Skip to main content
SearchLoginLogin or Signup

Turnover in Large US Policing Agencies Following the George Floyd Protests

JOURNAL OF CRIMINAL JUSTICE postprint | We examine whether police resignations and retirements significantly changed in the two years following public backlash related to the police murder of George Floyd. We employ Bayesian Structural Time Series to compare observed trends ...

Published onAug 05, 2023
Turnover in Large US Policing Agencies Following the George Floyd Protests


We examine whether police resignations and retirements significantly changed in the two years following public backlash related to the police murder of George Floyd. We employ Bayesian Structural Time Series to compare observed trends in each agency to synthetic counterfactuals using monthly staffing data from fourteen large municipal policing and sheriffs’ agencies in the US. In the two years since the Floyd protests began, large metropolitan agencies have experienced significant increases in resignations, retirements, or both. One agency was unaffected, two saw small improvements, and eleven saw between 2.2% and 16% excess loss of sworn full-time personnel when compared to the synthetic counterfactual. These results reaffirm the importance of understanding how agency operational and personnel patterns have shifted since the summer of 2020.

Post-print version August 5, 2023


Police executives have been expressing concern for the better part of three years about their ongoing struggle to recruit and retain officers (Smith, 2022). Empirical evidence appears to justify their concerns (Duret & Weithua, 2023). Inadequately staffed agencies may struggle to function efficiently or, worse, cause harm in their communities. Recently, the judge overseeing the Baltimore Police Department’s federal consent decree said staffing shortages are “at a level of severity that is bigger than the police department can solve” (Simms, 2023, para. 6). Meanwhile, questions have been raised about whether the five officers charged with murdering Tyre Nichols would have been assigned to the (now disbanded) SCORPION Unit – or even hired in the first place – had the Memphis Police Department not relaxed its hiring standards in 2018 in an effort to recruit more applicants (Johnson, 2023).

Among other things, police agencies have been characterized as institutions, as representing the “or else” of the government, and as “nothing more than a mechanism for the distribution of situationally justified force” (Bittner, 1970, p. 39). Ultimately, though, police agencies are a collection of individual employees tasked with providing some sort of service to their communities. If, for whatever reason, agencies are unable to employ a sufficient number of officers, they will invariably struggle to meet community service expectations. Accordingly, employee retention is a logical policy focus. The basic reality is that 911 calls keep coming in, whether or not there are sufficient police officers available to respond. Moreover, these calls are the exclusive responsibility of the police. As Lum et al. (2022, p. 255) note, “the amount and types of incidents for which people call the police are voluminous, with the vast majority not obviously transferable to other organizations or government sectors” (Lum et al., 2022, p. 255). Simply put, the level of staffing matters: Losing personnel places an inordinate strain on the ability of the officers in a given agency to appropriately and effectively acquit their duty.

Following the murder of George Floyd at the hands of a now-convicted and imprisoned police officer, millions of people around the country took to the streets to protest (Buchanan et al., 2020). A growing demand for police reform accompanied these protests. While certainly not new, the Floyd moment in policing was notable for prompting the most widespread protest movement in recent memory (Boudreau et al., 2022), wide-ranging policy change, mainstream political calls for defunding and even abolishing police agencies, increased (but perhaps localized) firearm violence (Boehme et al., 2022), and rapid declines in favorable opinions of the police (Reny & Newman, 2021).

It is not yet clear what long-term effects, if any, these post-Floyd changes will have on the institution of policing itself. Policing agencies have raised alarms about the severe negative impacts of the post-Floyd period on their ability to recruit and retain officers. On the one hand, skeptics might contend that these concerns have been raised before (New York Times, 1953), and that the institution of policing may well use such appeals to shield itself from meaningful critique and reform. Similarly, there is no specific formula that gives a universally agreed-upon “correct” number of officers required for any given agency (Hollis & Wilson, 2015). On the other hand, the media, practitioners, and other non-academic sources have provided a swathe of anecdotal evidence that a shift is underway (Arden, 2021; Kakade & Ramirez, 2022; MacFarquhar, 2021; PERF, 2021; Thorne, 2022). Reporting by the Marshall Project, for example, was initially skeptical of a crisis in police personnel losses (Weihua & Ilica, 2021) but later found that, compared to historical growth in the number of police officers, the 2020-2021 period was characterized by declines in total police personnel (Duret & Weithua, 2023). In this vein of thought, regardless of one’s beliefs about the “correct” amount of staffing, a significant shift downward in police personnel may have serious consequences for the communities, as police are often the only public institution required to respond to a number of community needs for which there is no formal agency in place to respond to (Bittner, 1970; Lum et al., 2022). In other words, regardless of what we believe police should do, a sharp reduction in operational strength absent any other readily available substitute is likely to have negative results for communities.

At least one academic study provides more rigorous evidence of increased turnover in policing, finding a short-term but large increase in the number of officers resigning from one agency in the six months following the start of Floyd-related protests (Mourtgos et al., 2022a). Whether this represents a generalized trend across agencies is unclear, as Mourtgos et al. situated their study in a capital city that saw significant protests over an extended period. The many open questions related to police staffing include whether the effect demonstrated in Mourtgos et al. extends to other types of municipalities, such as suburbs; whether different types of law enforcement agencies, such as sheriff’s offices, have been impacted; and whether there are shared characteristics among agencies that have, or have not, experienced large changes in officer turnover. Further, much extant research has been hampered by secondary data, such as that reported by federal agencies, which generally do not distinguish the mechanism for an officer’s departure (i.e., resignations, retirements, or involuntary separations). Such data are also typically aggregated to the yearly level, preventing researchers from modeling, for example, the effects of exogenous shocks that occur mid-year. Finally, federal reporting prevents differentiation between sworn officers and civilians, as with the Quarterly Census of Employment and Wages (US Bureau of Labor Statistics, 2023), which reports total employment in policing, thereby preventing researchers from disambiguating sworn officers from those working in civilian roles.

This study adds to the sparse literature on police turnover in the post-Floyd era by analyzing data from a diverse sample of agencies. We specifically focus on the two most prevalent types of leaving: voluntary resignations (i.e., before qualifying for a full retirement) and retirements. In undertaking this analysis, we partnered with fourteen large US agencies, including both municipal police and sheriff's agencies in both urban and suburban settings. We leverage a unique set of administrative data on police turnover at the monthly level to construct a time-series of how turnover shifted over the two years after the protests erupted following the murder of George Floyd. We employ Bayesian structural time-series (BSTS) methods to ask: Did police turnover increase substantially following the Floyd protests?

Literature Review

Police agencies have long complained they are under-resourced and over-tasked. One early news story cited police chiefs in New York (state) who claimed that a 120% spike in resignations was linked to several issues, including the scarcity of recruits due to the wartime draft, better working conditions and pay in the private sector, and the police minimum starting age of 25 being too high (New York Times, 1953). At a national level, the final report issued by President Johnson’s Commission on Law Enforcement and Administration of Justice (Douglas, 1967) calculated that, at that time, 50,000 additional officers would be needed across the country just to fill already-authorized agency strengths.1

In more contemporary times, similar moments of increased/widespread public concern about policing have raised questions about the effects of external events on police recruitment. Survey-based research of Texas police chiefs following the 2014 Black Lives Matter protests suggested respondents were concerned that their recruitment efforts would be negatively impacted (Copeland et al., 2022). However, the study was limited to cross-sectional, perceptual data. Further, it could not access administrative recruitment or retention data, and, to date, most research has failed to find empirical support for the so-called “Ferguson Effect.” However, research has linked the post-Ferguson era to increased expressed reluctance among college-aged students to join the policing profession (Morrow et al., 2019; Todak, 2017).

The Importance of Studying Police Turnover

One potential consequence of a lack of police turnover research (owing in part to the dearth of data we have outlined above) could be inadequately accounting for the mechanisms driving lower police activity, such as in studies linking increased violent crime to the George Floyd protest moment (Cassell, 2020). Whereas recent studies linked increased firearm violence in Minneapolis to the murder of George Floyd, that has not been the case with some comparison cities (Boehme et al., 2022). Models linking Floyd protests and crime may inadvertently mask an important causal factor: the number of officers available to deter, respond to, and investigate serious crime (Petersen et al., 2023). In the case of studies attempting to understand the mechanism(s) that explain(s) such effects, we must consider that police staffing is a potential confounding variable. As Tillyer and Eck note: “Theory suggests that changes in crime patterns can be explained by…the availability of effective handlers, guardians, and place managers” (2011, p. 181). In other words, a sudden reduction of police officers (i.e., capable guardians) may increase criminal offending and victimization (Cohen & Felson, 1979).

