Complex meologit Multilevel Model with Sample Design Weights - Cross-classification needed?

[Sue Sheffield]

Hi I have a complex Multilevel model with numerous models (sorry I can't provide a sample of the actual data because it's secure-level).

I have panel data which I have xtset (time(year), clustered within individuals).

Individuals can be in Region A,B,C, primary sampling unit (postcode sector) x,y,z and also be members of classification group 1,2, 3 etc because a region and


postcode sector are so big they could contain several different types of area (rural town, village, urban town, etc) - see "YO" or "PE" on map,

The levels of the model are:

1. REGION

2. Primary Sampling Unit << cross-classify? >> Britain-wide Geographical classification (Rural/Urban etc)

3. (Household)

4. Individual < sampling weight to generalise to population >

5. Time (Year)

So, the weights provided are STRATA, Primary Sampling Unit (postcode sector from e.g. 2008) and the longitudinal weights at individual level which adjust for attrition so that the survey panel remains representative of the general population.

Now, with Level 2, the psu's are primarily within region, but the Geog classification of rural, urban etc overlaps with PSU and REGION, but I'm not seeing how I can "cross-classify" i.e. allow individuals to be members of both region/psu and the rural-urban classification groups.

Anyone got any suggestions as to how I can cross-classify the level 2 group variables please?

(I also read that there should be weights at each level but I'm ignoring that for now due to model complexity).

Any advice welcome! Thanks.

[Sue Sheffield]

Hi, I have tried this model:

meologit ind2hosp age_squared i.sex i.Geog_classificn [pweight = indweight] || REGION: || psu : R.Geog_Classificn, (vce cluster: REGION)

I'm hoping this covers all the above bases!