I’m trying to compute an analogous version of Cohen’s D from a robust multiple linear regression following this answer here: https://stats.stackexchange.com/questions/171941/cohens-d-from-regression-coefficient.
The answer above suggests dividing the coefficient of interest by the residual standard error. I was wondering how to obtain the RSE in an lm_robust object. I cannot seem to retrieve residuals using the normal operators.
Hi there! If fit is your object created with lm_robust, I think you just want sqrt(fit$res_var).
Would be interested to hear more about the application, if you like. It sounds like something DeclareDesign is setup to handle
Awesome! Thanks Jasper. No application in particular – I have a significant coefficient on my treatment effect, but colleagues wanted to know if that effect was meaningful in some sense.
That makes total sense. I’ve always reported (regression-adjusted) effect sizes in terms of standard deviations of the unadjusted outcome in order to convey meaningfulness (e.g., this is equivalent to a 1/5 SD increase in the outcome). But I could see an argument for using the residual SD – if your regression’s doing well, it’s certainly going to be a bigger number!
This is correct, but if you have weights set, you should double check which scale the reported residuals are on.