3

The question is relatively simple although it may won't be possible to answer at all. In meta-analysis using random-effect model, we can choose between several different methods for estimating tau (e.g. DerSimonian-Laird, Maximum Likelihood, Restricted Maximum Likelihood etc.).

How can one choose between these different methods? Is that anyone recommended over other? Do the choice depends from the nature of data or outcome (i.e. different choices should be made for meta-analysis of Odds Ratios vs. meta-analysis of prevalence/incidence rates etc.)?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
user89547235
  • 476
  • 1
  • 3
  • 11
  • This article might be a starting point https://journals.sagepub.com/doi/10.3102/10769986030003261 but it is fifteen years old so there may well be later work on the topic. – mdewey Apr 01 '21 at 16:25
  • 1
    These articles are more recent overviews of the available estimators: [https://onlinelibrary.wiley.com/doi/10.1002/jrsm.1164](https://onlinelibrary.wiley.com/doi/10.1002/jrsm.1164), [https://onlinelibrary.wiley.com/doi/abs/10.1002/jrsm.1198](https://onlinelibrary.wiley.com/doi/abs/10.1002/jrsm.1198), and [https://onlinelibrary.wiley.com/doi/abs/10.1002/jrsm.1316](https://onlinelibrary.wiley.com/doi/abs/10.1002/jrsm.1316). Restricted maximum likelihood and the Paule-Mandel estimator are recommended nowadays. – User33 Apr 06 '21 at 06:43

0 Answers0