Police employment data has previously been subjected to significant empirical scrutiny to ascertain whether police rates of police employment affect crime. Recent studies have rigorously probed this connection, and most scholarship finds that fewer police officers translates to a diminished ability of agencies to control crime. Hur (2013) tests the correlation between crime and both voluntary and involuntary police turnover in 464 cities with more than 50,000 residents, combining socioeconomic data from the US census, employment data from Law Enforcement Management and Administrative Statistics (LEMAS), and 2003 Part 1 crime data from the FBI. Hur concludes that voluntary turnover is associated with increased violent and property crime, while involuntary turnover (i.e., firings) are unrelated to crime. However, those findings are limited to correlational relationships, leaving questions about causation unanswered. However, a later study builds a causal model of 1981-2018 crime and aggregate police employment data in 242 cities, and finds that for each additional ten officers employed by an agency there is one fewer homicide overall, and that this effect is doubled (two fewer homicides for every additional ten officers in place) in Black communities (Chalfin et al., 2020). These findings coincide with research leveraging a natural experiment in New Jersey’s two largest cities (Piza & Chillar, 2021). The authors took advantage of a mass layoff in one of two nearby and otherwise similarly situated agencies, and found that police layoffs were associated with significant increases to overall crime, violent crime, and property crimes.

One clear impact of having fewer officers available is that police operations are negatively impacted, particularly in the form of longer response times to citizen calls for help (Asher, 2023). A recent analysis revealed that when the Salt Lake City Police Department came under intense turnover pressure in 2020-2021, their ability to respond to emergency calls quickly suffered (Mourtgos et al., 2022b). The analysis provides breakdowns of the effects of adding additional officers by workshift. For example, an additional ten officers on an afternoon shift would equate to a 21% improvement in response times overall. As police response times grow slower, the negative impacts on the community increase. For example, researchers have recently shown that a 10% slower response time causes a 4.7% decrease in likelihood of clearing crime, and this effect is stronger for violent crimes (Blanes i Vidal & Kirchmaier, 2018). Shorter response times are also linked to an increased ability to apprehend suspects during in-progress burglaries (Cihan et al., 2012). Crime is not the only outcome affected by police response times, as slower response times have also been linked to increased traffic fatalities (Liu, 2022).

The cost of suddenly losing police officers extends beyond attenuated crime control, more traffic fatalities, slower response times, and lower odds of clearing crimes. Given that police agencies are public entities and have a mandate to spend public monies effectively and efficiently, there is also a high economic cost to police turnover. This is especially true with voluntary turnover, as retirements are constrained by mandatory years of service and thus more predictable with regard to in-city budgetary planning. The cost of turnover is at least partly a function of the high costs of hiring and training new officers, which includes background checks, physical and psychological testing, and training expenses (Hilal & Litsey, 2020). These costs are estimated to range from between one and five times the actual salary of an officer (Orrick, 2008). Other research supports the contention that sudden decreases are costly in civic terms. The “Great Recession” forced some cities to shrink the number of police personnel they employed (Wilson et al., 2010), and these decisions were found to adversely affect police “cohorts” over long periods of time (Wilson & Heinonen, 2012). Demand for policing remained high even while available personnel decreased, leading to mandatory overtime and increased complaints against officers for use-of-force incidents (Weichselbaum & Lewis, 2020).

When we put this in the context of the effects reduced personnel has on crime, it is plain that fewer officers translates to more violent crime (Chalfin et al., 2020; Piza & Chillar, 2021), which in turn leads to substantial financial costs that are borne by both individual victims of crime and by the broader community. As one might expect, the pressure on public coffers from crime costs related to police turnover are significant. For example, Heaton (2010) estimates that, in Los Angeles, the loss of just 1.2% of the LAPD’s sworn officers (n=90) would equate to approximately $42.5 million (2022 inflation-adjusted dollars) in increased costs as a result of crime.2

There seems to be little doubt that police turnover has such a significant impact on crime that it merits study on a national scale. Turnover has deleterious impacts at both the organizational and community level. Reductions in personnel inhibit an agency’s strength and cohesion, reduces productivity, increases the volume of public complaints, burdens city budgets with increased recruitment and training costs, and impedes service delivery (Wilson et al., 2010; Wilson & Heinonen, 2012). These are all measures of police performance, which is itself directly linked to public confidence and trust in the police (Ren et al., 2005). The loss of officers drains policing agencies of the “ability to engage in proactive problem solving and respond to calls for service, which in turn could contribute to increased crime and further erosion of public trust” (Mourtgos et al., 2022a, p. 11).

Police Turnover in the post-Floyd Era

The evidence above points to the critical importance of understanding if, and where, police turnover did surge in the post-Floyd era. A long line of media reports highlight police agencies reporting urgent shortages of sworn personnel, including in Seattle (Kakade & Ramirez, 2022), Detroit (Hunter, 2022), Salt Lake City (Harkins & Larsen, 2021), Portland (Arden, 2021), New York City (Thorne, 2022), and Minneapolis (Bailey, 2020). While we focus on US police agencies here, it is worth considering that reports stretch to non-domestic locales experiencing rapid increases in departures from police service, including the UK (Charman & Bennett, 2022).

In addition to media reports, professional membership organizations such as the Police Executive Research Foundation (PERF) have occasionally weighed in with observational, stylized reporting on their member agencies’ experiences with turnover. For example, a recent PERF report indicates that among 194 agencies that participated in a survey on turnover, they saw, on average, an 18% increase in resignations from 2020-2021, and a 45% increase in retirements, even while hiring activity declined by approximately 5% (PERF, 2021). While reporting has been primarily limited to non-peer-reviewed studies, most observers see the current turnover crisis as the worst of its type. Following his organization’s report on the police turnover crisis, PERF Executive Director Chuck Wexler commented that “We are in uncharted territory right now, policing is being challenged in ways I haven’t seen, ever” (Westervelt, 2021, para. 4).

Academics, however, have been slow to describe and explain what, if anything, is happening in police turnover today, and how it might differ from concurrent trends across the labor force. A lack of research, poor data quality and access, and the fractured nature of the country’s 18,000 police agencies, has given rise to competing explanations and, in some cases, a reluctance to acknowledge that there is a current problem in recruiting and retaining sworn police personnel. For example, in a journalistic report from the Marshall Project (Weihua & Ilica, 2021), the authors report that the concerns about a hike in police turnover following the Floyd protests were overblown, a conclusion that was based on internal analysis using Bureau of Labor monthly data. However, the data used in the report captured total law enforcement employment, including non-sworn personnel (which may have differential employment patterns and concerns related to turnover). Furthermore, the analysis did not compare pre- and post-Floyd periods; instead, it considered year-on-year­ total employment, and compared the effect of Covid-19 on total law enforcement employment with healthcare, restaurants, and education – all of which were dramatically affected by Covid-era shutdowns. In short, this comparison does not seem applicable in the policing context.

In contrast, a study from just before the Marshall Project report restricted its analysis to one large city in a western state capital (Mourtgos et al., 2022a). The authors found that resignations, but not retirements or involuntary departures (firing), saw a significant increase from June to December 2020, compared to the previous 60 months. Here, resignations during the post-Floyd period were up approximately 279% over baseline expectations, and the authors conducted a one-year forecast, predicting that the phenomenon could have “negative consequences throughout the organization for decades to come by disrupting workforce cohorts” (Mourtgos et al., 2022a, p. 16). While these results undoubtedly provide a solid basis for moving forward with more research, the study was limited to a single agency, which makes generalizing to other agencies challenging.

While academic disagreements continue, the volume of calls from police agencies concerned about their ability to operate effectively seems to have generated political momentum to increase federal help. In July 2022, President Biden unveiled his “Safer America Plan,” which promised $13 billion to help train 100,000 additional police officers (The White House, 2022). The policy is not yet finalized, and questions remain about the practicality of this initiative. For example, most media reports about a police recruitment and retention crisis have focused on the largest cities in the country; however, it does not necessarily follow that smaller agencies, or agencies that do not happen to exist in a major national media market, are experiencing similar trends. From a policy perspective, several possibilities exist. It might be that all police agencies are experiencing a turnover crisis. Or, it might be that only certain types of agency are affected – perhaps only large agencies in major cities are subject to the worst of the crisis, or perhaps municipal police departments, but not sheriffs agencies (or vice versa) are seeing these effects. It seems clear that the answers to these questions should inform how federal funding is distributed. If it is true that the post-Floyd era funneled police out of the profession in a uniform way across agencies, then it also seems reasonable that funds would be distributed uniformly. On the other hand, uniformly distributed funds run the risk of under- or over-compensating, depending on whether agencies are (or are not) experiencing the post-Floyd-era effect differently based on size, locale, or agency characteristics.

