Distinguishing Race Effects on Pre-Trial Release and Sentencing Decisions

Abstract

Racial disparities in court dispositions and sentences might reflect systemic biases toward minorities, but they might also stem from race group differences in legal or other extra-legal factors linked to a defendant’s risk for future criminality. Analyses of over 5,000 felony defendants from an urban Ohio jurisdiction revealed that significant main effects of a defendant’s race on release on one’s own recognizance (ROR), bond amounts, and prison sentences were rendered nonsignificant when controlling for legal factors, such as offense severity. Analyses of interaction effects, on the other hand, revealed that African American males age 18–29 experienced lower odds of ROR, higher bond amounts, and higher odds of incarceration in prison relative to other demographic subgroups, even with the inclusion of rigorous controls for legally relevant criteria. The relevance of these findings for understanding disparate treatment at different stages of case processing is discussed.

Keywords:

• racial disparities
• extra-legal disparities
• court dispositions
• sentencing

Acknowledgment

I thank the reviewers for their helpful comments.

Notes

1. The original formula is . Subtracting 2cov(b ₁, b ₂) from the term under the radical will subtract any overlapping variance in each coefficient resulting from the common population (Theil, 1971, p. 303).

2. Some readers might be concerned about including so many control variables as to possibly mask any “real” race effects on the outcomes. All of the legal factors examined here, however, are relevant to judicial decisions. Given that these measures are legally relevant whereas race is not, and given the possible overlap between these legal factors and race, it is important to give judges the benefit of the doubt and allow the legally relevant factors to account for as much variation in the outcomes as is possible. The alternative is potentially problematic, to permit race to account for as much of this variation and only then allow legal factors to explain what is left.

3. Felony sentencing under Ohio’s guidelines is dictated by five felony levels and, as such, has been referred to as a “single-grid” sentencing scheme (Kauder & Ostrom, 2008, p. 19). For the sample examined here, roughly 37% of the variation in imprisonment was accounted for by the felony classification measures and the 14 specific offense measures (based on the Nagelkerke R ² computed from a pooled logistic regression in SPSS 18.0). This level of prediction was superior to levels obtained with alternate offense rankings, including the summed ranks of convicted charges examined by Wooldredge et al. (2002) in their evaluation of sentencing practices before and after the implementation of Ohio’s sentencing guidelines (Nagelkerke R ² = .26).

4. The software used for the analysis was HLM6.02 (Raudenbush et al., 2004).

5. The odds of falling into the reference category (prison) were computed from the multinomial models as 1/[1+(P1/P3)+(P2/P3)], where the odds ratios P1/P3 and P2/P3 were computed from each equation by holding all other predictors constant at their means.

6. Although the odds ratio for bond amount might appear weaker relative to other coefficients, this is because the ratio reflects change in the outcome for a one-unit change on a ratio scale rather than on a binary scale. Dummy variables span only one unit, so the odds ratio for a dummy predictor describes the entire amount of change in the outcome across the entire range (one unit) of the binary scale. The impact of bond amount is strong when considering the range of the logged scale (0.00–14.63).

7. The absence of significant differences in the odds of imprisonment between African American males and white males found here counters Steffensmeier and Demuth’s (2006) finding of a significant difference in imprisonment between these groups in a sample of felony defendants processed in the largest US counties. This difference in statistical significance is likely due to a difference in sample sizes between studies, however, because the magnitude of difference between the odds ratios for their model constants (1.39 for white males vs. 1.40 for African American males) is actually smaller than the magnitude of difference in the odds ratios for the model constants presented for these two demographic sub-groups in Table 6 for “no prison/jail vs. prison” (1.68 for white males vs. .835 for African American males).