Thanks Jeff and everyone else for your response.

I solved it by generating the interaction term manually.

...

> Ulrich Atz <

[hidden email]> is using -margins- after -mlogit-, but Stata

> crashes in the middle of the -margins- command execution:

>

> > I can produce a crash with the following command (all dummy variables) after

> > a multinomial logit regression:

> >

> > margins i.us#i.highed#i.highrelig#(i.us#i.highrel), at(age=50)

> >

> > This one for example works -margins

> > (i.us#i.highed#i.highrelig)#i.us#i.highrel, at(age=50)-, though not as I

> > intend. (The idea was to interact the first three dummies with an

> > interaction term.)

> >

> > I know this is very little information and that the syntax may actually be

> > wrong, but I suppose an error would be more appropriate than a crash. I use

> > Stata/IC 11.1 for Mac (64-bit Intel).

> >

> > If the statalist is not the right place to report bugs, I apologise.

>

> We have determined that the crash is caused by more than one duplicate

> variable specified in an interaction; parentheses control the parsing order

> and can prevent the crash.

>

> This will be fixed in the next executable update.

>

>

> In the mean time Ulrich can avoid the crash by specifying a given factor

> variable only once per interaction.

>

> For example, Ulrich's example was

>

> . margins i.us#i.highed#i.highrelig#(i.us#i.highrel), at(age=50)

>

> The duplicate variables are 'i.us' and 'i.highrelig', and the parentheses add

> nothing to this interaction specification. The above syntax is equivalent to

>

> . margins i.us#i.highed#i.highrelig, at(age=50)

>

>

> In a side note Ulrich said:

>

> > (The idea was to interact the first three dummies with an interaction term.)

>

> If Ulrich meant to separately interact each of the dummies with another term,

> then the following syntax might be what Ulrich wants:

>

> . margins (us highed)#highrelig, at(age=50)

>

> This is a shorthand notation for

>

> . margins us#highrelig highed#highrelig, at(age=50)

>

> If Ulrich meant to compute the margins for each dummy in addition to their

> interaction with another term, then Ulrich can type:

>

> . margins (us highed)##highrelig, at(age=50)

>

> This is a shorthand notation for

>

> . margins i.us i.highed i.highrelig us#highrelig highed#highrelig, at(age=50)

>

> --Jeff

>

[hidden email]
Please access the attached hyperlink for an important electronic communications disclaimer: