Abstract

This study explores the correlates of use of force (UOF) in police responses to calls for service (CFS), drawing on situational, community, and organizational perspectives. Linear probability models were used to analyze CFS and UOF records for Cincinnati Police Department (CPD) from October 2014 to July 2021, along with demographic and crime data aggregated to census tracts. The type of CFS, as well as quicker response times and police-initiated actions, were associated with UOF. Broader situational variables (including nighttime, alcohol consumption, and high activity periods), as well as tract-level crime and demographic characteristics, showed little statistical associations. The study finds modest evidence of organizational influences on UOF outcomes, in the form of variation by police district.

Cite as: Miller, J., Guthrie, S., and Piza, E. (2024). Examining Use of Force Outcome in Police Calls for Service: Evidence From Cincinnati. Crime & Delinquency. https://doi.org/10.1177/00111287241298687.

Keywords: Policing; use of force; calls for service

A defining feature of policing is the power to use physical force (Bittner, 1970; Paoline, 2021). Types of force range from empty hand control techniques, through less lethal weapons such as batons, sprays or TASERs, to firearms (Paoline et al., 2021; National Institute of Justice, 2009). Deployed legitimately, police use of force (UOF) can help officers take a resistant suspect into custody or neutralize a threat to public or officer safety. But it is open to abuse (Chemerinsky, 2000) and, in addition to the physical or psychological harms it may cause (Bryant-Davis et al., 2017), it can erode police legitimacy (Weitzer, 2002) and exacerbate citizens’ legal estrangement (Bell, 2017).

While existing research has provided important insights into the causes and correlates of UOF, the current study asks a novel question: how do situational, organizational, and community factors explain the distribution of UOF across police responses to “calls for service” (CFS)? Answers to this question reframe traditional explanations of UOF, and the resultant analysis models an approach that is of practical relevance to police departments seeking to manage UOF. The article proceeds using operational data from Cincinnati Police Department (CPD).

Background

Police UOF is constitutionally governed by fourth amendment jurisprudence. It is thereby permitted to support the application of criminal law (e.g., through arrest) and to protect public order and officer safety (Harman, 2008). However, according to Graham v. Connor, 490 U.S. 386 (1989) clarifies, its use should be “objectively reasonable” from the perspective of a reasonable officer at the scene, in relation to factors such as crime severity, the threat posed by a person, and whether they are resisting or attempting to flee.

In practice, research suggests UOF is rare in the context of routine police work. For example, a study of six medium-to-large police departments found just 1.4 to 9.8 instances of UOF per 1,000 CFS (Paoline et al., 2021). Inevitably, however, it is far more common once a suspect is encountered. Terrill and Reisig (2003), for example, found that restraint techniques were used in 19% of suspect encounters, while Paoline and colleagues (2021) found it in between eight and 40% of arrests.

Explanations for UOF

UOF research has tended to focus on three broad types of explanations, corresponding with different research questions and units of analysis. One approach examines the police-suspect encounter and asks: how does the situational context of police-suspect encounters give rise to force outcomes? Empirical research highlights important situational predictors such as suspect behaviors (e.g., Garner et al., 2002; Paoline et al. 2021; Piza et al., 2023; Terrill & Mastrofski, 2002), offense type (Garner et al., 2002; Morrow et al., 2017), suspect demographics (Garner et al., 2002; Terrill & Reisig, 2003; Paoline & Terrill, 2007), and officer characteristics such as (lower) educational levels (Paoline & Terrill, 2007; McElvain, & Kposowa, 2008) or relative youth or inexperience (Garner et al., 2002; McElvain & Kposowa, 2004; Paoline & Terrill, 2007).

In a second approach, scholars compare geographical areas, and ask: what characteristics explain variations in UOF rates between them? Some studies find evidence for “racial threat” theory which suggests officers, alert to threats to dominant white interests, use force more in minority areas (Jacobs & O’Brien, 1998; Lautenshlager & Omori, 2019). Some adapt social disorganization theory (Shaw & McKay, 1942) to suggest that that in socially disorganized areas police step in to impose grater control, including UOF, an idea also supported by some empirical evidence (Lautenshlager & Omori, 2019; Nouri, 2021). Finally, researchers have applied Klinger’s (1997) ecological theory, which suggests less vigorous policing occurs in areas of high crime. The theory might be used to predict both lower and higher rates of UOF in these areas depending on whether UOF is a manifestation of, or an alternative to, policing vigor (Terril & Reisig, 2003). In practice, research tends to support the latter interpretation (Lautenschlager & Omori, 2019; Nouri, 2021; Terrill & Reisig, 2003).

A third perspective focuses on the police organizations. It asks: what features of organizations correlate with UOF rates? Empirical research highlights the role of police policies (Fyfe, 1979; Terrill and Paoline, 2017) and leadership (White, 2002) in UOF, which apparently convey different messages and modes of accountability around UOF.

