The Development and Validation of a Preliminary Model for Classifying Technology Facilitated Child Sexual...

The current study aimed to develop and validate a preliminary model for classifying self-reported TF-CSEA perpetration in general population samples. Data were based on census-weighted samples of adult men from Australia (n = 1,939), the U.K. (n = 1,506), and the U.S.A. (n = 1,473). Penalised regression with repeated 10-fold cross-validation were used to identify predictors in the Australian sample. The final model included thirteen items related to attitudes towards TF-CSEA, use of encrypted messaging apps, online behaviours, and mental health. The model demonstrated good discrimination (AUROC = 0.91), calibration (O:E = 1.10), and accuracy (Brier score = 0.04) in the Australian sample, with strong external validity in the U.S.A. (AUROC = 0.90; O:E = 0.99; Brier score = 0.05) and U.K. (AUROC = 0.85; O:E = 1.07; Brier score = 0.03) datasets. The model was translated into a nomogram, and net benefit decision curve analysis indicated that flagging men exceeding a probability threshold of 0.24 produced a net gain of 42, 17, and 43 true positives per 1,000 men assessed in Australia, the U.K., and U.S.A. Subgroup analyses demonstrated higher net benefits when applied to men who work with children. This model provides proof-of-concept evidence that TF-CSEA related behaviours can be classified within anonymous survey data using non-incriminating self-report indicators in general population samples.