The last two decades in Cognitive Science have seen the productive application of Causal Graphical Models (Pearl, 2000; Spirtes, Glymour & Scheines, 1993) to theories of human causal reasoning. Causal Graphical Models currently employ a minimality constraint: strictly prefer representations that posit fewer causal relations. This thesis discusses three major problems with minimality, that each limit Causal Graphical Models as a theory of human causal reasoning: Minimality minimizes away causal mechanisms, minimality is at odds with evidence that people are determinists about causal relations, and minimality does not account for the robust nonindependence effects found in human causal reasoning. The thesis presents an alternative to minimality, that uses a generative edge replacement rule to define a prior distribution over causal structures. It is shown that Causal Graphical Models, as amended by edge replacement , better account for mechanism, determinism, and nonindependence, than minimality. Edge replacement also makes several novel predictions, which are tested experimentally in later chapters. Because edge replacement and minimality make the most divergent predictions about basic representations, present early in learning, experiments involved preschool-aged children. Six experiments provide evidence that edge replacement may be a better model of basic causal representations than minimality.
Buchanan, David W.,
"Edge Replacement as a Model of Causal Reasoning"
(2011).
Cognitive Sciences Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://doi.org/10.7301/Z04M92TH