Mechanisms and Theory

Our first task must be to understand why, and how, different forms of turnover might be related to the Floyd-era turmoil. We review three possible, competing explanations for why police turnover would increase: heightened danger (real or perceived), organizational and financial factors, and shifts in the socio-political environment. Some reporting points to all three factors as creating a more challenging environment for officers (Ridderbusch, 2021). To be a convincing mechanism, there must have been something that shifted in the Floyd era beyond mere long-standing chronic factors known to affect police perceptions of their work environment. For example, though turnover was not specifically studied, the post-Ferguson moment in US policing resulted in commentary that a “Ferguson effect” was driving a “war on cops” (MacDonald, 2016). These claims have generally been roundly dismissed following empirical investigation, with scholars finding no “Ferguson effect” on lethal (Maguire et al., 2017) or non-lethal violence directed at officers (Shjarback & Maguire, 2021), the number of people shot by police (Campbell et al., 2018), officers’ willingness to engage with community members (Wolfe & Nix, 2016), or crime rates (Pyrooz et al., 2016). To put it plainly, the “Ferguson effect” failed to materialize.

However, despite conflation of the Ferguson and post-Floyd eras, they are separate events and could conceivably have had different impacts on the police workforce. One possible narrative is that the actual danger officers faced on the job increased in the post-Floyd moment, leading to more turnover. Previous research has suggested that agencies with higher likelihood for dangerous and hostile encounters faced more police turnover (Schuck & Rabe-Hemp, 2018). Boehme and Kaminski (2023) looked to the post-Floyd period in Indianapolis and New Orleans, and found that suspect resistance and use of force increased in both cities, while officer injuries increased only in the New Orleans police department.

Sierra-Arévalo et al. (2023) provide the clearest evidence to date for a period of “retributive violence” directed at police officers in the immediate aftermath of the Floyd protests. They detect a clear spike in gun violence directed at police officers in the three weeks following George Floyd’s murder, with the trend regressing to slightly higher-than-expected levels thereafter. If officers perceived that occupational danger had increased following the Floyd protests (see e.g., Nix et al., 2018), perhaps this could have marginally increased turnover. However, the relatively short-lived, diffuse nature of this effect suggests that any turnover effects would also be short-lived if officers recognized that the increased danger lasted only a few weeks. There is, of course, the possibility that they did not, and that the spike was enough to instill a factually inaccurate belief that the job was inherently more dangerous. If taken seriously, this phenomenon would suggest a “fear ratcheting” effect, wherein officers’ generalized perceptions of fear only increase over time. This seems an unsatisfactory, small, disparate (and at this time entirely theoretical) prompt for the type of sustained trend reported by agencies and media.

Though the post-Floyd period began just a few months after the COVID-19 pandemic took hold across the US, no evidence or study has emerged that COVID-related concerns have a clear theoretical link to a possible increase in the numbers of officers resigning or retiring from the profession. Despite ample evidence that COVID-19 had noticeable effects on crime (Koppel et al., 2023), officer response levels (Jennings & Perez, 2020), and public perceptions of police (Sandrin & Simpson, 2021), we are unaware of any empirical evidence that officers were leaving in greater numbers due to COVID-era concerns. Indeed, the sole study examining the acceptance and efficacy of the COVID vaccine in a police agency indicates that officers were accepting of the vaccine and that the vaccine was successful in vastly reducing infections, which had been producing elevated staffing pressure through quarantine mechanisms, but not leavism (Mourtgos & Adams, 2021). Nonetheless, authors have made the point that pulling apart the effects from overlapping COVID-19 and Floyd-era police reform movement may be less important than describing and understanding how the front-line reality of police departments and their operations have been affected (Mourtgos & Adams, 2023).

A longer line of evidence indicates that officer leavism is primarily driven by organizational, financial, and socio-political factors. On the financial front, scholars have long documented that police salaries are a key way for agencies to manage turnover intent (Schuck & Rabe-Hemp, 2018). However, financial and organizational stress are unsatisfactory explanations for a sudden increase in police turnover, as these are well documented, chronic stressors in the context of police employment (Schuck & Rabe-Hemp, 2018; Shane, 2010, 2019); nor are we aware of any information suggesting that pay or benefits saw sudden negative shifts in the immediate aftermath of Floyd. Still, given the relatively narrow scope of studies in this area, it is possible that some interactive effect from some or all of these economic conditions, alongside other factors such as heightened physical danger, could lead to increased leavism.

This said, a strong foundation of media and academic reporting suggests that the socio-political environment did shift in response to Floyd protests (Reny & Newman, 2021), and we know from prior research that police officers attend closely to such shifts. For example, one large study funded by the National Institute of Justice found that, in interviews with officers employed at over 100 diverse agencies, officers believed an already negative sociopolitical environment had grown even worse in recent years (Saunders et al., 2019). These officers reported that a combination of unfair community expectations and excessive media hostility made them believe they could not be successful. As is the case with financial and organizational stress, police complaints about the media and critical community perceptions are not new. Niederhoffer (1967) notes that an important part of police cynicism is linked to distrust of the media and other community stakeholders, and contemporary research has continued to validate this phenomenon in modern policing (Adams & Mastracci, 2019; Mourtgos et al., 2020; Mourtgos & Adams, 2019; Nix et al., 2020; Nix & Pickett, 2017).

Only one study has directly linked post-Floyd-era pressures to increased police turnover. In this instance, the study referred only to voluntary turnover in one city (Mourtgos et al., 2022a). The authors highlight the perception among officers that the socio-political environment shifted considerably in the aftermath of the protests following George Floyd’s death. Other studies support these perceptions, noting a broad shift in support for police reforms endorsed by the Black Lives Matter movement (Boudreau et al., 2022). The protests quickly politicized attitudes towards the police, with strong negative shifts among large segments of the US populace (Reny & Newman, 2021). We know that officers pay attention to media hostility and citizen disrespect, both of which are key antecedents of what Nix et al. (2020) term “perceived audience legitimacy” – that is, how officers think they are viewed by the public. Though under-theorized, an increase in police turnover is consistent with a decline in perceived audience legitimacy: If officers perceive they are being set up to fail, unfairly targeted by media, and increasingly vulnerable to poorly-framed narratives , they may well perceive a “no-win socio-political environment” and decide to quit (Mourtgos et al., 2022a, p. 17).

Procedure and Method

Previous scholarship on post-Floyd changes in police turnover found significant increases in resignations, but not retirements, in the six months following the protests (Mourtgos et al., 2022a). However, that study was limited to a single agency, leaving its generalizability an open empirical question. We derive two related hypotheses. The first is related to a specific form of turnover: resignations.

H1: Voluntary resignations increased significantly between June 2020 and September 2022.

The other form of turnover we analyze is monthly retirements at the agency level. Importantly, the sole empirical study on post-Floyd turnover found that retirements were unaffected in a single agency study (Mourtgos et al., 2022a). However, retirements are contingent on time served. That is, an officer who wants to retire must also consider whether their time in service meets the minimum requirements to be eligible for retirement. This differentiates retirements from resignations because resignations have no similar constraint. Therefore, it may be that the immediate six months studied in Mourtgos et al. was simply too short to pick up a retirement effect, and by studying a longer period, as we do here, officers who wanted to retire and had become eligible in the two-year post-Floyd period would be able to. Therefore, we test a similar retirement hypothesis:

H2: Retirements increased significantly in the period between June 2020 and July 2022.

To ensure enough count data for each of the types of turnover, we restricted our sampling frame to agencies with more than 100 full-time, sworn officers. Though such agencies account for less than 10% of police departments nationwide, they employ over 60% of all full-time officers (Goodison, 2022). Beginning in the fall of 2022, we began contacting agencies in order to request turnover data. Each author contacted agencies with whom they had pre-established research relationships, and shared the project scope and requirements.

Our final sample includes fourteen agencies. Twelve agencies were able to provide historical turnover data that distinguished between leaving types. These geographically dispersed agencies were: Alexandria PD (VA), Austin PD (TX), Burlington PD (NC), Denver PD (CO), Manchester PD (NH), Greenville County Sheriff’s Office (SC), Omaha PD (NE), Richland County Sheriff’s Office (SC), Riverwood County Sheriffs Office (Sunbelt), Salt Lake City PD (UT), Seattle PD (WA), and Wichita PD (KS).3 In addition, we were able to obtain publicly available staffing data from two agencies: Memphis PD (TN), and Chicago PD (IL). Table 1 (below) reports basic agency information, including population, full-time employees, and ratios of employees to population.

While this is a convenience sample, we intentionally sought out a diverse cross-section of large police agencies in the US, including police forces from capital cities, suburban areas, and sheriff’s departments from across the country. Another advantage of assembling this dataset is that it enables a more granular look at turnover – i.e., at the monthly level. This differentiates the current study from previous turnover studies that have relied on yearly counts (Hur, 2013). Such counts can mask intra-year variation and overlapping causal claims. For these reasons, the current study provides the fullest picture to date of how policing turnover may have shifted in the aftermath of Floyd.


Data Begins



Authorized Sworn

Officers 2020

Sworn FTE

per 1000 pop.

