Estimating the Reliability of Crime Data

• In Event: Advances in quantitative methods 3

Abstract

Police crime data is problematic; crimes that are never reported undermine its validity and differences in police recording practices affect its reliability. However, the overall prevalence of those problems is not well known. This is because when comparing recorded crimes against secondary data sources like crime surveys existing studies have assumed the alternative sources represent an error free gold standard. In this paper we make no such assumption, instead comparing crime records to crime survey data using a Multi-Trait Multi-Method model which treats both data sources as potentially unreliable. Prior research has also tended to focus on specific crimes, locations, and points in time. In contrast, we examine the quality of police recorded crime data for multiple crime types at different area levels from 2011 to 2019. Whereas crime rates derived from police data will systematically underestimate the true extent of crime, we find that they are substantially more reliable than crime rates derived from survey data. We also find that the reliability is higher for vehicle crime and burglary than for personal crimes and criminal damage, and that the reliability ratios across all crime types and measures of crime are gradually getting worse over time. These findings help us understand the types of crime analyses where police data could be used in an effective way (namely, comparisons across areas), and facilitate the adjustments of studies using police data for causal analysis.

Author

• Ian Brunton-Smith, University of Surrey