Model selection between the fixed-effects model and the random-effects model in meta-analysis

Ke Yang, Hiu Yee KWAN, Zhiling YU, Tiejun TONG*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

The common-effect model and the random-effects model are the two most popular models for meta-analysis in the literature. To choose a proper model between them, the Q statistic and the I2 statistic are commonly used as the criteria. Recently, it is recognized that the fixed-effects model is also essential for meta-analysis, especially when the number of studies is small. With this new model, the existing methods are no longer sufficient for model selection in metaanalysis. In view of the demand, we propose a novel method for model selection between the fixed-effects model and the random-effects model. Specifically, we apply the Akaike information criterion (AIC) to both models and then select the model with a smaller AIC value. A real data example is also presented to illustrate how the new method can be applied. We further propose the generalized AIC (GAIC) to reduce the large variation in the AIC value, and demonstrate its superiority through real data analysis and simulation studies. To the best of our knowledge, this is the first work in meta-analysis for model selection between the fixed-effects model and the random-effects model, and we expect that our new criterion has the potential to be widely applied in meta-analysis and evidence-based medicine.

Original languageEnglish
Pages (from-to)501-510
Number of pages10
JournalStatistics and its Interface
Volume13
Issue number4
DOIs
Publication statusPublished - 2020

Scopus Subject Areas

  • Statistics and Probability
  • Applied Mathematics

User-Defined Keywords

  • Akaike information criterion (AIC)
  • Common-effect model
  • Fixed-effects model
  • Metaanalysis
  • Model selection
  • Random-effects model

Fingerprint

Dive into the research topics of 'Model selection between the fixed-effects model and the random-effects model in meta-analysis'. Together they form a unique fingerprint.

Cite this