No interactions allowed in mixed logit regression (Stata 14 -mixlogit- and -gmnl- packages)

[Georges Farina]

Hello,
This is my first post here, please pardon me for any mistake or obvious errors, but I haven't found an answer to my problem in the forum.
I am working on Stata 14 on the analysis of a discrete choice experiment. I am using a mixed logit regression (-mixlogit- package) and am trying to include interaction terms between a factor variable (a characteristic of the respondent, named -know_water-) and one of the attributes of the choice experiment -inf_100-.
My code looks like so

global randvars "inf_50 inf_100 heat_20 heat_40 heat_60 red_car dist_0 dist_100 dist_500 sq"
mixlogit choice inf_100#i.know_water, group(chid) id(id) nrep(500) rand($randvars payment) ln(1)

I systematically get the error "interactions not allowed" however I change the code. I also tried it with the gmnl package, and it did not work (code below):

gmnl choice, group(chid) id(id) nrep(500) rand($randvars mpayment )ln(1) mixl

However, I have performed the same interactions with a multinomial conditional logit regression without problem, like so:

clogit choice $randvars payment inf_100##know_water, group(chid)

My -know_water- variable was encoded from string to a factorial variable successfully, since it worked with the conditional logit regression

encode know_origin_water, gen(know_water)

Is my code wrong ? Or is it the whole -mixlogit- package that does not allow for interactions ? In this case, would you have any recommendations for other packages on Stata14 ?

Thank you in advance

Georges Farina

[Hong Il Yoo]

Georges Farina Many community-contributed packages including -mixlogit- and -gmnl- do not recognise Stata's factor variable short-cuts. You can generate the interaction terms directly as new variables, and include them in your model specification. For example:

generate double inf_100Xknow_water = inf_100 * know_water  
mixlogit choice inf_100Xknow_water, group(chid) id(id) nrep(500) rand($randvars payment) ln(1)