Is Educational Achievement a Turning Point for Incarcerated Delinquents Across Race and Sex?

Her perspective also argues that violent crime among Black males is due to their attempt to negotiate masculine identities in the face of structural constrains that make claiming valued identities difficult. On the other hand, “females who bring to interactions traditional beliefs about gender are likely to define situations such that delinquency seems undesirable, because they will be likely to view law violation as inappropriate for their gender. Females who are less accepting of traditional gender ideologies, in contrast, define situations such that delinquency is a more likely outcome” (Heimer 1995: 150).

An anonymous reviewer noted that the school achievement and participation measures used herein can be considered and interpreted in several ways. Specifically, there is likely to be potential growth in skills coming from education but also positive labeling/renegotiation of labels insofar as prospective employers and partners react positively to individuals who have attained educational success. Of course, the two attributions are not mutually exclusive but, it is important to be explicit that educational experiences and attainment act upon re-offending in much the same way as other life events (marriage and employment) because they represent positive turning points/events that open doors for conventional success, broadly defined, and may re-orient previous crime trajectories.

The continuous measure of academic achievement while incarcerated was dichotomized at the mean to distinguish youth who had excelled in the classroom while incarcerated relative to those who had under-achieved academically based on the number of academic credits earned and the proportion of academic credits relative to the total of academic and elective credits. Similarly, those youth who had above average attendance in public schools post-release were differentiated from those with below average attendance. We performed several alternative cut-points as a sensitivity analysis, including a median split and another one in which the measures were trichotomized and cases in the upper third of the distribution were defined as high academic achievers and exhibiting high school attendance. The results using these alternative cut-points were substantively the same as those produced when using the mean as the cut-point.

We were unable to compare the initial cohort of cases to those analyzed using PSM because the dataset created when the PSM models were generated did not include the case identifier (which was originally prohibited from inclusion by confidentiality agreements). Only the variables used in the models and those generated in STATA, such as whether those cases that matched were in the control/experimental group and cases that did not match, were available. And although we are unable to determine precisely the overlap between the two samples, we are confident that the overlap was very good. Nevertheless, we compared the mean values of the covariates in the PSM models reported in Tables 3 and 4 across cases that matched versus those that did not. For Table 3, i.e., academic achievement while incarcerated and returning to school post-release, we found the following: Of the 11 covariates in the model, four of the mean differences across the matched and non-matched cases were not significant. Of those that were significant, the percentage difference in the mean was below 5% for one variable, two were between 5 and 10%, and four were above 10%. For Table 4, i.e., attendance in school post-release and re-arrest, the mean differences across the matched and non-matched cases were not significant for seven of the covariates, one was different by less than 5%, and three greater than 10% different. While these comparisons indicate some instances in which there is some level of variability across the matched and non-matched cases, we do not believe that they are egregious enough to prevent faith in the analyses, results, and conclusions with the appropriate caveats noted above.

The 11 covariates in Table 1 were used to conduct the propensity score matching. Balance across the treated and comparison groups was achieved as evidenced by no significant differences in the propensity scores within the strata, indicating exposure to the treatment (above average academic achievement and above average school attendance) is random. The covariates of age at release and length of incarceration could possibly be considered inappropriate for inclusion in the matching process because they could be considered occurring after the treatment assignment. For the following reasons, it was determined that retaining these variables was appropriate. First, the length of incarceration and age at release have consistently been shown to be important control variables in recidivism studies and excluding them would reduce the explanatory power of the models. Second, these measures would likely influence treatment assignment because both could impact how much education a youth receives while incarcerated and their level of attendance in school post release. Finally, research has included these same variables in propensity score matching analysis in examining the impact of the treatment effect of Supermax confinement on recidivism (Mears and Bales 2009).