How to get Effect Size on an Absolute Scale

I have a friend asking for help modeling something, and I suggested using a hierarchical logistic regression using brms. However, I’m not familiar with brms (I’m more familiar with Turing.jl); how do I get an effect size on an absolute scale? (e.g. if the intervention increases the probability of success from 5% to 6%, this is an effect size of 1%, rather than the change on the log-odds scale which will be closer to 0.2.)

1 Like

Hi @Carlos_Parada
With non-linear link functions there is no one true effect on the outcome scale as the effect is dependent on the entire model. This is a partial answer that I wrote for another question but that should apply here as well:

This is where the conditional_effects() function can be very handy. It shows you the effect you are looking for for your average sample. While it again doesn’t show the complete picture, as the effect looks different for data points further away from the mean, it gives you a quick look at the outcome scale.

I have also written a small introduction for how to interpret such effects here (Page 11, Introduction of result section) maybe that is helpful as well.