I am running a simple model in brms with default priors using two multi-level factors (community type, and a quantile grouping).
rate_temp_comm <-brms::brm(rate ~ Community + quantilegroup + (1|Site),
data = rates, family = gaussian, chains = 3,
iter = 4000, warmup = 1000)
My summary makes sense, and the differences between the groupings align with my understanding of the dataset:
Effects:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS
Intercept 0.01402 0.00467 0.00425 0.02336 1.00116 1707 1180
CommunityMix -0.00852 0.00218 -0.01289 -0.00422 1.00015 7491 4866
CommunityShrub -0.00720 0.00233 -0.01206 -0.00276 1.00019 7188 5095
quantilegroup2 -0.00027 0.00256 -0.00534 0.00474 1.00016 8475 6437
quantilegroup3 0.00178 0.00249 -0.00317 0.00666 1.00117 8613 6694
quantilegroup4 -0.00122 0.00240 -0.00611 0.00347 1.00049 7409 6101
So my questions are:
- When reporting the output of this model, I would ideally like to be able to report both the estimate and the CI for the whole factor. I can see here breakdowns of the levels within these factors and that's helpful but doesn't ease interpretation (when I want to be able to state whether Community type drives or does not drive differences in rate). Whether or not the levels overlap zero or not is interesting but I'd like to go further. Is there a brms version of an F statistic? Or is this a case for leave-one-out analysis?
- The intercept uses CommunityGrass and quantilegroup1 as the 'baseline' condition, if I'm reading this right. And instead of looking at the CI of the intercept, I should be more interested in the relative differences in intercept and quantilegroup2 and CommunityMix, respectively, correct? Just want to make sure I'm understanding this correctly! Any advice on interpretation (or additional analyses to parse this out) would be greatly appreciated. Thank you!
