using sum/deviation contrasts in #r (contr.sum()) can help with not centering whiteness in the race/ethnicity coefficients. (unless your point is to explicitly measure disparities against that standard, in which case the default contrasts are useful).
what are other ways to decenter whiteness in #regressions?
@healthstatsdude another way would be to fit/report regression models separately by race/ethnicity
More causal then regression, but I really enjoy the following paper by John Jackson https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8478117/
@PausalZ @mccarthymg yeah i’ve heard of doing this, but have only seen the following paper on it—do you know of other resources (esp r tutorials) out there on this? i feel a little tentative still, but that’s discomfort i can work to overcome!
@PausalZ @healthstatsdude Not specifically on health inequalities, but TJ Mahr’s mixed effects models posts have approaches that might be relevant if you need information about the different groups: https://www.tjmahr.com/plotting-partial-pooling-in-mixed-effects-models/
I was thinking more from the perspective that using random effects would allow you to decentre whiteness, since the fixed effects would no longer be based on a reference group
@mccarthymg @PausalZ that makes sense, thank you!
@PausalZ @healthstatsdude You’re welcome!
If you end up doing the separate regressions like Pausal suggested, and are taking a frequentist approach, the lmList() function from {lme4} makes that easy to do: https://www.rdocumentation.org/packages/lme4/versions/1.1-31/topics/lmList
@PausalZ @mccarthymg that’s cool! i haven’t seen that function before!
@PausalZ wow that paper looks fascinating, thank you!
@PausalZ @healthstatsdude Instead of the no pooling approach with separate models, another option would be the partial pooling approach with race/ethnicity as random effect (if you have enough levels for it).