Using problem-oriented policing to address police problems through the study of body-worn camera footage

Abstract

Herman Goldstein’s concept of ‘problem-oriented policing’ (POP) emphasized data gathering, analytics, and pattern identification to allow police to more effectively address problems faced by citizens in their communities. One of the most pressing problems in modern policing, however, is how departments should respond to accusations that police are not fair and consistent in their dealings with the public. In this article, the research team examined 1 year of officer body-worn camera footage, to create a roadmap of how the scanning, analysis, response, and assessment method of POP can be utilized to address public concerns by assessing and addressing how police interact with the public.

Keywords

Problem-oriented policing body-worn cameras

Herman Goldstein’s concept of ‘problem-oriented policing’ (POP) has become enormously influential since he first propounded the idea in a 1979 journal article and further elaborated on it in his 1990 book of that title. Goldstein’s paradigm emphasized the importance of outcome-oriented police planning and deployment of resources. This more nuanced deployment was to be informed by a constant intelligence and analysis loop. Comprehensive data gathering, sophisticated analytics, and pattern identification characterized his suggested intelligence process. The products of that process were to be considered strategically in assigning police resources and setting police objectives. Policing was to become proactive not reactive; ‘problem-oriented’ as opposed to process-oriented.

Goldstein’s impact can be measured in part by the fact that he has been cited over 100,000 times in criminal justice literature (‘Google Scholar’, 2019). POP features prominently in the mission statement of the Police Executive Research Forum and is the organizing principle of the George Mason University Center for Evidence-Based Crime Policy. Important successes in policing have been obtained through the POP model in places such as Newport News, Virginia, under the aegis of department of justice (DOJ) and National Institute of Justice (NIJ) funding (Eck and Spelman, 1987). This article argues that the proliferation of body-worn cameras (BWCs) among police agencies will provide a substantial source of original data that can be integrated into the POP process.

One of the most pressing problems in modern policing is how departments are to respond to accusations that police are not fair and consistent in their dealings with the public, particularly with regard to use of force against citizens of color. Since 2015, the Washington Post has maintained a digital database of every civilian shot and killed by police officers in the United States (‘Fatal force’, 2019). While the database fails to include several important distinctions regarding the legality or propriety of the officers’ actions, the very fact that the database exists serves as evidence that segments of the American public believe there are widespread problems centered around police/citizen interactions. Police departments are being pressured to ‘do something’ to ensure that their officers are treating people fairly and to ensure that officers are being held accountable for any improper actions before those actions result in someone’s death. Police managers versed in problem-oriented or data-driven policing are beginning to seek data that can help them develop policies and trainings to satisfy public demand and reduce the likelihood of citizens being harmed during police encounters.

In this article, we purport that the proliferation of BWCs among police agencies provide a substantial source of original data that can be integrated into the POP process. Most studies of BWCs to date situate the cameras as an intervention and focus on the anticipated effects of their use (e.g. reductions in officer use of force and reductions in citizen complaints). The present study extends the examination of BWCs beyond the mere impact of their presence and demonstrates that the cameras themselves provide a rich source of jurisdiction specific data that can be used by proactive departments to identify and address policing problems as well as increase transparency and trust with the communities that they serve.

What follows is an exemplar of how one police department was able to harness data from their BWC program in order to more thoroughly study the interactions between police and the local citizens. This provided department leadership with information that could be incorporated into training and policy as well as an opportunity to increase transparency and trust with the local community.

The research team examined 1 year of BWC footage from a police department which has been using BWCs since 2012. The data collected from the footage were utilized to create a more complete picture of the number and nature of problems that occur during police/citizen interactions in a community with a large Hispanic population. While relatively small in scale, this study can serve as a guide to other agencies who may wish to use a data-driven or problem-oriented approach to identify patterns in how police interact with the public and, when necessary, develop policies or training designed to help officers interact with the public more professionally. Hopefully, this could have the larger effect of bolstering public trust in local policing agencies.

Literature review

Problem-oriented policing

The idea of POP was introduced in Herman Goldstein’s 1979 article but was more widely received after the publication of his book on the subject in 1990. In Problem-Oriented Policing, Goldstein defined problem analysis as the process of obtaining knowledge by reviewing existing work, evaluating new data, and then using that knowledge to inform police practice. A 2003 report from the Police Foundation similarly defined problem analysis as a process ‘in which formal criminal justice theory, research methods, and comprehensive data collection and analysis procedures are used in a systematic way to conduct in-depth examination of, develop informed responses to, and evaluate crime and disorder problems’ (Boba, 2003: 2).

In 2018, Goldstein reiterated the important aspects of POP by saying that it called for police to ‘identify specific problems the public expected them to handle, to dig deeply into understanding each problem, and to think freshly and creatively about the best possible tailor-made response’ (Goldstein, 2018: 1). In a 2018 speech, Goldstein also acknowledged that when he initially conceived of this idea, he did not adequately consider how policing agencies would gain access to enough people with the research and assessment skills needed to fully implement this (often data-driven) approach. Several solutions to this oversight have been implemented over time, one of which is for policing agencies to establish relationships with university-based researchers (Goldstein, 2018). While both Goldstein and the Police Foundation group expressed a desire for police departments to be able to develop problem analysis skills internally, the Police Foundation report notes that for some agencies or problems, this may not be possible or desirable, and outside researchers may be needed. In any case, the overarching goal is for data-based knowledge to inform police practice.

