![Sample our Law journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5942/TOC-Subject-Sample-2021-Law.jpg"})
Abstract
Punishment scholars acknowledge the fundamental importance of both prosecutorial discretion and jurisdictional variation in punishment; yet, little is known about the extent to which charging practices vary across court contexts. This is especially true in the federal criminal justice system. This research investigates this issue by linking four years of charging data from the Administrative Office of the U.S. Courts (AOUSC) to corresponding data from the U.S. Sentencing Commission (USSC). It incorporates unique information on district court contexts from a variety of sources to investigate jurisdictional variation in charge reductions. Findings provide evidence that federal charging practices vary across district courts. In particular, several structural court characteristics are significantly related to the likelihood of charge reduction for similarly-situated defendants. Results from this study are interpreted through the lens of contemporary legal perspectives on court communities and suggest several fruitful directions for additional research on the social contexts of criminal prosecution.
Notes
1 The decision to file for a motion requesting a substantial assistance departure is controlled exclusively by the prosecutor, but federal judges maintain control over the magnitude of the sentencing discount. Substantial assistance departures are particularly important in that they allow offenders to be sentenced below not only the minimum guidelines recommendation, but also below any applicable mandatory minimum sentence requirements.
2 The time period was purposefully selected in part to allow for an examination of the consequential 2005 U.S. Supreme Court’s Booker decision (including 2 years of data before and after the decision), and also to provide for possible temporal lags in case processing in which earlier years of charging data (e.g. 2006) could be linked to later years of sentencing data (through 2009) where necessary. The timeframe also restricts the data to a single Presidential Administration (under George W. Bush) which is important because federal charging practices can shift under new administrations.
3 In total, 346,043 cases were prosecuted in the U.S. district courts between 2003 and 2006. However, only cases that are convicted and sentenced by federal district judges are recorded in the USSC data. Public-order crimes are overrepresented in the unmatched cases because they include certain traffic offenses that are rarely sentenced in federal court, but other case characteristics were similar for matched and unmatched cases (see Johnson (Citation2014) for a full discussion of linking procedures and matched comparisons). The remaining unmatched cases may reflect data entry errors, failure on the part of the matching algorithm to identify unique cases, or the fact that some cases were never turned over to the USSC (Adams & Motivans, Citation2003).
4 Immigration cases are omitted for several reasons. Specifically, they are geographically concentrated in southwestern districts, involve an overwhelming majority of non-citizens and a large number of undocumented immigrants, and are frequently subject to alternative case processing options such as “fast-track” departure programs and summary deportation proceedings (USSC, Citation2003). Duplicate analyses (available by request) that included immigration offenses produced similar substantive conclusions. The District of Columbia is excluded because it is in many ways unique; it is its own circuit court, and DC Superior Court cases are often heard in federal court.
5 We rely on a binary measure to capture the receipt of a charge reduction for two reasons. First our substantive interest is in district-level variation in the use of charge reductions; and second, the majority of cases involve reductions of only 1 level in severity, which results in limited variability in a continuous charge reduction measure.
6 To provide an example, if there are five filing charges and four terminating charges, the highest statutory severity level among the five filing charges is compared to the highest severity level for the four remaining charges. Although it is also possible for prosecutors to increase the statutory severity of charges from filing to termination, this is very rare in the data, with less than 1% of cases involving an increase in statutory maximums.
7 Cases in which lesser-included charges were dropped without affecting the severity of the top charge were not originally coded as charge reductions, nor were cases where statutory fine amounts were lowered without affecting maximum prison terms. To investigate these coding decisions, we examine alternative model specifications in which reductions in the number of charges are also coded as charge reductions, and in which reductions in maximum fine amounts are coded as charge reductions. This increases the percent of cases scored as charge reductions from 15.3% to 17.3% and 18.5%, respectively, but it has little impact on our overall patterns of results or substantive conclusions.
8 Trial cases were included in the analysis because, even though charge reductions are uncommon in trial cases, they can and do occur. Charge reductions in these cases may reflect evidentiary issues, successful defense motions, or partial acquittals on some charges. Supplemental models restricted to cases disposed of only by guilty pleas produce equivalent findings in all models.
