When summarizing studies in a meta-analysis, the methodological quality of the studies can vary widely. This variation is often described by assigning quality scores to studies. This paper presents a probability model for the effect of quality on a summary effect measure. The model motivates criteria used to assess quality, and can be used to incorporate quality into the calculation of a summary effect. We derive a number of summarization methods and compare them for simulated data and an actual set of studies.
Copyright 1999 John Wiley & Sons, Ltd.