Since its introduction, POP has largely been viewed as an approach used by police to address crime problems in their communities (Braga, 2008; Corsaro et al., 2013; Eck and Spelman, 1987; Maguire et al., 2015). The Newport News Police Department, for example, was selected by the NIJ to serve as a POP pilot program (Eck and Spelman, 1987). They developed a problem analysis guide that officers could use to create measurable outcomes. This guide became known as the SARA model. SARA stands for scanning the environment to identify problems, analyzing data about these problems to find patterns or trends, responding to the problem as a result of the conclusions drawn, and assessing the efficacy of that response. As a result of their use of the SARA model, burglaries in a particular apartment complex were reduced by 35%, and robberies in a particular district of their city were reduced by 40% (Eck and Spelman, 1987). In the late 1990s, the US Department of Justice developed a grant program to promote the use of POP in agencies across the United States, and they required that all applicants for those grants use the SARA model (Maguire et al., 2015). Since the Newport News pilot program, this data-driven approach to policing has been used to address crime problems ranging from open-air drug markets to speeding (Corsaro et al., 2013; Maguire et al., 2015). We assert that the principles of POP can now be used to address a different kind of concern: public distrust of the police. This may be particularly effective when data are collected and analyzed by external researchers.

Public distrust of the police and officer BWCs

The study of citizen perceptions of the police is not new. The Kerner Commission, which was formed in the late 1960s to investigate the 1967 race riots in the United States, cited Gallup Poll findings at the time as part of the cause for the riots. Those Gallup Poll findings suggested that 35% of Black or African American men believed that there was police brutality in their neighborhoods, while only 7% of White men felt similarly. They concluded that this perception led Black people to resent the police, which eventually contributed to outright violence (Kerner et al., 1968). Smith and Hawkins (1973) studied perceptions of police in Seattle Washington in the early 1970s. They found that White individuals generally held favorable perceptions of the police, while Non-White people did not, even when controlling for previous arrest and victimization (Smith and Hawkins, 1973).

While the general finding that White individuals have a more favorable view of the police has persisted through time (e.g. see: Engel, 2005; Gabbidon and Higgins, 2009; Hurwitz and Peffley, 2005; Schuck et al., 2008; Worrall, 1999), more recent news coverage of officer involved shootings has again caused researchers to examine this topic. In 2017, LaVigne, Fontaine, and Dwivedi went door to door to interview residents in the most disadvantaged (highest crime and lowest income) neighborhoods in six US cities. Only 11.9% of the 1278 people interviewed were White, while 66.3% were Black, and 10.6% were Latino/Hispanic. In addition, a majority (58.6%) were female. The researchers found that only 34% of individuals in these neighborhoods believed that police always or almost always try to help the people they deal with, 30.2% believed that police respect people’s rights, 30% believed that police treat people with respect, and only 26% believed that the police make fair and impartial decisions in the cases that they deal with.

Theodore and Habans (2016) examined how increased immigration enforcement has impacted perceptions of police for both immigrant and nonimmigrant Latinos in four very populous counties across the United States. Their results suggest that

a substantial portion of the Latino populations of Cook, Harris, Los Angeles and Maricopa Counties are reluctant to voluntarily contact the police to report a crime or to provide information about crimes, specifically because they fear that police will inquire about the immigration status of themselves, their friends or their family members. (Theodore and Habans, 2016: 985)

Nadal et al. (2017) examined both general perceptions of police and perceptions of police behavior during recent interactions, both in college students and in members of the general population, who were recruited through social media. They found that Black respondents viewed the police more negatively than did White respondents. They also found that race was the second most commonly cited reason (the first was situational factors, such as being near an area when a burglar alarm went off) given for why people believed that they had had a recent encounter with the police.

As the Kerner commission noted in 1968, these negative public perceptions often lead to a distrust of the police, which can, in turn, lead to an increase in violent interactions between the two groups. In an effort to avoid problematic police/citizen interactions, many police departments began to implement BWC programs. In 2018, a report from the NIJ listed the five potential benefits of BWCs as (1) better transparency, (2) increased civility, (3) quicker resolution, (4) corroborating evidence, and (5) training opportunities (Chapman, 2018). While all of these benefits could have a significant impact on the nature of policing in the United States, research on the use of this technology has not focused on all five areas with equal vigor. Most commonly, studies on BWCs focus on whether the cameras reduce the number of interactions that involve the use of force, on whether they reduce the number of complaints filed against police departments, or on whether they increase public trust in policing agencies (Ariel et al., 2014; Ariel et al., 2016; Crow et al., 2017; Headley et al., 2017; Jennings et al., 2015). Conversely, there are relatively few studies that have examined whether BWCs help to corroborate evidence for more successful courtroom proceedings (Bakardjiev, 2015; Jennings et al., 2015; Sommers, 2016), and even fewer that focus on how the content of the videos can be used to make or change policy, or to educate both officers and the public.

Studies that have focused on the effects of the initial implementation of BWCs have generally shown positive results. Ariel et al. (2014), for example, examined the effects of BWCs on use of force complaints, in the Rialto, California Police Department. The researchers employed both experimental and control groups of officers, and an interrupted time series design, to examine this outcome. Over the course of the observation year, they found that officers who did not wear cameras were twice as likely to use force than officers who did. They also detected a statistically significant reduction in use of force incidents among officers of the same shift, when they compared the numbers before and after the cameras were in use.

