- The selection of a model must be based solely on the question of which model fits the distribution of effect sizes, and takes account of the relevant source(s) of error. When studies are gathered from the published literature, the random effects model is generally a more plausible match. (Borenstein et. al. 2007)
- Heterogeneity among the studies was assessed by calculating the Cochran Q and the I2 statistic. A Q statistic with a value of p < .1 or an I2 statistic > 50% was considered to indicate significant heterogeneity between studies. When significant heterogeneity was present, a random-effects model of analysis was used; otherwise a fixed-effects model of analysis was used.
The debate question is: what method should be used?
Hi Jamil,
The second paragraph is one of the common mistakes in conducting a meta-analysis. It has been recommended that researchers should avoid selecting fixed vs. random effects model based on the result of heterogeneity. Actually, In most cases using the random effects model is the best choice. However, the results of random effects analysis and fixed effect analysis would be almost identical if you had few studies included in the meta-analysis.
Fixed-Effect Versus Random-Effects Models (Introduction to Meta-Analysis By Michael Borenstein, Larry V. Hedges, Julian P.)
https://www.meta-analysis.com/downloads/Intro_Models_optim.pdf
Thank you Mahdi Rostami Haji Abadi i for the explanation.
I am convinced of what you are saying, which Borenstein explained in his articles, book, webinars, and workshops.
Most of the meta-analysis that i read are selecting fixed vs. random effects model based on the result of heterogeneity. That what made me open this debate, Specially that i have a case in the meta analysis that I am working on where the p-value i= 0.154 in random effect and p value =0.002 in fixed effect (Q=30.73, I^2=28.40, P=0.101)
So choosing the right effect is important here.
Meta-Analysis | Common mistakes and how to avoid them | Fixed effect vs. random effects
https://www.youtube.com/watch?v=3_sbitgMeUg
thank you Mahdi Rostami Haji Abadi , I have already watched this video
See discussion below:
https://www.researchgate.net/post/Comment_on_Serghiou_S_Goodman_SN_Random-Effects_Meta-analysis_Summarizing_Evidence_With_Caveats_JAMA_2019_3213301-302
Hi Jamil,
Very often, and wrongfully so, the model choice in a meta-analysis is determined solely by the level of statistical heterogeneity. In my opinion, this should not be the case.
The research question and the probability of having a common effect across all studies should also be considered. For instance, in clinical studies with diverse patient groups the presence of clinical heterogeneity should prioritize a random effects model.
I hope you find this useful.
Regards,
I disagree. A random effects model probably is the wrong approach in 2019 and should be replaced by the iVhet model
Thank you for this information Suhail A. Doi.
Jamil Mourad, perhaps also of significance in a meta-analysis is the decision of whether or not there is any 'common effect' to estimate in the first place. In the presence of significant qualitative effect modification across studies, as opposed to quantitative effect modification, I would be skeptical on the utility of any pooled estimate.
Agree with that but to be more precise, the biological plausibility of an underlying common effect is a prerequisite for meta-analysis so if there is the possibility that "true effects "vary across studies, then meta-analysis should not be attempted.
Article Advances in the Meta-analysis of heterogeneous clinical tria...
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis