Zero variance question

[mdommers]

I have a large dataset of 80,000 records spread over 5 years and with records from both males/females and reception/year 6 pupils. I have successfully fitted a 2 level model with school and individual pupil as the two levels. This model seem to produce a reasonable fit although the variance attributable to the school is lower than I had expected.

I have then broken down the 80,000 records into 20 distinct datasets giving the following

Year 1, male, reception
Year 1, male, year 6
Year 1, female, reception
Year 1, female, year 6
Year 2, male, reception,
...
Year 5, female year 6

Using these datasets I have then fitted separate and multilevel models. All of these seem OK and produce reasonable fits with slightly more variance attributable to the school than in the full model, apart from one which gives a figure of 0 for the amount of variance attributable to the school, which I cannot believe is true. I have tried allowing the variance to go negative although it still ends up at 0.

Any advice or thoughts would be very grateful.
Many thanks,

[mdommers]

I am still having this problem and would welcome any thoughts other forum members may have as to how I can solve it.

[billb]

Hi Mdommers,
I could take a look if you send me your worksheet (william.browne@bristol.ac.uk).
Regards,
Bill.