meta-analysis

A meta-analysis is, in the Cochrane Handbook's words, the statistical combination of results from two or more separate studies. It normally sits inside a systematic review, which is the wider job of searching for every relevant study, deciding which of them qualify, and assessing how well each was conducted.

Pooling has two attractions. Individual trials often finish without enough participants to settle a question, and combining them narrows the confidence interval around the estimate. Pooling also makes disagreement visible: where studies point in different directions, or differ by more than chance would explain, the analysis reports that as heterogeneity rather than burying it inside an average.

A network meta-analysis is a variant that links studies sharing a common comparator, so it can estimate how two treatments compare even where few trials tested them against each other directly. That reach is bought with inference: part of the answer is calculated rather than observed.

A pooled number is not automatically better than the studies underneath it. It inherits their weaknesses, so a meta-analysis of small, poorly conducted trials yields a precise looking estimate of a distorted effect. That is why reviews report the certainty of the evidence beside the number, and why an analysis restricted to the best conducted trials is often more informative than the headline result.