Jennings et al. (2015) evaluated the impact of BWCs used by officers in the Orlando Police Department. A group of 46 officers were randomly assigned to wear the cameras, while 43 did not wear them. Officers with BWCs had significantly fewer response-to-resistance incidents and serious external complaints than officers who did not wear cameras. In addition, response-to-resistance incidents reduced 53.4%, and serious external complaints dropped 65.4% in the same officers before and after wearing the cameras. Ariel et al. (2017) examined the impact of BWCs on complaints against the police across 10 police departments located in two English-speaking countries. They found that the use of cameras ‘dramatically reduces the incidence of complaints lodged against police officers, thus illustrating that the treatment effect, first detected in a relatively small force in Rialto, carried a strong external validity’ (p. 302).

Crow et al. (2017) performed one of the only studies to directly measure citizens’ perceptions of officer BWCs. Surveying residents of two Florida counties, they found that people were generally supportive of officer use of these devices. They felt that cameras would improve police and resident behavior and were relatively unconcerned with the increased potential for invasions of privacy. They did, however, find that ‘Non-Whites, younger respondents, and those with more concern about crime all perceived less benefit of BWCs’ (Crow et al., 2017: 604).

The existing literature demonstrates that implementation of BWC programs may increase transparency, increase the civility of both officers and citizens, and provide a means by which to eliminate, reduce, or more quickly respond to complaints against the police. Yet we contend that important benefits of BWCs remain to be examined. The present study illustrates one manner in which the utility of BWCs can be extended beyond the impact of their presence on specific outcomes. Specifically, the substantial amount of footage generated by BWCs provide department specific data that can be useful for conducting local problem analyses. This research can then be used to inform police planning and training as well as to increase transparency and trust with the public.

Current study

Data

Data for the current project come from BWC footage obtained by officers of the Aransas Pass Police Department (APPD). Aransas Pass is located on the Texas Gulf coast, approximately 150 miles southeast of San Antonio. The community of roughly 8300 individuals is highly diverse, with a diversity index of 65 (‘DataUSA…’, 2017). Like much of south Texas, Aransas Pass has a small African American population (2.37% in 2017), but 49% of the population is Hispanic (‘DataUSA…’, 2017). While the crime rate has slowed in recent years, APPD still reports high levels of drug- and gang-related crime in the area. Police interactions with Hispanic citizens have received relatively little attention in the racial profiling literature, as the majority of the research focuses on encounters with African American citizens. Aggressive immigration reforms in the 1990s and, more recently, public concern over proposed measures such as Arizona’s Senate Bill 1070 (Cooper, 2010; Ferrell, 2004; Nowicki, 2010) have highlighted the need for increased research on the nature of police interactions in Hispanic communities. This has become even more evident since the November 2016 presidential election.

At the time this footage was drawn, the APPD employed 29 police officers. The Chief of the APPD, Eric Blanchard, began his policing career in 2000, the period in which POP was being discussed and implemented in departments nationwide. While APPD does not benefit as greatly from data-driven approaches as many larger metropolitan departments, they do subscribe to the problem-oriented approach. Aransas Pass was also an early adopter of BWC technology. The department began their BWC program in 2012 when fewer than 10% of police departments nationwide had begun using the technology (Police Executive Research Forum, 2018: 29), and by 2014, every officer on the force had been outfitted with a camera. This is in contrast to most police departments who were in the early stages of implementation or pilot testing during the same time period.

Department policy requires all officers to turn on their cameras at the start (or prior to the start, if they have notice) of any interaction with a member of the public, regardless of the nature of that interaction. At the conclusion of the contact, the camera is shut off and a new recording is initiated for the next encounter with a citizen. At the end of each shift, officers are required to dock their cameras at the police station and upload their videos. Failure to comply with this policy can result in disciplinary action, including termination. During the period under study, most officers were equipped with AXON flex cams, which are worn either attached to the arm of glasses or directly to the ear. This includes all patrol officers and patrol sergeants. Higher level supervisors are issued AXON body cams, which are usually worn either on a chain around the neck or otherwise affixed to the shirt near the center of the chest.¹ While most police departments archive nonevidentiary footage for only 60–90 days (Miller et al., 2014: 17), Aransas Pass retains all footage for at least 1 year, significantly longer than required by state law. While exceptions are made for footage that has been selected for internal review or is attached to a case file as evidence, most videos are uploaded on the day they are filmed, retained for 1 year, and then automatically deleted. Videos are uploaded and securely stored through Evidence.com, a digital evidence management system that allows for both storing and sharing data.

Video sampling

The research team was granted access to all videos uploaded to Evidence.com for a 1-year observation period which spanned from May 28, 2016 to May 27, 2017. Over 30,000 videos were uploaded during this time period. Since watching all of the footage was neither practical nor possible, the decision was made to watch a randomly drawn sample of videos from the database. To obtain that sample, the year was divided into four 3-month time periods. This was done because Aransas Pass is a coastal community, and the researchers felt that the city was likely to have a larger number and variety of people during the summer months, compared to the winter. By dividing the year into four separate time frames, the sample was guaranteed to have equal footage from every season. Statistical software (SPSS) was used to select a random sample of 600 videos, 150 from each 3-month period. When a sampled video did not include an interaction between an officer and a citizen, that video was removed from the study and replaced with another randomly selected video from the same time period. Videos in the sample ranged from 35 seconds to 30 minutes and 50 seconds.² They were uploaded by all 29 officers, and every shift was represented.

