Three worked guides to questions hidden behind a headline: which events changed, what a non-inferiority margin allows, and whether a pooled result deserves confidence.
3 articles · About 19 minutes in total · Read at your own pace
A combined outcome can tell a useful story while leaving its most important chapter unresolved. Work through a first-event example and learn what to look for in the component results.
Read the guide · 6 min“Not unacceptably worse” is a specific statistical question. Four invented confidence intervals show why it is different from equal, better, or simply not statistically different.
Read the guide · 6 minA pooled estimate is the end of a chain of decisions. Trace the question, search, independent studies, bias and uncertainty before deciding what the headline deserves.
Read the guide · 7 min