Many biases have been observed in probabilistic reasoning, hindering the ability to follow normative rules in decision-making contexts involving uncertainty. One systematic error people make is to neglect base rates in situations where prior beliefs in a hypothesis should be taken into account when new evidence is obtained, incomplete explanations for the phenomenon have impeded the development of effective debiasing procedures or tools to support decision making in this area. In this research, we show that the main reason behind these judgment errors is the causal representation induced by the problem context. In two experiments we demonstrate that people often possess the appropriate decision rules but are unable to apply them correctly because they have an ineffective causal mental representation. We also show how this mental representation may be modified when a graph is used instead of a problem narrative. This new understanding should contribute to the design of better decision aids to overcome this bias.