Video coding method

The research team developed a coding instrument specifically for this project. The coding instrument provided a convenient method for recording key information from each video including: who initiated the contact, the type of interaction, the outcome of the interaction, the race/ethnicity of the officer, the race/ethnicity of the citizen, the gender of the officer, the gender of the citizen, whether the citizen showed any signs of impairment from substance use, whether force was used and type of force (if any), whether anyone was injured during the interaction, and whether the interactions occurred during the day or nighttime. Once a video segment was selected for the study, it was reviewed and coded by a member of the research team.

Research questions

The study was designed to provide the police department with the following information:

Who is more commonly responsible for initiating police/public interactions (the police or the citizens)?

What is the nature of the police/public contacts seen in the camera footage?

How are most police/public interactions resolved?

How often are people arrested?

How often is force, other than that needed for an arrest in which the citizen was compliant, used?

What is the impact of race on police/citizen interactions?

What is the impact of gender on police/citizen interactions?

Analytical strategy

These research questions are addressed using descriptive analyses and multilevel logistic regression. Descriptive statistics provide an overview of the nature of the police/public contacts that occurred in Aransas Pass, Texas, during the 1-year study period. Multilevel logistic regression was selected to identify statistically significant predictors of various interaction outcomes including arrest, receiving a ticket, and the resolution of a police/citizen interaction with a simple conversation. Multilevel modeling was necessary because there were 29 officers performing 600 police/citizen interactions. With fewer officers than interactions, officer characteristics, including race and gender, are not unique or independent in each contact. Put simply, most of the officers performed more than one of the interactions in the sample. To account for this regression violation, and to prevent the potential inflation of estimates of statistical significance (Menard, 2010), multilevel modeling is being used to nest interaction incidents in APPD officers (Raudenbush et al., 2004). Accordingly, level 1 variables include characteristics of the interactions, including citizen race and gender, whether the incident occurred in daylight or darkness, the nature of the interaction, and whether the citizen showed signs of alcohol or drug impairment. Level 2 variables include officer characteristics such as race and gender. These models are used to address research questions four and five.

For the multilevel logistic regression models presented here, model statistical significance is tested using the model χ ² (G _M) statistic. Model substantive significance is determined using the likelihood ratio, or McFadden, R ². The Wald test statistic is used to determine the statistical significance of predictors, and predictor substantive significance is determined using fully standardized regression coefficients (see Menard, 2010, for a full description of how these are calculated).³

Variables for logistic regression models

Three multilevel logistic regressions models were run for the current study. Dependent variables for those models include different police/citizen interaction outcomes. The first of those outcome variables is whether the citizen was arrested as a result of the police/citizen interaction. That variable was coded dichotomously, with each citizen either being arrested (1) or not arrested (0). Additional models were run to determine predictors of receiving a written citation (1 = the citizen received a written citation or 0 = the citizen did not receive a written citation), and for conversations (1 = the interaction ended with only a conversation between the citizen and the officer or 0 = the interaction ended with something other than a conversation).

Independent variables in the models include the type of interaction, whether the interaction occurred in daylight or darkness, the race and sex of the officer, the race and sex of the citizen, and whether the citizen showed any indication of being under the influence of a substance. The type of interaction originally had several distinct response categories, but because of the relatively small numbers for some of the interaction types, categories were combined. For the current analysis, all traffic stops are one dichotomous variable (1 = yes, the interaction was a traffic stop of some kind and 0 = no it was not a traffic stop). All interactions that were initiated due to the officer being dispatched are combined into a separate dichotomous variable (where again 1 = yes and 0 = no), and other or unknown interaction types are computed into a third dichotomous variable (1 = yes and 0 = no). Interactions begun via an officer being dispatched to the scene are the largest category. Accordingly, these cases are being used as the reference category.

Aransas Pass has only Anglo and Hispanic officers with no officers of any other racial/ethnic categories. Accordingly, those two classifications were coded into two dummy variables: Officer_Anglo and Officer_Hispanic (1 = yes and 0 = no, for both), with Anglo being used as the reference category. While Aransas Pass citizens have slightly more racial diversity, only 11 African American citizens were seen in the police/public interactions captured on the selected videos. As a result, those individuals were combined into the ‘other’ category, and citizen race was broken into three dummy coded variables: Citizen_Anglo, Citizen_Hispanic, and Citizen_Other_Unknown. Again, these variables were coded as 1 = yes and 0 = no, and the largest category (Anglo) was used as the reference category. Citizen and officer sex are both coded as 0 = female and 1 = male. A predictor is also included in the multivariate models that measures whether a citizen in a police/citizen interaction shows any sign of being under the influence of a substance (1 = yes and 0 = no), and finally, a variable was included which indicated whether the event occurred during daylight (1) or darkness (0). It should be noted that during the summer months in south Texas there are more daylight hours than darkness.

Results

Table 1 presents the frequencies of several variables examined in this research. It should be noted that the frequencies presented in the table often represent the more detailed breakdown of the variables before they were combined, as described above, for use in the logistic regression models. This was done to show the most complete picture of what is occurring in Aransas Pass, even if some of the categories contain too few cases to be used for more advanced analyses.

