Question about common coefficients under multivariate model

Welcome to the forum for MLwiN users. Feel free to post your question about MLwiN software here. The Centre for Multilevel Modelling take no responsibility for the accuracy of these posts, we are unable to monitor them closely. Do go ahead and post your question and thank you in advance if you find the time to post any answers!

Remember to check out our extensive software FAQs which may answer your question: http://www.bristol.ac.uk/cmm/software/s ... port-faqs/
Post Reply
syhi2000
Posts: 1
Joined: Sat Apr 22, 2017 10:51 pm

Question about common coefficients under multivariate model

Post by syhi2000 » Sat Apr 22, 2017 11:02 pm

I try to run the multivariate multilevel model with common coefficient to a predictor under MCMC estimation. However, when I try to run the model with MCMC estimation, the results with MCMC are substantially different from IGLS estimation and also the results look strange.

For example, in the equation attached data set, the coefficients of pre-score were 0.442, 0.409, and 0.233 for each outcome under IGLS and common coefficient of pre-score was 0.275 under IGLS. On the contrary, the coefficients of pre-score were 0.440, 0.409, and 0.231 for each outcome under MCMC and common coefficient of pre-score was 0.007 under MCMC.

I feel weird because, as I know, the fixed effect is not substantially different depends on estimation method. I know it might be my data problem but I would like to ask if there is technical issue of MLwiN program that can explain the reason of the results.

ChrisCharlton
Posts: 890
Joined: Mon Oct 19, 2009 10:34 am

Re: Question about common coefficients under multivariate model

Post by ChrisCharlton » Mon Apr 24, 2017 10:21 am

Have you checked the MCMC chains to ensure that they have converged to a stable distribution?

ChrisCharlton
Posts: 890
Joined: Mon Oct 19, 2009 10:34 am

Re: Question about common coefficients under multivariate model

Post by ChrisCharlton » Wed Apr 26, 2017 1:57 pm

I checked this with Bill, who wrote the MCMC algorithm, and apparently it was not designed to allow common coefficients in multivariate models. As a workaround you can use Metropolis-Hastings for the fixed effects (via Model->MCMC->MCMC Methods), which does appear to work.

I will look into providing a warning for this in future versions of the software.

Post Reply