How to frame the model depends on what hypothesis you want to test, and in
particular whether you expect the slopes of x and y to vary by couple, or
just the overall intercept.
If you use ths syntax -xtmixed z x y || couple: gender-, you are forcing
gender to have zero effect on average, and also you are building a model
with heteroscedasticity: the residual variance in males differs from that
in females. Now, if that is what you want, fine.
But also be aware that you have to guess correctly which sex has the
higher residual variance. Coding gender with male = 1 and female = 2, if
the females have greater residual variance, you will get results. But if
the females have lower residual variance, then you are trying to estimate
a model where the variance component associated with the random gender
effect is negative--and the estimation will never get there and you will
just iterate until you hit maxiter and never converge.
Also, if you are going to ultimately try to interpret the covariance
structure among random effects, you will need to pay attention to the
centering of the x and y variables.
Not knowing your problem, my guess is that you probably want one of the
following, as these are more commonly encountered situations:
xtmixed z x y || couple:
xtmixed z x y || couple: x
xtmixed z x y || couple: y
xtmixed z x y || couple: x y
xtmixed z x y gender || couple:
xtmixed z x y gender || couple: x
xtmixed z x y gender || couple: y
xtmixed z x y gender || couple: x y
Good luck.
Clyde Schechter
Assoc. Prof. of Family & Social Medicine
Albert Einstein College of Medicine, Bronx, NY, USA
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
