Race and Officer Decision Making: Examining Differences in Arrest Outcomes between Black and White Officers

Abstract

Since the 1960s, one of the major reform efforts in law enforcement has been to increase the number of Black Americans within police agencies and on patrol in the streets. The general premise behind these efforts has been that increased diversity will improve police–community relations and will decrease biased police behavior, particularly against Black citizens. Policies seeking to reform policing through increasing the numbers of African American officers have been implemented with little empirical evidence that an officer's race (or ethnicity) is actually related to their behavior towards citizens, in particular arrest decisions. Using data from systematic social observations of police–citizen encounters in Cincinnati, OH, this study examines the influence of officer race on arrest outcomes, focusing on the behavior of Black officers. Findings suggest that officer race has direct influence on arrest outcomes and there are substantive differences between White and Black officers in the decision to arrest. In general, White officers in our study were more likely to arrest suspects than Black officers, but Black suspects were more likely to be arrested when the decision maker was a Black officer.

Keywords:

• officer race
• arrest
• police discretion
• citizen race

Acknowledgments

This work was supported by the National Institute of Justice Grant Number 96‐IJ‐CX‐0075. Points of view are those of the authors and do not necessarily represent the view of the US Department of Justice or the National Institute of Justice. A previous version of this article was presented at the annual meetings of the Academy of Criminal Justice Sciences, Las Vegas, Nevada, March 2004. We wish to thank Christopher Lowenkamp, Kenneth Novak, and Lawrence Travis for their comments and suggestions.

Notes

1. It has also been shown that officer race results in no preferential or deferential treatment, and in some instances the opposite occurs. Some research has found that Black officers may behave more coercively in Black communities (Banton, 1964; Leinen, 1984). Also, research on officers' ability to gain citizen compliance with police requests for orderly and legal behavior suggests that same race dyads between officers and citizen do not necessarily result in more successful police control of citizens. Mastrofski and his colleagues (1996) found that White officers were significantly more likely to get compliant responses from minority citizens than minority officers were, and minority officers were significantly less likely to gain compliance when they interacted with White citizens. Furthermore, same‐race dyads had no effect on the likelihood of police ability to gain citizen compliance (see also McCluskey, Mastrofski, & Parks, 1999).

2. A common interpretation of this finding is that because Black officers are assigned primarily to Black communities with higher rates of crime, Black officers are more involved in situations that necessitate higher degrees of coercive control such as arrest or use of force (Alex, 1969; Geller & Karales, 1981; Leinen, 1984).

3. Alex's (1969) seminal work is widely cited in discussions of African American police officers. However, the concept of double marginality has been challenged. According to Campbell (1980), there is little support for Alex's double marginality proposition. Unfortunately, none of the research on the issue of double marginality adequately documents the actual behavior of African American officers, particularly when it comes to arrest situations.

4. In a survey study of 167 African American police officers from the Milwaukee (Wisconsin) Police Department, Barlow & Barlow (2002) found that 1 in 10 respondents reported that they engage in racial profiling, suggesting that Black officers are not immune to race‐based policing. Racial profiling was defined in the survey as “when race is used by a police officer or a police agency in determining the potential criminality of an individual” (Barlow & Barlow, 2002, p. 344).

5. There are substantial differences between the process and requirements for becoming a police officer versus a correctional officer or a judge. Additionally; their respective work responsibilities and role constraints differ. While these differences are important, the potential for racial identity to influence occupational behavior is a common issue.

6. Very often officers would ask observers about the well‐being of other observers in the project (“How is Bob doing? I see him here all the time.”). This suggests that the officers were not threatened by the presence of observers. Also, officers would make comments to observers along the lines of “I know you rode with Officer Jones, and she said you were all right?” Thus, observers gained some level of legitimacy vicariously through conversations that officers had with one another.

7. Of the observations conducted with non‐White officers, 99.7 percent of the observations were conducted with African American officers. Only 0.3 percent of the observations were conducted with Asian officers and no observations were conducted with officers identified as Hispanic or any other racial/ethnic group. Given the focus of this analysis, and small number of observations involving Asian officers, encounters involving Asian officers were excluded from this study.

8. While the empirical research to date suggests that officer‐level correlates in general yield limited influence in explaining arrest decisions, we explored the effects of several of the correlates available in the data: officer gender (male versus female officer), officer level of educational attainment (high school versus some college), and patrol assignment (beat officer versus community policing officer). Consistent with the extant research, these particular variables did not substantively improve the explanatory power or findings of any of the models we estimated (Riksheim & Chermak, 1993; Worden, 1989, 1990). We did not deem them necessary as control variables; therefore, only officer length of service is included in the final models presented in this study (Table 2). However, as duly noted by the reviewers, the results from our models incorporating more officer characteristics should be transparent as they may be useful to future research examining officer behavior, and they are available from the lead author upon request.

9. This operationalization of evidence assumes all evidence criteria are given equal explanatory value. In other words, it is a measure of the quantity, not the quality, of evidence. Unfortunately, the existing data did not allow for further analysis of evidence quality. In encounters where the citizen was arrested, observers coded the presence of evidence prior to the arrest.

