The -mixed- command for multi-level modelling with a binary dependent variable

[Owen Wallbanks]

Dear all

I'm working with three level nested data:

• Individuals, nested within...
• Nuclear families (their parents and siblings), nested within...
• Extended families (their uncle/aunt and first cousins)

My dependent variable is a dummy for whether the individual is married (1=yes), so I have been using the -melogit- command. However, I was wondering whether the following:

mixed married [explanatory variables] || extendedfamilyid: || nuclearfamilyid:, mle

Would in effect fit a multi-level linear probability model?

Many thanks
Owen

[Leonardo Guizzetti]

Yes it would, though arguably it is better not to do so since you have -melogit- readily available.

[Owen Wallbanks]

Thank you for your speedy response Leonardo! I was just thinking about consistency. I have two explanatory variables of interest. One is number of siblings which is at the nuclear family level (L=2) and the other is birth order which is at the individual level (L=1). Naturally, you can't control for nuclear family fixed effects and estimate the coefficient on family size, since the two are perfectly correlated which is why I use a multi-level model instead. However, I was also thinking of then using nuclear family fixed effects to hone in on the birth order effect. To do this I would de-mean my variables with respect to nuclear family averages, which would transform my marriage variable from a dummy to a continuous variable ~[0,1], so logistic regression is no longer appropriate. I'm not sure whether this inconsistency is a problem?

Thanks again
Owen