Comparison of linear and nonlinear models in explanation of dishonest behavior

Abstract

Dishonest behavior impacts various sectors, including environmental protection, production quality, finance, and taxation. This study compares linear and nonlinear models for explaining behavioral data obtained through a laboratory experiment with economics students at the University of West Bohemia.Participants chose between honest production at a higher cost or dishonest production to save costs. The experiment varied inspection probabilities and introduced punishment for dishonesty or rewards for honesty. Personality traits (MBTI) and risk aversion (Holt-Laury measurement) were also assessed. Both linear and nonlinear (GAM, neural networks) models produced similar results. Increased inspection significantly reduced dishonesty (p < 0.01), while punishment and reward had no significant effect (p > 0.10). Thinking-oriented individuals were more prone to dishonesty (p ≈ 0.05), and higher risk aversion correlated with lower dishonest behavior (p ≈ 0.10). All models achieved a similar power to predict dishonest behavior.