Imagine I fit an ordinal model to the responses of one group of subjects (Group A), given on a 7-point scale, in which their responses are predicted by a continuous Predictor:
m.groupA <- brm(Response ~ Predictor, data = ratings.groupA,
family = cumulative(link="probit"))
In this model, the positions of 6 thresholds would be estimated ( Intercept[1]
, Intercept[2]
, etc.) plus a slope for Predictor
.
A similar model is then fitted to the responses of another group of subjects (Group B), and a different set of (6+1) estimates is obtained.
Imagine further that the slope for Predictor
, as well as the distances between threshold positions look quite different in the two Groups.
I would then like to fit the following model to compare Predictor
in the two Groups:
m.interaction <- brm(Response ~ Predictor * Group, data = ratings.combined,
family = cumulative(link="probit"))
However, the distances between threshold positions are now estimated for the combined dataset (so far as I understand it), and I can no longer capture that the two sets of thresholds are quite differently placed in the two groups.
What would be the best way to account for such differences?
(I tried the model Response ~ 0 + Group
, hoping to get 12 intercepts - 6 thresholds times 2 groups, but this produces an error)