Reassessing the Influence of Criminal History in Federal Criminal Courts

Abstract

Prior sentencing research indicates that defendants with more extensive criminal histories receive more punitive dispositions and that criminal history influences sentencing decisions over and above its influence on the guideline recommended sentence. To date, these additional effects of criminal history have almost exclusively been treated as linear effects. However, there are plausible reasons to expect that criminal history could have curvilinear effects on sentencing outcomes that taper off at higher scores. The purpose of this paper is to explore the potential curvilinear effects of defendant criminal history on incarceration, sentence length, and downward departure decisions in federal criminal courts. The findings suggest that criminal history has curvilinear effects on each of these sentencing outcomes. As criminal history category increases, defendants receive more severe sentences, net of other factors, but only up to a certain threshold level, at which point criminal history effects taper off and even reverse.

Keywords:

• Criminal history
• federal court
• focal concerns
• punishment
• sentencing

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes

1 Criminal History 1 is associated with 0 or 1 criminal history points. Criminal History 2 is associated with 2 or 3 criminal history points. Criminal History 3 is associated with 4, 5, or 6 criminal history points. Criminal History 4 is associated with 7, 8, or 9 criminal history points. Criminal History 5 is associated with 10, 11, or 12 criminal history points. Criminal History 6 is associated with 13 or more criminal history points.

2 The extant literature on criminal history’s effect on downward departure odds is mixed, with some studies finding that increases in criminal history decrease the odds of downward departure (Johnson et al., 2008), some finding that increases in criminal history increase the odds of downward departure (Hartley et al., 2007), and some finding null effects, net of other factors (Ortiz & Spohn, 2014; Spohn & Brennan, 2011). Notably, outside of Ortiz & Spohn (2014), these studies of departures provide almost no exploration of criminal history as a quadratic term.

3 Ulmer (2000) included a curvilinear effect of criminal history in a model using direct measures of offense seriousness, offense seriousness squared, and criminal history. He found that criminal history did have curvilinear effects; therefore, a threshold effect did exist. However, this effect was found without controlling for the presumptive sentence in the model.

4 For example, United States citizens cannot be charged with offenses such as being in the country illegally (Demuth, 2002). Eligibility for deportation limits the judges sentencing options in regards to downward departures (Byrne & Turner, 2010). Furthermore, the accuracy of the measure of criminal history for non-citizens is a concern (for further discussion see Demuth, 2002).

5 In addition, cases sentenced in the Northern District of Iowa (n = 519) are excluded from the analysis because their inclusion would not allow for model convergence using HLM (to account for clustering of cases by federal district). Specifically, models failed to converge due to a lack of variance in the incarceration outcome in the Northern District of Iowa (98% of criminal sentences in the Northern District of Iowa resulted in an incarceration sentence). However, supplemental analyses using clustered standard errors in STATA that included Northern Iowa cases were able to converge and produced results nearly identical to those shown here.

6 Supplemental analyses using the Heckman two-step correction were done to account for potential selection bias into the sentence length models. However, the inverse mills ratio used to account for selection was highly collinear to pretrial detention (r ≈ 0.80). Collinearity with inverse mills ratios in models has been a longstanding concern among researchers using the Heckman two-step correction (see, Bushway, Johnson, & Slocum, 2007; Feldmeyer & Ulmer 2011; Feldmeyer, Warren, Siennick, & Neptune, 2015). As a result the models without the Heckman correction are presented.

7 Ulmer (2000) found that both offense seriousness and criminal history had curvilinear effects on incarceration and sentence length decisions when controlling for the other (using criminal history, criminal history squared, offense seriousness, and offense seriousness squared), without controlling for the presumptive sentence. However, Ulmer (2000) did not analyze the influence of criminal history squared or offense seriousness squared on incarceration and sentence length decisions while controlling for presumptive sentence. Research after Ulmer (2000) has primarily used criminal history as a control variable alongside the presumptive sentence due to collinearity concerns when using presumptive sentence and offense seriousness in the same model (Feldmeyer & Ulmer, 2011; Johnson et al., 2008). In the current research, a similar reasoning exists for not including a squared measure of offense seriousness. The 43-point offense level is highly correlated with the guideline minimum (r = 0.9119; p < 0.001). Meanwhile, criminal history is only weakly-to-moderately correlated with the guideline minimum (r = 0.3715; p < 0.001). Taken together, due to collinearity concerns and to answer the specific research question the curvilinear effect of criminal history is modeled but the potential curvilinear effect of offense seriousness is not.

8 Supplementary analysis was conducted using criminal history category as a series of dummy variables, with the first criminal history category as the reference group. The findings were substantively the same as using it as a continuous measure. This is discussed more in depth in the supplemental analyses portion of the results.

9 Feldmeyer and Ulmer (2011) control for acceptance of responsibility sentencing discounts. However, in the current data, acceptance of responsibility was highly correlated (r > 0.60) with mode of disposition. As such, we exclude the acceptance of responsibility from models.

10 Criminal history and criminal history squared were highly correlated but that should be expected as criminal history squared is a product of criminal history. In assessing multicollinearity, we excluded criminal history squared from the diagnostics. We followed the same procedure for age and age squared. Notably, adding the quadratic term for criminal history did not substantively change any of the effects of our multivariate models, and it was not closely associated with any other predictors.

11 According to the Interactive Source Codebook from the United States Sentencing Commission (2016), downward departures occurred in 49-52% of federal cases annually between 2014 and 2016. The larger percent of downward departed cases in our sample is likely a product of the exclusion of cases which could not receive downward departures (Zone A cases) or were unlikely to receive downward departures (Armed Career Criminals/Dangerous Sex Offenders Against Minors; non-citizens).

Additional information

Notes on contributors

Bryan Holmes

Bryan Holmes is a Ph.D. student in the School of Criminal Justice at the University of Cincinnati. His research interests include sentencing, pre-trial court processes, and the use of discretion by criminal justice actors.

Ben Feldmeyer

Ben Feldmeyer is an Associate Professor of Criminal Justice at the University of Cincinnati. His research focuses on criminal behavior and criminal sentencing across demographic groups, social class, and macro-level contexts. His recent work has appeared in Criminology, Journal of Research in Crime and Delinquency, Social Science Research, and The Sociological Quarterly.