Exploring the distribution of UOF in routine policing

Notwithstanding prior research, no study has examined the routine police task, in the form of a CFS response as the unit of analysis.. This allows us to ask: What features of these responses lend themselves to UOF outcomes? Addressing this requires refashioning existing theory to further ask: What kinds of calls bring police in contact with suspects? What situational circumstances lend themselves to tension and conflict once suspect encounters are underway? And what community and organizational settings more often prime officers to use force when meeting suspects on a CFS? (We use the term “suspect” loosely here to describe citizens suspected of criminal or other violations, or otherwise disturbing the peace.)”

Situational features

We expect UOF to arise almost exclusively when officers encounter suspects. These likely include calls in which suspects are identified in advance and/or have less likelihood of fleeing the scene before police arrive. We speculate these would include: stops, search or arrest warrants (which mostly imply contact with suspects), responses to active or recent (rather than completed) crimes, responses to domestic disturbance (where suspects are often in their homes), and responses to mental health incidents (in which individuals involved may not have committed an offense and/or may lack the presence of mind to avoid the police). Similarly, we would expect suspect contact to be more frequent when officers arrive quickly in responding to calls or when they directly initiate actions in the field.

We further theorize that some types of CFS have additional situational features that make UOF more likely when suspect encounters do occur. These would include violent crimes, given that officers may seek to neutralize a public safety risk or retaliate for violent provocation. We also suggest that mental health calls will fit this category if officers find these situations threatening or unnerving. Some of these ideas are echoed in research indicating that domestic and mental health calls lead to more officer injuries and/or assaults (Hirschel et al., 1994; Uchida et al., 1987), that civilian fatalities often involve suspects experiencing mental or emotional disturbance (Klinger & Slocum, 2017), and that UOF is more likely in an officer-suspect encounter when a violent offense is suspected (Garner et al., 2002; Morrow, et al., 2017).

Additionally, it is possible that temporal and spatial aspect of the task environment may affect UOF. Following a deterrence logic, we speculate that UOF will be more common during the nighttime, when visibility is less and there are fewer people on the streets providing natural surveillance that might inhibit officers otherwise ready to resort to force (Jacobs & Cherbonneau, 2019). Additionally, we suggest UOF will be more common during the evenings, weekends, public holidays, and key sports events when there is heavier alcohol consumption. This could reduce citizens’ inhibitions and induce more provocative behaviors, as suggested by prior research on UOF in police-public encounters (Freidrich, 1980; Terrill & Mastrofski, 2002) and on violence in general (Plant, et al. 2002; Clarke & Eck, 2016; American Addiction Centers, n.d.; Dawson, 1996; Lau-Barraco et al., 2016; Riordan et al., 2022; West et al., 2012; Young et al., 2004). Finally, UOF may be more common during busier periods, when officers are rushing between multiple jobs, and their patience and judgment is strained. Research finds evidence for such an effect for police officers experiencing stress in simulations (Haller et al., 2014).

Community context

In line with community-level explanations, which explain variations in area rates of UOF according to community characteristics, we suggest the same dynamics may affect the tendency for police to resort to UOF in specific CFS responses. This leads to the idea that UOF is more common when police respond to CFS in neighborhoods with higher local crime rates (Lautenshlager & Omori, 2019; Klinger, 1997), higher levels of minority representation (Jacobs & O’Brien, 1998; Nouri, 2021) and/or greater levels of social disorganization (Shaw & McKay, 1942).

Organizational setting

Finally, as previously described (Fyfe, 1979; Terrill and Paoline, 2017; White, 2002), departmental policy and leadership, which seem to shape UOF rates, may also specifically shape the tendency of officers to use force during a CFS. This would suggest a lower rate of UOF outcomes in organizational settings where policies against UOF were more restrictive, and where leadership messages are less permissive. In practical terms, this would likely manifest in differences in UOF outcomes for CFS across organizational units (e.g., police precincts), or variations through time if UOF policies have evolved.

The Current Study

Our study tests associations between a set of situational, organizational, and community measures and UOF, informed by the discussion above, within CPD. This city, the third largest in Ohio, has a population of approximately 309,000, of which 50.7 percent is White, 42.3 percent Black, 3.8 percent Hispanic or Latino (U.S. Census Bureau, 2022). It had a 2021 UCR violent crime rate of 6.1 per 1,000 residents (https://www.cincinnati-oh.gov/police/crime-statistics1/), higher than the national average of 3.99 in 2020 (https://crime-data-explorer.app.cloud.gov). CPD is the main law enforcement agency in the city, employing about one thousand sworn officers, as well as 125 civilian employees (https://www.cincinnati-oh.gov/police/about-police/). The city has experienced a number of high profile UOF scandals and episodes of reform, including a fatal officer shooting of an unarmed black male in 2001 leading to riots (Momodu, 2017). This was followed by a Department of Justice investigation leading to a Memorandum of Agreement committing the CPD to reforms, including de-escalation training (Pilcher, 2020).