Alexandria PD






Austin PD






Burlington PD






Chicago PD






Denver PD






Manchester PD






Memphis PD






Greenville SO






Omaha PD






Richland SO






Riverwood SO






Salt Lake City PD






Seattle PD






Wichita PD






Table 1: Agency Descriptives

Note: Population from 2020 census. PD = police department; SO = Sheriff’s Office


We use agency-level, monthly administrative data on turnover to test the hypotheses that resignations and retirements increased following the Floyd protests. We observed some 20,756 turnovers across fourteen agencies in the period studied. We first present raw linear trends for all agencies. We turn to Bayesian Structural Time Series modeling to allow cross-agency comparisons and calculate the magnitude of turnover in each unique agency context. We approach this analysis in two phases. We first estimate synthetic counterfactuals for each jurisdiction for the post-Floyd period. We then evaluate the difference in means between the observed data and synthetic counterfactual data at the monthly level for the entire post-Floyd period. The synthetic counterfactual approach allows us to estimate a total difference between observed turnover and turnover in a counterfactual world where Floyd was not murdered—the synthetic turnover after May 2020 being a probabilistic function of previous trends.

As a first step, we plotted and inspected the raw data: the monthly number of resignations and retirements in each agency. Figure 1 plots the linear trends for resignations, while Figure 2 plots the linear trends for retirements. In each grid of plots, the trends are decoupled at the intervention point (i.e., June 1st, 2020). These visualizations provide descriptive but crude evidence of whether the post-Floyd environment was associated with an increase, decrease, or steady pattern in resignations or retirements. Figure 1 shows that there was a pattern of increased resignations in ten agencies, while Figure 2 suggests that seven agencies saw increases to retirements.

While these preliminary observations are informative, they are simple pre-post comparisons and insufficient for several reasons. First, this kind of naive comparison cannot account for seasonality in the trends; moreover, the varying lengths of pre-intervention data across agencies render this initial overview somewhat simplistic. For example, while Chicago PD’s open data goes back to 2002, the pre-intervention period in other agencies ranged from 2012 to as late as 2017. This feature of the data makes raw linear comparisons unwise because shorter pre-intervention measures will naturally increase our uncertainty levels (and vice-versa). Second, the patterns are only suggestive, and are therefore only capable of demonstrating a general direction (increased or decreased), without providing solid ground for understanding the exact magnitude of difference between pre- and post-trends (should such a difference be present). Further, each agency is assumed to have unique turnover patterns, such as for retirements that are necessarily a product of the original hire date, or regular hiring which is commonly linked to fiscal calendar considerations, which vary widely across public organizations. Finally, because some agencies are very large, their counts of turnover will necessarily be much larger than smaller agencies, further confounding our ability to make raw, naïve comparisons. Therefore, to better inform policymakers about how different agencies are experiencing turnover, we must derive estimates for the magnitude of any detectable effect in the context of each agency, prior to making comparisons. We focus on this undertaking in the next analytic phase.

Figure 1: Linear Trends, Resignations

Figure 2: Linear Trends, Retirements

Analysis and Method

We leverage our data using Bayesian structural time series (BSTS) analysis. This method has been used previously to investigate police turnover in a single agency (Mourtgos et al., 2022a). Bayesian methods assume a probability distribution for each parameter coefficient rather than assuming one ‘true’ population parameter. The ability to acknowledge and leverage uncertainty through Bayesian inferential methods allows researchers to better guard against false-negative results, inflated false-discovery rates, and inflated effect sizes (Barnes et al., 2019). Moreover, Bayesian methods require researchers to better account for the data-generating process and allow for testing the proposed model against the observed data, thus providing greater confidence in the findings (McElreath, 2020).

BSTS is a specific form of Bayesian analysis which we use to develop counterfactual distributions. In other words, this method allows us to develop a probability function of what would have been expected to happen in the absence of the Floyd protest movement. BSTS models are best described as observation equations, linking observed data with an observed latent state and transition equation. The transition equation describes the latent state’s development over time (Brodersen et al., 2015a; Mourtgos et al., 2022a). The observation equation is defined as:

yt =ZtTαt+ εty_{t\ } = Z_{t}^{T}\alpha_{t} + \ \varepsilon_{t}

where yt y_{t\ }is a scalar observation, ZtZ_{t} is an output vector, and αt\alpha_{t} is the unobserved latent state. The transition equation is defined as

αt+1 = Ttαt+ Rtηt\alpha_{t + 1}\ = \ T_{t}\alpha_{t} + \ R_{t}\eta_{t}

where αt\alpha_{t} is the unobserved latent state, TtT_{t} is the transition matrix, and RtR_{t} is the control matrix. RtηtR_{t}\eta_{t} accounts for seasonality. The error terms (εt\varepsilon_{t} and ηt\eta_{t}) are Gaussian and independent (Brodersen et al., 2015b).

BSTS models use historical time series data on an outcome of interest to forecast a synthetic counterfactual of that outcome’s post-intervention trend while transparently integrating the uncertainty of such forecasting. Because BSTS models produce predictions based on prior observations (i.e., the observed trend’s current state, t=k, informs estimates of the trend’s next state, t = k + 1), the resulting counterfactual can incorporate unobserved data generation processes reflected in the observed, historical data. Of course, there is inherent uncertainty in accounting for every possible unobserved variable that may influence police turnover. However, unlike other forecast models which produce point estimates, the posterior probability distributions produced by a BSTS transparently present the uncertainty of the model. Thus, the means of the posterior probability distributions which we present can be better assessed with information about the entire probability distribution of the posterior inference, such that readers can assess for themselves their level of confidence in our results.

A BSTS model was estimated for each jurisdiction’s turnover measures (voluntary resignations and retirements). To estimate the BSTS model, all data available within each jurisdiction up to June 2020 was incorporated (each jurisdiction provided different periods of time—see Table 1 for the test period associated with each jurisdiction). Each model used a local level trend state component which assumes the trend is a random walk4, along with a weakly informative empirical prior placed on the sigma parameter. Ten thousand Markov Chain Monte Carlo (MCMC) iterations were simulated to fit each model. The resulting models are probabilistic estimations of ‘normal’ turnover (both resignations and retirements) for each jurisdiction. We use these models to forecast a synthetic counterfactual for what turnover ‘should’ have looked like without the intervention of the events surrounding Floyd’s murder (as with pre-Floyd periods of time, each jurisdiction provided varying lengths of post-Floyd data points—see Table 1 for the time period associated with each jurisdiction).

Next, to quantify the aggregate effect of Floyd’s murder on each agency’s turnover, we evaluated the difference in means between the observed data and synthetic counterfactual at the monthly level for the entire post-intervention period.5 If we denote the set of observed data as D, consisting of all observed and synthetic counterfactual values, a descriptive model has the following form, where µ is the mean of the observed and synthetic counterfactual values; 𝜎 is the standard deviation of the observed and synthetic counterfactual values; and ν\nu (nu) is a normality parameter shared by the two measures.6

p(D|μobs, σobs, μsyn, σsyn, ν) × p(μobs, σobs, μsyn, σsyn, ν)p(D) =p(μobs, σobs, μsyn, σsyn, νD)\frac{p\left( D \middle| \mu_{obs},\ \sigma_{obs},\ \mu_{syn},\ \sigma_{syn},\ \nu \right)\ \times \ p(\mu_{obs},\ \sigma_{obs},\ \mu_{syn},\ \sigma_{syn},\ \nu)}{p(D)}\ = p(\mu_{obs},\ \sigma_{obs},\ \mu_{syn},\ \sigma_{syn},\ \nu|D)

Because the integral is impossible to compute in many Bayesian models, the posterior distribution is estimated by generating a large representative sample through MCMC methods. This difference in means method yields complete distributional information about the parameters of both measures, with µobs - µsyn being the causal effect of interest (Angrist & Pischke, 2009; Kruschke, 2013).


Figure 3 visualizes the observed and counterfactual trends pre- and post-Floyd for resignations. Figure 4 shows the same for retirements. The overwhelming majority of agencies experienced resignation levels above the predicted post-Floyd counterfactual, whether in the form of a substantial immediate jump in resignations or an increasing trend over time. While the effect on retirements appears more muted and has exceptions, the overall story is the same: the levels of retirement are above what the probabilistic counterfactual would predict. Exact statistical information is reported in Table 2 and Table 3, where the tables collect each jurisdiction’s experienced aggregate change in Resignations and Retirements.

Our analysis indicates that the police turnover crisis is real and has had a significant impact on law enforcement agencies. Resignations (31.8%) and retirements (63.9%) form the largest types of turnover across all time periods and agencies, while terminations form a relatively small amount (4.4%). The data show that the majority of the cities studied have seen an increase in both the number of resignations and retirements. Alexandria, Austin, Chicago, Denver, Memphis, Salt Lake City, Seattle, Riverwood County, and Wichita all show an increase in resignations. Additionally, Austin, Chicago, Denver, Omaha PD, Richland County, Seattle, and Wichita have all experienced an increase in retirements. Our findings are consistent with the professional and media reporting on the police turnover crisis. The two years following the Floyd protests are characterized by an increase in police turnover, with the majority (78%) of the agencies experiencing sharply increased resignations, retirements, or both. However, the size of this effect was not consistent across all agencies, and the turnover crisis appears to affect agencies differently.

