multivariate multiple membership models

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
schillsg99
Posts: 11
Joined: Sat Mar 09, 2013 6:11 pm

multivariate multiple membership models

Post by schillsg99 »

Hello:

I am attempting to fit a multivariate multiple membership models. I am following the same procedures as for a univariate multiple membership model with weights and higher level id's in consecutive columns. The weights are weight1, weight2 and the ids are picx1 picx2 where the picx's indicate teacher id's. I have two responses - a mathscore and a reading score. After fitting the 3 level model with the responses at level 1, the student ids at level 2 and the teacher id picx1 at level 3, I switch to MCMC estimation and estimate the three level model. Then I go the Model/MCMC/Classifications and indicate multiple classification at level 3. For number of columns I indicate 2; for weights start column I indicate weight1; and for ID start column I indicate picx1. When I hit done I get a pop-up box 'Error detected by MLN' 'multiple membership column wrong length(XCLA)'. I realize that MLwiN sets up a picx1_long column whereas I do not have a picx2_long column. However, if I hit OK in the error box and proceed by hitting the start button the MCMC estimation seems to proceed normally with the 3rd level variance/covariance matrix estimates becoming slightly larger as I would expect.

Am I OK here or do I need to set up long versions of the ID's and weights? I notice that MLwiN does not set them up after the estimation is finished.

Steve
ChrisCharlton
Posts: 1292
Joined: Mon Oct 19, 2009 10:34 am

Re: multivariate multiple membership models

Post by ChrisCharlton »

MLwiN will not create the required long version of the IDs and weights for multiple membership models when using a multivariate column, so you will need to create these expanded versions yourself. When doing so you will also need to ensure that the new columns containing these appear consequentially in the worksheet.
schillsg99
Posts: 11
Joined: Sat Mar 09, 2013 6:11 pm

Re: multivariate multiple membership models

Post by schillsg99 »

I tried this but the results don't look right. When I fit a univariate multiple membership model for reading I get a 3rd level variance component of 10.436.

When I fit a multivariate multiple membership model for reading and math I get a 3rd level variance component of 5.676. In both cases I have a reading pretest as a fixed covariate.

The way that I created a long version of picx1 and picx2 was to repeat them twice for each 2nd level unit which is exactly what happened when MLwIN created the long version of the 2nd level unit. I did the same with the weights. I then copied these into c101 and c102 for the 3rd level id's and c103 and c104 for the weights.

I am a bit perplexed.

Steve
ChrisCharlton
Posts: 1292
Joined: Mon Oct 19, 2009 10:34 am

Re: multivariate multiple membership models

Post by ChrisCharlton »

We have looked into this further and believe that the information that we provided earlier on how to set this up was incorrect. Could you try the following to see whether you get more plausible results?:
  • Set up the multivariate model as for the single-membership model
  • Switch to MCMC
  • Open Model->MCMC->Classifications
  • Tick "Multiple Classifications level..." appropriate for the level at which you have multiple-membership
You will now notice that as well as the normal question regarding the starting column for weights it will also ask for an ID start column
  • Choose the non-expanded versions of these columns (making sure that they are consecutive in the worksheet)
  • Click done
  • Click Start to run the model
MLwiN will not create expanded versions in the worksheet, but we believe that the underlying data structures will take the multiple membership into account. We intend to double-check the code to confirm this.
schillsg99
Posts: 11
Joined: Sat Mar 09, 2013 6:11 pm

Re: multivariate multiple membership models

Post by schillsg99 »

That is what I had done originally. It does give reasonable results. The only problem is that when I set up the multiple membership model in the classification window and click done, I get a pop-up window that says "multiple membership column wrong length(XCLA)". If I click OK for that pop-up and proceed, the program estimates the model and the results are what I would expect - the variance components get slightly larger.

Steve
ChrisCharlton
Posts: 1292
Joined: Mon Oct 19, 2009 10:34 am

Re: multivariate multiple membership models

Post by ChrisCharlton »

It appears that the error message is spurious and can be ignored, and we believe that the software is fitting the model correctly. We intend to fix this error checking behaviour in the next MLwiN release.

As this particular model set-up hasn't been extensively used you may want to double-check the results by getting MLwiN to generate the BUGS code for the model (via Model->MCMC->WinBUGS options) and then comparing the results obtained by using this.
Post Reply