The current written UOF policy (City of Cincinnati, 2021) incorporates core elements of contemporary good practice (e.g., International Association of Chiefs of Police, 2020), including a commitment to: use no more than reasonable force, rely on explicit factors to determine objective reasonableness; prioritize the sanctity of human life; use de-escalation principles; intervene when excessive force is used by a colleague; and promptly administer first aid following a UOF incident. Notably, UOF policies have expanded over recent years. For example, commitments to intervene with colleagues and provide first aid were added in 2017, while the objective reasonableness and sanctity of life elements were added in 2019.

Data and matching

We used a variety of data sources to create an analytic dataset, drawing in particular from Cincinnati’s open data portal (https://data.cincinnati-oh.gov/). Central were CFS records, derived from the CPD computer aided dispatch (CAD) system. Prior research suggests CFS records tend to under-record proactive police tasks (Lum, et al., 2020). Notwithstanding, a third of the substantive CFS records seemed to reflect proactive police tasks. Our analysis focused on 1,832,908 CFS records (pre-matching) for the period October 2014 to July 2021 that involved a police response, after excluding those recording primarily administrative or non-patrol or response functions. We additionally used a CPD UOF database focusing on 2,701 unique incident records (pre-matching) for the October 2014 to July 2021 period. A total of 2,410 UOF records were matched to CFS records, a match rate of 89.23%. We also used CPD data on 256,445 crime incidents (pre-matching) for years from 2014 to 2020. Finally, to create measures of community demographic and socio-economic characteristics, we obtained census tract-level data from the American Community Survey (ACS), using five-year estimates for 2015 to 2019.

UOF records were matched to CFS records primarily using a common CFS incident number, supplemented by matches on date and daily CFS event number for a few cases. CPD data were geographically matched to 2010 census tracts, to which ACS data were already aggregated, using an address locator generated for Hamilton County (in which Cincinnati is situated) using street centerlines. Spatial joins were conducted using the 2010 Hamilton County census tracts, current Cincinnati city boundaries (obtained from the Cincinnati data portal) and the address fields of CFS and crime records. Because individual addresses in CPD data were partially masked, typically with the last two digits of building numbers replaced with "XX," we created an algorithm to replace these addresses with a mid-point address within the conceivable address range. This mostly involved substituting “XX” with “50,” but was adjusted where the street segment file suggested streets had missing address sequences. It is likely that some addresses closer to the edges of tracts, therefore, will have been misclassified to an adjacent tract. Using this method, the CFS data saw a 93.94% match and crime data saw a 96.15% match, all of which had geocoding rates well above the recommended minimums of past empirical research (e.g., Andresen et al., 2020).

Because the city boundary is not perfectly coterminous with census tract boundaries, we further reduced our dataset to focus only on the 93 matched census tracts that had at least 90% overlap with the Cincinnati city boundary (leaving out 57 matched census tracts). This produced a reduced analytical dataset of 1,499,539 records, a shrinkage of 18.19% from the initial matched dataset. A subsequent geographical match of the full and final dataset was made to an administrative police shape file allowing records to be matched to police districts, producing 1,498,594 records with police district information (a further 0.06% shrinkage) and affects analysis using the police district variable.

Measures

We used two operationalizations of our dependent variable. First was an all-encompassing binary UOF measure according to whether a CFS record matched with a UOF record (0.14% of CFS records). This encompassed: “use of force investigation” (4.49% of UOF records) which were among the most serious UOF contacts (according to conversations with Cincinnati officials) and seemed to include most events involving police firearms; “weapon discharge at an animal” (1.78% of UOF records); “taser-beanbag-pepperball-40mm foam” (49.90% of UOF records); “chemical irritant” (2.17% of UOF records); “injury to prisoner” (24.28% of UOF records) referring to instances where a suspect is injured, though not necessarily by force (for example if a they had fallen or ingested narcotics when arrested); and “non-compliant suspect/arrestee” (17.37% of UOF records) involving more minor levels of force such as “balance displacement”, “hard hands”, or pressing of someone against a car while handcuffing. Recognizing that the aggregate variable likely included some non-UOF incidents (notably “injury to prisoner,” and perhaps “use of force investigation”) and may have been patchy on coverage of lower end use force (“non-compliant suspect/arrestee) our second dependent variable operationalization included only clearly defined use of non-lethal weapons (“taser-beanbag-pepperball-40mm foam” and “chemical irritant”) (0.07% of CFS records). This allowed for a sensitivity analysis of core results.

To represent the theoretically relevant call characteristics, a categorical call type variable was created, based on reclassifying 189 discreet CFS codes to 18 theoretically relevant groupings (more details of call types included in these categories is provided in an online Appendix). Categories include stops, warrants served, responses to specific problems (such as domestic, mental and disorder issues), responses to violence (excluding domestic and mental health related violence accounted for in the aforementioned categories), as well as responses to a variety of other crime categories and problems. We should note, based on conversations with technical staff in Cincinnati (Boudrie, personal communication), that about 22% of call types are reclassified following response. However, this takes place among the 189 categories and would not necessarily produce changes across our derived 18 categories (for example, the most common change is from “domestic violence” to “family trouble,” which would not change our aggregate coding from “domestic response”).

