I carried out a Dersimonian-Laird random-effects meta-analysis using Cochrane (RevMan) software to create a summary effect estimate of weighted mean difference. I notice that the distribution of study effects (n=20) is negatively skewed. Do you recommend using the Permutation Method for non-parametric RE meta-analysis (e.g. Stata)? How skewed do the estimates have to be to require this form of RE meta-analysis. Thanks.
Thanks. The number of homogeneous studies with moderating characteristics precludes moderator analysis. We are being very clear about sources of heterogeneity and contextual differences in how estimates were derived but really wanted an 'average' summary effect for groups of studies (given above caveats). I was happy with DL approach but was concerned about assumption of normal distribution of effects for this RE approach due to negative skew of our studies. I haven't really found a good example but in this situation should non parametric RE MA be used (eg permutation method)?
Hi
What was your effect measure? OR, Risk ratio, rate ration, mean diff and ...?If your effect measure is an relative measure such as OR and RR, you can transform it in log scale and then pooled it. I propose CMA software for your condition.
The effect estimates are weighted mean differences (relates to an before and after measurements for an intervention reducing exposure to pollutants - particulate matter and CO).
So, i think skewness is not important in your research. Normality is not an assumption for each type of meta-analysis. CMA software can help you for better assessment of the data.
We have a solution for this we call the IVhet model of meta-analysis that makes no assumption regarding "distribution of true effects". This can be accessed through MetaXL - see link below
www.epigear.com
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis