Lethal Use of Force around 2020 Police Brutality Protests in Phoenix, Arizona
Introduction
Public safety and police use of force have been socially and politically salient issues for decades in the United States. The average person wants to feel safe in their community; however, many also acknowledge that law enforcement can have disproportionately negative effects on disadvantaged communities. Over the last few decades, United States history has been punctuated by over a half-dozen instances of police brutality so heinous that they sparked protests that garnered nationwide attention. In response to the 2020 killing of George Floyd, protests erupted across the country during a tense late spring marked by stay-at-home orders in the midst of a global pandemic. Although the victims of police use of force tend to be people from marginalized communities with less stable backgrounds, the popularity of protests against police brutality make this an increasingly salient issue at the national level.
Law enforcement in Phoenix, Arizona is notably violent. Immediately after the killing of George Floyd in late May, people gathered at the State Capitol to protest police brutality, including the contemporary police killing of Dion Johnson in Phoenix. Phoenix police arrested more than 100 protestors during the first weekend of protests (Staff 2020b). These protests lasted for weeks with local activists calling for increased civilian oversight of police and more transparency into police officers’ use of force history (Holmes 2020). Weeks-long protests with frequent clashes with the police were met with relatively modest use of force policy changes. PhxPD Chief of Police Jeri Williams suspended the use of the carotid control method of restraining someone in custody in favor of a more “compassionate” method that did not put pressure on the carotid artery (Staff 2020a).
By 2022, the Mapping Police Violence project ranked the Phoenix Police Department (PhxPD) #1 in use of lethal police force when compared to other major metropolitan police departments (Archer 2022). By mid-June 2024, the US Department of Justice Civil Rights Division had released a comprehensive report on their investigation of PhxPD. This report was initially triggered in 2021 by reported violations of the Americans with Disabilities Act and the constitutional rights of unhoused people. Chief Williams retired after a scandal in which over a dozen 2020 protesters were falsely accused of criminal gang activity which could lead to several years in prison; she was replaced by Interim Chief Michael Sullivan in 2022 (DoJ 2024). The 2024 DoJ report identified several systemic issues with the PhxPD that have resulted in deaths, injuries, and violations of constitutional rights. The report found that PhxPD used unnecessarily brutal force; they routinely unlawfully detained and arrested homeless people; they discriminated against marginalized racial groups and people with disabilities with regards to how they enforced the law; and they violated protesters’ rights to free speech (DoJ 2024). The 2020 protests in Phoenix represented a convergence of national and local sentiment regarding modern policing.
Literature Review
Environmental factors affecting police use of force
There is some evidence that more police brutality occurs in areas with higher rates of violent crime. Lee’s study uses self-reported use of force data from the Austin Police Department (APD) from January 1, 2006 to December 31, 2007. At the micro-level (within a 3,000-ft radius of the encounter), the number of violent crimes in the area increased the level of force police used. The same relationship was not observed at the meso level, based on the APD command area shape files. (Lee, Vaughn, and Lim 2014) Seyvan Nouri studied the effect of local neighborhood characteristics on the prevalence of use of force in an unspecified mid-sized city in the Southwest. He concluded that areas with higher crime rates and more calls for service were associated with more incidents of police use of force; unsurprisingly, these areas also tended to be disadvantaged. (Nouri 2021) The SLMPD study also concluded that more police shootings happened in block groups with middling levels of violent crime than those with high levels of violent crime (Klinger et al. 2015).
As with research on violent crime, there have also been studies on how other environmental and structural factors affect incidents of police use of force. Kane studies how precinct and division level factors like racial segregation, transience, and socioeconomic status impacted NYPD rates of police misconduct from 1975 to 1996. He concludes that poverty, transience and the percent of Latino residents were directly associated with higher rates of police misconduct. At the time, the percent of Black residents did not account for any variation in police misconduct. (Kane 2002) Kane and White continue the analysis of 1,500 officers (amounting to 2pct) from Kane’s 2002 NYPD data set. This study focuses on officers who were fired from the NYPD because of misconduct. Of those, only 4.8% (119) were fired over reports of excessive force. The most common offenses were related to administrative work and drugs. (Kane and White 2009) This finding suggests that, in large police departments like the NYPD, only a small percentage of police officers are expected to lose their employment because of reported brutality.
