Dealing with multiple tests is often described as a “burden”. This likely originates from the fact that controlling family-wise error rate requires more stringent thresholds as the number of tests increases. However, modern analyses prefer to control the FDR33, which does not depend on the number of tests34. Consequently, the term “burden” is inaccurate and unhelpful. We believe that the results of many tests in many conditions should instead be viewed as an opportunity—an opportunity to learn about relationships among underlying effects, and make data-driven decisions to improve inferences. This will inevitably, it seems, involve modelling assumptions, and the challenge is to design flexible models that work well in a range of settings. The methods presented here represent a major step towards this goal.
