Estimating a multilevel simult equation model

[russellandrew]

Dear Listers,

I have estimated a multivariate model with three outcomes (Language, Math and Science) using a national Grade 6 survey. I am examining the contested issue of resources and their impact on outcomes. In the second stage, I wanted to follow up on the advice and modelling strategy in the Steele et al (2007) paper and proceeded to set up the bivariate response model where outcome and resources are modelled jointly (class size and outcome seperately for each of the three outcomes). However, as soon as I enter covariates into the bivariate response model, I get the IGLS/RIGLS numeric warning pertaining to the fixed part of the model. So two questions
1. If outcome is a pupil level variable and class size is a school level variable, how can I specifiy that so that the former has random intercepts across students, schools and districts (LEA in UK speak) while class size only has school and district random intercepts? is this the crux of the problem because I would have thought that a multivariate and a bivariate model is specificed in the same way in MLwiN
2. Do I set up this model differently from the multivariate model? I would have thought not seeing that both are variants of the multivariate model

I have sorted my data as per the multilevel structure: observations in students; these nested into schools and schools nested in ditricts.

Any help will be hugely appreciated.

Many thanks,
Russell Wildeman

[Lydia]

I can't help with your second question, but in response to your first question:

When you run a multivariate response model, just like when you run a single response model you specify random intercepts by clicking on cons and ticking the boxes for all the levels that you want a random intercept for. With a multivariate model, each response has a different cons, which in your case should be called something like class_size.cons, Language.cons, Math.cons, and Science.cons. When you click on one of these and tick the boxes, random intercepts are specified for that response only, not for all the responses in the model. So in your case you can click on class_size.cons and tick just the school and district boxes, then click on Language.cons and tick all three boxes, and the same for Math.cons and Science.cons.

Hope that helps!

[russellandrew]

Thanks Lydia for the response. Much appreciated, Russell