When Choice of Data Matters: Analyses of U.S. Crime Trends, 1973–2012

This study uses UCR and NCVS crime data to assess which data source appears to be more valid for analyses of long-term trends in crime. The relationships between UCR and NCVS trends in violence and six factors from prior research are estimated to illustrate the impact of data choice on findings about potential sources of changes in crime over time. Crime-specific data from the UCR and NCVS for the period 1973–2012 are compared to each other using a variety of correlational techniques to assess correspondence in the trends, and to UCR homicide data which have been shown to be externally valid in comparison with other mortality records. Log-level trend correlations are used to describe the associations between trends in violence, homicide and the potential explanatory factors. Although long-term trends in robbery, burglary and motor vehicle theft in the UCR and NCVS are similar, this is not the case for rape, aggravated assault, or a summary measure of serious violence. NCVS trends in serious violence are more highly correlated with homicide data than are UCR trends suggesting that the NCVS is a more valid indicator of long-term trends in violence for crimes other than robbery. This is largely due to differences during the early part of the time series for aggravated assault and rape when the UCR data exhibited consistent increases in the rates in contrast to general declines in the NCVS. Choice of data does affect conclusions about the relationships between hypothesized explanatory factors and serious violence. Most notably, the reported association between trends in levels of gasoline lead exposure and serious violence is likely to be an artifact associated with the reliance on UCR data, as it is not found when NCVS or homicide trend data are used. The weight of the evidence suggests that NCVS data represent more valid indicators of the trends in rape, aggravated assault and serious violence from 1973 to the mid-1980s. Studies of national trends in serious violence that include the 1973 to mid-1980s period should rely on NCVS and homicide data for analyses of the covariates of violent crime trends.