Predicting CEO Misbehavior from Observables: Comparative Evaluation of Two Major Personality Models

Abstract

The primary purpose of this study is to demonstrate how publicly observable pieces of information can be used to build various psychological profiles that can be utilized for the prediction of behavior within a risk analysis framework. In order to evaluate the feasibility of the proposed method, publicly available interview data is processed from a sample of chief executive officers (CEOs) using the IBM Watson Personality Insights service. The hypothesis-that group membership gives rise to a specific selection bias-is investigated by analyzing the IBM Watson-derived personality profiles at the aggregate level. The profiles are represented by two major theories of motivation and personality: the Basic Human Values and the Big Five models. Both theories are evaluated in terms of their utility for predicting adverse behavioral outcomes. The results show that both models are useful for identifying group-level differences between (1) the sample of CEOs and the general population, and (2) between two groups of CEOs, when a history of rule-breaking behavior is considered. The predictive performance evaluation conducted on the current sample shows that the binary logistic regression model built from the Basic Human Values outperforms the Big Five model, and that it provides a practically more useful measurement of individual differences. These results contribute to the development of a risk analysis method within the domain of information security, which addresses human-related risks.