Mi estimate error for ML multinomial

[guobl]

I tried run ML multinomial model using runmlwin with MI data generated by realcom, but get error message as "mi estimate: omitted terms vary
The set of omitted variables or categories is not consistent between m=1 and m=2; this is not allowed. To identify varying sets, you can use mi xeq to run the command on individual imputations or you can reissue the command with mi estimate, noisily".
mi xeq can't work for runmlwin so I ran ml multinomial model for each MI dataset one by one (runmlwin), no problem at all. I also check each MI dataset, all outcome have 3 categories. any help please? my code is
mi estimate, cmdok: runmlwin outcome2 cons, level2(psytm: cons) level1(patient) discrete(distribution(multinomial) link(mlogit) denom(cons) basecategory(2))

[guobl]

I could get mi xeq working by
forvalues i=1/10 {
mi xeq `i':runmlwin outcome2 cons, level2(psytm: cons) level1(patient) ///
discrete(distribution(multinomial) link(mlogit) denom(cons) basecategory(2)) rigls nopause
}
the code worked ok for each file. if I use
mi estimate,noisily cmdok:runmlwin outcome2 cons, level2(psytm: cons) level1(patient) ///
discrete(distribution(multinomial) link(mlogit) denom(cons) basecategory(2)) rigls nopause
it would stoped after modelling for m=2 dataset. I can't tell the reason, the only cause I am wondering is the level parameter estimates of m=1, which are 0 for both VAR anc Cov

------------------------------------------------------------------------------
Random-effects Parameters | Estimate Std. Err. [95% Conf. Interval]
-----------------------------+------------------------------------------------
Level 2: psytm |
var(cons_1) | 0 0 0 0
cov(cons_1,cons_3) | 0 0 0 0
var(cons_3) | 0 0 0 0
------------------------------------------------------------------------------
Is this the reason why MI could not combine the values and issue me an error message? or other reason please? any help will be highly appreciated.

[GeorgeLeckie]

Dear Guobl,

We are not sure what the source of the error is. However, if you are getting zero estimates for the variances and covariances on any given imputed dataset then you should certainly not proceed. You want to try to work out what aspect of your data is causing this more fundamental problem first. Only return to attempting to combine the 10 sets of results once you are sure that each set gives sensible results.

Best wishes

George