A fundamental goal of cognition is to reason about the causal properties of the physical and social worlds. However, as Hume (2004/1748) observed, knowledge of causality is puzzling because although events are directly observable, causal connections between them are not. Hume's puzzle has both philosophical and psychological aspects. The philosophical puzzle is how causal knowledge can be justified – that is, when should people infer causality? Hume argued that this problem is simply unsolvable – causality can never justifiably be inferred. But this leaves the psychological puzzle. Whether defensibly or not, people routinely do infer causality from experience; the puzzle is to understand what principles underlie these causal inferences.

Hume believed that this psychological problem was solvable: He suggested, in essence, that people infer causality from constant association or, in statistical terms, correlation. However, inferring causality from correlation is fraught with peril. One particularly serious difficulty concerns the theme of this book: sampling. Biased samples can induce numerous correlations that are spurious from a causal point of view; and, conversely, can lead to no correlation, or anticorrelation, where there is a causal link between events.

From Hume's skeptical perspective, this difficulty might not seem important. Indeed, if causal knowledge is unjustified and unjustifiable, there is really no question of whether causality is inferred correctly, or incorrectly: The associations between events are all there is. From a modern perspective, however, such skepticism seems untenable.