We also created a response status variable, indicating whether the CFS action was officer-initiated (i.e. it was proactive officer action rather than a response to a call) and (when it was not) the response time (in minutes). The variable was based on the time elapsed between the time the record was created (based on time stamps in the CFS data) and the arrival of the primary police unit. Those scoring zero were classed as “officer-initiated,” while those above zero were grouped into different response-time bands. There was one exception to this: for “warrants issued” we coded all records as officer-initiated because any time gap was understood as reflecting an administrative rather than operational process. While we are interested in how quicker and slower response times may affect UOF outcomes, we recognize that response time may be confounded with the urgency of calls if police respond more quickly to volatile and high-risk calls. While our call type measure goes some way to control for this, it still includes more and less urgent CFS even within common categories. We therefore included an additional call priority measure using CPD’s contemporary CFS priority rating score applied across the study period (M = 0.00; SD = 5.47). This was mean centered for each call type value, so as not to confound the hypothesized relationship between the latter and UOF outcomes.

We generated three variables to signal time periods we theorized were more prone to UOF owing to situational factors. First, a binary nighttime variable indicated police arrivals during hours characterized by darkness. This relied on published sunset and sunrise times for Cincinnati for the years 2014 to 2021 (on www.timeanddate.com/sun/usa/cincinnati) and generalized monthly night hours from the 14th day of February and the 15th day of every other calendar month. Second, a high alcohol consumption variable flagged CFS arrival times between 6.00pm and 6.00am on all Friday and Saturday nights (going into the next day) along with Christmas Eve and Christmas Day, New Year’s Eve, St. Patrick Day, Thanksgiving, and Cinco De Mayo, holidays hypothesized to coincide with greater alcohol consumption. This variable additionally flagged potential high alcohol periods around home game days for the Cincinnati’s Bengal football team and the Super Bowl, wrapping in time periods at least two hours before and two to three hours after games were scheduled. Finally, we created a variable that measured the level of demand on CPD officers (M = 1.00; SD = 0.56) by calculating, for each CFS record, how many reactive police tasks there had been within the same (of five) police districts within the prior hour. These measures were then divided by the mean activity level for each district, thus expressing the demand variable as a ratio compared with the average district demand. This produces a variable that varies through time and across districts.

We developed tract-level measures to approximate relevant neighborhood community variables. Specifically, we used ACS data to estimate the proportion of the tract population that was either non-Hispanic Black or Hispanic (M = 0.49; SD = 0.28; M = 0.04; SD = 0.05). We calculated the tract violent crime rate (M = 452.52; SD = 315.38) and other crime rate (M = 825.34; SD = 586.34) per 1,000 population based on CPD crime counts across the period and ACS estimated tract population. Given the potential undercount in tracts that did not fully overlap with CPD boundaries we further adjusted crime estimates by dividing them by the proportion of the tract that was within the city boundary. We also calculated tract-level social disorganization measures using ACS data, including: racial and ethnic heterogeneity (M = .43; SD = .15) based on the Herfindahl Index (Lautenshlager & Omori, 2019); concentrated disadvantage (M = .47; SD = 1.82) using a principal components score of the percentages of households that are female headed with children, people and families with income below the poverty line, labor force members unemployed, and households with public assistance or food stamps (α = .85); and residential instability (M = .52; SD = 1.44) using a principal components score of the percentages of homes renter occupied, housing units empty, the population in a different home from a year prior (α = .58; while this alpha score is less than ideal, for the whole of Hamilton County α = .71).

To measure variables that might shed light on organizational influences on UOF, we generated dummy variables for the five Cincinnati police districts in which CFS events took place. We recognize this is not a direct measure of organizational influence, because it says nothing about the organizational characteristics of the different districts. However, any statistical variation in UOF by district in our models will be net of measures of social disorganization, minority representation and crime rates, making organizational characteristics at least a plausible explanation. The measures relied primarily on pre-existing district codes in the CFS database, supplemented by our own match to the police district file for missing records. Additionally, we created annual dummy variables, from 2014 to 2021 to assess potential changes over time. Again, while not a direct measure of organizational change, if the progressively more restrictive UOF policies across the time period had affected UOF, we would expect to see some declines over time in UOF outcomes, holding other variables constant.

Analysis

Using Stata 18.0, we first compare distributions of CFS responses and UOF incidents across independent variables. Subsequently, we create multivariate models to examine the relationship between our independent variables and UOF. In order to generate readily interpretable model coefficients, we calculated linear probability models with robust standard errors clustered by census tract to accommodate clustering and measurement at this second level. We considered multilevel mixed models to address the multilevel data structure, but these did not readily converge. Given the large number of statistical tests associated with our research hypotheses, and the associated risk of inflated false positive reporting, we also included an adjusted p value using the Benjamini-Hochberg procedure (Thissen et al., 2002).