Another study on St. Louis Metropolitan Police Department (SLMPD) police shootings between 2003 and 2012 found that racial segregation and median income at the block group level have no direct relationship with police shootings (Klinger et al. 2015). However, in a later study Leung-Gagné analyzed deadly use of force data from over 700 local police departments spanning 2008-2017. Police department deadliness is strongly correlated with White-Black neighborhood segregation and being located West of the Mississippi River. PhxPD was not included in this analysis, but Tucson ranked as the sixteenth most deadly police department in the data set. It is also worth noting that SLMPD is the deadliest police department in their 2008-2017 data set. (Leung-Gagné 2024) The conclusions of the Leung-Gagne article suggest that PhxPD could be a prime example of a particularly violent police department with a degree of segregation and is in the Western part of the country.
There has been some literature on the effects of the 2020 protests on crime rates and lethal police force. Roman et al. studied incident-level data on four major U.S. cities including Austin, TX. They concluded that there was no change in violent crime rates in Austin despite a decline in police activity in the aftermath of 2020 protests (Roman et al. 2025). Travis Campbell created unique data sets with police killings data and crowd size estimates for Black Lives Matter protests in 2020 to study the effects of the protests across multiple cities on police killings at the national scale. The conclusion was that protests led to “reduced lethal force” overall (Campbell 2024). Boehme and Kaminski compared incidents of suspect resistance, police use of force, and officer injuries in the Indianapolis Metropolitan Police Department (IMPD) and the New Orleans Police Department (NOPD) before and after the death of George Floyd. There was a significant increase in all incidents except IMPD police injuries. (Boehme and Kaminski 2023)
Situational factors affecting police use of force
There are also many individual-level factors and factors unique to each situation that affect the chances of a police encounter resulting in injury or death for any parties involved. MacDonald et al. study on Miami-Dade Police Department (MDPD) reported differences in use of force by calls for service from 1996 to 1998. This study categorizes suspect resistance and police use of force into 4 levels each, stopping short of lethal force, use of firearms or other weapons on either side of the interaction. The study concludes that police are more likely to use force in response to property crime calls as opposed to domestic disturbances. (MacDonald et al. 2003) Miller et al. found that the rate of Cincinnati Police Department (CPD) use of force incidents also varies by the call for service type. Mental health calls and calls related to violence were associated with the highest rates of use of force incidents. Outside of calls for service, shorter response time and an officer initiating the encounter are also associated with more use of force. Other situational and community factors such as the time of day and “racial threat” had no effect on police use of force. (Miller, Guthrie, and Piza 2024)
There have been instances where police departments were able to reduce instances of police use of force. Prenzler et al. studied seven cases (2 in Western US, 2 in Eastern US, and 3 outside the US) where police departments were able to reduce police use of force. One of the most relevant individual-level strategy recommendations was for early intervention systems to identify and reform officers with histories of complaints. Other recommendations included training, more explicit policies, and more accountability for officers who exhibited poor behavior. (Prenzler, Porter, and Alpert 2013) Engel et al. conducted a stepped-wedge randomized control trial (RCT) to determine the effect of the Integrating Communications, Assessment, and Tactics (ICAT) de-escalation training on police use of force and resulting injuries in the Louisville Metro Police Department (LMPD). ICAT training is specifically designed to teach law enforcement officers how to de-escalate situations where the suspect is behaving unpredictably and unarmed or armed with anything that is not a gun. The study found that just over 1,000 LMPD officers receiving the ICAT de-escalation training was associated with a significant reduction in police use of force and police/civilian injuries. (Engel et al. 2022)
The efficacy of hot spot policing
For years, researchers have studied and, to date, reaffirmed the effectiveness of hot spot policing in reducing violent crime in urban areas (Braga et al. 2019). While members of the community may enjoy lower crime rates, they are not necessarily insulated from the risk of being antagonized, injured or even killed by local police officers patrolling high crime areas. Generally, high-crime areas also tend to be socioeconomically disadvantaged areas where people from marginalized communities live (Nouri 2021). Research into optimal levels of policing high-crime areas suggests that violent crime can increase in some areas when certain levels of proactive policing are not maintained (Koper, Wu, and Lum 2021). However, there is some evidence that more police shootings occur in areas with only medium levels of violent crime (Klinger et al. 2015). If police officers are primed to consider some calls (Miller, Guthrie, and Piza 2024) and some neighborhoods (Lee, Vaughn, and Lim 2014) to be especially dangerous, there is always the distinct possibility that they overestimate the amount of force necessary to physically control a suspect. In situations where police were called to resolve a mental health crisis, those tended to be the incidents where police used the most force (Miller, Guthrie, and Piza 2024).
