Use of External Representations in Reasoning about Causality

This research investigated if diagrams aid in deductive reasoning with formal causal models. Four studies were conducted exploring participants' ability to discover causal paths, identify causes and effects, and create alternative explanations for variable relationships. In Study 1, abstract variables of the causal model were compared to contextually grounded variables and causal models presented as text or diagrams were compared. Participants given abstract diagrams did better in most tasks than participants in the other conditions, who all did similarly. Studies 2 and 3 compared causal models expressed in text to diagrammed causal models, and compared models using arrows to models using words when connecting variables. Participants who had arrowheads replaced with words made more errors than participants in other diagram conditions. Diagrammed causal models led to better performance than did other conditions, and there was no difference between different text models. Studies 4 and 5 tested the hypothesis that predictive reasoning (from cause to effect) is easier than diagnostic reasoning (from effect to cause). The two studies did not find any such effect.