I have a brief question about some of my analyses and would appreciate any clarification. The first analysis requires the use of hurdle gamma for the family function and the second requires the use of bernoulli for the family function. From other help pages, the hu estimates from the hurdle gamma models are logistic regression estimates that indicate the odds of reporting a zero compared to a one (which is flipped from most interpretations). I thought this interpretation might carry over to brm models with bernoulli as the family type; however, it seems, based on other help pages, that this type of model indicates the odds of a one compared to a zero (more common interpretation). Am I right about these different interpretations, or am I completely off base about one of these?
Hey, sorry for not responding sooner. Since your post didn’t have the “brms” tag, I didn’t get an email notifier for it (I have changed the tag now).
You are right with your interpretation. hu and zi parameters are exceptions in the sense that higher predictor values indicate lower response values. This is not ideal in the greater picture, but I think may feel more natural in the context of zero-inflated and hurdle models (and now it’s too late to change that anyway).
1 Like