I² and Q (heterogeneity)

Q tests whether the studies differ from each other more than chance within the studies would suggest. I² gives, in percent, the part of the total variation in effects that stems from real differences between the studies rather than chance. Rule of thumb based on Higgins and colleagues (2003), who name 25, 50 and 75% as markers: below 25% low, from 25% moderate, from 50% substantial, from 75% high.

Example: I² = 95%, Q(8) = 123.7, p < .001: almost all of the variation is due to real differences between the studies (length of stay, Normand 1999).

Caution: With few studies I² is very imprecise; that is why the interval is shown. A small Q with few studies does not prove that the studies are alike. I² is a proportion, not a measure of how large the differences are: with very precise studies I² becomes large even when the effects lie close together. How far the true effects spread is shown by τ and the prediction interval.