Rational causal induction from time

A longstanding focus in the causal learning literature has been on inferring causal causal relations from contingency data. This format abstracts away from time by collecting independent instances, aggregating over independent entities or across regular demarcated trials. In contrast, individual causal learners encounter events in their daily lives that unfold in a continuous temporal flow without distinct experimental trials or demarcation. Consequently, the process of learning causal relationships in naturalistic environments remains comparatively less understood. In this paper, we develop a rational framework that foregrounds the role of time in causal learning. We work within the Bayesian rational analysis tradition, while departing from past analyses of causal inference by linking causal influence with dependence between events in continuous time via the Poisson-Gamma distribution family, rather than co-incidence of variable states across independent trials (i.e. contingency). We show that our account parsimoniously explains the human preference for causal explanations that involve shorter, more reliable and more predictable causal influences. Furthermore, we show that it provides a unified explanation for human judgments across seven experimental datasets from the causal learning literature.