Reasoning about data

We report an experiment in which people observe causal data that follow either conjunctive or disjunctive decomposition functions. Among reasoners who generalize their empirical function (conjunction or disjunction) to novel stimuli in the same domain, they nonetheless apply causal invariance to the 'whole cause' level. This appropriate 'switch' between levels of representation is consistent with having analytic knowledge of causal invariance that guides causal learning.

Given the same prior knowledge and training data, people make different intuitive causal judgments according to their perception of the outcome variable type as either continuous or binary. Our causal invariance hypothesis explains why this reasoning is adaptive to our representation-dependent mind.