Where to look to find out why: Rational information search in causal hypothesis testing

Many of our causal beliefs rest partly on covariation data, that is, the joint frequencies of effects and their hypothesized causes. The focus of this dissertation is on how people should and do search for such data in causal inference, which is modeled here as a problem of hypothesis testing.; Much research on covariation judgment and causal inference has been devoted to the questions of how people use frequency data to judge relations between binary variables and whether their judgments conform to statistical norms. It is here argued that these norms fail to reflect the goals of real-world causal inference. Furthermore, expecting people to conform to them requires making unrealistic assumptions about data availability in natural environments, in which information search is often limited.; Some theorists have successfully used the technique of rational analysis to justify apparently irrational patterns of information search in hypothesis testing which closely parallel judgment biases identified in covariation judgment and causal inference. In particular, it has been proposed that people apply a general-purpose strategy for information search that maximizes the expected informativeness, expressed in terms of Bayesian measures of diagnosticity, of the data that might be revealed. This strategy works well only when the absolute and relative probabilities of the effect and the hypothesized cause meet certain assumptions.; In the present analysis of specifically causal hypothesis testing, I explore two criteria in terms of which people's information search may be evaluated and described. The first is expected test diagnosticity. The second is the epistemic goal of the causal inference, specifically, whether the decision-making context requires a prediction or a diagnosis. Based on this analysis, I (1) hypothesize that in testing causal hypotheses people appropriately adapt their search strategies to changes in cause and effect probabilities, even when default probability assumptions are violated; (2) hypothesize that people likewise adapt their search strategies to whether the decision-making task implicitly demands prediction or diagnosis; (3) show how adaptability in search can be explained by boundedly rational strategies for estimating test diagnosticity that eschew complex Bayesian computations; and (4) test predictions derived from the analysis in three experiments.