Heterogeneity
Also known as: Between-study heterogeneity
When studies of the same question genuinely disagree in their results.
What it means
Heterogeneity in a meta-analysis is variation in true effects across studies that exceeds what sampling error alone would produce, signaling that the studies are estimating somewhat different things. It can stem from differences in populations, doses, settings, outcome measures, or design quality. Statistics such as Cochran's Q and the I-squared index quantify how much of the observed variation is real rather than random, guiding whether pooling is even appropriate. Substantial heterogeneity shifts the goal from a single summary effect toward explaining the variation, often through subgroup analyses or meta-regression.
Examples
A drug helps strongly in younger patients but barely in older ones, so the trials' effects scatter far more than chance predicts.
Trials of a classroom reading programme scatter from large gains to none; averaging them into one number hides that the gains come only from schools with trained teaching assistants.
Studies of remote work report effects from clearly positive to clearly negative — a signal to ask what differs between these workplaces, not to pool them into a single meaningless average.
First described in Meta-analysis methodology; Higgins & Thompson I-squared (2002).