MCMC estimation

Welcome to the forum for runmlwin users. Feel free to post your question about runmlwin 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!

Go to runmlwin: Running MLwiN from within Stata >> http://www.bristol.ac.uk/cmm/software/runmlwin/
Post Reply
janna
Posts: 9
Joined: Thu Jun 03, 2010 5:53 pm

MCMC estimation

Post by janna »

Hi,

I am currently running some analyses with runmlwin and mcmc estimation. Since mcmc options are not included in the help file yet, is there anything more to regard than the burn in / chain length and starting values option ?

Currently I set up a hierarchical three level model with PISA data (28 countries, about 9 000 schools, 200 000 students). I started with a burn in of 500 and a chain length of 5 000. The trajectory plots showed high autocorrelation for the intercept and the level three predictor, so I increased both paramters to 5 000 and
500 000. Since these models take a lot of time, is it possible to adjust the thinning factor?

And is it possible, besides the diagnostic plots, to get the numerical diagnostics (Brooks-Draper, Raftery-Lewis)?

I would also be interested in estimating cross classified models with mcmc in the future!

Thanks a lot and best wishes,
Janna
GeorgeLeckie
Site Admin
Posts: 432
Joined: Fri Apr 01, 2011 2:14 pm

Re: MCMC estimation

Post by GeorgeLeckie »

Hi Janna,

We are currently developing the MCMC functionality of runmlwin and we about half way through doing this. This is why we have not yet documented the options. The options are also subject to change as we continue our development and receive feedback from users like you!

We therefore currently only encourage experienced MLwiN users to play around with these options. Users should always check that the options are working as they expect (e.g. by comparing their results to those obtained through using MLwiN in the traditional way).

Yes you can currently specify a thinning option:

... mcmc(burnin(5000) chain(50000) thinning(50)) ...

To get numerical MCMC diagnostics, use the post-estimation mcmcsum command that we have also written. This is currently bundled with runmlwin so it should already be installed on your computer. For example, to get the MCMC diagnostics for a level 2 variance parameter for an intercept variable cons, type:

. mcmcsum [RP2]var(cons)

To specify cross-classified models use the cc option:

... mcmc(burnin(5000) chain(50000) cc)

Hope this helps

George
roberto
Posts: 3
Joined: Tue Apr 12, 2011 7:23 am

Re: MCMC estimation

Post by roberto »

Hi George, would you also be able to specify the "parameter expansion" feature of MCMC through runmlwin :?:

Cheers, Roberto
ChrisCharlton
Posts: 1351
Joined: Mon Oct 19, 2009 10:34 am

Re: MCMC estimation

Post by ChrisCharlton »

This is currently implemented as a sub option of the MCMC option i.e. mcmc(paex(2 3)) to specify parameter expansion at levels 2 and 3. As the MCMC functionality is still being developed the exact syntax may still be subject to change in the future.
roberto
Posts: 3
Joined: Tue Apr 12, 2011 7:23 am

Re: MCMC estimation

Post by roberto »

ChrisCharlton wrote:This is currently implemented as a sub option of the MCMC option i.e. mcmc(paex(2 3)) to specify parameter expansion at levels 2 and 3. As the MCMC functionality is still being developed the exact syntax may still be subject to change in the future.
Hi guys, has this option currently changed? it doesn't seem to work anymore in the updated version of runmlwin :?:

Roberto
ChrisCharlton
Posts: 1351
Joined: Mon Oct 19, 2009 10:34 am

Re: MCMC estimation

Post by ChrisCharlton »

This has moved into the level() sub options so now has the following form:

runmlwin variables, level2(variables, parex) level1(variables) initsp mcmc(options)
Post Reply