We study causal narratives – narratives which describe a (potentially incorrect) causal relationship between variables. In a series of experiments across a range of data-generating processes, we show that externally provided causal narratives influence decisions in ways inconsistent with rational theory. Instead, decisions are generally consistent with a behavioral theory, but there is significant heterogeneity in responses driven by the fact that subjects construct their own causal models of the data-generating process. To examine these homegrown models directly, we ask subjects to observe a dataset and offer advice to future subjects. The resulting homegrown narratives reveal that many subjects mistake correlation for causation, leading both themselves and those who receive their advice to make systematic errors.