Figure 3: Resignations

Figure 4: Retirements


Observed Mean, post-Floyd

Counter-factual Mean, post-Floyd


Aggregate Change Above/Below Expected

95% HDI

Difference Probability






[.23, 1.34]







[.88, 2,73]







[-.36, .67]







[11.37, 21.51]







[1.54, 4.03]







[-.53, .51]







[1.29, 3.75]







[-1.21, .14]







[-.19, .54]







[-1.18, 1.06]







[-.22, 3.39]







[1.75, 5.69]







[1.11, 2.84]







[.67, 2.26]


Table 2: Resignations, BSTS Modeling


Observed Mean, post-Floyd

Counterfactual Mean, post-Floyd


Aggregate Change Above/Below Expected

95% HDI

Difference Probability






[-.63, .02]







[1.88, 5.36]







[-.03, -.02]







[-3.50, 23.10]







[.66, 3.78]







[-.08, .08]







[-1.81, 1.35]







[-.019, -.018]







[.08, 2.65]







[-.11, 1.10]







[-.41, 1.40]







[2.79, 6.13]







[-.49, .64]







[-.06, .91]


Table 3: Retirements, BSTS Modeling


Appropriately staffing the nation’s police agencies has been an intermittent focus of practitioners and policymakers for several decades. In the aftermath of large protests against both the lawful and unlawful use of lethal force by the police in the last decade (Buchanan et al., 2020), agencies around the country have raised the alarm because they are unable to effectively mitigate simultaneous decreases in recruitment and increases in resignations and retirements. By documenting substantial turnover increases amongst the large majority of studied agencies, we have demonstrated the likely veracity of this agency perspective. Our results suggest difficult times ahead for policing: The scope of these intersecting crises has only recently become apparent and agencies are already struggling to fulfill their operational commitments; historic increases in violent crime have left cities scrambling for answers in their quest to control increasing crime, and all the while the deficit in staffing itself contributes to that increase in crime. Indeed, Lewis and Usmani (2022) have recently made the case that the U.S. is underpoliced given its levels of violent crime, arguing that our carceral system places too much emphasis on incarceration and not enough on providing public safety through policing.

The primary conclusion from our analysis is that police resignations and retirements have substantially increased in the two years following the Floyd protest movement. Eleven of the fourteen agencies we examined show a volume of post-Floyd turnover above the synthetic counterfactual. The outlier agencies here are either smaller suburban agencies, or in one case, a sheriff’s department. Two of these agencies, Burlington PD and Greenville SO, saw slightly lower resignations or retirement numbers, while Manchester PD saw no substantial shifts in either form of turnover. These results support news stories and jurisdictions’ repeated warnings of critical deficits in the police workforce.

Turnover in the Agency Context

Putting the turnover effects into the context of the individual agencies is important. Large agencies might be expected to lose more officers in absolute terms, even while relatively smaller agencies are experiencing turnover effects more strongly. To place our findings in a generalizable context, we summarized resignation and retirement effects by agency. Based on authorized strength (full-time law enforcement officers), we calculated the total loss of officers by turnover type, and the proportion of officers lost. These results are reported in Table 4. Importantly, these excess loss numbers are surplus loss – that is, they are added to whatever “normal” turnover a given agency experienced.

In real terms, across our fourteen studied agencies, an additional 1429 officers were lost over what normal turnover would predict, equating to a surplus 5.4% loss of authorized strength. Note this reflects a conservative loss estimate, as no agency in this study was operating at full authorized strength. For example, although Seattle had 1422 authorized sworn positions in 2020, the agency’s actual sworn strength that year was just 1094, and trending downward. This means that the “felt loss” of 228 officers (see Table 4 below) would equate to an additional 20.8% of sworn personnel. Therefore, while we use authorized strength to ensure consistent reporting across agencies, policymakers must also consider operational strength when calculating surplus loss.

These results demonstrate the consistent phenomenon of increased turnover nationally and how that effect has been experienced differentially. First, eleven of the fourteen agencies studied here experienced significant increases in the numbers of officers either resigning or retiring in the post-Floyd period through 2022. Only one agency, Manchester PD, had no substantial effect on either resignations or turnover. One type of agency appears to have suffered most – large, metropolitan policing agencies. Austin, Chicago, Denver, and Seattle all experienced substantial increases to both resignations and retirements. Austin, Denver, Seattle and Salt Lake City also experienced the highest levels of disruption, with between 6.7% to 16% excess losses of sworn personnel. However, these overall numbers should not be interpreted as meaning that lower overall loss means the agency is not struggling. The losses we report here are in addition to “normal” attrition, or the levels of resignations and retirements that would have been predicted based on the pre-2020 observations. For example, if Chicago PD would normally expect around 5% attrition, their 4.5% excess loss rate would equate to near double expectations.

Table 4: Summarizing Agency-level Effects


Excess Resignations

Excess Retirements

Real Loss

Authorized Strength 2020

Proportion Lost/Gained





































Greenville County


















Richland County






Riverwood County






Salt Lake City
























Note: “Real Loss” sums only statistically significant resignations and retirements, at the a = 0.1 level.

Policing Operations, Activity, and Crime Control

A sudden and sustained increase in police turnover has a number of implications for police operations. It can lead to a shortage of experienced officers, making it difficult for the department to investigate crimes and maintain public safety effectively. Personnel loss can also lead to decreased morale among the remaining officers, who may be required to work longer hours to compensate for the lack of personnel. Additionally, a high turnover rate can be costly for the department, as they may need to devote more resources to recruiting and training new officers. These cascading shifts can also create a cycle of instability within the department (Wilson & Heinonen, 2012), as new officers may not have the experience or support they need to carry out their duties effectively.

There are also potential impacts on crime control. A robust body of research has shown modest gains in crime control and public safety flow from both the number of officers in a community (Chalfin et al., 2020) and proactive forms of policing (Braga et al., 2019). However, the practical definition of “proactive policing” varies, and a growing number of scholars are building an evidence base on how to best deploy police officers to effectively reduce violent crime (Lum et al., 2020). For example, one recent study considers how varying the type of police proactivity in an area affects later crime rates (Carter & Wolfe, 2022). The authors found that in Seattle, proactive police stops for suspicious circumstances led to lower assault rates, while decreases in robbery rates were associated with three types of proactive stops (suspicious circumstance stops, premise checks, and traffic stops).

Related to police activity, we should also consider the implications of sharp police turnover, and falling overall staffing, on the ability of agencies to effectively control crime. While some critics correctly observe that we do not have unimpeachable and precise estimates for how each form of police activity directly affects specific forms of crime, most experts agree proactive policing can effectively deter crime (Weisburd & Majmundar, 2018). Increased police staffing, for example, is associated with net decreases in homicide, and that effect is doubled in Black communities (Chalfin et al., 2022). And while the relationship between police staffing and aggregate crime is well established, it is reasonable to ask whether sudden shocks, such as those documented here, may lead to larger magnitude crime effects that persist for many years. For example, the sudden layoff of 13% of the Newark Police Department (NJ) led to immediate increases in both violent and property crime, and those increases became more substantial in each of the five years following the layoffs (Piza & Chillar, 2021).

As research continues and evidence mounts, scholars can positively impact police operations by capturing what makes policing more effective while concurrently reducing negative outcomes. However, proactive policing requires some room for maneuver on a day-to-day basis. That is, it requires sufficient staffing, such that calls for service do not overwhelm the agency’s capacity, and officers can use non-call-for-service time to engage in proactive policing. In instances where agencies are struggling with turnover crises, the outcomes are manifest: Fewer officers in operation leads to slower response times for service calls (Mourtgos et al., 2022b), and reduces the amount and type of proactive policing that can occur.

Personnel Management

In conversations with command staff at a university police department, one commander noted that a concern of their administration was that the intense pressure of increased turnover would result in problematic employees being brought into the agency. In response, a sort of mantra had developed in the agency: “Better a hole than an asshole.” This epithet highlights a potentially serious, but currently unsubstantiated consequence of the operational pressures that can arise in the aftermath of high turnover. When facing intense turnover pressure, agencies may well attempt to compensate by relaxing recruitment standards (Shjarback, 2023). In turn, this could create further public outcry as less qualified officers engage in what inevitably would be lower-quality policing (Adams et al., 2023). In other words, while some believe a downturn in the numbers of US police officers will result in improved outcomes – speculating that the bad apples have left the barrel12 - the reality may be quite the opposite. The fact is that there is no compelling theory as to why officers more prone to misconduct or other bad acts would be more likely to leave an agency under operational pressure. On the other hand, officers more deeply committed to procedurally just, democratic, and constitutional policing – all goals of progressive policing – could flee what is perceived as a failing department. In a hiring environment where agencies prioritize attracting experienced officers from other agencies, highly performing officers might be more likely to leave failing agencies, as they would be more likely to get hired elsewhere compared to officers prone to misconduct or other problems. The urge to leave may even be exacerbated by mandatory overtime (Kanik, 2023) and the dissolution of non-patrol units (Kuhn, 2023), which often follow severe personnel shortages. This phenomenon was recently witnessed during Florida Governor DeSantis’ trip to Illinois, during which he urged Chicago officers to transfer to Florida agencies (Hickey, 2023). One former Chicago PD officer who made the move to Florida pointed to staffing pressure and the resultant long shifts as the reason (para. 11): “If you can’t staff, then you can’t put enough officers on the street…that’s where your twelve-hour days are coming from. Morale is definitely going to suffer.”

