Causal judgments are stubborn. If people learn about twocorrelated variables B and C, and judge that B causes C, theytypically stick to that judgment even when contradictoryevidence comes to light. One form of contradictory evidenceis that a third variable A causes both B and C, explaining thecorrelation. This paper extends prior work showing thatsimply presenting statistical evidence that A is the commoncause of both B and C does not lead to belief change about B.However, if first subjects learn to categorize phenomena bytheir underlying causal relationships (i.e., as exemplars of acommon cause category), then they can transfer their categoryknowledge to properly interpret the evidence. They recognizethat A is the common cause of B and C and revise their beliefabout B. These results suggest that teaching abstract causalcategories is a promising strategy to help revise false beliefs.

The ability to form relational categories for objects that share few features in common is a hallmark of humancognition. However until recently, neuroimaging research largely focused solely on how people acquire categories defined byfeatures. In the current electroencephalography (EEG) study, we examine how relational and feature-based category learningcompare in well-matched learning tasks. Building on a previous functional magnetic resonance imaging study by our labo-ratory, we capitalise on the rich temporal information offered by EEG. Focusing on the neural dynamics of how people learncategory memberships over individual trials in an experimental task, we investigate how these single trial dynamics modulatecomputational estimates from decision-making modelling frameworks. Specifically, by sorting participants’ individual trialsby their position in the experimental sequence we observe striking relationships between EEG dynamics (e.g., frontal thetaoscillations and P300 component) and feature-based and relational categorisation behaviour.