0

I'm running a one-way Bayesian GLM (in brms in R) between three groups (Class), using a heteroskedasticity adjustment:

brm(bf(Var ~ Class, sigma ~ 0 + Class), data = df)

When I plot the condition effects it's very clear that there are group differences, but I'm not sure how to "test" this. When I look at the output, it's also clear that the CIs don't overlap. I almost feel like maybe I don't need a formal test because the CIs don't overlap, but not sure if that applies.

I'm guessing this is really simple and I'm just a noob (true). Any recommendations conceptually or in terms of packages?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
DreadPiratePotato
  • 1
  • @DemetriPananos, yes, it does. Until I got to the part with the second Gelman paper and now I'm unsure again. What I took from this was that I need to add a random-effects term for individuals beyond just the groups, maybe think about using pd/ROPE/HDI without reifying the need for them even if I don't, and interpret carefully. Am I on the right track? – DreadPiratePotato Mar 17 '21 at 01:52
  • Yes, I think you're on the right track. – Demetri Pananos Mar 17 '21 at 02:05

0 Answers0