ABSTRACT

To create ten risk models (five types of sex derived from the Sexual Aggression in Prison (SAP), separately for male and female inmates), we used exploratory exhaustive Chi-Square Automated Interaction Detector (CHAID). Due to our relatively small sample size, analyses were considered exploratory. We therefore used a cross-validation method which divides the sample into ten folds and verifies the original risk model. This method produces a risk estimate as well as a cross-validated risk estimate that represents the average of all risk estimates across the ten sub-samples. We allowed as few as five cases per cell.