Bayes or Pascal? The computations underlying motivated reasoning

The construct of motivated reasoning has inspired an influential body of research. However, most theories of this construct are expressed in a verbal form. This is somewhat limited in light of contemporary research in cognitive science that emphasizes the insight afforded by employing computational modeling. To address this, the paper introduces a computational model of motivated reasoning. The model builds on previous accounts of belief formation based on Bayesian inference by adding computations concerning value or utility. The result is an interpretation of motivated reasoning as being akin to a process reflecting an unconscious Bayesian decision, in a way that is reminiscent of the famous Pascal’s wager. This framework is broadly consistent with empirical evidence, especially about the effect of loss function asymmetries on probability judgments, about the confirmation bias, and about the backfire effect. Moreover, it is compatible with evolutionary explanations of motivated reasoning that interpret this phenomenon as ensuing from self-deception. The model helps understanding the computational principles behind the concept of motivated reasoning. Moreover, it facilitates the comparison between perspectives that downplay motivated reasoning and theories that emphasize its role. This may inform empirical research aimed at establishing the real contribution of motivated reasoning during belief formation.