Stouffer Method for Meta-Analysis

From Genome Analysis Wiki
Jump to: navigation, search

This is a method for meta-analysis of data across studies that requires only a p-value, sample size and estimated direction of effect for each study. We rediscovered it when implementing the METAL Program. The method is extremely convenient because it can easily execute meta-analyses even when different analytical approaches were used in each individual study.

Advantages of the Approach

The Stouffer method results in only a small loss of efficienty compared to methods that use estimated effect sizes and standard errors as input, and is more robust to than these methods when there are differences in analytical approach between studies (for example, changes in measurement scale can render an analysis based on estimated effect sizes and standard errors useless, but should not impact a p-value based analysis). It is also extremely useful in situations where effect size estimates cannot be obtained for some models (such as when a logistic regression analysis is attempted in case control data but a rare risk factor is observed only in cases or only controls).

Original Reference

Stouffer, S. A., Suchman, E. A , DeVinney, L.C., Star, S.A., Williams, R.M. Jr (1949). Adjustment During Army Life. Princeton, NJ, Princeton University Press.