Hot spot policing is one of the most widely adapted modern policing techniques. Hot spot policing assumes that the majority of crime in a city occurs within small geographic areas, not evenly or randomly distributed geographically. David Weisburd calls this phenomena the law of crime concentration at place. For several large cities, crime concentration (25% or 50% of reported incidents) at the street-segment level falls within a narrow bandwidth where the band is the percentage of street segments experiencing high-crime rates. According to his 2015 analysis, crime concentration generally remains within that narrow bandwidth over at least 9 years. In contrast, the number of incidents can fluctuate greatly from one year to another. (Weisburd 2015) Hot spot policing has been found effective at reducing violent crime. Anthony Braga’s 2019 meta-analysis collected 65 quasi-experimental and randomized experimental studies, resulting in the comparison of 78 tests of hot spot policing. Nearly 80% (62) of tests found hot spot policing to be effective in reducing crime. Although no studies used Phoenix Police Department data, there were three studies from Arizona. All three Arizona studies saw decreases in crime outcomes. (Braga et al. 2019)
However, not all hot spot policing tactics are equally effective. Tregle et al. studied the effect that hot spot policing has had on the number of Dallas Police Department arrests, officer-initiated traffic stops and routine investigations. One of their main research goals was to determine if increased police presence in hot spot areas would necessarily mean an increase in police-citizen interactions. They concluded that areas that received “offender focused” hot spot treatments, which involved serving warrants for repeat offenders and covert surveillance, had significant increases in the numbers of arrests, officer-initiated traffic stops and routine investigations. (Tregle, Tillyer, and Smith 2025) Koper et al. focus on one administrative division of a major metropolitan area’s police department over the course of 4 unspecified years. They conclude that crime hot spots require more proactive policing than low-crime spots. However, serious crime can increase in hot spots when there is reduction in proactive policing. There may be increases in minor crime in low-crime areas if there is a substantial decline in proactive policing in those areas. (Koper, Wu, and Lum 2021)
Hypothesis
The hypothesis interrogates the effect of various calls for service hot spots after the beginning of the 2020 George Floyd protests on the Phoenix Police Department’s (PhxPD) use of deadly force on civilians. Because of PhxPD’s post-2020 reputation as a notably deadly police department, the expected result is that there would be a significant increase in the use of deadly force in all calls for service hot spots. This hypothesis contrasts with the existing literature which asserts that police brutality protests (particularly Black Lives Matter protests) led to less police use of lethal force. This analysis is purely descriptive. The null hypothesis is that there is no relationship between calls for service hot spots and PhxPD use of deadly force, particularly after June 2020. The protests began on May 28, but the beginning of the protests for this study is June 1 to coincide with the governor-issued statewide curfew which resulted in clashes between protesters and the police the night of May 31 (Staff 2020b).
Data Sources
This analysis relies primarily on officers’ use of force (16,170 obs, 57 variables) from the City of Phoenix Open Data Portal and is supplemented by calls for service (289,452 obs, 8 variables) and violent crime (82,791 obs, 8 variables) data from the same portal (Phoenix Open Data 2025). Much of the data has been collected from as early as 2015 to the present; however, each data set is filtered to include a roughly equivalent date range around June 1, 2020, from 2018 to 2023. Irregularities across the data sets suggest changes in data collection methods over the years. This may explain more complete records in more recent years. The Phoenix open data sources are supplemented by block-level population estimates from the 2020 US Census (Bureau n.d.). Block level is the most granular geographic level available for population estimates. Using geocoding, the incident sites were assigned to their corresponding census block. The data set is limited by the self-selection bias of its source; some level of under-reporting is anticipated in reports on officer use of force.