Results

Table 1 shows the distributions of police calls and the two measured UOF outcomes (aggregate UOF and less-than-lethal UOF) across independent variables. Overall, it indicates some notable differences in the distributions of CFS overall than UOF incidents, while differences between aggregate and less-than-lethal UOF are modest.

The single most frequent CFS call type was directed patrol, followed by disorder responses, and then stops. Other (relatively) common activities include responses to accident/fire/health threats and violence (other), and investigations. Slightly fewer than half of CFS responses took place during nighttime, and about one in six occurred during periods of high alcohol consumption. There is little variation in UOF activities by year (taking into account that 2014 and 2021 are only partial years).

Table 1

Distribution of police tasks and UOF outcomes by independent variables

^{Note: “Other violence” excludes any violence calls associated with the mental health or domestic violence response categories.}

The distribution of UOF events, while similar to CFS in general, shows some important differences. Most notable are differences related to call type: pursuits, stops, investigations, responses to mental health issues, other violence, and property crimes are substantially overrepresented in both aggregate and less-than-lethal UOF records. Meanwhile, traffic enforcement, crime report follow-ups, alarm responses, directed patrol, and accident/fire/health threat responses are notably underrepresented. Also, there are noticeably higher proportions of UOF for reactive police responses that occur in under five minutes, while much smaller proportions are seen for responses of half an hour or more. Proportions of UOF for officer-initiated police tasks are also slightly larger than for police tasks overall, and UOF incidents are more heavily concentrated among high priority calls than are CFS overall. Tract characteristics also show some modest differences, with slightly higher proportions of UOF incidents, compared to CFS, in tracts with higher rates of black representation, concentrated disadvantage, and violence. There are also some differences by police district. Other variables show limited differences.

Table 2 presents the results from multivariate linear probability regression models that test the associations of the independent variables with UOF. In keeping with our earlier discussion, we rely on Benjamini-Hochberg adjusted p values (Thissen et al., 2002) in our assessment of significance. Consistent with the descriptive analysis, the model indicates highly significant coefficients for most task police call categories. Here, the reference category is low level traffic enforcement (focused on parking and traffic hazards and violations, but not traffic stops), chosen because we would expect this to have low levels of UOF based on prior theorization (low chances of suspect contact, few situational risk factors for escalation). The overall call type variable is also significant when testing whether all coefficients are equal (adjusted p < .001).

Table 2

Linear-probability regressions of UOF outcomes for CFS responses (N = 1,486,431)

^{Notes: “p-adj” provides the Benjamini-Hochberg adjusted p values. “Other violence” excludes any violence calls except those associated with mental health or domestic violence response categories. Standard errors are cluster robust by census tract. Stata’s linktest was not significant (p > 0.05) for either model, suggesting correct model specification. Only census tract crime rate variables (violent and other crimes) had collinearity above 5 (though below 10). A summed crime variable in place of these two crime measures was substituted in the models in a sensitivity test (not presented here) and this reduced all collinearities to below 5 but left substantive results essentially unchanged, and the combined crime variable was non-significant in both models.}

Figure 1 presents model margins for CFS types as percentage probabilities for both aggregate and less-than-lethal UOF. The figure excludes “pursuits,” which dwarfed all other categories (estimating UOF in a whopping 16.8% of cases for aggregate UOF and 11.9% for less-than-lethal UOF). It also substitutes a zero for negative probabilities that were estimated for directed patrol for both UOF measures and warrants for less-than-lethal UOF. Overall, if shows that the two UOF measures were similar in terms of their relative likelihood across CFS categories. Moreover, the figure reveals that some, but not all, the call types we anticipated would be more prone to UOF in fact were. These included mental health and (other) violence, and responses to active rather than completed crimes (with crime report follow-ups having much lower UOF rates than responses to the active crime categories of other violence, property crime, drug crime, and other crimes/violations). Less consistent with expectations, domestic calls and stops were closer the middle of the distribution, and warrants were at the low end. It is also notable that property crime calls were towards the higher end of the distribution, which had not been anticipated. The very large rate of UOF in the case of pursuits had not been anticipated either.

Figure 1

Model estimated UOF probabilities across CFS types (excluding “pursuits”; negative estimated marginals presented as zero)

Coefficients in Table 2 also confirm our expectations that police-initiated CFS are more prone to UOF than reactive calls, and that shorter reactive response times are more prone to UOF than long response times. Thus, coefficients for shorter response periods were larger in both models. Moreover, compared with response times of 30 minutes or above, calls under five minutes or officer initiated-calls were statistically significant in both models compared with a 30 minutes or more response time; and a separate test of equality of coefficients additionally shows the variable overall is significant in both models (adjusted p < .001). Importantly, this association was detected after controlling for the urgency of the call. This serves to offset the possibility that an association was detected simply because response status serves as a proxy for the volatility or risk of the call type.

