Four level problem

[PeterTymms]

I am trying to model data in which I have pupils nested in classes which are nested in school which are nested in juridictions

Numbers are 87,361 pupils in 4591 classes in 2911 schools in 11 jurisdiction.

My problem, of course, if with the fourth level. I want to estimate the vaiance at that level and there are really too few jurisdictions to do this easily. In a null model MLWiN gives estimates but as I add explanatory variables it crashes. I've tried adding new variables one at a time and clicking "more". I've tried different estimation procedures. But all to no avail. Then I thought that I could use dummies instead of the fourth level and get the variance from of the separte measures of the jurisdictions. But is this approprate? Is there a different way?

Peter Tymms

[ChrisCharlton]

We are not sure why you are experiencing crashes with your original model, however if you can send an example worksheet with instructions on how to reproduce the crash we will look into it.

[ChrisCharlton]

In case anyone else comes across this problem: It appears that there is a bug in MLwiN where it has problems generating estimates if some of the higher level units contain no observations after taking into account missing data. The workaround for this is to omit any records where a predictor has a missing value prior to running the model.