Table 1.

Overall frequencies and percentages.

	Frequency	Percent	Cumulative percent
Who initiated the contact?
Citizen	221	36.8	36.8
Officer	229	38.2	75
Other/unclear	150	25	100
Interaction type
Traffic stop—Speeding	68	11.3	11.3
Traffic stop—Reckless driving	17	2.8	14.2
Traffic stop—Other	87	14.5	28.7
Officer dispatched—City ordinance	18	3.0	31.7
Officer dispatched—Property crime	50	8.3	40
Officer dispatched—Persons crime	26	4.3	44.3
Officer dispatched—Vice offense	10	1.7	46
Officer dispatched—Other/unknown	107	17.8	63.8
Officer stopped pedestrian	12	2.0	65.8
Citizen flagged down officer	32	5.3	71.2
Other/unknown	173	28.8	100
Incident outcomes
Citizen received directions	4	0.7	0.7
Officer and citizen had conversation	298	49.7	50.3
Citizen was victim/officer rendered aid	69	11.5	61.8
Citizen received written ticket	72	12	73.8
Citizen arrested	52	8.7	82.5
Citizen had force threatened against them	2	0.3	82.8
Citizen had force used against them	1	0.2	83
Other	102	17	100
Officer race/ethnicity
Caucasian	20	69	69
Hispanic	9	31	100
Officer sex
Female	3	10.3	10.3
Male	26	89.7	100
Citizen race/ethnicity
Caucasian	298	49.7	49.7
Hispanic	200	33.3	83
African American	11	1.8	84.8
Other/unknown	91	15.2	100
Citizen sex
Female	242	40.3	40.3
Male	358	59.7	100
Signs of substance use impairment shown by citizens
No impairment indicated	548	91.3	91.3
Slurred speech	7	1.2	92.5
Impaired balance	5	0.8	93.3
Individual admission	9	1.5	94.8
Failed sobriety test	1	0.2	95
Multiple indications	30	5.0	100
Citizen use of force
No force used	594	99	99
Verbal threat	3	0.5	99.5
Use of hands	1	0.2	99.7
Other/physical noncompliance	2	0.3	100
Officer use of force
No force used	596	99.3	99.3
Verbal threat	3	0.5	99.8
Use of hands	1	0.2	100
Did the interaction occur in the daylight or after dark?
Daylight	362	60.3	60.3
Dark	238	39.7	100

From Table 1, we can see that by a slim majority, the largest group of police/citizen encounters is initiated by an officer (38.2% vs. 36.8%). For interactions in which the initiator is unknown, there are two principal reasons. The first is that the recording officer was a backup officer, and, therefore, not the first on the scene. The second includes cases in which the citizen was brought to the police station after an event. The footage selected for this study begins with the booking process, and since the researchers do not know how the initial process began, these cases were coded as ‘other’. There are additional, rarer, events but these are the most common reasons that contact initiations were coded as ‘other’ or ‘unknown’. The most common type of traffic stop was for ‘other’ reasons, which were overwhelmingly related to brake, tail, or trailer lights being out. Speeding was the second most common reason that an APPD officer initiated a traffic stop. When officers were dispatched to a scene, the two most common reasons were ‘other’ and for a property crime. Again, ‘other’ deserves some explanation. This classification of police/citizen interactions included a wide variety of events such as building alarms that needed to be checked (in all of these seen for this study, none of them indicated an actual break-in to a building), wellness checks on people, suspicious persons reports called in to dispatch, and other more random things. For instance, one case included a family dispute in which a mother called the police because her ex-husband would not return the child’s medication. In many of these cases, officers were talking to people, calming them down, getting information, and very often giving information on how to proceed legally from there.

Table 1 also indicates that police/citizen interactions, overall, are most likely to end in a conversation, as opposed to any other outcome. When you include giving directions, 302 of the 600 viewed interactions ended in this manner. An additional 11.5% of interactions involved an officer rendering some sort of aid to a victim (everything from filling out a report, to an investigation, to medical assistance until EMS arrived). Only 12% of people who interacted with the police during the study received a written ticket, 8.7% were arrested, and only three incidents (0.5%) of the 600 involved a citizen who had force threatened or used against them. None of that force was lethal. In some cases, the outcome of an interaction was listed as ‘other’ (17%). Those cases included things such as jail releases, feeding citizens in holding cells, and other more random events such as working with a citizen to rescue a kitten trapped in a clothing donation bin. In those cases, the police interacted with a citizen, but the result varied from kitten rescue to jail release.

Table 2 illustrates the results of the multilevel model for arrest as a dependent variable. These results show that when all predictors are examined together, they do have a statistically significant impact on whether a citizen was arrested (p = 0.000). That impact, however, is very weak. Substantive significance measures indicate that the model explains only 3.3% of the variance in arrest. These weak predictive values are due to the fact that only one variable in the model has a statistically significant relationship with the outcome. Specifically, if a citizen shows signs of being under the influence of a substance, they are statistically significantly more likely to be arrested than if they do not show those signs. Other than that, no other variable has a significant impact on arrest. When one considers this result either from the perspective of police officers or from citizens in this jurisdiction, this is largely good news. It suggests that neither the race or sex of the officer nor the race or sex of the citizen predict whether a citizen will be arrested. Given state laws regarding driving and being in public under the influence of a substance, it is not a serious concern that officers are arresting those who show signs that indicate they are, indeed, under said influence.

