Predicting unethical behavior from interview responses : machine learning models versus human judges

Abstract

How can we evaluate peoples moral character? Judging someones moral character can be a difficult task, especially through only short interactions such as an interview. In this thesis, I examined the possibility of using machine learning techniques to predict peoples propensity to commit certain unethical behavior based on analyzing their responses to interview questions aimed at testing their moral character. I experimented with a number of machine learning algorithms and text analysis techniques and created models for predicting unethical behavior based on the interview response texts. The model results are then compared to 1. human judge ratings of the interviewees moral character and 2. human judge predictions of the interviewees tendency to cheat based on reading the same interview responses. Overall, we showed that machine learning models can explain parts of the variance in unethical behavior that were not explained by human judge ratings.