www.cmm.bristol.ac.uk/forum

Hello. I am fitting a multivariate 2 level model, where both responses are proportions (ie grouped binary data). When I try running the Model using MCMC to obtain a Bayesian DIC (for model comparison purposes later) I get the message "MCMC Error 0421:Can only use Gibbs sampling with 0/1 data. Please change to MH". Ch19 of the MCMC manual refers to the DIC not being available for mixed response models (eg normal/binomial). Would someone be able to advise if this applies here, or whether with amending by the code I can obtain a DIC?. Many thanks for any insights anyone can offer !

You should be able to fix the error that you are seeing by opening the Model->MCMC->MCMC Methods window and then changing the estimation method to one of the MH options. The -runmlwin- equivalent to this is using the univariatemh or multivariatemh options within mcmc_method (see the help file for more details).

MLwiN is unable to calculate a value for the DIC if any of the responses in the mixed response model are non-normal, so yes unfortunately this applies here. There is no way around this limitation.

Thanks very much for your reply. That's what I thought, so wanted to double check. Keep up the good work!