946303_en-US

liliconcom

Cohen's d Effect Size from Mixed Model Output

Hello,

I am wondering how I can calculate a Cohen's d effect size from my mixed model output (attached), especially in the case of an interaction effect? I have seen some online resources saying to use the least squares mean difference divided by the model root mean square error in order to do so for main effects. However, when doing this, I sometimes get an effect size that appears inflated based on the data. I am using JMP Student Edition 19.0.3.

Any advice would be greatly appreciated, thank you!

- Ava

Basic Data Analysis and ModelingData Exploration and VisualizationRe: Cohen's d Effect Size from Mixed Model Output

Hi @AvaSchwartz7 : I'm no expert in Cohen's d, but there is a thoughtful discussion here.

https://stats.stackexchange.com/questions/179098/cohens-d-for-2x2-anova-interaction/179102#179102

And, for a Mixed Model, another issue is what to use as your estimate of sigma.  Rather than RMSE (as with a fixed effects model), I'd think, in general, you'd want to use the Total SD (square root of the total variance). 

And, FWIW, why not use confidence intervals instead?  Cohen's d, in my view, is too simplistic. Others may disagree...

 

Edit: And, generally speaking, Cohen's d (and confidence intervals) for main effects would not be appropriate when interactions are present (yes, it can be calulated but it could be very misleading).