I hope there's somebody who can help me. I know the basic concepts of hierarchical logistic regression and I am confused.

Here is the situation: 7500 schoolchildren in 170 schools from 8 countries. I would like to examine the relations between building characteristics and health. The health outcomes are dichotomous (yes/no) or ordinal (no, yes and less than 3 episodes, yes and 3 or more episodes per week). By study design, a hierarchical logistic model 3 levels seems relevant but some colleagues said me to perform a 2 levels model and to adjust for the country at level 2. Indeed, there is no other covariables at level 3. The small sample size and the complexity of 3 levels model are the main reasons advanced but, is it really accurate to perform a 2-levels model? Is it true to adjust for country at level 2 instead of taking into account a higher level?

Furthermore, they said me that it is not really reasonable to build a hierarchical ordinal model for the ordinal health outcome due to same reasons (small sample size and complexity)... should I transform this variable into a binary variable? But some information seems lost...

Thank you in advance for your help! ]]>

I wonder (and this is a long shot) if anyone knows something similar for Australian Census Data from the ABS. In terms of geodata I want to feed the API postcodes, lat/lon, or local government areas. ]]>