Dear Statalist,

I have data on individual outcomes where the individual is a member of a subgroup (e.g. department) and a group (e.g. firm). I have analysed this (cross-sectional) data using an ordered logit model, as the outcome variable takes discrete values. The variable of interest is the percentage of individuals on a certain contract. I control for the group and subgroup, to remove "fixed effects". In Stata, this takes the form:
[CODE] ologit outcome percentage i.group i.subgroup, vce(robust) [\CODE]

It was suggested that I try a multilevel ologit model at both the group and subgroup levels (with the levels as: group>subgroup>individual). I understand that I can do this using the meologit command in Stata, which I believe allows the slope of the group and subgroup to vary.

My question revolves around whether I still need to control for group and subgroup in the meologit model or whether this is not necessary if they are contained as a random effect. In other words, is the following regression a sufficient extension of the ologit model above, or is there no need to control for group and subgroup?:

[CODE] meologit outcome percentage i.group i.subgroup || group: || subgroup: [\CODE]

Additionally, would I need to add any variables after group: and subgroup:? I have experimented including percentage but the coefficient of interest doesn't change.

Any thoughts on this would be much appreciated. I've read a few papers around the use of multilevel ologit and I am pretty sure that the meologit specification above is sufficient but am not 100% sure about controlling for group and subgroup so thoughts/comments/confirmation are most welcome!

Thanks and best wishes,
Rhys