Table 2.

Multilevel logistic regression model for arrest as an interaction outcome.

Dependent variable	Model statistics	Independent variables	Unstandardized coefficients (b)	Standardized coefficients (b*)	Significance (Wald statistic)
Arrest (0 = not arrested; 1 = arrested)	G _m = 48.436 df = 9 (p = 0.000) $R_{L}^{2}$ = 0.033	Officer sex (male)	0.039	0.004	0.971
		Hispanic officer	−0.235	−0.034	0.693
		Citizen sex (male)	0.339	0.053	0.658
		Under influence?	2.094	0.188	0.020
		Daylight?	−0.167	−0.026	0.824
		Hispanic citizen	0.564	0.085	0.525
		Other/unknown race citizen	−1.124	0.000	0.493
		Traffic stop	−0.731	−0.162	0.451
		Officer dispatched	−1.134	−0.113	0.133

Table 3 illustrates the results of the multilevel model that predicts whether citizens will receive a written ticket as the result of an interaction with police. Again, the model is statistically significant (p = 0.000) but weak ( $R_{L}^{2}$ = 0.082). As with the previous model, there is only one independent variable in the model that has a statistically significant relationship with the dependent variable. In this case, if the police/citizen interaction is the result of a traffic stop, then the citizen is more likely to receive a written ticket. This result makes some sense. Even though tickets can be issued for a number of nontraffic-related reasons, they are the primary tool used by officers for those violations. Again though, the results indicate that neither the sex and race/ethnicity of the officer nor the sex and race/ethnicity of the citizen impact whether a citation is issued.

Table 3.

Multilevel logistic regression model for a written ticket as an incident outcome.

Dependent variable	Model statistics	Independent variables	Unstandardized coefficients (b)	Standardized coefficients (b*)	Significance (Wald statistic)
Ticket (0 = no; 1 = yes)	G _m = 126.993 df = 9 (p = 0.000) $R_{L}^{2}$ = 0.082	Officer sex (male)	0.162	0.016	0.904
		Hispanic officer	0.357	0.051	0.655
		Citizen sex (male)	0.094	0.015	0.830
		Under influence?	−1.249	−0.112	0.650
		Daylight?	0.372	0.058	0.537
		Hispanic citizen	0.152	0.023	0.745
		Other/unknown race citizen	0.207	0.025	0.739
		Traffic stop	3.218	0.463	0.000
		Officer Dispatched	0.138	0.021	0.854

Table 4 examines the predictors of the most common, and positive, outcome of a police/citizen interaction: a simple conversation. Again, when all predictors are considered together, they do have a statistically significant (p = 0.007) impact on the dependent variable. That impact, however, is the weakest of the three models. Together, these predictors explain a little over 1% of the variance in this outcome. There is a marginally significant (0.05 < p < 0.10), negative, relationship between a citizen showing signs of being under the influence and a police/citizen encounter ending with only a conversation. In this case, if a citizen is showing signs of being under the influence, the outcome is less likely to be a conversation. We know from Table 2 that those citizens are more likely to be arrested. Again, the null results in this case are perhaps more important than the significant findings. Unless a citizen is intoxicated, most interactions with the police end in a conversation regardless of race, ethnicity, sex, the type of interaction, or the time of day that the interaction occurred. This is good news for citizens in this jurisdiction.

Table 4.

Multilevel logistic regression model for conversation as an interaction outcome.

Dependent variable	Model statistics	Independent variables	Unstandardized coefficients (b)	Standardized coefficients (b*)	Significance (Wald statistic)
Incident ends in conversation (0 = no; 1 = yes)	G _m = 22.535 df = 9 (p = 0.007) $R_{L}^{2}$ = 0.012	Officer sex (male)	−0.280	−0.038	0.808
		Hispanic officer	0.124	0.025	0.625
		Citizen sex (male)	−0.146	−0.032	0.420
		Under influence?	−1.301	−0.161	0.058
		Daylight?	−0.229	−0.050	0.647
		Hispanic citizen	−0.134	−0.028	0.703
		Other/unknown race citizen	0.337	0.057	0.310
		Traffic stop	0.044	0.009	0.857
		Officer dispatched	0.155	0.033	0.614

Discussion

The APPD subscribes to the problem-oriented approach to policing. When faced with a new wave of public trust issues, the department relied on those principles to conduct a jurisdiction specific problem analysis. Utilizing the SARA framework, the city of Aransas Pass SCANNED for and found the current problem. Specifically, national coverage of a wave of officer involved shootings around the country led to widespread decreases in public trust of the police (Jones, 2015). While there were no noteworthy cases of police misconduct in Aransas Pass, the department took the proactive step of soliciting an independent study to determine if there were problems in how officers were interacting with the public. Aransas Pass was in the advantageous position of fully outfitting their entire police force with BWCs prior to most departments nationwide.

The footage from the cameras provided a rich jurisdiction specific source of data that could be unobtrusively collected and ANALYZED by an independent research team. The researchers observed and coded 600 police/citizen interactions, spanning an entire year, all shifts, and all officers within the department. Interactions varied from funny, to sad, to frightening, to potentially deadly. In all, they represented what officers do on any given day in Aransas Pass. The analysis revealed that officers were, overall, performing their duties quite well. They had a good rapport with citizens and often engaged them in conversation. They treated people of all races, ethnicities, and genders with respect and often gave procedural information to citizens that helped them to better understand the local criminal justice system. Citizens, in turn, also commonly treated APPD officers with respect and, as the data show, arrest was rare, and the use of force was exceedingly rare.

