It’s becoming clear that a lot of published research is unreplicable and untrustworthy. How do incorrect findings occur, and how can we predict whether the results of a particular study are likely to be sound?