Hello,
I am a new user of brms (coming from MLwiN) and am getting disagreement between the two programs when estimating the same model. I would like to check that I am specifying my model correctly in brms, as it is more complex than those offered in the brms paper and that is the most likely source of the error.
For simplicity sake: I am intending to specify a multiple membership multilevel model, where students are clustered in schools, and they occasionally move between schools. Let’s say that while there are 20 schools in this study, any one student would attend maximum of 3 schools. I would like to include a student-level predictor (“stud”) as well as a school-level predictor (“schl”). “schl1” - “schl3” represent the value of that predictor for the schools attended by each student. The outcome(Y) is binary (0/1). Weights are assumed equal for now.
I currently have the following model:
Y|trials(1) ~ 1 + stud + (1 + mmc(schl_1, schl_2, schl_3) | mm(school1, school2, school3))
However, I have also seen folks on here model multiple membership models with some sort of average cluster-level variable (schl_avg):
Y|trials(1) ~ 1 + stud + schl_avg + (1 + mmc(schl_1, schl_2, schl_3) | mm(school1, school2, school3))
I must admit I’m not certain why that is. I think the first model is only random effects, while the second is random + fixed effects? I would appreciate clarification, and any corrections to my models.
Thank you.