Contrary to our expectations, however, UOF did not vary significantly according to other situational variables, namely nighttime hours, hours of high alcohol consumption, or periods of higher officer activity. Nor were any of the community-level variables significant: UOF outcomes for CFS are apparently unrelated to tract variations in crime rates, minority representation, or indicators of social disorganization.

Finally, there was some limited evidence that district mattered for UOF. Primarily, this involved significant coefficients for police district 2 in both models (compared to district 1; adjusted p < 0.001 for both models); with a test of equality for police district dummies significant in both models (adjusted p < 0.001). While not a direct test of organizational influences, this at least may suggest a role for police leadership and policy in shaping UOF outcomes. Meanwhile, CFS year showed little relationship to UOF in the main model. While year showed significance (adjusted p < .050) in the non-lethal model, the size (and significance) of the individuals coefficients does not indicate a trend towards reduced UOF, as we would expect following the tightening of UOF policy over the period.

Discussion and conclusion

In contrast to prior research that has examined UOF within police-public encounters already underway, or aggregate UOF rates in specific communities or policing organizations, the current study breaks new ground by examining associations between UOF and a set of theoretically relevant variables across police responses to CFS. Drawing on insights from prior literature, it theorizes that the tendency for UOF outcomes are influenced by: the type of CFS, because it affects the likelihood of police coming into contact with suspects; the situational circumstances of the task environment because they affect the potential for tension and conflict when suspect encounters occur; and the community and organizational settings of CFS, because these shape officers’ policing styles.

Applying this logic, we correctly anticipated that UOF would vary according to the call type. We also correctly anticipated that mental health and violent calls would see higher rates of UOF, as would responses to active (rather than completed) crime events, though domestic calls, police stops and warrants had somewhat lower UOF rates than we expected. Meanwhile, officer-initiated encounters and faster reactive incident responses were more prone to UOF, as anticipated.

Meanwhile, there was little evidence of broader situational variables associating with UOF. That is, nighttime, periods of high alcohol consumption, or periods of high police activity—situations we thought would present triggers for UOF—showed no relationship with UOF. Similarly, community variables associated with racial threat, ecological, or social disorganization perspectives had no associations with UOF outcomes. Police district showed some modest statistical associations with UOF outcomes, offering some indirect evidence of organizational characteristics perhaps affecting UOF. However, OUF outcomes did not change over time providing little evidence that recent tightening of UOF policies had affected the way CFS responses unfold.

From a theoretical point of view, the findings suggest that UOF decisions may respond principally to immediate situational threats or challenges tied to the call character, but not to broader temporal and social situational factors, or community or “neighborhood” contexts, though organizational influences may be relevant, based on indirect evidence. Overall, perhaps UOF is, by its nature, a stable and predictable phenomenon calibrated to the broad characteristics of calls, irrespective of policy attempts to influence it or the shifting scenery that sits behind CFS events of the same generic type. In turn, this may reflect features of a police culture that are resistant to influence: the emphasis on group loyalty and social isolation alongside a distrust of citizens and a need to assert control and authority, as documented by police scholars, may overwhelm other sources of influence on UOF decisions (Brown, 1988; Paoline, 2003).

However, the null findings for some variables may have more prosaic explanations. For situational variables, they may simply reflect differences between suspect experiences and general population experiences. For example, patterns of suspect intoxication that previous research suggests are a predictor for UOF within police-public encounters (Freidrich, 1980; Terrill & Mastrofski, 2002) may not correlate well with the general public’s alcohol use that we sought to estimate. We should also acknowledge that existing research supports the idea that organizational policy can affect UOF levels (Fyfe, 1979; Terrill & Paoline, 2017), so it may be that policy reforms in Cincinnati were less profound or less well implemented than we have seen in other settings.

Alongside theoretical insights, the research makes a practical contribution to the monitoring and potential management of UOF in police work. It provides an example of how police data can be analyzed to call attention to officer activities prone to UOF, which may ultimately help police policy efforts to mitigate against its overuse. In the current analysis, for example, high rates of UOF for mental health responses might usefully prompt a review of related policies. Recent years have seen important innovations in this area that might inform such an effort, including specialized training and crisis intervention teams (Wood & Watson, 2017). The analysis also finds a strikingly high rate of force for suspect pursuits (though these incidents overall are very infrequent). This might also usefully prompt review, paying attention both to recording practices (in case this is in part an artefact of recording) as well practices and policies (e.g., Alpert et al., 1996; Kenney & Alpert, 1997). And our finding that officer-initiated tasks appear more prone to UOF than reactive tasks raises questions about how CPD (and other agencies if this is generalizable) should manage proactive policing strategies in an era that promotes proactive policing as an evidence-based strategy (Lum et al., 2020) but which may lead to higher rates of UOF.

