The Non-Constant Effect of Defendant Sex across Criminal History: How and When Does Criminal History Condition Sex Disparity at Sentencing?

Abstract

The goal of this study is to examine how, and when, criminal history conditions the effect of defendant sex on incarceration and prison sentence length decisions in Minnesota state courts. Results suggest that sex differences in sentencing are largely concentrated amongst those who have extensive criminal histories, bypassing those who have little or no criminal history. Moreover, criminal history’s aggravating contextual effect on sex often depends on the dependent variable examined and type of offense committed. Study results shed light on the situations where extralegal disparities are “hidden” in the nuances of case characteristics.

Keywords:

• Criminal history
• criminal justice
• sex
• punishment
• sentencing
• women

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Notes

1 In 2016, the Minnesota Sentencing Guidelines Commission implemented a new Drug Sentencing Grid with different offense level patterns than the standard sentencing grid (MSGC, 2019). To remain parsimonious and examine person, property, and drug defendants sentenced on the same sentencing grid—we use data spanning pre-2016.

2 Supplemental models which top-coded the values that fell over 470 months (n = 22) were run and the results of the models were identical to those presented.

3 Supplemental models which include the presumptive sentence measure unlogged were run and the results were substantively identical. However, the Pseudo R² was greater in models which use the logged term so the models with the logged presumptive sentence measure are provided in this paper. Results of the models with an unlogged presumptive sentence are available upon request from the corresponding author.

4 In the MSGC codebook, the plea measure is described as “Plea Entered? Yes/No.” This led us to investigate whether this means that a plea deal happened or whether it was just the verbalization of a plea. To look into this, we went to the MSGC annual reports which are based on the data in this study. In the 2015 MSGC Annual Report, they state that “97% of felony convictions were obtained without a trial.” This matches the “plea” variable in the MSGC Monitoring dataset which suggested that 3.38% of cases had a “Not Guilty Plea” entered.

5 There is disagreement in the sentencing literature about whether to control for departures in multivariable models. While some studies do include departure controls (Albonetti, 1997; Doerner & Demuth, 2010; Feldmeyer & Ulmer, 2011), recent legal research has advocated against it (Fischman & Schanzenbach, 2011; Starr & Rehavi, 2013). Specifically, recent research argues that by controlling for departure status, temporal ordering is not kept and that the decision to depart is not separate from the disposition decision. We recognize this disagreement so we ran supplemental sentence length models with a control for durational departure from the Minnesota Sentencing Guidelines. Results from the models were substantively similar regardless of whether a departure control was included. The results presented are without departure controls but models with departure controls are available upon request from the corresponding author.

6 We account for inter-county correlation using the clustered sandwich estimator in Stata 17. This approach is similar to using a multi-level model with no level 2 predictors: treating observations between counties as independent but relaxing the independency requirement within counties.

7 We did explore using OLS regression with the Heckman correction term for selection bias. Similar to recent sentencing research which has noted significant collinearity between the Heckman correction term for selection bias (Bushway et al., 2007; Feldmeyer et al., 2015) and legal factors, the inclusion of the inverse mills ratio to the model introduced collinearity concerns (Presumptive Sentence VIF = 6.88, Criminal History VIF = 6.84, Mills Ratio VIF = 11.48). The inability to use the Heckman correction term in OLS models is yet another justification for the tobit approach. See Footnote 8 for collinearity diagnostics for analytic sample.

8 The mean VIF was 1.37 with a maximum of 1.80.

9 This sentence length statistic is only amongst those sentenced to prison. The average sentence length for the whole analytic sample (including those whose prison sentence length is 0) is 11 months (SD = 30.98).

10 Presentence detention has been found to be a predictor of sentencing outcomes in past empirical research so we include it in this study. It should be noted that the Minnesota Sentencing Guidelines use presentence detention when calculating presumptive sentence—thus some of this variable may already be picked up by the presumptive sentence measure. There is precedent for using a variable although a guideline indices already, theoretically, controls for it (see Ulmer, 2000). We ran models with and without presentence detention in the model and the results were substantively similar.

11 These models include controls for all variables in the main analysis. However, control variable estimates are omitted from these tables in the text for brevity. Control estimates for the offense-specific models are available upon request from the corresponding author.