?Here is an example of writing the log likelihood function for
?Logit with dependent variables that are shares.
?It largely follows Example 9.6.7 in the User's Guide.
?Say you have 3 choices with shares S1-S3 and NC is the number of
?observations per cell (i.e. the denominator of the shares),
?with covariates X1 and X2.

FRML SHLOG LOGL = NC*( S2*XB2 + S3*XB3 - LOG(1 + EXP(XB2) + EXP(XB3)));
FRML XB2 B20 + B21*X1 + B22*X2;
FRML XB3 B30 + B31*X1 + B32*X2;
PARAM B20-B22 B30-B32;
EQSUB SHLOG XB2 XB3;
ML(HITER=N,HCOV=N) SHLOG;

?It's important that HITER=N is used in this case.  If HITER=B was used,
?the share would get squared in the VCOV calculation, which would not be
?a proper way to weight things.