Thanks for getting back quickly, Rich! I willed this to work, but
unfortunately, -mlbeta- simply ignored nocons again and promptly dropped
the same dummy variable due to multicollinearity. Your other idea is
admissible, as I might want to model the effect of the dummy variables
through the dispersion (although not normally), so I also tried:
. ml model lf mlbeta_ll (`lhs'=`rhs', noconstant) (`dispersion',
noconstant) `if' `in', `robust' `svy'
If you are losing variables to collinearity, that is another issue. Add
the parameter collinearity to the ml model command, e.g.
