Proceedings of the Annual Meeting of the Cognitive Science Society

Predictive causal reasoning requires estimating the probability of an effect given a cause, P(e|c). This estimate is context-dependent: a change in the causal structure may render a previous estimate biased. We investigate whether reasoners can discern estimation-relevant from estimation-irrelevant structural changes of causal models. Using the Causal Bayes Net framework, we compare situations where adding a causal link is either estimation-relevant (e.g., introducing a confounder) or irrelevant. In an experiment ( N=204), participants successfully discerned these cases, adjusting estimates in the normative direction only when structural shifts warranted it. However, we also found that few participants provided normative point-estimates. We identified two dominant heuristic strategies: a "normative change direction" strategy, where reasoners adjusted estimates but overemphasized alternative causes, and a persistent "alternative cause neglect" strategy, where participants ignored alternative causes entirely. Our findings suggest that while people recognize whether structural shifts require re-estimation, they struggle with the numeric integration of multiple causal pathways.