10. Approximately 99 percent of the non‐White citizens were African American. There were very few observed encounters between police and Hispanics (0.6 percent), Asians (0.2 percent), American Indians (0.1 percent) and “other” racial groups of citizens (0.1 percent). As such, citizens with these racial/ethnic characteristics were categorized as non‐White. Hispanic is an ethnicity, not a separate race. In fact, Hispanic citizens could be classified as either White or Black. However, the original data collection instrument coded Hispanic citizens as a separate race. Observers were not able to code citizens as White‐Hispanic or Black‐Hispanic. As such, for the purposes of this research Hispanic citizens will be classified as “non‐White.”

11. Several other operationalizations of citizen demeanor have been used in the extant research on the influence of demeanor on arrest outcomes. According to Lundman (1994, p. 637), “There is no basis for arguing that one representation is superior to another.” In the current data, different measurements of the same construct revealed high levels of intercorrelation (see also Novak et al., 2002). Most recent research has operationalized demeanor as a dichotomous variable, measuring citizen behavior as either polite or disrespectful as the differences in citizen demeanor appear to be a matter “of kind rather than degree” (Worden, Shepard, & Mastrofski, 1996, p. 330). “In other words, ordinal scales may fail to capture the threshold of antagonism that would most likely affect an officer's behavior” (Novak et al., 2002, p. 93).

12. Aggregate‐level crime rates may influence the vigor of officers' responses (Klinger, 1997). In communities with high crime rates, particularly violent crime and serious crime, officers may be more prone to actuate arrests due to perceptions by officers that crime is prevalent in the area and that they should arrest in order to deter criminal activity. Klinger (1997) comments, however, that in order for officers to make an arrest (or act with greater vigor) in communities with high levels of crime, the instant offense must meet a “seriousness threshold.” In these communities, arrests for minor offenses may not be seen as worth the effort because officers become desensitized by the large amount of crime. Harmful levels of multicollinearity were detected when crime rates were included in our estimated models. As was the case in Novak et al.'s (2002, p. 94) research using the same data, “Communities with high levels of disorganization also had correspondingly high levels of Part I and Part II crimes, therefore making our structural characteristics proxies for aggregate‐level crime rates.”

13. The overall number of officer–suspect encounters and the number of encounters per community vary considerably, which makes it difficult to estimate stable hierarchical models with acceptable levels of bias (Terrill & Mastrofski, 2002). A small number of officers in the sample had numerous suspect contacts and a large number of the suspect contacts occurred in a small number of neighborhoods. Of the 136 officers in the total sample, 17 officers had 10 or more encounters, accounting for 36 percent of the police–suspect encounters (N = 614). Approximately 82 (28 percent) of the other officers in the data had four or fewer suspect contacts. Fifty percent of the observed encounters occurred in 10 of the 44 neighborhoods in Cincinnati. As Lowenkamp and his colleagues (2003) note, a consensus has not been articulated regarding the minimum number of “level 1” observations (police–suspect encounters in this case) necessary to generate stable within‐ and between‐aggregate parameter estimates (e.g., compare Jang, 2002; Reisig & Parks, 2000; Roundtree & Clayton, 2001; Roundtree & Land, 2000; Roundtree, Land, & Miethe, 1994; Sampson, Raudenbush, & Earls, 1997; Wooldredge, Griffin, & Pratt, 2001). However, there must be enough observations to meet a major assumption of HLM techniques—the assumption of normally distributed error terms within and across levels of analysis—to estimate reliable hierarchical models (see Bryk & Raudenbush, 1992; Kreft & de Leeuw, 1998).

14. We would like to thank the anonymous reviewers for their thoughtful comments on this issue. It is plausible that certain officers within the data could unduly influence the prediction outcomes. There is a possibility of correlated error terms among officers as 15 of the 136 officers in the data accounted for 32 percent of the observations. While the ideal would be to have data that contained a larger number of observations at all levels, thereby increasing the reliability of parameter estimates from hierarchical modeling, which is the more appropriate analytical technique for these data, given the nature of the data we believe that our logistic regression models are also appropriate. The results from our logistic regression models are largely congruent with our HLM analyses, particularly when it comes to significant differences based on officer race; therefore, we present the findings from the logistic regression models. Results from our hierarchical models are available from the lead author upon request.

15. The following equation is used to test for significant differences between parameter estimates in Models B and C:

16. For the purposes of the −2 log likelihood test, to conduct the test of no difference between White and Black officers properly it was necessary to exclude officer race from the model with all officers. Accordingly, officer race was not presented in the final Model A (see Table 2). We would like to note that when officer race is included in Model A, officer race is significant (b = −0.945; p = 0.002; exp(B) = 0.388), indicating that when an encounter involves a White officer, suspects are significantly more likely to be arrested. While the inclusion of the officer race variable in Model A would, of course, alter the values of the other parameter estimates in the model, including officer race in Model A does not change the direction or significance level of any of the examined variables in the model.

17. Using the logit coefficients presented in Table 2, the following formula was used to calculate the arrest probabilities: P ₁ = exp(Z ₁)/(1 + exp(Z ₁)), where

Additional information

Notes on contributors

Robert A. Brown

Robert A. Brown is an Assistant Professor in the School of Public and Environmental Affairs at Indiana University Purdue University Indianapolis. His research interests include officer decision making and police control mechanisms.

James Frank

James Frank is an Associate Professor in the Division of Criminal Justice at the University of Cincinnati. Recently he has directed police observation studies examining the work routines of street‐level officers in the Cincinnati Police Division and 21 small‐town police agencies. His research interests include officer decision making and citizens' attitudes toward the police.