Belief revision, a process in which one revises one's current belief in the light of new information, is an essential component of human abductive reasoning. The order effect, a phenomenon in which the final belief is significantly affected by the temporal order of information presentation, is a robust empirical finding that is not compatible with normative theories such as Bayes' theorem. In this article we explore, both empirically and computationally, how and why the order effect occurs. Both a tactical abductive reasoning task and a learning paradigm (the UECHO model) show that although a recency effect occurs at the beginning of the training, it decreases and disappears as the training continues. We conclude that when little is known about the uncertainty and the dynamics of the environment, the order effect results from one's coherently and dynamically adaptive expectations of the statistical properties of the environment.