Hello everyone,
I am conducting a meta-analysis and there are two major outliers in the results that seem to be confounding the results of the publication bias analysis.
We ran Egger's test and the trim and fill procedure for both all studies, and with outliers removed. When outliers are included, Egger's test is sig. (intercept: 7.43, p < .01) and the trim and fill procedure adds 4 studies to account for bias - however due to the outliers, two of the effect sizes that are added are large, negative effects (g = -3 and -6), and there is significant heterogeneity remaining (95.7%)
When the 2 outliers are removed, Egger's test is non-sig and the trim and fill does not add any studies. Heterogeneity is moderate (50%).
If anyone has any recommendations on how to deal with outliers in this case, your input would be much appreciated! Is it best-practice in this case to report the results both with and without outliers to be transparent?
Ivan Lvov Hi, what we decided to do was to report both untrimmed (with outliers) and trimmed (no outliers) results in our publication, with an emphasis on the trimmed results. We also discussed the outliers in the discussion section. You can see our results here if helpful:
Article Dialectical Behaviour Therapy Skills Training Groups for Com...
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis