Chi-square test, Tau or I-squared? I guess I shouldn't need to use them all, but for meta-analysis, with relatively few studies (less than five per set) which would be most important?
The best way to interpret heterogeneity in meta-analysis is to compare tau squared to its empiric distribution. The distribution of tau squared for all Cochrane meta-analyses has been published for continuous outcomes by Rhodes et al. (https://www.jclinepi.com/article/S0895-4356(14)00348-5/fulltext) and for binary outcomes by Turner et al. (Article Predictive distributions for between-study heterogeneity and...
). You can look at the distribution for the relevant outcome to your analysis, based on the type of outcome. The authors further categorised their results by the exact effect measure, as less objective effect measures will be expected to have greater inconsistency. Compare your distribution for tau squared to the distributions here and decide whether that seems high or not high by comparison. For example, you may decide that heterogeneity is high if it is in the fourth quartile by comparison.Contrary to popular belief, I squared does not quantify heterogeneity and can be particularly misleading when there is a small number of studies as you have here. Tau2 represents the absolute between-study variance and is a better/truer index of heterogeneity.
Tau measures the absolute heterogeneity, while I-squared measures the relative heterogeneity with respect to the total variability. Both are useful.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis