www.cmm.bristol.ac.uk/forum

I have estimated a 3-level model, with respondent as level 1, interviewer as level 2, and (27 European) countries as level 3. Now that I see the parameters for the respondent variables (beta's), and the interviewer variation (sigma²_v0), I know these apply for the whole 'EU-27' area, and I'm realising it might not be as meaningful. Now I am more curious what the effects are in each of the separate countries, and where the differences are. My solution would be to estimate a two level model for each country. So my question is, is there a type of 'BY-statement'in MLwiN which I can use to estimate the model for each country, or do I need to do it manually for each country? Or, do you see a way to derive the information I am looking for from the parameters of the 3-level model I have already estimated, or a modified specification of this model?

The following is the equation I have used so far:

There is no automatic way of doing this in MLwiN, however you could get the same effect by excluding all but the current group that you are interested in using the EXCL command, or through the exclusion option of the hierarchy viewer window. Alternatively, if you have access to Stata then the -runmlwin- command (http://www.bris.ac.uk/cmm/software/runmlwin/) supports Stata's by keyword.