Yuri Vorokhta

Hi Syed,

I disagree with the statement that you cannot use different response variables in the analysis. In fact, you can combine all the data including multiple response variables in the same analysis and also consider the variable itself as a moderator. Also, depending on the data available you can use this approach to determine which variable is more sensitive in order to detect differences. However, you might consider if both variables are linearly related: If they are (e.g. meters and inches; talking about distances), for sure you can combine them; however, if they are not (e.g. X and logX) it could be convenient not to combine them because heterogeneity might increase sharply.

Picking up the variable that shows the bigger differences could not necessarily mean more sensitivity for the meta-analysis: not only the difference but also the confidence interval of the effect size will determine the sensitivity, and it depends on the variability: Variable A could show bigger differences than variable B, but also higher variability. Rather than choosing one, it would be better to run an independent meta-analysis for each of the variables.

I agree with Dr. Diego A. Martinez " it would be better to run an independent meta-analysis for each of the variables."

*Falavigna et al. (2012) demonstrated that the r coefficient between HADS-D and BDI was 0.714 (p

I would suggest including all variables (despite different measures) while presenting them in a subgroup analysis. If you'd like to pool the overall ES, not just ES for each subgroup, different measures will yield high heterogeneity, and you should report them and take care when interpreting the results. Also, better stick to SMD, and not change to MD.

If two measures predominate, consider performing a multivariate/multilevel meta-analysis. This would allow you to incorporate both scores as well as investigate their agreement and explore any potential reasons for discordance.

The metaSEM package in R does this well.