I have fit a cumulative ordinal model using brms. The code for model formulation is below. The issue I have is with conditional_effects().
My understanding is that conditional_effects() by default will plot the mean of the posterior samples of the expected value/mean of the posterior predictive distribution as calculated by posterior_epred(). However if I calculate directly the means using posterior_epred() and summarising they differ from the estimates__ provided in conditional_effects().
I believe these estimates are the probabilities of having an outcome at that particular category and should sum to 1 across all categories for given conditions, they do from posterior_epred() but not from conditional_effects().
This is not the case for a gaussian family mixed effects model for example, where the estimates were the same subject to sampling error. Is there an issue with the estimates being calculated for ordinal models in conditional_effects() or am I misunderstanding what is being calculated here.
I am not able to provide the data to replicate the example as it arises from a clinical trial, I apologise. The snippets of code hopefully provide some insight into how one would recreate this, newdata would need to match the conditions within conditional_effects()
Any comments/help would be greatly appreciated
model_fit <- brm(WHOScore ~ (1 + time | TNO) + CareStatusID_txt + Age + time + TrtID_txt + TrtID_txt:time, data = model_data, family=cumulative("logit"), iter = 5000, file = 'modelfit') conditions <- make_conditions(model_fit , c('CareStatusID_txt', 'time')) conditional_effects(model_fit, effects = TrtID_txt, conditions, categorical = TRUE) posterior_epred(model_fit, newdata) %>% data.frame() %>% summarise_all(mean)