Measuring the Impacts of Everyday Police Proactive Activities: Tackling the Endogeneity Problem

The effect of real-world proactive policing on crime

“…Crime data come from SPD's Records Management System (RMS), which contains information on all officially reported criminal incidents known to the police. Similar to other observational studies of proactive policing (e.g., Rosenfeld & Fornango, 2014, 2017Wu et al, 2021), the primary outcomes of interest violent crimes such as robbery and aggravated assault, and property crimes such as larceny, burglary, and motor vehicle theft. All crimes were aggregated annually, measured at the beat level, and converted into rates per 10,000 residents in Seattle police beats (see Figure 2).…”

“…Although measuring these relationships at a larger jurisdictional level and over a longer period fills a critical gap in the literature, more favorable observational evidence exists at micro-levels of analysis (e.g., , and others have shown that deterrent effects on crime may last only a week or two (Wu et al, 2021). Accordingly, more research is needed that can leverage the benefits of administrative datasets with enhanced spatial information to test these looming differences in the findings of proactivity on crime across spatial and temporal levels of analysis.…”

“…First, we estimated the effect of proactive policing on crime within police beats by up to two years. Police behavior in one year can influence crime up to two years later across multiple levels of spatial analysis (i.e., blocks, cities, states) (Marvell & Moody, 1996;8 We employed CLPMs over the more popular general methods of moments (GMM) estimators used to identify deterrent effects of proactivity on crime for two important reasons (e.g., Beck et al, 2022;Wu et al, 2021). First, empirical simulation evidence suggests that CLPMs exhibit less bias than the GMM estimators in small sample settings (N ≤ 100), which is advantageous given the current sample of police beats (N = 51) (Allison et al, 2017).…”

“…Second, CLPMs are estimated through the maximum likelihood method, which allows them to review a host of goodness-of-fit indices when making empirically informed decisions about their modeling procedures. This is not the case with GMM and is a consistently cited shortcoming of the approach in the empirical (Allison et al, 2017;) and applied literatures regarding the effect of police on crime (Rosenfeld & Fornango, 2014, 2017Wu et al, 2021). Simpson & Hipp, 2019).…”

The Racialized Consequences of Jail Incarceration on Local Labor Markets

“…They are often used in macro-level contexts when there is not a theoretically relevant instrumental variable available, but the dataset has a higher number of panels than time periods, as in this study (Agnew et al, 2011;Beck & Goldstein, 2018). The system GMM has been used effectively in a number of important recent studies of crime and criminal justice (see, for example, Manning et al (2018), Rosenfeld & Fornango (2014), or Wu, Koper, & Lum (2021) for recent uses of this GMM dynamic panel model approach in similar macro-level longitudinal criminal justice analyses).…”