search

JHart96 / bram

A compact Bayesian workflow for brms.
2 stars 0 forks source link

Add option to perform Average Treatment Effect #4

Closed DanFranksYork closed 1 year ago

DanFranksYork commented 1 year ago

ATE (causal inference literature). Rather than just contrasts in a 2-category variable, use the new_data option in the posterior to get the ATE (e.g. between yes and no). Contrasts would take the estimated distribution for the actual data for yes and also for no and then take the difference. ATE takes the difference in distributions between the actual data with all of that covariate set to yes, and the data with all covariates set to no.

JHart96 commented 1 year ago

Am I right in thinking this is the default behaviour of marginaleffects()? If so, this is already what we use.

DanFranksYork commented 1 year ago

I thought that average treatment effects were different to marginal effects. Plus I didn't know you could do contrasts with marginaleffects?