A longitudinal study of mental health symptoms in young prisoners: exploring the influence of personal factors and the correctional climate

In Portugal, criminal responsibility starts at the age of 16.

The prisoners that did not participated in wave 2 and wave 3 (20 %, n = 15) had been released or transferred to another facility. Also, among the prisoners who remained in prison, some data on the BSI and PEI are missing at wave 2 and wave 3 (12 %, n = 8 and 17 %, n = 10, respectively) because they were serving disciplinary measures, they were in diligences in outside facilities, or simply because they did not want to participate. To evaluate the influence of attrition bias we run bivariate logistic regressions predicting attritions through the independent variables of this study. Prisoners who drop-out from the study were more likely to have a shorter prior time served in prison (z = −2.36, p = .018). With regard to the instruments, no variable was significant at the 5 % level, neither at wave 2 nor wave 3.

The BSI is not designed for prison populations and some items require slight adaptation to make sense in this context. Minor modifications were made in two items related to Phobic Anxiety, but respecting the underlying symptom being explored.

The translated PEI was previously administered to a random sample of 10 prisoners to test its face validity. Based on their comments, two items were substituted by others from the same dimension included in the longer PEI version (see [24]).

A pooled linear regression analyses with clustered standard errors was used instead of an ANOVA or a random-effect model because, in our data, this method provided predicted values closer to the observed values than the other two methods.

We thank the reviewer for this pertinent comment.

These analyses were performed on the GSI scale only because regressing all BSI subscales on the independent variables would have generated an extensive list of results, and the findings would not be much different. Including the predictors of Model 2 in Table 5 in multiple regression analyses, the direction of the effects was the same across all BSI subscales and all models were significant (p < .001). However, educational level was not significantly related with the Obsessive-Compulsive, Depression, Hostility, Paranoid Ideation, and Psychoticism subscales. Sentenced status was not significantly related with Hostility.

We developed an ordered logit regression model for each missing item, using the missing item as an outcome and the other items of the BSI or PEI as predictors. After fitting the best model, marginal effects were calculated to predict the prisoner’s score on the missing item based on the score of the items included in the model.

Likelihood-ratio tests were used to compare the fit of the pooled regression model with the random-effect model (random-effect of the intercept – initial level), and to compare the random-effect model with the mixed-model (random-effect of time). Because the mixed model did not prove to be a better fit to the data (i.e., patterns of mental health symptoms did not significantly vary over time across prisoners), the simpler random-effect model was used.