I have the summary results of a linear regression model (including 229 independent variables that categorised it in 7 moderator variables. This is including t-statistic and standard error for each study for independent variables. I want to determine effect size every 7 moderator variables through CMA ver. 2. Is this possible? if yes how? Please, guide me.
Hi Mohammad,
That's not really possible - CMA and most other programs only work with correlation coefficients. You can try Meta-Essentials, which uses partial correlations or semi-partial correlations based on data from regression models to compute a meta-analysis. (disclaimer: I am a co-developer of the software, but its free Excel-based). You can also do such analysis easily in R, for example with the metafor package.
Don't confuse this with the idea of meta-regression, which is a regression-based technique to analyze the influence of moderators on the main effect in the meta-analysis.
Best,
Robert Suurmond
Hi Robert,
Thanks a lot for your help.
The phrase of moderator variables maybe not correct here. It would be better explaining that I really do not have exact definition for moderator variables. I categorized many independent variables that collected from several studies (regression studies) such as the total volume of sale, sale volume of species, diameter tree, number of tree in hectare. I named them as sale size variable (as a moderate variable) and I similarly repeated for other variables. Then, I calculated the effect size of each variable and parameter of fixed effects and random effect model under excel sheet without using Professional software such as CMA, R. In details, I converted t-statistics of each independent variables to r effect size. Indeed, I used this paper.
"Meta-analyses and Forest plots using a microsoft excel spreadsheet: step-by-step guide focusing on descriptive data analysis".
https://bmcresnotes.biomedcentral.com/track/pdf/10.1186/1756-0500-5-52
However, I find that the meta-analysis process is thus not possible for regression studies that reporting variable regression coefficients and t-statistics. It is true?
Best wishes
Yes, i'd say that's true generally speaking. You can use that data to convert it into an approximate correlation coefficient (partial or semi-partial), which you in fact have done if you 'converted t-statistics into r effect size'. You then conduct the meta-analysis on partial correlation coefficients, which are similarly distributed but conceptually distinct from 'zero-order' correlation coefficients.
We build on the idea in the paper you cite in the Meta-Essentials software.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis