Data Analysis
Before conducting the statistical analysis, the researcher ran the descriptive statistics to check for any abnormalities in the data that may interfere with the validity of the results. At this time, data missingness of over 10 percent was detected in the dependent variable (Was the offender a convicted sex offender at the time of arrest?). To address data missingness, the researcher excluded cases with missing values for the dependent variable. Nine hundred forty-six cases were excluded from the study due to missing data, resulting in a final sample size of 105. Upon inspection, all other variables appeared suitable for entry into the analysis and bivariate correlations were run. Among independent and control variables, no Pearson’s Correlation value exceeded .8, suggesting that the assumption of multicollinearity was met and it was appropriate to proceed with the hierarchal logistic regression.
The results of the binary logistic regression showed nonsignificant results for Block 1, Block 2, and Block 3 (See Table 1). Block 1 demonstrated overall statistical non-significance (p=.209; Nagelkerke R-Square = .081). The Hosmer and Lemeshow test of goodness of fit shows non-significant results, suggesting that observed event rates matched the expected event rates (p=.215). Each of the control variables, sex (p=.999), Age (p=.238) and Race (p=.657) yielded non-significant results. These findings suggest that none of the control variables significantly predicted whether or not the offender had previously been convicted of a sexual offense.
Block 2, including variables on the nature of the crime, also yielded non-significant results (p= .098; Nagelkerke R-Square= .211). Although overall non-significant, the increase in the Nagelkerke R-Square value suggests that the addition of the nature of the crime variables in Block 2 improved the independent variables’ ability to explain variation in the dependent variable. For Block 2, the Hosmer and Lemeshow results show a goodness of fit between the observed and expected events (p=.854). The variables entered, “Did the case involve the sexual exploitation of a minor?,” “Was this a child pornography distribution to minors case?,” “Was this a child pornography production case?,” and “Was this a child pornography possession case?” all demonstrated non-significant findings (p > .05).
Block 3 included all personal history variables and also reported non-significant findings (p=.062; Nagelkerke R-Square=.342). Once more the increase in the Nagelkerke R-Square value suggests that as personal history variables were added to the model, it became a more parsimonious model. Hosmer and Lemeshow results demonstrate non-significant results, suggesting a goodness of fit between the expected and observed values (p=.561). Each variable entered in Block 3 including, “Did the offender have a problem with drugs and alcohol at the time of the crime?,” “Did the offender have a diagnosed mental illness at the time of the crime?,”
“Did the offender have any prior arrests for nonsexual offenses?,” and “Did the offender have an illegal occupation at the time of the crime?” all yielded non-significant results (p > .05). Ultimately, Block 1, Block 2, and Block 3 all reported non-significant results overall and for each variable therein suggesting no significant prediction power for offender recidivism.
Table of Contents
- A Study of Recidivism Among Online Sexual Predators
- Literature Review
- Gaps in the Literature
- Methodology
- Data Analysis
- Discussion and Conclusion
- Appendix
- ENDNOTES