9 The civilian labor force unemployment rate was also examined but is not included because the reliability of the four-item scale was lower (α = 0.71). Alternative model specifications that included the unemployment rate produced identical findings.
10 Racial and ethnic minority populations are combined to capture aggregate threat dynamics associated with all minority groups. Separate measures were examined but not included because the vast majority of federal districts have negligible populations of both Asian and Native Americans, and because there was a strong positive correlation between percent Hispanic and size of the court that introduced multicollinearity into the statistical model. Examination of the final correlation matrix for the level 2 variables demonstrated no problematic levels of collinearity with other measures.
11 Senatorial voting patterns have to be measured at the state rather than district level. The consequence is that variation on this measure will be reduced, making statistical significance tests more conservative for this measure.
12 Preliminary models using three levels of analysis, first with cases nested in districts and circuits, and second with cases nested in districts and states, revealed no evidence of significant circuit-level or state-level variation in charging practices (τDistrict = 339, p<.001; τCircuit = 018, p < .05; τDistrict = 330, p>.001; τState = 056, p > .05), so two-level models are utilized and reported.
13 Because the logistic model lacks a meaningful level-1 variance component, the estimate for the intraclass correlation (p = .11) assumes that there is a continuous latent variable underlying the binary response outcome, which follows a logistic distribution with individual level variance VI equal to π2/3 (see Snijders & Bosker, Citation1999). This approach is consistent with prior multilevel sentencing research (see e.g. Johnson, Citation2006).
14 Explained variance is computed as a proportionate reduction in error (PRE) measure that compares residual variance from the individual-level models to the total variance from the unconditional model (see Snijders & Bosker, Citation1999: 99–109).
15 All individual-level predictors were specified as random coefficients because their effects varied significantly across courts, with the exception of the Native American race, number of dependents, and educational attainment variables, which did not vary and were therefore specified as fixed coefficients. Full results for all random effects in the model are available by request.
16 The coefficient for black defendants approached statistical significance (p = 10), and alternative model specifications (not reported) including an interaction term for young (i.e. under 30), male, minority (i.e. black or Hispanic) defendants also demonstrated that this group was marginally less likely to receive charge reductions (b= −0.05, SE =0.03, p = 052).
17 Alternative model specifications including immigration offenses demonstrated that immigration crimes were the least likely to result in charge reductions. When immigration crimes are included and specified as the reference group, drug offenses are 2.6 times more likely to result in charge reductions. In part, this reflects the fact that roughly two-thirds of immigration offenses have maximum terms of only 1–2 years, which means that prosecutors would have to reduce felony immigration charges to misdemeanors in order to lower their statutory maximums. This illustrates the point made by Wright and Engen (Citation2006) that the “depth and distance” of the penal code can shape charging decisions in important ways.
18 Additional supplemental analysis showed that charge reductions operate primarily by moving defendants into lower cells in the federal sentencing guidelines; specifically, a charge reduction reduces the presumptive sentence recommendation by 32% on average, which translates into actual sentences that are 28% shorter when compared to similar defendants who do not have their initial charges reduced.
19 In multilevel regression models, statistical tests for individual-level predictors are based on the total sample size (n = 198,603), whereas for district-level predictors they are based on the number of districts (n = 89). For this reason, significance tests for contextual variables have less statistical power; as such, we consider marginally significant results (p≤.10) when reporting and discussing district level effects in the text and tables.
20 The lone exception involved a significant effect for political conservatism in supplemental analyses in which reductions in the number of charges were included in the dependent variable. As noted above, this likely reflects the fact that conservative districts had a significantly higher mean number of initial charges, providing greater opportunity to reduce the number of charges in these court environments.
21 Other temporal changes could be confounded with the Booker decision. For instance, the observed effect is also consistent with earlier changes that accompanied the Ashcroft Memo, which was released in September of 2003 and instructed federal prosecutors to only charge the most serious readily provable offense, with written supervisory approval required for exceptions.
22 A target letter is used to inform a putative defendant that they are the target of a federal investigation and may be a prelude to pre-charge negotiations. A deferred prosecution, or pretrial diversion agreement, involves the release of the defendant under specific supervision conditions in which the U.S. Attorney agrees to decline prosecution if the defendant successfully completes the period of supervision. See U.S. Attorney’s Manual sections 9–11.151 and 9–22.200 for additional detail.