6

What is the difference between the classic meta-analysis (that aggregates effect sizes from a sample of studies to a summary effect size), meta-regression analysis and moderator analysis?

As I understood moderator analysis is used to explain heterogeneity in a meta-analysis by regressing a vector containing the effect sizes of each study on specific variables that could explain the heterogeneity (e.g. publication year,...). But what's the difference between moderator-analysis and meta-regression?

Thanks for your help!

franz
  • 79
  • 1
  • 3
  • 1
    Your question could be improved by providing a citation or two pertinent to "meta-regression" and, perhaps to the use of moderator analysis within the meta-regression context. While effect moderation (aka effect modification aka interaction effects) is common enough in applied data analysis, I wonder if you are interested in a very specific kind of application? The "edit" link in the lower left of your question can be used for you to clarify your question along these lines. – Alexis Sep 02 '14 at 19:10

1 Answers1

14

Here are some suggestions for definitions that may help to clarify the terminology:

  • Meta-analysis: A general term to denote the collection of statistical methods and techniques used to aggregate/synthesize and compare the results from several related studies in a systematic manner.
  • Moderator analysis: In the context of a meta-analysis, this refers to using some kind of method in an attempt to find and account for systematic differences in the size of the effect or outcome that is being meta-analyzed.
  • Meta-regression: This is one possible way of conducting a moderator analysis, where we regress the observed effect sizes on one or multiple study characteristics.

There are other ways of conducting a moderator analysis. For example, one could simply subgroup studies based on a categorical moderator (using dummy variables in a meta-regression model is quite similar). Not commonly used, but one could also consider mixture models or clustering techniques.

Wolfgang
  • 15,542
  • 1
  • 47
  • 74