When the findings were released to the Chief, the RESPONSE was twofold. First, there were a few minor things found that, if corrected, could improve the public image of officers in the department. For example, one officer got frustrated directing traffic and was muttering complaints that could potentially be overheard by passengers in passing vehicles. This was not a major transgression, and observers might even sympathize with the officer, but cases such as these were marked and the Chief reviewed them with the officer, asking him or her to see it from a citizen’s perspective. This served to remind the officers that the footage is being watched, and that, even on bad days, everything they do affects public opinion. Second, since the data showed that APPD officers are largely fair and consistent in their interactions with the public, particularly with regard to things such as race/ethnicity and gender, that information was shared with the community. Local and national news outlets ran stories on the project, and the research team was able to explain the methods and results of the problem analysis. The police department was able to post the major findings on their website and also share the results through social media (Facebook and Twitter). The Chief has reported an overwhelmingly positive response from the community. The findings from the study, as well as the transparency demonstrated by the department, have bolstered APPD’s reputation as a progressive and community-oriented law enforcement agency.

The final step in the SARA model is to ASSESS the efficacy of the response in place. To do this, a second round of footage will need to be reviewed to see if police/citizen interactions are still being conducted properly. In addition, a survey of Aransas Pass citizens could be conducted to better determine their perceptions of APPD officers, both generally and after interacting with police. It would also be worthwhile to survey officers to determine what impact the study had on their interactions with the public. Although very few problems were identified during the period under study, APPD has indicated an interest in studying public perception of how the police are performing.

As with any study, this one has limitations. The primary limitation is that this research has been conducted in a small community, with a relatively small population of some minority groups. For example, Aransas Pass has very few African American residents. While it is informative to study police/public interactions in a population that has a high percentage of Hispanic individuals, it would be equally informative to study a population that has higher percentages of other racial/ethnic groupings, more language diversity, and potentially more diversity in terms of sexual orientation or representation. In addition, smaller agencies can potentially be monitored more closely by supervisors, because there are fewer ranks between Chief and line officer, and fewer officers to observe in general. Therefore, this type of study might serve additional functions in larger agencies that have a greater number of officers in supervisory positions but less familiarity with the daily work of each officer under their command.

Additionally, the variance explained by each of the predictive models in this study is low. While it serves as a relief to the Aransas Pass community that many of the variables included were not statistically significant predictors of police encounter outcomes, it also indicates to researchers that there may be variables with greater explanatory power. It is very possible that these variables would be apparent upon a more individual or case-by-case level examination than is conducted here. For example, in each interaction between the police and public, there are more qualitative, subtle factors at play. The tone of an officer or a citizen’s voice, for example, might affect how the corresponding party responds, which might, in turn, affect the outcome of that interaction. Such factors could have an effect across all groups or classes of people, or their impact may be moderated by factors like race and gender. While more arduous for coding schemes, these variables are likely to hold significant explanatory power. Future research should be conducted within larger agencies, hopefully within a wide variety of communities, and include quantitative or qualitative assessments of these additional components.

While this study was performed in a single jurisdiction, it serves as a roadmap of how the SARA method of POP can be utilized to address public concerns by assessing and, if necessary, addressing how police interact with the public. The recent proliferation of police BWCs provides an abundant source of jurisdiction specific data that can be analyzed by external researchers unobtrusively and with relative ease. This provides police departments with an accessible mechanism for determining whether the local police force is interacting fairly and legally with the public. Where problems are identified, departments can act swiftly to address them. Where problems are sparse and police are consistent in their dealings with citizens, these studies provide a powerful opportunity to demonstrate transparency and gain public trust.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported with funding from the College of Liberal Arts at Texas A&M University-Corpus Christi and from the City of Aransas Pass.

ORCID iD

Wendi Pollock

Notes

References

Ariel

Farrar

Sutherland

(2014) The effects of police body-worn cameras on use of force and citizens’ complaints against the police: a randomized controlled trial. Journal of Quantitative Criminology 31: 509–535.

Ariel

Sutherland

Henstock

, et al. (2016) Increases in police use of force in the presence of body-worn cameras are driven by officer discretion: a protocol-based subgroup analysis of ten randomized experiments. Journal of Experimental Criminology 12: 453–463.

Ariel

Sutherland

Henstock

, et al. (2017) ‘Contagious accountability’: a global multisite randomized controlled trial on the effect of police body-worn cameras on citizens’ complaints against the police. Criminal Justice and Behavior 44(2): 293–316.

Bakardjiev

(2015) Officer body-worn cameras-capturing objective evidence with quality technology and focused policies. Jurimetrics 56(1): 79–112.

Boba

(2003) Problem Analysis in Policing. Washington: Police Foundation.

Braga

(2008) Problem-oriented Policing and Crime Prevention, 2nd ed. Monsey: Criminal Justice Press.

Chapman

(2018) Body-worn cameras: What the evidence tells us. Available at: https://nij.ojp.gov/topics/articles/body-worn-cameras-what-evidence-tells-us (accessed 10 January 2019).

Cooper

(2010) Arizona immigration law target of protest. MSNBC. Available at: https://www.msnbc.msn.com/id/36768649/ (accessed 10 Janary 2019).

