ABSTRACT
We apply a cumulative disadvantage framework to examine racial inequality in the criminal justice system for drug defendants. Using State Court Processing Statistics data for the period 1990–2006 (N = 34,814), we estimate probit, multinomial probit, and OLS models to examine racial disparities in pretrial detention, adjudication, sentence type, and sentence length. We find that disparities in sentencing are not considerably large, particularly in sentence length. Larger disparities occur earlier in the process, in more discretionary stages, and through indirect pathways. In a criminal justice system that “nickel and dimes” racial inequality, examining this inequality should occur through multiple stages in the court process, rather than at a single stage.
Notes
1. While the SCPS sample is representative of the most populous counties in the United States, this does not make the dataset representative of the United States as a whole. States with more populous counties (and higher populations in general), such as California, Florida, New York, and Texas, are therefore overrepresented compared with lower-population states. For example, no counties in more rural states, including Alaska, Arkansas, Idaho, Iowa, Kansas, or Montana are represented.
2. This measure is calculated as:, where G is the proportion of each racial/ethnic group j out of J groups, subtracted from 1. In this case, we use four racial/ethnic groups: non-Hispanic Black, non-Hispanic White, non-Hispanic other, and Latino/Hispanic.
3. We also estimated these models as multilevel models and with fixed effects for county and compared the results to the original models. In the case of the fixed effects, we compared the adjudication and sentencing models as probit outcomes due to computational limitations of the software. In both cases, the results are similar to the models with clustered standard errors.
4. One problem with the Heckman correction is that the inverse Mills ratio often results in multicollinearity with other independent variables. Bushway, Johnson, and Slocum (Citation2007) suggest that the best solution (and for model identification) is to have exclusion restrictions.
5. All predicted probabilities are estimated with the mean value for each racial/ethnic group, because we feel that it is a more realistic illustration of the chances of defendants receiving a particular outcome. Similar disparities persist even when holding all other variables at the overall mean; for example, Whites have a 28.5% chance of pretrial detention, compared with about 37.5% for Blacks and 40% for Latinos.
6. Given concerns about collinearity between the selection terms, we ran variance inflation factors (VIF) for the selection term in the sentence length models. The VIF for the selection term in the prison sentence length model is 14.79, probation length is 12.59, and jail is 5.87. O’Brien (Citation2007) suggests that multicollinearity should be examined in the context of other factors that may affect the variance of the coefficient. In this case, the sample size is large and the explained variance in y does not change much when the selection term is included.
Additional information
Funding
Notes on contributors
Marisa Omori
Marisa Omori is an Assistant Professor of Sociology at the University of Miami. Her research focuses on the racialization of crime control, including racial inequality within criminal justice institutions, and drug use and punishment. Specifically, her work investigates questions of how racial inequality is created and maintained within the criminal justice system, and how context and place matters for this inequality. She has published in Law & Society Review, Crime and Public Policy, Theoretical Criminology, Journal of Drug Issues, and Crime & Delinquency.