Margins/probabilities after unlabelled choice model -cmmixlogit ..., nocons-

[Jeremy van Dijk]

Dear all,

I see that this question has more or less been raised a couple of times and no real answer has been provided, so I am trying again and hoping for more help.

I am analysing the responses from a choice experiment using -cmmixlogit- in Stata 16. The choice alternatives are unlabelled, thus I use the option -nocons- to avoid estimating ASCs. In order to include case variables of interest and still avoid the meaningless ASCs, I therefore interact the case variables with a key attribute of interest. As an example (similar style, but made-up variable names):

cmset id alt
cmmixlogit chosen  ///
    i.comfort##i.income_groupl i.comfort##i.rural_urban i.comfort##i.hh_size  /// case vars
    car_range max_speed co_2e  ,  ///  choice attributes
    rand(  trip_cost  ln_subscription_price )   ///  random vars
    intmeth(halton) intpoint(100) nocons

This all runs fine, but then afterwards I want to estimate the predicted probabilities of choosing a particular comfort-level, as well as the impact the other variables, eg. income_group, have upon that. When I try to run -margins- I receive the error: alternative A is not in the fitted model; error in option outcome() r(458). This error remains if I try to estimate for different outcomes. i.e. - margins, outcome(3) - produces: alternative C is not in the fitted model; error in option outcome() r(458). It seems to me to be an issue because of the missing ASCs. But I realise perhaps I am mistaken or just need to tweak my commands.

So,
Q1: Is it possible to use margins for unlabelled choice analysis?
Q2: If so, how; or if not, what alternatives are there to complete the analysis I'm after?

Thank you for any help and clarity you can provide.

[Tom Schuette]

Hi Jeremy,

have you found any success in your efforts to analyze unlabelled choice experiment data? I am in the same situation as you are and there still aren't any real answers to this questions on here.

Thank you for any update on this.
Tom