![Sample our Social Sciences journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110822/TOC-Subject-Sample-2021-Social-Sciences.jpg"})
Abstract
By virtue of their ability to capture evidence of criminal behavior, body-worn cameras (BWCs) have been associated with enhanced investigations and prosecutions. To date, however, research on these outcomes has been inconsistent, and there has been no attempt to systematically review or synthesize the results of these studies. To address this gap, we conducted a systematic review and meta-analysis of 12 experimental and quasi-experimental studies examining the impact of BWCs on prosecutorial and court-related outcomes. In aggregate, we find no significant effects of BWCs across any reported outcome measure, however, we find that studies focused on domestic violence offenses are associated with significant and large treatment effects across most outcome measures. While these results show promise, extant domestic violence studies are generally quasi-experimental in nature, and thus the effect of crime type cannot currently be separated from that of research design. Policy implications and directions for future research are discussed.
Acknowledgements
We would like to thank Drs. David Weisburd and David Wilson for their input on prior versions of this manuscript. We would also like to thank Dr. Cynthia Lum for discussing this review with us and providing conceptual feedback.
Disclosure Statement
No potential conflict of interest was reported by the author(s)
Notes
1 While a single incident can have multiple types of outcomes (e.g., prosecuted, convicted, sentenced), a standard incident cannot have multiple forms of the same outcome at different points in time.
2 Eligible incidents or cases were required to be criminal in nature. We excluded one study that was otherwise eligible as it tracked the outcomes of civil cases (see Zamoff, Citation2019)
3 Initially, we also analyzed “Investigations and case clearances”, which primarily included sanction detections and investigative arrests. Due to considerable uncertainty surrounding the identification of investigative outcomes, however, we ultimately excluded these measures. Interested readers may contact the authors directly for more information about these analyses.
4 The term “NEAR/3” is a proximity operator that connects words to the left and right of the operator within the specified range (e.g., 3 words).
5 These reviews included Lum et al. (Citation2019) and Lum et al. (Citation2020)
6 These journals included: Criminology, Journal of Experimental Criminology, Journal of Quantitative Criminology, Criminology and Public Policy, Justice Quarterly, Policing, Police Quarterly, Policing and Society, British Journal of Criminology.
7 Some emails were returned as undeliverable, and thus not all authors were successfully contacted.
8 This inevitably involves the synthesis of effects across differing units of analysis (e.g., convictions by officer, convictions by day, convictions by case). While Lipsey and Wilson (Citation2001) caution against such an approach, this caution is in reference to more traditional effect size metrics such as Cohen’s d. Cohen’s d is measured in standard deviation units, which are dependent on the unit of analysis. In contrast, the RR/RIRR is measured in proportional units, which are not dependent on the unit of analysis (Wilson, Citation2022). Indeed, recent work has used these measures to synthesize effects across various units of observation and analysis (see Lum et al., Citation2020). For more information, see Appendix C.
9 All models were estimated using the natural log of the RR/RIRR. Results were then exponentiated for ease of interpretation. When exponentiated, the RR/RIRR can be converted to a percentage point change by subtracting 1 and multiplying by 100.
10 The one eligible study that our searches did not identify was an unpublished report (Groff et al., Citation2018). Of note, one additional study was recommended to us by subject-matter experts but was published after our eligibility cut-off date (Todak et al., Citation2022). This study analyzes data stemming from the same RCT as two studies already included in this review, however (Huff et al., Citation2021; White et al., Citation2021).
11 The anticipated effect of low camera activation and technological issues would be to limit the overall use of BWCs. If BWCs are an effective evidentiary tool, we would expect this to dilute any potential treatment effect. Likewise, if BWC officers are present at control incidents, we might expect the quality of the evidence collected at control incidents to increase. If so, this would reduce any existing contrast in case outcomes between treatment and control groups.
12 Note that we define increases in downstream outcomes as effects favorable to treatment. While the desired direction of effect could be debated, increases in these outcomes would be consistent with the direction of effect expected from BWC adoption (see Bakardjiev, Citation2015).
13 For the Petersen et al. (Citation2021) study, this outcome was originally reported as “No Action” cases, where no action represented a case rejection by the prosecution. Given that this is the inverse outcome of the decision to charge or prosecute, we simply reverse coded this effect for the purposes of this review.
14 One study (Groff et al., Citation2018) included a measure of processing time from which we could not obtain an effect size.
15 All moderator analyses were conducted using the analog to the ANOVA method, which tests for significant differences in mean effect sizes between levels of categorical variables (see Lipsey & Wilson, Citation2001).
16 Across all models, the results of quasi-experimental studies were also significantly different from 0 (as indicated by the 95% confidence intervals), while the results of experimental studies were only significant for guilty outcomes and processing time (though these effects were in the opposite direction as those of quasi-experimental studies).
17 BWCs in domestic violence studies were also associated with a statistically significant 80% increase in processing time. However, only one domestic violence study analyzed processing time (Morrow et al., Citation2016), and the authors note that this effect may have been due to changes in administrative management.
18 The residual heterogeneity for each of these models (i.e., the variance remaining after accounting for crime type) was also nonsignificant, with residual I2 values ranging from 0.00-0.08%. This suggests that, after accounting for crime type, there was very little remaining between-study variability.