Revisiting the Racial Threat Thesis: The Role of Police Organizational Characteristics in Predicting Race‐Specific Drug Arrest Rates

Abstract

Previous research examining the relationship between structural factors and drug arrest rates has neglected the role of the police organization. A central proposition of racial threat theory is that indicators of a threatening Black population will be associated with law enforcement actions as a form of social control. In order to fully test this proposition, however, organizational aspects of law enforcement beyond size of the police force must be considered. Hence, the present study examines police organizational factors as direct predictors of race‐specific drug arrest rates but also as potential moderators of the effects of structural factors on drug arrest rates. Using data from 260 cities, we find that police organizational factors matter, both directly and as moderators of the association between racial economic competition and Black drug arrest rates. Consistent with expectations derived from racial threat and organizational theory, we find that racial threat measures are associated with Black drug arrest rates under conditions of relatively low organizational control.

Keywords:

• racial threat
• social disorganization
• drug arrests
• police bias

Notes

1. We utilized data at the Census place level because such a unit was roughly approximate to the jurisdictional boundaries of a single, metropolitan police agency. Out of 1543 Census places with at least 1500 Black residents, arrest data and police organizational data for a single, metropolitan police agency were available for 260 places.

2. Arrest data used in this study are from a separate data release, entitled Uniform Crime Reporting Program Data [United States]: Arrests by Age, Sex, and Race, Summarized Yearly, 2000, released in 2004. The arrest information is compiled by the FBI and cross‐checked to provide as accurate information as possible regarding arrest information by age, sex and race. However, as discussed by others using UCR data (see Ousey & Lee, 2004; Parker, Stults, & Rice, 2005), not all cities report data every month in a given year. However, in our sample of cities, there is a great deal of consistency in the monthly reporting of data, similar to others utilizing this data (Ousey & Lee, 2004). Our sample reported an average of 11.6 months of data. Nonetheless, we conducted additional analyses (available from the first author) that excluded cases that did not report 12 months of data and the resultant estimates were very similar and there were no substantive differences in the results reported in this study.

3. The problems with including dichotomous measures in principal components analyses have been discussed in detail elsewhere (Kim & Mueller, 1978).

4. The male marriage pool index is used as a proxy for the availability of partners for marriage and/or stable relationships.

5. The formalization index failed to load strongly on any single factor while the training variable did not load well with any other variables. Hence, these two measures, along with unionization (a three‐category variable) are included in the analyses as separate predictors.

6. Variance inflation factors (VIF) were calculated for all the estimated models and for no analysis did the VIF for an individual variable exceed 4, indicating that multicollinearity did not significantly impact our findings (Gujarati, 1995, p. 339). Furthermore, an examination of the correlation matrices (Appendices 3 and 4) show that none of the correlations are greater than .70; Walker (1999) has noted that when this is the case, multicollinearity is generally not a problem in regression. Additionally, an examination of leverage revealed that only six cases exceeded Huber's (1981) suggested cutoff of h between .2 and .5 for each of the race‐specific models. For Black arrest rates, the six cases were: Santa Barbara, CA; Houma, LA; Baltimore, MD; Atlantic City, NJ; Chattanooga, TN; and Memphis, TN. For the white arrest rates, the six cases were: Phoenix, AZ; Houma, LA; Atlantic City, NJ; Union City, NJ; and Chattanooga, TN. However, a re‐analysis of the models with these cases deleted did not significantly alter the core findings presented in this study, including the conditional relationships involving racial threat measures and police organizational factors.

7. An additional variable that was also considered in other analyses (not reported), percent Hispanic residents, was dropped from the analysis because it failed to demonstrate predictive utility.

8. Like prior research in this vein (e.g., Parker, Stults, & Rice, 2005), we also found evidence of heteroskedasticity (Breusch–Pagan test of independence χ ² = 23.82, p = .000 and χ ² = 25.11, p = .000 for the baseline and full models, respectively) and transformed the dependent variables to their natural logarithm forms to reduce this problem (Maddala, 1977). While the full model including the transformed dependent variable for Black drug arrest rates still indicated evidence of heteroskedasticity using the Breusch–Pagan test (χ ² = 4.50, p = .034; χ ² = .87, p = .350 for the model including the transformed white drug arrest rates), other post‐estimation tests, including White's test of heteroskedasticity suggest that such concerns were greatly reduced with these transformations.

9. We also considered two other police organizational variables that capture structural complexity (spatial and functional differentiation) but did not find that the interaction terms comprised of these measures and the racial threat indicators were significant predictors of the dependent variables (analyses available upon request).

10. Additional analyses (not reported but available upon request) were also conducted, using only marijuana arrests as the dependent variable. The results paralleled those using overall drug arrests as the dependent variable, including the similar interactions being found significant in the marijuana only arrest models.