Sex Disparities in Sentencing and Judges’ Beliefs: A Vignette Approach

ABSTRACT

The predominant theories on courtroom decision-making explain extralegal disparities with the courtroom actors’ use of stereotypes. We conducted a vignette study on a sample of judges at the Chinese National Judges College, manipulated the sex of the defendant in each vignette, then asked the subjects for the recommended sentence. The survey also contained a series of questions on the beliefs about the causes of crime, the patterns of criminal behaviors, and the effectiveness of punishment. We found that the judges recommended significantly less harsh sentences for the female homicide defendant, but recommended significantly harsher sentences for both female defrauding and drug trafficking defendants. We also found little evidence that the perception and belief variables were confounders behind the observed sex disparities.

KEYWORDS:

• Sentencing
• women as offenders
• courts

Acknowledgments

We thank the anonymous reviewers for their comments, and Siyu Liu and Becky Leung for their input. Early results of the study were presented at a seminar at Southern Illinois University, and we thank the comments from the audience as well. All errors remain our own.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15564886.2021.1947427.

Notes

1. Throughout the paper, we use “sex” to refer to the biological sex of people, and “gender” to refer to the cultural norms and expectations associated with people’s biological sex. In the overwhelming majority of criminal justice studies, researchers can only measure the sex of people (and therefore sex disparities in case outcomes), but many theories (including the main ones used in this study) are based on gender roles and norms. The two concepts are especially prone to confusion in the criminal justice context, and researchers sometimes use them interchangeably. While an in-depth differentiation between the two concepts is beyond the scope of our study, we aim to use the two terms in a precise manner.

2. Although both theoretical frameworks were much more commonly associated with racial disparities in sentencing (Ulmer, 2012), many studies on sex disparities or the race-sex interaction effects also utilized them, since the blameworthiness and risk components are compatible with the themes of gender-centered theories (see Bontrager et al., 2013). For simplicity, the paper focuses on the gender-centered theories (i.e., chivalry and “the evil woman”) whenever possible, and only applies the more general perspectives when necessary.

3. Some studies used the term “selective chivalry” to denote a perspective closer to “the evil woman” hypothesis (Embry & Lyons, 2012; Rodriguez et al., 2006). Throughout the paper, we stick to the original scopes of the selective chivalry perspective.

4. Out of all independent variables used in the analysis, none was missing for more than 12 subjects (4.41% of completed surveys). Three of the questions on the comparison between offenders had a “not sure/does not know” option. Instead of dropping the subjects who selected those options (between 6.6% and 11.2% of completed surveys), we kept them as a separate category. Out of the 272 judges, 17 (6.25%) had a missing recommended sentence type or length for one or more vignettes. We imputed the missing values with the median length for each of the crime types (120 months for homicide, 72 months for robbery, 36 months for defrauding, and 84 months for drug trafficking). We also tried to leave the fields as missing and used listwise deletion, and the substantial findings were very similar.

5. The original instrument also contained a belief question about defendants’ recidivism patterns, and sentence recommendations if the defendants had prior records. We did not utilize those questions in the current study because we believed they fall under a different scope, and the more appropriate way to explore these issues is in a different study.

6. The vast majority of defendants receiving life imprisonment do get released after serving approximately 20 years through commutation. Most defendants receiving a suspended death sentence will first have their sentence reduced to life imprisonment, then eventually get released in a similar fashion. While China did enact a life without parole article in 2015 (before our survey), it is unlikely to be applicable to any of the defendants depicted in the vignettes. For additional information on life sentences in China, see Smith and Jiang (2019); for additional information on the suspended death penalty, see Trevaskes (2013). Throughout the nearly 1100 judge-vignette combinations (272 judges * 4 vignettes), the death penalty without a suspension was only recommended twice, both for the drug trafficking scenario, which suggests its coding should have no impact on three of the four vignettes, and only a minimal impact for drug trafficking.

7. A pairwise correlation analysis found a low-to-modest level of correlation among the perception and belief variables (|r| ≤ .45 for all pairs), suggesting that the variables were capturing different dimensions of judges’ beliefs.

8. Our coding scheme for both variables here were 1 = strongly agree and 4 = strongly disagree, so each additional point on the scale was one point away from the belief in prisons’ functions.

9. We first added each and only one sex*perception (responsible party, the three perceived differences, the two beliefs in prisons’ functions) interaction term to our main model at a time, then also estimated one overall model with all interaction effects included, neither found noticeable interaction effects.

10. For the first two sensitivity tests, we only present the coefficient for the male defendant since the key purpose was to examine if the sex disparities remained. We present the full parsimonious models, because the purpose was to examine whether the reduction of parameters affected all coefficients.

11. If a judge selected “does not know/not sure” for one or more of the perceived difference questions (17 total), we dropped the judge’s response to that question, and used their response to the other perceived difference questions with a valid value to calculate the average score. For example, if a judge answered “very different” (1) to one question, “almost the same” (3) to another question, and “not sure” (4) to a third question, the average score would be 2 rather than 2.67. We also estimated the models using only judges who gave a response other than “not sure” for all questions and did not find meaningful differences in the results.

Additional information

Funding

This work was supported by the Beijing Social Science Fund [16FXC031].