Linking Staffing to Police Reform

It is also important to note the ancillary negative effects of decreased police staffing. In the first instance, it is possible that such shortfalls could hamper criminal justice reform. One potential outcome of the sustained pressure that comes with high turnover is a lowering of disciplinary standards. An agency executive facing a disciplinary decision may be more inclined to overlook violations by personnel, which would be both inherently problematic and a hindrance to positive reform. While some potential disciplinary issues could be penny-ante, there is undoubtedly room for serious consequences, including opting not to terminate an employee who, all else being equal, would have been let go from the agency if the agency were better-staffed.

The potential for marginal changes in disciplinary firings remains a mostly theoretical point, and a difficult one to test empirically, given the relative rareness of terminations we observe in the data. However, one of the largest agencies in the country, Chicago PD (IL), provides a unique look at twenty years of staffing data. We provide a preliminary test of this question by looking at termination trends before and after the Floyd protests. Appendix Figure 2 offers preliminary evidence that the stress of sudden and exaggerated trends in resignations and retirements threatens pro-organizational outcomes, such as the termination of under-, non-, and mal-performing officers. However, we caution against drawing firm conclusions when interpreting this evidence. For example, an alternate compelling theory is simply that officers who are more likely to generate complaints through aggressive policing tactics choose to “lay low and avoid trouble” (Van Maanen, 1973, p. 4) during periods of intense public scrutiny. We are not able to differentiate these mechanisms in the present data, and further investigation using comparison agencies is needed before it will be possible to draw any translatable conclusions.

There is also evidence that staffing levels adversely affect an agency’s ability to reform. Underpolicing, and an ongoing staffing crisis in Baltimore Police Department, has been implicated in holding up progress in the city’s attempts to comply with federal consent decree mandates. Judge James Bredar, who is overseeing the mandate, recently called police staffing “the most daunting challenge” and was unequivocal in voicing his concern: “I cannot overstate the seriousness of this issue,” he said. “The problem is at a level of severity that is bigger than the police department to solve” (Simms, 2023, para. 6).

Rapid hiring may also exacerbate misconduct issues, as younger officers are more likely to use force or commit serious misconduct (McElvain & Kposowa, 2004; Wolfe & Piquero, 2011); moreover, their inexperience contributes to racially disparate policing (Pryor et al., 2020). This is an under-theorized area; however, at its core is the notion that a lack of adequate personnel could lead to more misconduct and use-of-force which, in turn, could worsen police-community relations. The Memphis police department, which our analysis demonstrated experienced significant additional staffing stress in the post-Floyd period, was already facing significant staffing pressure after losing approximately 20% of its sworn officers since 2011. Facing mounting pressure to address high levels of violent crime, the city responded by lowering hiring standards in 2018 (Daves, 2018), and loosening the requirements for younger officers joining specialty squads, such as the now-disbanded SCORPION unit (Lucas, 2023). Five members of that unit, all with just two to five years of experience, are now facing second-degree murder charges in the killing of Tyre Nichols.

It is important to note that these are unsubstantiated theories, albeit ones rooted in straightforward logic. While we have outlined the core effects of sudden resignations and retirements, more research is required regarding the complex interplay between recruitment and retention. While both have independent effects, agencies facing simultaneous and rapid increases in resignations and retirements are likely to be the locus of the theoretical impacts on recruitment and disciplinary outcomes outlined in this section.

Competing Explanations

Considering what we have proposed above, one competing explanation should be noted. Recent economic reporting supports the contention that overall public employment decreased by 1.2% between March 2020 and March 2021 (Duret & Weithua, 2023). While both private and public sector employment were down immediately after the start of the Covid-19 crisis, the private sector has mostly returned to its previous employment levels; the public sector, however, continues lagging its pre-pandemic levels. Jobs data indicate that an increase in voluntary resignations from public service is a key driver of this overall trend in public employment. Accordingly, one possible explanation for this phenomenon is that the trends we report here are less a function of policing and, instead, reflect broader public sector labor patterns.

Still, other studies have attempted to differentiate between policing and other public service employment. They have found that while “governments are understaffed and unequipped to deal with further losses,” these losses are concentrated in policing, as “law enforcement is the biggest group driving resignations and retirements” (Farmer, 2022, paras. 7–11). Further, as we discussed earlier, data limitations make interpreting these trends extremely challenging. The most significant limitation in leveraging this information is that the national law enforcement employment data does not differentiate between sworn and civilian employment. Even more importantly, these analyses are aggregated at the national level and are therefore blind to employee movement within a profession. For example, police executives at large urban agencies have blamed “lateral transfers” – where already employed officers move to a different agency – as being responsible for many voluntary resignations at their agencies (Marcius, 2022). The pressure for already-trained officers has led to agencies competing in the police labor market by offering ever-escalating signing bonuses for officers willing to transfer. In the national payroll data analyzed by Duret and Weithua (2023), such lateral movement would not show up as employment loss. This phenomenon gains further support when considering the magnitude range of our findings. We see a 16% surplus loss in Seattle PD, and rather large losses in Salt Lake City, Denver, and Austin. At the same time, suburban and smaller agencies such as those in Burlington and Manchester, experienced relatively moderate shifts. The contrast supports the idea that inter-agency lateral transfers could create a mirage of minor loss noted at the yearly, national level across the profession.

Our results contribute to the overarching discussion about police staffing in several ways. First, we have shown that turnover patterns vary by agency, and that those patterns must be considered within the unique context of each agency in order to understand the impact of increased resignation and retirement fully. Policy interventions must therefore be differentially targeted, depending on the size of the agency and the specific staffing pattern that is manifesting (resignations, retirements, or both). Second, we report estimates of the magnitude of any turnover trends and compare those estimates to those provided through reporting of national aggregate trends. Clearly, the largest urban police departments are experiencing dramatic increases in resignations in the aftermath of the post-Floyd protest movement, and these losses far outpace previous national estimates of 1.2% for all public employment (Duret & Weithua, 2023). The gap between the national estimates and those reported here suggests that lateral transfers play a significant but unascertainable role in the staffing puzzle confronting large US agencies.


The research team, with varying degrees of both academic and practitioner experience, shares at least one consistent perspective: we are skeptical of our ability to generalize from single, or even small numbers of agencies. We recognize this flaw in the present study. While we have provided the largest study to date on a much-commented upon “crisis” in policing, we still only offer insight into fourteen agencies, all employing 100 or more sworn officers. As with any public policy concern, particularly concerning the police, we urge continued efforts to standardize administrative data and make the resulting unstylized raw data available for public use. Only in this way will researchers be able to provide a complete and accurate picture of staffing issues and, beyond that, the accompanying effects on public safety and crime.

Another key point is that we have no way of knowing who departs from these agencies. One possible narrative is that turnover is not linked to specific, or intersectional, officer identity characteristics. However, it is at least possible that increased voluntary resignation is correlated with age, sex, or race characteristics. These are potentially troubling connections: While a great deal of progress is still to be made (Ba et al., 2022), recent decades have seen agencies making progress in diversity and hiring. However, it is not clear what the future holds in this regard. There may be an exodus of diverse hires; but the reverse is just as possible. Kochel (2022) explored the experience of officers working the Ferguson protests in 2014, and found that the same Black officers who bore the brunt of abuse on the skirmish lines expressed the highest levels of resilience and the lowest sensitivity to her measure of the “Ferguson Effect.” While heartening on the surface, it also demonstrates the limitations of research to date. Let us be clear: we should hope that the “good” officers are resilient to the post-Floyd pressures in policing and will stay in service and rebuild agencies into whatever better version of policing awaits. However, we must also be forthright regarding the question and the research undertaken to date: we have no idea what awaits. With the current study supporting and building on the initial findings reported in Mourtgos et al. (2022a), the balance of evidence should allow confidence in just one thing: that more cops have been calling it quits since George Floyd was murdered.

Our efforts here cannot prescribe the correct level of police staffing. Still, the post-Floyd protest era has been characterized by increased police turnover across eleven of the fourteen agencies studied. Recent scholarship suggests that at the national level, police staffing levels are inadequately balanced against the crime control and public safety demands that agencies face (Lewis & Usmani, 2022). However, any agency's ideal staffing level is difficult to pin down. Hollis and Wilson (2015) analyze the staffing levels of 15,917 US agencies and found wide variation that resisted easy categorization. Communities with similar characteristics varied a great deal in the number of officers employed. This is yet another under-theorized area of study that could greatly benefit communities as they seek to better calibrate their favored approach to public safety and crime control.


