Eric L. Piza, David N. Hatten, Jeremy G. Carter, Jonas H. Baughman, and George O. Mohler (2023)

Policing: A Journal of Policy and Practice

*This study was funded by a grant from the National Institute of Justice (grant number 2019-R2-CX-0004)

Key Takeaways

• GDT alerts occurred a median of 93 seconds before the first 9-1-1 call for service (CFS)
• GDT alert locations were a median of 234.91 feet from the location reported via CFS
• In more than 26% of cases, GDT and CFS were geocoded to different street segments that do not intersect, meaning that officers responding to the CFS location would be a meaningful distance away from where the gunshot occurred
• Regression analysis findings suggest time savings and spatial precision decrease in cases more conducive to citizen reporting

Research Summary

Gunshot Detection Technology (GDT)  is expected to impact gun violence by accelerating the discovery and response to gunfire. GDT consists of networks of acoustic sensors that detect and identify the location of gunfire in real time. This can help generate police response to shooting scenes quicker than when gunfire is reported by citizen calls to 9-1-1. GDT should further collect more accurate spatial data, given gunfire locations are assigned to the coordinates measured by the acoustic sensors rather than addresses reported second hand by callers to 9-1-1. However, little research has focused specifically on the level to which GDT offers such benefits.

The current study is a partnership between a multi-university research team and the Kansas City, Missouri Police Department (KCPD). KCPD data systems were triangulated to identify gunfire events reported by both GDT and a 9-1-1 call for service (CFS), with 2,946 such incidents included in the analysis. The study focuses on the time savings and spatial precision offered by GDT as compared to CFS over the first nearly 5 years of the program (9/14/2012-5/9/2017). Time savings measures the number of seconds between the GDT alert and the first CFS reporting the same gunfire event. Spatial precision measures the linear feet between the location detected by the GDT alert and the location reported by the CFS.

GDT generated an average time savings of 125.44 seconds, with a median of 93 seconds. To contextualize this value, police respond to reported gunfire in a median time of a 223 seconds according to KCPD data. Following arrival on scene, EMS responses have a median of 78 seconds, and the median time to the nearest trauma center is 480 seconds. The 93-second time savings represents nearly 12% (93 of 781 seconds) of the response and travel time.

The average level of spatial precision was 433.91 feet, with a median of 234.91 feet. In more than 50% of cases, GDT and CFS locations were geocoded to different street segments. In more than 26% of cases, GDT and CFS locations were geocoded to different street segments that do not intersect, meaning that officers responding to the CFS location would be a meaningful distance away from where the gunshot occurred.

Regression models, which incorporated 18 variables that could theoretically influence the reporting of gunfire, identified situational characteristics that influence GDT performance. The pattern of statistically significant variables suggests time savings decreases in cases that are more conducive to citizen reporting. For example, multiple gunshots detected and ambient population were negatively related to time savings. This suggests that the additional noise generated by multiple gunshots and more people on-street to hear such noises may lead to citizens calling 9-1-1 quicker. A similar theme was found in the spatial precision analysis. Levels of shots fired CFS and firearm-related crime reported on the street segments were consistently associated with decreased spatial precision. The relative proportion of residential parcels on a street segment was negatively associated with spatial precision. Taken together, this suggests that residents of street segments with high levels of illicit firearm activity may be better positioned to identify the source of gunfire.

Eric L. Piza, Kevin T. Wolff, David N. Hatten, and Bryce E. Barthuly (2023)

Health & Place

*This study was funded by a grant from the Bureau of Justice of Assistance and administered by the Institute for Intergovernmental Research

Key Takeaways

• Group-based trajectory analysis classified block groups in Passaic County, New Jersey according to drug overdose trends from 2015 through 2019
• A mixed-effects panel negative binomial regression model examined environmental and neighborhood characteristics associated with annual overdose counts
• Block groups were classified across 3 groups: low and stable, low with moderate increase, and elevated and increasing
• All but 1 of the elevated and increasing block groups were spatially contiguous within a single city
• Concentrated disadvantage exhibited the largest effect size in the regression models
• Most variables positively associated with overdose levels were built environment measures

Research Summary

Drug overdose has emerged as a national public health emergency in the United States over the previous decade. Prior spatial analyses have generated important insights into the problem of drug overdoses. However, spatial analyses of drug overdoses typically incorporate cross-sectional designs that are unable to measure the developmental trends of high overdose areas. Cross-sectional designs are further unable to account for within unit differences over time, which can bias estimates of independent variable effect. 

The current study sought to address gaps in the literature through a spatial analysis of drug overdoses in Passaic County, New Jersey from 2015 through 2019. This study is an outgrowth of an action research partnership between a multi-university research team and the Paterson, NJ Coalition for Opioid Response and Assessment (COAR). The mission of COAR is to develop data-driven, multi-agency responses to the overdose crisis in City of Paterson, NJ. COAR stakeholders anticipated county-wide resources would need to be mobilized to successfully address the opioid crisis in the Paterson. As such, COAR’s analysis efforts began with an assessment of overdoses throughout the entirety of Passaic County.