Finally, we should place these findings in the context of a number of methodological limitations. These relate, in particular, to our reliance on administrative police data. Insofar as the analysis demonstrates a strategy that police departments themselves could undertake to monitor UOF across CFS, it is appropriate that we use administrative data. However, as a method for accurately measuring underlying concepts of interest, it has limitations. Notably, CAD systems are structured to manage policing responses and resources, and UOF data systems are used to track administrative case records. Neither are primarily designed to support social scientific measurement. For example, our dependent variable did not easily differentiate types of UOF, and included some suspect injuries that did not arise directly from UOF. Similarly, our efforts to measure demands on officers could not control for the number of officers available at any given time. Moreover, pragmatic coding choices made by operators, dispatchers, and police personnel, in the context of daily operational policing, will routinely involve some level of error. In particular, we should expect some under-recording of police tasks, particularly in relation to proactive interventions (Lum et al., 2022), which might upwardly bias our estimates of UOF rates for officer-initiated tasks. We should also expect some under-recording of UOF incidents, which may further affect our estimates. Finally, while our research indicated limited changes over time in UOF outcomes, none of our data points predate the killing of Michael Brown in Ferguson, MO in 2014 which catalyzed significant national interest in this issue, and likely organizational responses.

At the very least, this study takes an important first step in examining UOF across CFS and CFS contexts. Research that improves upon the methods used here would be beneficial. This could involve replicating the analysis presented here in other police agencies and, optimistically, could use observational data to supplement administrative records to promote better measurement across a wider range of relevant constructs. We look forward to further research that advances the field in this way.

References

Alpert, G. P., Kenney, D., Dunham, R., Smith, W., & Cosgrove, M. (1996). Police pursuit and the use of force. National Institute of Justice (NIJ Publication No. 164833). Washington, DC: US Government Printing Office.

Andresen, M. A., Malleson, N., Steenbeek, W., Townsley, M., & Vandeviver, C. (2020). Minimum geocoding match rates: An international study of the impact of data and areal unit sizes. International Journal of Geographical Information Science, 34(7), 1306-1322.

Bell, M. C. (2017). Police reform and the dismantling of legal estrangement. The Yale Law Journal, 2054-2150.

Bryant‐Davis, T., Adams, T., Alejandre, A., & Gray, A. A. (2017). The trauma lens of police violence against racial and ethnic minorities. Journal of Social Issues, 73(4), 852-871. https://doi.org/10.1111/josi.12251

Bittner, E. (1970). The functions of the police in modern society: A review of background factors, current practices, and possible role models. National Institute of Mental Health.

Chemerinsky, E. (2000). An independent analysis of the Los Angeles police department's board of inquiry report on the Rampart scandal. Loy. LAL Rev., 34, 545.

City of Cincinnati (2021). 12.545 Use of force. (Revised January 28). Cincinnati: Author. https://www.cincinnati-oh.gov/sites/police/assets/File/Procedures/12545.pdf

Clark, R. V., & Eck, J.E. (2016). Crime analysis for problem solvers in 60 small steps. Washington, DC: Office of Community Oriented Policing Services.

Dawson, D. A. (1996). Temporal drinking patterns and variation in social consequences. Addiction, 91(11), 1623-1635. https://doi.org/10.1046/j.1360-0443.1996.911116234.x

Fyfe, J. J. (1979). Administrative interventions on police shooting discretion: An empirical examination. Journal of Criminal Justice, 7(4), 309-323. https://doi.org/10.1016/0047-2352(79)90065-5

Garner, J. H., Maxwell, C. D., & Heraux, C. G. (2002). Characteristics associated with the prevalence and severity of force used by the police. Justice Quarterly, 19(4), 705-746. https://doi.org/10.1080/07418820200095401

Harmon, R. A. (2008). When is police violence justified. Nw. UL Rev., 102, 1119.

Hirschel, J. D., Dean, C. W., & Lumb, R. C. (1994). The relative contribution of domestic violence to assault and injury of police officers. Justice quarterly, 11(1), 99-117. https://doi.org/10.1080/07418829400092151

International Association of Chiefs of Police (2020) National consensus policy and discussion paper on use of force. (Originally published October 2017). https://www.theiacp.org/sites/default/files/2020-07/National_Consensus_Policy_On_Use_Of_Force%2007102020%20v3.pdf

Jacobs, B. A., & Cherbonneau, M. (2019). Carjacking and the management of natural surveillance. Journal of Criminal Justice, 61, 40-47. https://doi.org/10.1016/j.jcrimjus.2019.01.002

Jacobs, D., & O'Brien, R. M. (1998). The determinants of deadly force: A structural analysis of police violence. American Journal of Sociology, 103(4), 837-862.  https://doi.org/10.1086/231291

Klinger, D. (1997). Negotiating order in patrol work: An ecological theory of police response to deviance. Criminology, 35, 277–306. https://doi.org/10.1111/j.1745-9125.1997.tb00877.x

Klinger, D. A., & Slocum, L. A. (2017). Critical assessment of an analysis of a journalistic compendium of citizens killed by police gunfire. Criminology & Pub. Pol'y, 16, 349. https://doi.org/10.1111/1745-9133.12283

