www.cmm.bristol.ac.uk/forum

Have you tried running this with fewer responses? You don't say how many rows you have in your data, or number of categories in each response but it may be that there isn't enough data to support this model.

As you are fitting a multivariate model there will already be a level-1 identifier included for the response indicator. When setting up the model you would therefore need to specify three levels, with patient at level-2 and clinic at level-3. MLwiN checks that the identifier variable you set at leve...

Bill gives some guidance on slide 57 of a workshop presentation that he gave some years ago, available at: https://www.ukdataservice.ac.uk/media/307220/presentation4.pdf The MRC Biostatistics unit, who originally developed the BUGS software, also give some suggestions: https://www.mrc-bsu.cam.ac.uk/...

You can obtain the MCMC chains as a dataset with the following command: mcmcsum, getchains Once you have this you can use the standard Stata data manipulation commands to process it, i.e. gen , sort and centile . You ought to get the same results as before as this should be equivalent to what runmlw...

The way that the credible intervals for this would be calculated is as follows: Fetch the MCMC chain for the parameter Exponentiate the values to convert to odds-ratios Sort the chain Extract the 2.5% and 97.5% quantiles I suppose that if the chain contained a few extreme values then this could resu...

That does sound a bit strange. Have you tried plotting the chains? What do the results look like it you don't convert them to odds-ratios?

Sorry, there is no change here, and this is not currently in our in our future plans.

I suspect that your problem is that you have forgotten to multiple the fixed-part parameters by their associated data, so you'll want something more like: generate prediction = [FP1]cons*cons + [FP1]male*male + [FP1]agegroup2*agegroup2 + [FP1]agegroup3*agegroup3 + [FP1]agegroup4*agegroup4 + [FP1]age...

As you have found, running predict after runmlwin will only give you the fixed-part prediction. If you want a prediction with both fixed and random effects then you will need to manually request that the higher-level residuals are returned via the residuals() option and then use these to extend the ...

You may find the following post of interest, as it explains the negative-binomial options within MLwiN:
Negative binomial multilevel model