Computational principles underlying people's behavior explanations

There are often multiple explanations for someone’s behavior, but people generally find some behavior explanations more satisfying than others. We hypothesized that people prefer behavior explanations that are simple and rational. We present a computational account of behavior explanation that captures these two principles. Our computational account is based on decision networks. Decision networks allow us to formally capture what it means for an explanation to be simple and rational. We tested our account by asking people to rate how satisfying several behavior explanations were (Experiment 1) or to generate their own explanations (Experiment 2). We found that people’s responses were well predicted by our account.