I am trying to compute the overall posterior differences between groups (A - B = 0) from a
brmsfit object using the
hypothesis function but I am unsure I am performing it correctly with the interaction term so I would love a bit of advice.
The structure of my model is as follows:
prop_mod=brm(prop_links~0+PolOrder*clim+(1|Reference/Network), family=Beta(link = "logit"), inits=0, prior=propprior, data=g.sub,cores=4)
Prop_links= proportion of network links
PolOrder = Pollinator insect order (6 groups: Bee, Coleoptera, Diptera, Hymenoptera, Lepidoptera and Syrphidae)
clim = Climate zone (5 groups: A, B, C, D, E)
As such, I am interested in differences between
PolOrder group within each
climate A (the reference level of
clim) I have performed the following when considering bees (the reference level of
PolOrder) with Syrphidae:
hypothesis(prop_mod,alpha=0.05, c("PolOrderBee - PolOrderSyrphidae = 0")
And within a different
climate I have done as follows, adding in the interaction:
hypothesis(prop_mod,alpha=0.05, c("(PolOrderBee + climB) - (PolOrderSyrphidae + climB + PolOrderSyrphidae:climB) = 0"
These appear to give me sensible results but id love some clarification as similar posts I have read aren’t exactly like my case.
Also, am I missing something in the
hypothesis output or is it possible to get a numeric value how far the expected value under the hypothesis lies outside the 95%-CI as well the *
Thanks in advance,
Please also provide the following information in addition to your question:
- Operating System: MacOS High Sierra
- brms Version: 2.5.0