For calculating summary effect size in meta analysis do i need to group correlation/beta coefficient between 2 variables which are significant or even non significant ones. Eg I have 20 studies and in 15 studies the Beta coefficient between sy PE-BI(variables) is significant but non significant in 5 studies. While uploading data in software should I only include correlation coefficients of 15 studies and then report all stars.
Hello Stanny Dias As far as I know and working with SPSS software, it is better to enter the effect size of all correlations (significant or non-significant) in the study.
You need to include all 20. Selecting just 15 significant ones will bias the results. Many years ago the it was common practice to set NS effect sizes as zero in the analysis (which is almost the same as what you propose, but worse) and this is now recognized as a deeply flawed approach.
A major problem in meta-analysis is publication bias, so further excluding NS published effects would bias the results of your analysis (overestimating the effect and possibly underestimating variability). It would result in additional bias over and above the publication bias inherent in the literature.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis