Brms multi-membership model with ordinal outcome and multiple imputation


I am new to stan, brms, but have some previous JAGS/BUGS knowledge.

We have been looking into multiple-membership models with brms and have seen that it is possible to use this framework for ordinal outcomes. We are working with some education data which has school level clustering and two time points, however, the individuals change school between time point 1 and 2 (uk primary level at t1, and uk secondary level at t2) and there is around 25% attrition from t1 to t2. Is it possible to use imputation with data at three levels for this type of model please?

q1_base_each <- brm(target ~ 1 + x1 + x2 + group*ks*subj + (1|yp_id) + (1|mm(schl_id_ks2, schl_id_ks3)), family = cumulative("probit"), data = sats_imp1)

where x1 and x2 are control variables, group is a group effect, ks is the keystage indicator, subj is taught subject categorical variable (maths, English, science).

Unfortunately, we don’t yet have the study data or simulated data.

I’m using the following system setup:

  • Operating System: Windows 10 enterprise (Version 10.0.19044 Build 19044)
  • brms Version: 2.17.0

thanks for any advice.

1 Like

If you want to use multiple imputed data sets with brms, you may want to look at the brm_multiple function. If that is not what you are after, could be please clarify your question a bit more?

1 Like

Thank you for your reply. Our challenge is more that we cannot find a way to run the imputations with a multiple membership structure, and wondered if anyone else had succeeded in this given the possibility of fitting multiple membership models with brms. We also considered imputation during model fitting, directly within the brms framework, but suspect this is not possible given binomial/categorical variables - is this correct please?

1 Like

I understand. yes you are right. this is not possible with brms itself at the moment.

1 Like