Lau-Barraco, C., Braitman, A. L., Linden-Carmichael, A. N., & Stamates, A. L. (2016). Differences in weekday versus weekend drinking among nonstudent emerging adults. Experimental and Clinical Psychopharmacology, 24(2), 100–109. https://doi.org/10.1037/pha0000068238 

Lautenschlager, R., & Omori, M. (2019). Racial threat, social (dis) organization, and the ecology of police: Towards a macro-level understanding of police use-of-force in communities of color. Justice Quarterly, 36(6), 1050-1071. https://doi.org/10.1080/07418825.2018.1480792

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. https://doi.org/10.1177/1098611119896081

Momodu, S. (2017, October 21). The Cincinnati riot (2001). Blackpast. https://www.blackpast.org/african-american-history/the-cincinnati-riot-2001/

Morrow, W. J., White, M. D., & Fradella, H. F. (2017). After the stop: Exploring the racial/ethnic disparities in police use of force during terry stops. Police Quarterly, 20(4), 367-396. https://doi.org/10.1177/1098611117708791

National Institute of Justice, (2009, August). The use-of-force continuum. Department of Justice August, National Institute of Justice. https://nij.ojp.gov/topics/articles/use-force-continuum

Nouri, S. (2021). Police use of force at street segments: Do street-level characteristics matter?. Journal of Criminal Justice, 77, 101862. https://doi.org/10.1016/j.jcrimjus.2021.101862

Paoline III, E. A., & Terrill, W. (2007). Police education, experience, and the use of force. Criminal Justice and Behavior, 34(2), 179-196. https://doi.org/10.1177/0093854806290239

Paoline III, E. A., Terrill, W., & Somers, L. J. (2021). Police officer use of force mindset and street-level behavior. Police Quarterly, 24(4), 547-577. https://doi.org/10.1177/10986111211025523

Piza, E., Connealy, N., Sytsma, V., and Chillar, V. (2023). Situational factors and police use of force across micro-time intervals: A video systematic social observation and panel regression analysis. Criminology 61(1), 74-102. https://doi.org/10.1111/1745-9125.12323

Riordan, B., Merrill, J. E., Ward, R. M., & Raubenheimer, J. (2022). When are alcohol-related blackout Tweets written in the United States? Addictive Behaviors, 124, 107110–107110. https://doi.org/10.1016/j.addbeh.2021.107110

Shaw, C. R., & McKay, H. D. (1942). Juvenile delinquency and urban areas. University of Chicago Press.

Terrill, W., & Mastrofski, S. D. (2002). Situational and officer-based determinants of police coercion. Justice Quarterly, 19(2), 215-248. https://doi.org/10.1080/07418820200095221

Terrill, W., & Paoline III, E. A. (2017). Police use of less lethal force: Does administrative policy matter?.Justice Quarterly, 34(2), 193-216 https://doi.org/10.1080/07418825.2016.1147593

Terrill, W., & Reisig, M. D. (2003). Neighborhood context and police use of force. Journal of Research in Crime and Delinquency, 40(3), 291–321. https://doi.org/10.1177/0022427803253800

Thissen, D., Steinberg, L., & Kuang, D. (2002). Quick and easy implementation of the Benjamini-Hochberg procedure for controlling the false positive rate in multiple comparisons. Journal of educational and behavioral statistics, 27(1), 77-83.

Uchida, C. D., Brooks, L. W., & Kopers, C. S. (1987). Danger to police during domestic encounters: Assaults on Baltimore County Police, 1984-86. Criminal Justice Policy Review, 2(4), 357–371. https://doi.org/10.1177/088740348700200404

UCR (2020). https://crime-data-explorer.app.cloud.gov/pages/explorer/crime/crime-trend

U.S. Census Bureau. (2022). QuickFacts: Cincinnati city, Ohio. https://www.census.gov/quickfacts/cincinnaticityohio

Weitzer, R. (2002). Incidents of police misconduct and public opinion. Journal of Criminal Justice, 30(5), 397-408. https://doi.org/10.1016/S0047-2352(02)00150-2

West, J., Hall, P., Hanson, C., Prier, K., Giraud-Carrier, C., Neeley, E. and Barnes, M. (2012). Temporal variability of problem drinking on Twitter. Open Journal of Preventive Medicine, 2, 43-48. https://doi.org/10.4236/ojpm.2012.21007

Wood, J. D., & Watson, A. C. (2017). Improving police interventions during mental health-related encounters: past, present and future. Policing and Society, 27(3), 289-299. https://doi.org/10.1080/10439463.2016.1219734

White, M.D. (2002). Identifying situational predictors of police shootings using multivariate analysis. Policing: An International Journal of Police Strategies & Management, 25(4), 726–751. https://doi.org/10.1108/13639510210450659

Online Appendix

Table A1

Police CFS types

Table A2 Correlation matrix of independent variables

Please see correlation matrix here:

https://www.dropbox.com/scl/fi/c2ry1j34qz1fcaifjcl14/correlation-matrix.xlsx?rlkey=azgtdir9vz2ufvdkz4zge6fme&dl=0