In-Car Cameras and Police Accountability in Use of Force Incidents

Abstract

New policing technologies have generated solutions to many policing issues. In particular, portable camera systems (in-car or body-worn) have been offered as a tool to address the issue of police excessive use of force. It has been argued that police camera systems increase transparency in law enforcement and deter both police officers and citizens from engaging in undesirable behaviors during encounters. However, the question of how effective these technologies are in increasing the accountability of police departments still remains unanswered. Some argue that the use of camera systems to record police behavior does not create a significant reduction in excessive use-of-force complaints or does not serve as an effective accountability tool as expected. From this perspective, this study explores the impact of in-car camera usage on police use-of-force investigations. This research examines the impact of in-car cameras on the total, dismissed, and sustained excessive use-of-force complaints against 891 police departments in the USA with more than 100 sworn officers. We employed Law Enforcement Management and Administrative Statistics (LEMAS) 2007 dataset to conduct this analysis. We utilized negative binomial regression analysis in STATA 15 to examine whether the adoption of vehicle camera systems by police agencies has an impact on dismissed and sustained complaints of inappropriate use-of-force. We found that the adoption of in-car cameras correlates with the number of dismissed cases; however, we did not find any significant relationship between in-car camera usage and sustained cases. Police departments using in-car camera systems are more likely to dismiss citizen complaints, rather than sustaining them. We concluded that video footages generated by in-car camera systems are inadequate in producing evidence to back up the complainants’ claims or in generating proof of excessive use of force for further investigation. Our findings suggest that police departments should not solely rely on in-car cameras if they want to enhance accountability and unearth police misconduct within their department.

Appendices

Appendix 1. Histogram of Use-of-Force Complaints

Fig. 4

Total number of use-of-force complaints

Appendix 2. The Comparison of Poisson vs. Negative Binomial Model

Table 4 Vuong test for negative binomial model (number of total force complaints)

The equidispersion assumption of the Poisson probability distribution was not met in this analysis, because descriptive statistics for the variable total number of complaints revealed that the variance 23.716 is larger than the mean (11.571), indicating slight overdispersion. The rate parameter and variance for the constant-only Poisson model of variable total number of complaints is the exp 3.54, which is 34.46. The rate parameter for the constant-only negative binomial model is 379.4 and variance for this model is 0.458.

Appendix 3. The Comparison of Standard Negative Binomial Model vs. Zero-modified Negative Binomial Model

The Vuong test (z = 0.76, p = 0.2244) is not statistically significant, and inflation parameter is also not statistically significant. Thus, we fail to reject the null hypothesis that standard negative binomial model provides an overall better fit as good as the non-modified version. The Vuong test results for sustained and dismissed complaints also indicate that the zero-modified negative binomial model does not improve the fit: For sustained complaints: The Vuong test (z = 0.51, p = 0.3034) is not statistically significant. For dismissed complaints: The Vuong test (z = 0.07, p = 0.4727) is not statistically significant.

To better understand the fit of different models, we also produced a scatterplot of deviations of the predicted probabilities from the observed proportions (Fig. 5).

Fig. 5

Scatterplot of Poisson, negative binomial, zero-modified Poisson and zero-modified negative binomial models (for total number of use-of-force complaints)

In the scatterplot shown in Fig. 5, the deviations from the horizontal line at zero gives us an idea on the better model. Figure 5 illustrates that all models appear to be performing well at high values, but the Poisson model performs most poorly—particularly at low values. The Poisson and zero-modified Poisson models have less deviations after 5. However, they perform quite poor below 5. We can also see that the negative binomial and the zero-modified negative binomial provide identical predictions. Consistent with the Vuong test findings outlined above, the non-modified negative binomial model is the best model. Furthermore, as it is shown in Fig. 5, the negative binomial regression model is effective in reproducing the sample proportions for total number of use-of-force complaints. We have continued with the negative binomial regression model since it provides the most parsimonious model among all.