We first conducted a group-based trajectory analysis to classify block groups according to their overdose trends. To our knowledge, this is the first application of group-based trajectory analysis in the drug overdose literature. A mixed-effects panel regression model then identifies the built environment and neighborhood characteristics associated with overall overdose levels. Overdose data were provided by the New Jersey State Police (NJSP), which tracks state-wide drug overdoses as part of the national Overdose Detection Mapping Application Program (ODMAP).

The group-based trajectory analysis identified three groups with distinct drug overdose trends: low and stable (72% of block groups), low with moderate increase (24% of block groups), and elevated and increasing (4% of block groups). Areas in the elevated and increasing group accounted for the majority of overdoses with an average of 76.2 incidents over the five-year period. The year-to-year average in overdose events increased dramatically among this small number of block groups, from an average of 1.75 in 2015 to an average of 26.5 in 2019. The block groups in this trajectory grouping were highly clustered, with all but one spatially contiguous within the City of Paterson. This indicated overdose prevention resources could be highly focused within the geographies suffering from the most disproportionate levels of drug overdose.

In the regression analysis, concentrated disadvantage exhibited the strongest effect. This suggests that recent policy proposals to substantially increase investment in community institutions and general community wellbeing as a public safety strategy may also support overdose prevention efforts. Nonetheless, most statistically significant variables positively associated with overdose counts were built environment measures (liquor stores, health care facilities, vacant parcels, and public land parcels). These findings suggest certain types of land usage may provide targets for proactive social outreach efforts or may benefit from place-based policy solutions such as vacant lot greening.

Eric L. Piza, Nathan T. Connealy, Victoria A. Sytsma, and Vijay F. Chillar (2022)

Criminology

*This study was funded by the Charles Koch Foundation, Policing and Criminal Justice Reform program

Key Takeaways

• Systematic social observation of body-worn camera footage and panel regression analysis tested the effect of police officer and civilian actions on police use of force across 5-second intervals
• The most influential variables are related to authority maintenance
• Additional variables reflecting procedural justice, civilian resistance, and bystander presence significantly influence when police use force occurs during civilian encounters
• Certain variables influence use of force at a distinct point in time whereas others exert influence over multiple time periods
• Overall findings support theoretical perspectives considering use of force as a transactional event

Research Summary

Prior research has consistently found that police-citizen encounters involving force typically extend across fairly lengthy time periods. During these extended periods, police officers must rapidly assess a series of dynamic situational factors, requiring the use of cognitive shortcuts that can influence later use of force decisions. Situational factors, and the cognitive shortcuts they motivate, can ebb and flow throughout a police-civilian encounter, differentially influencing the likelihood of force at various points. Within this context, empirical evaluations require measuring and assessing what happened in the seconds or minutes that preceded force as well as how the actions and behaviors of officers and the civilians they engage with influence the probability of force being used. Unfortunately, common data sources typically only allow for the cross-sectional analysis of use of force events, precluding the use of longitudinal methods that can sufficiently measure transactional events.

The current study reports on a systematic social observation (SSO) of body-worn camera (BWC) footage and panel regression analysis in Newark, NJ. The study sample includes 91 use of force events recorded on BWCs in Newark, NJ between December 2017 and December 2018. This study period reflects the pilot phase of the NPD’s BWC deployment. Rather than analyze encounters as cross-sectional moments in time, we parsed out the entirety of each individual use of force event into a series of five-second time intervals to examine encounters longitudinally. Data on the actions of police officers, suspects, and bystanders, which were collected during the SSO, were scored into their appropriate corresponding time interval(s) based on the start and/or end time of the given action or behavior. The unique structure of the dataset allows for a new way of examining the situational nature of police-citizen encounters, which to our knowledge has yet to be done in use of force research.

All model covariates were categorized into theoretical constructs: active civilian resistance (n=4), authority maintenance (n=5), procedural justice (n=5), and bystander presence (n=3). Four different interval-based operationalizations we applied to the model covariates.  The instant operationalization has start and end times that occurred instantaneously and simultaneously (e.g., a civilian shoving an officer). The active operationalization marks all of the consecutive interval(s) at which an action was actively occurring (e.g., officer explaining the reason for suspect detainment). The post-occurrence operationalization assigns the next six intervals (30 seconds) after a recorded instance of an action or behavior a value of ‘1’ to explore their potentially delayed effect. The lasting operationalization captures the number of times an action or behavior occurs, and the influence of its occurrence on the remaining duration of the event.

Significant effects were observed for 2 of 4 civilian resistance variables, 3 of 4 authority maintenance variables, 4 of 5 procedural justice variables, and 1 of 3 bystander presence variables. Post-hoc analyses found that the 6 covariates most predictive of increased use of force likelihood were authority maintenance variables. These 6 authority maintenance variables accounted for more than 65% of the regression model’s predictive capacity. This finding supports prior theoretical perspectives arguing that police use of force largely results from officer attempts to maintain authority over civilians during face-to-face encounters. Findings further indicate that the nature of the temporal influence greatly differed across variables. For example, civilians disobeying an officer’s calm command increases the likelihood of force in the instant it occurs, whereas officers explaining detainment increases the likelihood of force in the post-occurrence period. This finding supports theoretical perspectives considering the actions and reactions of police officers and civilians as key factors in determining exactly when use of force occurs during police-civilian encounters.