ABSTRACT
The central aims of this experiment were to compare observational and interventional learning of a simple causal chain, and to ascertain whether people represent their interventions in accordance with the normative model proposed by Pearl (2000). In the observation condition people treated putative causes as independent, and systematically selected the wrong model. In the intervention condition performance improved, in particular greater sensitivity was shown to the relevant conditional independencies. However, participants’ likelihood judgments approximated the observed frequencies rather than reflecting the appropriate causal model.
