Metaanalyses are observational studies: how lack of randomization impacts analysis.

Abstract

Although metaanalyses are often syntheses of many experiments, they are themselves observational studies. As such, when performing or reading metaanalyses, we must consider the sources of bias that we usually expect in an observational study. The main two sources of bias for metaanalysis are publication bias and systematic heterogeneity. I consider the nature of both of these sources, methods to detect this bias, and ways to correct for the bias.