Fig. 1: Race-based sentencing disparity in the federal court system.
Each point represents mean sentencing disparity (in months) as estimated from a regression model that conditions the outcome by adding an additional explanatory variable (given on the x-axis) to the model immediately to the left. The one exception is the “Grid Cell” model, marked with an asterisk (*) on the x-axis. This model combines the previous four explanatory variables into a single one representing the cell of the United States Sentencing Grid onto which each defendant is placed, as the grid cell sets the presumptive sentence. Disparities are shown as the average difference in sentences given to defendants who are Black (blue circles), Hispanic (yellow triangles), and another racial identity (ARI, red squares), each as compared to white defendants. The leftmost model, labeled “Baseline,” provides the average difference in sentences between racial groups irrespective of any and all other factors. The rightmost model demonstrates average disparities for minoritized defendants-especially Black and ARI ones-that remain unexplained. Table 1 reports the values of the estimates represented here along with p-values and confidence intervals. Table 2 provides a more detailed specification of the 12 regressions. The analysis is based on N = 518,721 sentencing records. Model diagnostics, including F-statistics, the associated degrees of freedom, the associated p-value, as well as raw and adjusted r2 values appear in our permanent data repository (Topaz, 2023). Each of these F-statistic p-values is numerically indistinguishable from zero, and the adjusted r2 values range from 0.02 for Model (1) up to 0.79 for Model (12).
