Predicting Violent Crime Rates for the 2010 Redesign of the National Crime Victimization Survey (NCVS)

The National Crime Victimization Survey (NCVS) is a major crime survey for the United States. The survey collects data on several types of crimes, including the broad categories of violent crime and property crime. The 2010 redesign of the NCVS can potentially improve the efficiency of the survey if the level of crime can be predicted well by external data. Previously, we reported initial success in predicting the level of crime at the county level based on the Uniform Crime Reporting (UCR) System. A more finegrained analysis shows, however, far greater success for property crime than for violent crime. This paper extends the previous results to examine the underlying associations more thoroughly. We find that the largest single component of violent crime in the UCR, aggravated assault, fails to add to the predictive accuracy of a regression equation including the other violent crime components of the UCR, rape and robbery. We believe this finding has implications for interpretation of the UCR. We extend the analysis to include demographic characteristics from the census and the ACS, and we examine the ability of tract-level characteristics to predict individual victimizations.