Modeling Causal Generalization with Particle Filters

Psychological experiments have shown that human performance on traditional causal reasoning experiments can be greatly influenced by different pretraining and postraining conditions. In this paper we present a Bayesian theory of sequential learning that captures observed experimental results [1] . We implement our theory using the particle filter algorithm, and show that model selection and model averaging are able to capture the respective effects of preand posttraining. In addition, we model the highlighting effect observed in [5] using a particle filter algorithm as an approximation to exact statistical inference, in accord with the limited computational capacity of human cognition. We find that the inferential approximation based on particle filters predicts the highlighting effect.