Construction of a deception resistant questionnaire for alleged child sexual abuse victims : enhancing the Finnish Investigative Instrument of Child Sexual Abuse (FICSA)

Abstract

The prevalence of child sexual abuse (CSA) is seemingly decreasing, but at the same time the number of reported CSA cases is rising, indicating that also unfounded CSA allegations are being processed by the authorities. In CSA cases, experts must deal with large amounts of information that needs to be evaluated and integrated to reach an objective conclusion. CSA investigations are difficult, and their outcomes can have serious consequences for all parties involved. Thus, it is crucial that CSA investigations are carried out properly. However, studies have shown that best-practice guidelines are often overlooked by CSA experts. The Finnish Investigative Instrument of Child Sexual Abuse (FICSA) is a decision-making support tool, created to address some of the issues in CSA expert evaluations. FICSA offers a statistical approach to deal with the complex information in CSA investigations, by estimating the probability for the alleged CSA being happened based on the child’s background information.

The current study aimed at developing FICSA further, by adding and testing a deception resistant feature in it. Two gender-specific questionnaires with the original FICSA questions and additional “trap” questions were constructed. The trap questions were selected from variables in the victimization survey that FICSA is based on. The trap questions were not statistically related to CSA but seemed like it by their nature. We instructed 275 first-grade high school students to answer the questionnaire trying to simulate being CSA victims. Their responses and the responses of 278 real CSA victims from the victimization study were combined to build three Naïve Bayes classifiers with different combinations of question sets for each gender, including a cross-validation procedure, to separate between these two participant groups. The performances of the built classifiers were compared to find the best set of questions for deception control.

The most efficient classifier had excellent diagnostic utility for both genders (AUC = 0.93 for boys and AUC = 0.91 for girls). Thus, the selection of trap questions was successful, and the built classifier was able to separate real victims from the simulator group. Moreover, this suggests the classifier being useful in the investigation process of CSA cases, offering a statistical approach to use at the starting point of the investigation. By being able to identify a possibly false allegation at an early stage of a CSA investigation, authorities’ resources can be better directed towards allegations that are more probably true. This provides a possibility to help more CSA victims, more efficiently.