Dependent Variable
The dependent variable is a dummy variable that indicates whether a use of force incident involved lethal force as the highest force applied. Figure 1 shows the contingency tables for the independent and control variables split by the deadly force dummy. Overall, 3.72% of the observed use of force incidents resulted in the use of deadly force. For the independent variables, the percentage of incidents that were both in a hot spot and involved deadly force ranges from 0.21% (Mentally Ill/Intoxicated Persons) to 2.15% (Trespassing). Of the control variables, aggravated active aggression (3.16%), the subject being armed with a gun (2.22%), the incident occurring on a weekday (2.30%), and the incident occurring after June 2020 (2.67%) were the factors with the greatest shares of lethal force incidents.
Figure 2 shows the means and standard deviations for continuous contextual variables. Population size, number of officer force incidents in the previous year and number of beat incidents in the previous year are log-transformed in the model to account for census blocks where the population is 0 and police officers with few and infrequent incidents.
Independent Variables
The independent variables are census-block level hot spots where the year-over-year rate of calls increased. The hot spots are segmented by four broad groups of service calls: trespassing, violent crime, property crime and mentally ill or intoxicated persons. Figures 3, 5, 7, and 9 are kernel density estimate (KDE or “hot spot”) maps of Phoenix, AZ; below each map is a paired t-test for the mean difference in the hot spot incidents at the block-level. Each map represents the density of calls before and after June 2020 relative to the 2020 population. The hot spots are generated using the Getis-Ord Gi* (gi-star) statistic and its corresponding p-values to show only statistically significant (p < 0.05) hot spots relative to the area.
Lethal Force Hot Spots
Figure 3 shows maps for lethal use of force incidents in calls for service hot spots before and after June 2020. The pre-event map shows large, light-to-moderate hot spots around the Central City, Encanto and Camelback East urban villages. After the event, the largest and heaviest hot spots were closer to Alhambra and North Mountain, north of I-10. The table in Figure 4 shows that there was a significant increase in the mean difference of lethal force incidents in calls for service hot spots between the two time periods at the 95% confidence interval.
Trespassing CFS Hot Spots
Figure 5 shows hot spot maps for lethal force incidents in trespassing calls hot spots. Trespassing calls were included in the independent variables because of the frequency with which those calls’ hot spots overlapped with lethal use of force hot spots. The pre-event hot spots are small and dense, running broadly along either side of I-10, from Maryvale to Camelback East. The post-event hot spots are larger and lighter. After June 2020 those hot spots span from Grand Avenue to North Mountain in a relatively narrow corridor between N 27th and N 19th Avenues. Figure 6 shows that there was also a significant increase in the mean difference in lethal force incidents in trespassing calls hot spots.
Property Crime CFS Hot Spots
Figure 7 shows hot spot maps for lethal force incidents in property crime calls hot spots. Before June 2020, the larger, denser hot spots were mostly below Camelback Road, around the Central City urban village; the notable exception is the large dense spot in North Mountain. After June 2020, the hot spots are smaller and lighter. Figure 8 shows a significant increase in the mean difference in lethal force incidents in property crime calls hot spots. This suggests that property crime calls may be less geographically concentrated after June 2020.
Violent Crime CFS Hot Spots
Figure 9 shows hot spot maps for lethal force in violent crime calls hot spots. The pre-event hot spots were small, ranging from light to moderate. The post-event hot spots are larger and denser; they generally land between N 27th and N 19th Avenues, from Central City to North Mountain. Figure 10 shows a significant increase in the mean difference in lethal force incidents in violent crime calls hot spots.
There is no paired hot spot map or t-test results for calls to mentally ill and intoxicated subject hot spots because the sample size was too small for kernel density estimation.
