Reasoning about complex causal systems is central to everyday cognition, and yet our knowledge of such systems is often incoherent, in that individual (or small sets of) causal relationships are not integrated to form more comprehensive mental models. Two factors are hypothesized to have large roles in determining when, why, and how causal knowledge is integrated: (a) the use of correlational information to distinguish among multiple possible integrated models and (b) the use of fragmentary knowledge in reasoning. Two studies examine how adults use correlational information in integration. These studies distinguish between a hypothesis derived from Bayesian network theory and an alternative hypothesis in which learners interprete correlational information in a way that enables them to acquire incomplete or shallow models. Two further studies track the development of these abilities in young children. Finally, it is hypothesized that using fragmentary knowledge for different kinds of reasoning leads to different kinds of integrated models. This hypothesis is tested in two studies. ? ?