Corsaro

Brunson

McGarrell

(2013) Problem-oriented policing and open-air drug markets: examining the Rockford pulling levers deterrence strategy. Crime & Delinquency 59(7): 1085–1107.

10.

Crow

Snyder

Crichlow

, et al. (2017) Community perceptions of police body-worn cameras: the impact of views on fairness, fear, performance, and privacy. Criminal Justice and Behavior 44(4): 589–610.

11.

DataUSA: Aransas Pass, TX (2017). Available at: https://datausa.io/profile/geo/aransas-pass-tx/ (accessed 10 January 2019).

12.

Eck

Spelman

(1987) Problem-Solving: Problem-Oriented Policing in Newport News. Washington: Police Executive Research Forum.

13.

Engel

(2005) Citizens’ perceptions of distributive and procedural injustice during traffic stops with police. Journal of Research in Crime and Delinquency 42(4): 445–481.

14.

Fatal force (2019). Available at: https://www.washingtonpost.com/graphics/2019/national/police-shootings-2019/ (accessed 20 May 2019).

15.

Ferrell

Jr (2004) Immigration enforcement: Is it a local issue? Police Chief 71(2): 36–40.

16.

Gabbidon

Higgins

(2009) The role of race/ethnicity and race relations on public opinion related to the treatment of blacks by the police. Police Quarterly 12(1): 102–115.

17.

Goldstein

(1979) Improving policing: a problem-oriented approach. Crime & Delinquency 25(2): 236–258.

18.

Goldstein

(1990) Problem-Oriented Policing. New York: McGraw-Hill.

19.

Goldstein

(2018) On problem-oriented policing: the Stockholm lecture. Crime Science 7(13): 1–9.

20.

Google Scholar: Herman Goldstein (2019). Available at: https://scholar.google.com/scholar?hl=en&as_sdt=0%2C44&q=Herman+Goldstein&btnG= (accessed 10 January 2020).

21.

Headley

Guerette

Shariati

(2017) A field experiment of the impact of body-worn cameras (BWCs) on police officer behavior and perceptions. Journal of Criminal Justice 53: 102–109.

22.

Hurwitz

Peffley

(2005) Explaining the great racial divide: perceptions of fairness in the US criminal justice system. The Journal of Politics 67(3): 762–783.

23.

Jennings

Lynch

Fridell

(2015) Evaluating the impact of police officer body-worn cameras (BWCs) on response-to-resistance and serious external complaints: evidence from the Orlando Police Department (OPD) experience utilizing a randomized controlled experiment. Journal of Criminal Justice 43(6): 480–486.

24.

Jones

(2015) In US, confidence in police lowest in 22 years. Available at: https://news.gallup.com/poll/183704/confidence-police-lowest-years.aspx (accessed 10 January 2019).

25.

Kerner

Lindsay

Harris

, et al. (1968) Report of the National Advisory Commission on Civil Disorders. Washington: Special report to the President.

26.

LaVigne

Fontaine

Dwivedi

(2017) How do people in high-crime, low-income communities view the police? Available at: https://www.urban.org/sites/default/files/publication/88476/how_do_people_in_high-crime_view_the_police.pdf (accessed 11 May 2019).

27.

Maguire

Uchida

Hassell

(2015) Problem-oriented policing in Colorado Springs: a content analysis of 753 cases. Crime & Delinquency 61(1): 71–95.

28.

Menard

(2010) Logistic Regression: From Introductory to Advanced Concepts and Applications. Thousand Oaks: Sage.

29.

Miller

Toliver

and Police Executive Research Forum (2014) Implementing a Body-Worn Camera Program: Recommendations and Lessons Learned. Washington: Office of Community Oriented Policing Services.

30.

Nadal

Davidoff

Allicock

, et al. (2017) Perceptions of police, racial profiling, and psychological outcomes: a mixed methodological study. Journal of Social Issues 73(4): 808–830.

31.

Nowicki

(2010, July 25) Arizona immigration law ripples through history, US politics. Arizona Republic. Available at: http://www.azcentral.com/arizonarepublic/news/articles/2010/07/25/20100725immigration-law-history-politics.html (accessed 17 March 2018).

32.

Police Executive Research Forum (2018) Cost and Benefits of Body-Worn Camera Deployment. Washington: Police Executive Research Forum.

33.

Raudenbush

Byrk

Cheong

, et al. (2004) HLM6: Hierarchical Linear & Nonlinear Modeling. Lincolnwood: Scientific Software International.

34.

Schuck

Rosenbaum

Hawkins

(2008) The influence of race/ethnicity, social class, and neighborhood context on residents’ attitudes toward the police. Police Quarterly 11(4): 496–519.

35.

Smith

Hawkins

(1973) Types of citizen-police contacts, and attitudes toward the police. Law & Society Review 8(1): 135–152.

36.

Sommers

(2016) Will putting cameras on police reduce polarization? The Yale Law Journal 125(5): 1304–1362.

37.

Theodore

Habans

(2016) Policing immigrant communities: Latino perceptions of police involvement in immigration enforcement. Journal of Ethnic and Migration Studies 42 (6): 970–988.

38.

Worrall

(1999) Public perceptions of police efficacy and image: the ‘fuzziness’ of support for the police. American Journal of Criminal Justice 24: 47–66.