The authors shared equally in the development of this manuscript, including design, data collection, analytic decisions, and prose. Author order was determined by a series of arm-wrestling contests at the 2022 annual meeting of the American Society of Criminology. As Seneca notes, crimes often return to their teacher, and all errors remain Geoff Alpert’s.


Adams, I. T., & Mastracci, S. H. (2019). Police Body-Worn Cameras: Development of the Perceived Intensity of Monitoring Scale. Criminal Justice Review, 44(3), 386–405.

Adams, I. T., Nix, J., & Mourtgos, S. M. (2023, February 7). Memphis police numbers dropped by nearly a quarter in recent years – were staffing shortages a factor in the killing of Tyre Nichols? The Conversation.

Angrist, J. D., & Pischke, J.-S. (2009). Mostly harmless econometrics: An empiricist’s companion. Princeton University Press.

Arden, A. (2021, October 6). PPB officer numbers continue to drop. Here’s why some say they’re leaving. KOIN.Com.

Asher, J. (2023, January 9). Police Are Taking Longer To Respond [Substack newsletter]. Jeff-Alytics.

Ba, B., Kaplan, J., Knox, D., Komisarchik, M., Mariman, R., Mummolo, J., Rivera, R., & Torres, M. (2022). Who are the Police? Descriptive Representation in the Coercive Arm of Government.

Bailey, H. (2020, November 16). Minneapolis violence surges as police officers leave department in droves. Washington Post.

Barnes, J. C., TenEyck, M. F., Pratt, T. C., & Cullen, F. T. (2019). How powerful is the evidence in criminology? On whether we should fear a coming crisis of confidence. Justice Quarterly.

Bittner, E. (1970). The functions of police in modern society. US National Institute of Mental Health.

Blanes i Vidal, J., & Kirchmaier, T. (2018). The Effect of Police Response Time on Crime Clearance Rates. The Review of Economic Studies, 85(2), 855–891.

Boehme, H. M., & Kaminski, R. J. (2023). Suspect resistance, police use of force, and officer injuries in a post-Floyd era: An analysis of two large police departments. Police Practice and Research.

Boehme, H. M., Kaminski, R. J., & Nolan, M. S. (2022). City-Wide Firearm Violence Spikes in Minneapolis following the Murder of George Floyd: A Comparative Time-Series Analysis of Three Cities. Urban Science, 6(1), Article 1.

Boudreau, C., MacKenzie, S. A., & Simmons, D. J. (2022). Police Violence and Public Opinion After George Floyd: How the Black Lives Matter Movement and Endorsements Affect Support for Reforms. Political Research Quarterly, 75(2), 497–511.

Braga, A. A., Turchan, B. S., Papachristos, A. V., & Hureau, D. M. (2019). Hot spots policing and crime reduction: An update of an ongoing systematic review and meta-analysis. Journal of Experimental Criminology, 15(3), 289–311.

Brodersen, K. H., Gallusser, F., Koehler, J., Remy, N., & Scott, S. L. (2015a). Inferring causal impact using bayesian structural time-series models. Annals of Applied Statistics, 9(1), 247–274.

Brodersen, K. H., Gallusser, F., Koehler, J., Remy, N., & Scott, S. L. (2015b). Inferring causal impact using Bayesian structural time-series models. The Annals of Applied Statistics, 9(1), 247–274.

Buchanan, L., Bui, Q., & Patel, J. K. (2020, July 3). Black Lives Matter May Be the Largest Movement in U.S. History. The New York Times.

Campbell, B. A., Nix, J., & Maguire, E. R. (2018). Is the number of citizens fatally shot by police increasing in the post-Ferguson era? Crime & Delinquency, 64(3), 398–420.

Carter, T., & Wolfe, S. (2022). The effect of real-world proactive policing on crime. SocArXiv.

Cassell, P. G. (2020). Explaining the Recent Homicide Spikes in U.S. Cities: The “Minneapolis Effect” and the Decline in Proactive Policing. Federal Sentencing Reporter, 33(1–2), 83–127.

Chalfin, A., Hansen, B., Weisburst, E. K., & Williams Jr., M. C. (2022). Police Force Size and Civilian Race. American Economic Review: Insights, 4(2), 139–158.

Chalfin, A., Hansen, B., Weisburst, E. K., & Williams, Jr., Morgan C. (2020). Police Force Size and Civilian Race (Working Paper No. 28202; Working Paper Series). National Bureau of Economic Research.

Chalfin, A., & Kaplan, J. (2021). How many complaints against police officers can be abated by incapacitating a few “bad apples?” Criminology & Public Policy, 20(2), 351–370.

Charman, S., & Bennett, S. (2022). Voluntary resignations from the police service: The impact of organisational and occupational stressors on organisational commitment. Policing and Society, 32(2), 159–178.

Cihan, A., Zhang, Y., & Hoover, L. (2012). Police Response Time to In-Progress Burglary: A Multilevel Analysis. Police Quarterly, 15(3), 308–327.

Cohen, L. E., & Felson, M. (1979). Social Change and Crime Rate Trends: A Routine Activity Approach. American Sociological Review, 44(4), 588–608.

Copeland, C., del Carmen, A., & Semukhina, O. B. (2022). Revisiting the Ferguson effect: Law enforcement perception of recruitment in the post George Floyd era. International Journal of Police Science & Management, 24(3), 261–272.

Daves, M. (2018, June 26). MPD makes changes to college requirements for recruits. Https://Www.Actionnews5.Com.

Douglas, J. D. (1967). The Challenge of Crime in a Free Society: A Report by the President’s Commission on Law Enforcement and Administration of Justice. JSTOR.

Duret, D., & Weithua, L. (2023, January 21). It’s Not Just a Police Problem, Americans Are Opting Out of Government Jobs. The Marshall Project.

Farmer, L. (2022). The Great Resignation’s Impact on Local Government. Rockefeller Institute of Government.

Harkins, P., & Larsen, L. (2021, January 22). Exclusive: Why police resignations have spiked in Salt Lake City and what happens next. The Salt Lake Tribune.

Heaton, P. (2010). Hidden in Plain Sight: What Cost-of-Crime Research Can Tell Us About Investing in Police. RAND Corporation.

Hickey, M. (2023, February 21). Growing number of Chicago Police officers are taking up Florida on invitation to move. CBS Chicago 2.

Hilal, S., & Litsey, B. (2020). Reducing police turnover: Recommendations for the law enforcement agency. International Journal of Police Science & Management, 22(1), 73–83.

Hollis, M. E., & Wilson, J. M. (2015). Police staffing levels: Disaggregating the variation. Policing: An International Journal of Police Strategies & Management, 38(4), 820–839.

Hunter, G. (2022, August 31). Detroit is losing a cop nearly every day: “A lot of officers are saying: Screw it.” The Detroit News.

Hur, Y. (2013). Turnover, Voluntary Turnover, and Organizational Performance: Evidence from Municipal Police Departments. Public Administration Quarterly, 37(1), 3–35.

Jennings, W. G., & Perez, N. M. (2020). The Immediate Impact of COVID-19 on Law Enforcement in the United States. American Journal of Criminal Justice, 45(4), 690–701.

Johnson, J. (2023, February 22). ‘Defund the Police’ Led to Lower Standards. Wall Street Journal.

Kakade, A., & Ramirez, Q. (2022, July 7). New recruiting plan unveiled as Seattle Police Department reports 30-year low in officers. King5.Com.

Kanik, H. (2023, January 18). Short-staffed Los Gatos police department loses two more officers. The Mercury News.

Kochel, T. R. (2022). Policing Unrest: On the Front Lines of the Ferguson Protests. NYU Press.

Koppel, S., Capellan, J. A., & Sharp, J. (2023). Disentangling the Impact of Covid-19: An Interrupted Time Series Analysis of Crime in New York City. American Journal of Criminal Justice, 48(2), 368–394.

Kruschke, J. K. (2013). Bayesian estimation supersedes the t test. Journal of Experimental Psychology: General, 142(2), 573–603.

Kuhn, A. (2023, January 5). Merced police disbanding key enforcement team due to staffing shortages, says acting chief. Merced Sun-Star.

Lewis, C., & Usmani, A. (2022). The Injustice of Under-Policing in America. American Journal of Law and Equality, 2, 85–106.

Liu, C. (2022). Exploration of the police response time to motor-vehicle crashes in Pennsylvania, USA. Journal of Safety Research, 80, 243–253.

Lucas, F. (2023, January 31). Could Lower Standards for Police Recruits Breed Future Misconduct? The Daily Signal.

Lum, C., Koper, C. S., & Wu, X. (2022). Can we really defund the police? A nine-agency study of police response to calls for service. Police Quarterly, 25(3), 255–280.

