Understanding and modeling the convergence of the UCR and NCVS

Abstract

The initial discrepancies between the UCR and NCVS data sets decreased with time, and the two data sets are assumed to have converged for most of the crime categories. Different definitions and methods have been used by the studies that have tested and explained the convergence between the two data sets, and those studies often reported different results. The two objectives of this study include understanding and explaining the convergence between the two data sets. The data for the study have been drawn from multiple sources. A multiple analytic strategy is used to test the convergence, and autoregression models with relevant predictor variables are estimated to explain the convergence. Graphic and correlational analyses support the convergence between the two series for all categories. However, the cointegration test indicates that the series are cointegrated for burglary and are in the process of converging for robbery and violent crime. The rate differences between the two data sets have been greatly affected by the percentages of reporting crimes to the police. Therefore, the convergence tests were repeated after adjusting the UCR rates for reporting, but the results did not differ substantially. The results of autoregressive models show that the increase in number of police officers and the methodological changes in the NCVS in 1992 are significant factors that reduced the divergence between the UCR and NCVS data sets. The determining convergence largely depends on the definition of convergence. In this case, a perfect convergence, in which the two series overlap and move together, is neither possible nor desirable because the UCR and NCVS use nonidentical measurements to measure nonidentical sets of crimes. The study provides important research, policy, and methodological implications and suggests future research directions on the subject.