Making decisions often requires quick and intuitive assessments of the likelihood of unobserved events. People make such judgments with little effort, quickly and systematically. The work reported in this dissertation evaluates the hypothesis that people's facility for intuitive likelihood assessment is based on their ability to build and evaluate causal models of the scenarios they are reasoning about (the 'causal model conjecture'). The work contrasts reasoning about two fundamental inferential forms: Reasoning forward from causes to effects ('prediction') and reasoning backward from effects to causes ('diagnosis'). The main findings are as follows: Both predictive and diagnostic reasoning display sensitivity to many of the appropriate causal variables in approximately the right way, providing support for the causal model conjecture. However they also diverge in a variety of empirical phenomena; when making predictions, people tend to be 'myopic' failing to think about relevant alternative causes. When making diagnoses they think more broadly, considering alternative causes and adjusting their judgments accordingly. Diagnosis is also slower, more difficult, and develops later than prediction. To account for the results I outline a causal model theory constrained by non-normative processing principles. The broad conclusion of the work is that accurate descriptive models of intuitive likelihood assessment will have to accomplish two things: (a) account for people's sensitivity to the normative principles that distinguish prediction and diagnosis and (b) model the cognitive processes that support these inferences to account for errors of judgment.
Fernbach, Philip M.,
"Causal Models in Prediction and Diagnosis"
(2010).
Cognitive Sciences Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://doi.org/10.7301/Z0WH2N72