Control variables
The control variables include temporal variables (days since the beginning of the protests, post-event flag, months, weekday flag and daytime flag); environmental variables (high-volume use of force police beats, rising violent crime hot spots, and block-level population); demographics for civilians (race, age and gender) and police officers (race and gender); and situational factors (civilian weapons, civilian resistance, highest final charge, officer use of force history, and other force applied).
Figure 11 is another hot spot map to contrast the violent crime service calls with the reported violent crime rates. Figure 11 shows hot spot maps for lethal force in reported violent crime hot spots. Compared to the violent crime calls hot spots, the pre-event lethal use incident violent crime hot spots are larger and denser, although they fall in the many of the same general areas of the city. The post-event hot spots are less dense and largely fall between N 27th and N 19th Avenues. Figure 10 shows a significant increase in the mean difference in lethal force incidents in reported violent crime hot spots.
Empirical methods & Results
Figure 13 includes the multilevel mixed-effect logistic regression model summary for incidents of deadly officer use of force in different hot spots. The units of analysis are use of force incidents between 2018-01-01 and 2023-07-27. The odds-ratios provide context to the differences in situations that occur in and outside of lethal force hot spots such as the demeanor of the person in custody and the responding officers’ actions and when incidents are more likely to occur.
Taking all variables into account, the model suggests that the odds that lethal police force is applied was significantly impacted when the incident took place in a post-event hot spot for trespassing (odds ratio [OR] = 0.147; p < .001), property crime (OR=0.143;p<.001), or mentally ill or intoxicated persons (OR=11.623;p<.05) calls for service. Those results do not support the hypothesis that there would be an increase in the odds of officers using lethal force in hot spots after the protests; however, the odds-ratio for the mentally ill/intoxicated calls hot spots may be attributed to a post-2020 effect. The exception is the violent crime service calls hot spot variable where there was no significant effect. Trespassing calls hot spots were the only independent variable that had any significant effect (OR=8.026;p<.001) on the odds of lethal force applied regardless of when the incident occurred in the time frame.
Predictably, the contextual variables that presented the greatest odds that lethal force would be applied were the subject resisting with aggravated active aggression (OR=50.869;p<.001), the subject being armed with a gun (OR=16.130;p<.001) and the incident taking place after June 2020 (OR=34.621;p<.001). Variables where the odds of applied lethal force were the lowest for subjects who resisted in passive (OR=0.372;p<.001) and active (OR=0.363;p<.001) manners, and where soft empty hand force (OR=0.195;p<.001) was applied. Suspects in their 20s also had significantly lower odds (OR=0.269;p<.001) of being met with lethal force with all other variables held constant.
Limitations
This analysis is missing several variables that are commonly cited as important influences on police behavior regarding use of force. Data on whether the subject was injured or died as a result of the force used was not collected for a large part of the relevant time frame. Also, there was no information on whether encounters resulting in lethal police force were initiated by officers. Other City of Phoenix portal data could be used to supplement the data used here. This analysis did not include hot spots for reported property crimes in order to prioritize violent crime reports and calls for service. In this case, total population data was applied at the census block level; however, this did not allow for race data to be supplemented for the use of neighborhood racial segregation as a contextual variable.
This is an analysis of only one major U.S. city after widespread police brutality protests. The significance of geographic proximity and clustering grants another dimension for analyzing crime and use of force data. A comparison with a similarly-sized city or a city in the same region could provide more information on how protests against police brutality can impact policing.
Conclusion
PhxPD is a unique case study because of its status as a notoriously deadly police department. However, the alternative hypothesis that the odds of being met with lethal police are greater in calls for service hot spots was not supported. A use of force incident in most of the various hot spots might, in fact, be less likely to be involve lethal force. According to this analysis, the odds that police use of force would be lethal decreases in hot spots for trespassing and property crime calls for service. Although the odds of a police force incident involving lethal force increases in mentally ill/intoxicated persons hot spots, the sample size is too small to draw meaningful conclusions. Other more significant factors could be included to this analysis to determine if the results hold.