methods (304874)

Keywords: National Crime Victimization Survey, National Incident-Based Reporting System, Small-Area Estimation, Uniform Crime Reports, Administrative Data, Hierarchical Bayesian Modeling, Data Fusion

The two main crime measures in the US are the Uniform Crime Reports (UCR) sytem, which collects data on police-reported crimes, and the National Crime Victimization Survey, which is a nationally-representative survey that asks about victimization, whether or not the crime was reported to the police. The UCR system has good national coverage but by definition does not cover unreported crime, while the NCVS collects data on all crime but can only provide national estimates.

We compare two methods of synthesizing the UCR and NCVS data to produce estimates for small areas (counties). The first method is a resampling technique that uses NCVS data to estimate the proportion of crimes reported to the police in each county, then scales up the number of crimes reported through the UCR. The second technique uses Bayesian hierarchical modeling to estimate police reporting rates. Appropriate variance estimates are produced with each method. We discuss the strengths and weaknesses of each method, and provide suggestions for practical application and extensions.