Lum, C., Koper, C. S., Wu, X., Johnson, W., & Stoltz, M. (2020). Examining the Empirical Realities of Proactive Policing Through Systematic Observations and Computer-Aided Dispatch Data. Police Quarterly, 23(3), 283–310.

MacDonald, H. (2016, February 22). Stat crimes matter: How researchers try to obscure the existence of the Ferguson effect. City Journal.

MacFarquhar, N. (2021, June 11). Departures of Police Officers Accelerated During a Year of Protests. The New York Times.

Maguire, E. R., Nix, J., & Campbell, B. A. (2017). A war on cops?: The effects of Ferguson on the number of U.S. police officers murdered in the line of duty. Justice Quarterly, 34(5), 739–758.

Marcius, C. R. (2022, December 9). N.Y.P.D. Officers Leave in Droves for Better Pay in Smaller Towns. The New York Times.

McElreath, R. (2020). Statistical rethinking: A Bayesian course with examples in R and Stan (2nd ed.). CRC Press.

McElvain, J. P., & Kposowa, A. J. (2004). Police officer characteristics and internal affairs investigations for use of force allegations. Journal of Criminal Justice, 32(3), 265–279.

Morrow, W. J., Vickovic, S. G., Dario, L. M., & Shjarback, J. A. (2019). Examining a Ferguson Effect on College Students’ Motivation to Become Police Officers. Journal of Criminal Justice Education, 30(4), 585–605.

Mourtgos, S. M., & Adams, I. T. (2019). The rhetoric of de-policing: Evaluating open-ended survey responses from police officers with machine learning-based structural topic modeling. Journal of Criminal Justice, 64(C), 1–1.

Mourtgos, S. M., & Adams, I. T. (2021). COVID-19 vaccine program eliminates law enforcement workforce infections: A Bayesian structural time series analysis. Police Practice and Research, 0(0), 1–9.

Mourtgos, S. M., & Adams, I. T. (2023). Police Proactivity in an Era of Pandemic and Protest. In M. Deflem (Ed.), Crime and Social Control in Pandemic Times (Vol. 28, pp. 207–224). Emerald Publishing Limited.

Mourtgos, S. M., Adams, I. T., & Nix, J. (2022a). Elevated police turnover following the summer of George Floyd protests: A synthetic control study. Criminology & Public Policy, 21(1), 9–33.

Mourtgos, S. M., Adams, I. T., & Nix, J. (2022b, August 24). Linking the Workforce Crisis, Crime, and Response Time. Police Chief Magazine.

Mourtgos, S. M., Mayer, R. C., Wise, R. A., & O’Rourke, H. (2020). The Overlooked Perspective of Police Trust in the Public: Measurement and Effects on Police Job Behaviors. Criminal Justice Policy Review, 31(5), 639–672.

New York Times. (1953, September 2). Police Turnover in State Soaring; 120% Rise in ’52 Laid to Low Pay and Long Hours—Social Security Link Opposed. The New York Times, 49.

Niederhoffer, A. (1967). Behind the shield: The police in urban society. Doubleday.

Nix, J., & Pickett, J. T. (2017). Third-person perceptions, hostile media effects, and policing: Developing a theoretical framework for assessing the Ferguson effect. Journal of Criminal Justice, 51, 24–33.

Nix, J., Pickett, J. T., & Wolfe, S. E. (2020). Testing a Theoretical Model of Perceived Audience Legitimacy: The Neglected Linkage in the Dialogic Model of Police–community Relations. Journal of Research in Crime and Delinquency, 57(2), 217–259.

Nix, J., Wolfe, S. E., & Campbell, B. A. (2018). Command-level Police Officers’ Perceptions of the “War on Cops” and De-policing. Justice Quarterly, 35(1), 33–54.

Orrick, D. W. (2008). Recruitment, Retention, and Turnover of Police Personnel: Reliable, Practical, and Effective Solutions. Charles C Thomas Publisher.

PERF. (2021). Police Workforce Survey 2021.

Petersen, K., Weisburd, D., Fay, S., Eggins, E., & Mazerolle, L. (2023). Police stops to reduce crime: A systematic review and meta-analysis. Campbell Systematic Reviews, 19(1), e1302.

Piza, E. L., & Chillar, V. F. (2021). The Effect of Police Layoffs on Crime: A Natural Experiment Involving New Jersey’s Two Largest Cities. Justice Evaluation Journal, 4(2), 176–196.

Pryor, M., Buchanan, K. S., & Goff, P. A. (2020). Risky Situations: Sources of Racial Disparity in Police Behavior. Annual Review of Law and Social Science, 16(1), 343–360.

Pyrooz, D. C., Decker, S. H., Wolfe, S. E., & Shjarback, J. A. (2016). Was there a Ferguson Effect on crime rates in large U.S. cities? Journal of Criminal Justice, 46, 1–8.

Ren, L., Cao, L., Lovrich, N., & Gaffney, M. (2005). Linking confidence in the police with the performance of the police: Community policing can make a difference. Journal of Criminal Justice, 33(1), 55–66.

Reny, T. T., & Newman, B. J. (2021). The Opinion-Mobilizing Effect of Social Protest against Police Violence: Evidence from the 2020 George Floyd Protests. American Political Science Review, 115(4), 1499–1507.

Ridderbusch, K. (2021, June 11). Violence, stress, scrutiny weigh on police mental health. Georgia Health News.

Sandrin, R., & Simpson, R. (2021). Public assessments of police during the COVID-19 pandemic: The effects of procedural justice and personal protective equipment. Policing: An International Journal, 45(1), 154–168.

Saunders, J., Kotzias, V., & Ramchand, R. (2019). Contemporary Police Stress: The Impact of the Evolving Socio-Political Context. Criminology, Criminal Justice, Law & Society, 20(1), 35–52.

Schuck, A. M., & Rabe-Hemp, C. E. (2018). Investing in people: Salary and turnover in policing. Policing: An International Journal of Police Strategies & Management, 41(1), 113–128.

Shane, J. M. (2010). Organizational stressors and police performance. Journal of Criminal Justice, 38(4), 807–818.

Shane, J. M. (2019). Stress Inside Police Departments. Routledge.

Shjarback, J. A. (2023, January 29). Tyre Nichols’ tragic death happened despite police reforms enacted to prevent it. NJ.

Shjarback, J. A., & Maguire, E. R. (2021). Extending Research on the “War on Cops”: The Effects of Ferguson on Nonfatal Assaults Against U.S. Police Officers. Crime & Delinquency, 67(1), 3–26.

Sierra-Arévalo, M., Nix, J., & Mourtgos, S. M. (2023). The “War on Cops,” Retaliatory Violence, and the Murder of George Floyd.

Simms, B. (2023, January 26). Judge overseeing BPD consent decree concerned about officer recruitment. WBAL.

Smith, M. (2022, December 25). As Applications Fall, Police Departments Lure Recruits With Bonuses and Attention. The New York Times.

The White House. (2022, July 21). FACT SHEET: President Biden’s Safer America Plan. The White House.

Thorne, K. (2022, October 24). NYPD officers quitting in record numbers amid growing issue over New York City crime. ABC7 New York.

Tillyer, M. S., & Eck, J. E. (2011). Getting a handle on crime: A further extension of routine activities theory. Security Journal, 24(2), 179–193.

Todak, N. (2017). The Decision to Become a Police Officer in a Legitimacy Crisis. Women & Criminal Justice, 27(4), 250–270.

US Bureau of Labor Statistics. (2023). QCEW Data Files: U.S. Bureau of Labor Statistics. US Department of Labor.

Van Maanen, J. (1973). Observations on the Making of Policemen. Human Organization, 32(4), 407–418.

Weichselbaum, S., & Lewis, N. (2020, June 9). Support For Defunding The Police Department Is Growing. Here’s Why It’s Not A Silver Bullet. The Marshall Project.

Weihua, L., & Ilica, M. (2021, September 1). Police Say Demoralized Officers Are Quitting In Droves. Labor Data Says No. The Marshall Project.

Weisburd, D., & Majmundar, M. (Eds.). (2018). Proactive Policing: Effects on Crime and Communities. National Academies Press.

Westervelt, E. (2021, June 24). Cops Say Low Morale And Department Scrutiny Are Driving Them Away From The Job. NPR.

Wilson, J. M., Dalton, E., Scheer, C., & Grammich, C. A. (2010). Police recruitment and retention for the new millennium. Center on Quality Policing.

Wilson, J. M., & Heinonen, J. A. (2012). Police Workforce Structures: Cohorts, the Economy, and Organizational Performance. Police Quarterly, 15(3), 283–307.

Wolfe, S. E., & Nix, J. (2016). The alleged “Ferguson Effect” and police willingness to engage in community partnership. Law and Human Behavior, 40(1), 1–10.

Wolfe, S. E., & Piquero, A. R. (2011). Organizational Justice and Police Misconduct. Criminal Justice and Behavior, 38(4), 332–353.

No comments here
Why not start the discussion?