Reference class in latent class logit

[Daniel PerezT]

Hello Statalist.

I'm using the command lclogit2 (written by Hong Il Yoo) to fit a model with the results of a discrete choice experiment. I have a dependent variable (choice, [0,1]) and some independent variables which are the characteristics of the product (price [10,20,30,40], format [on-site, digital], etc).

I've fitted the following latent class logit model (treating the different levels of the characteristics as dummies -and omitting one of the levels- and with 4 classes).

lclogitml2 choice, id(id) group(group) rand(format1 price1 price2 price3 ...) nclasses(4)

Now, I have some sociodemographic characteristics of the respondents. I would like to include them as covariates in the model (I know I can do it with the option membership) but if I do it, the interpretation becomes more difficult. It is because class 4 is selected as the reference class, and by the coefficients of the latent class model, I know that class 4 shows an extreme behavior. Is there a way to set a different class as a reference class (for example, class 2)? If not, is there any other alternative?

Thank you so much in advance.

[Hong Il Yoo]

Daniel PerezT Thanks a lot for your interest in -lclogit2-. Please note that the membership model coefficients can be interpreted similarly as -mlogit- and -fmlogit- coefficients. Thus, you can easily re-normalise your coefficients in the post-estimation stage. For example, let's suppose that you would like to re-normalise the coefficients by setting the coefficients for Class 2 to 0 to identify the remaining coefficients. You can achieve this by typing:

nlcom (Share1_cons: [Share1]_cons - [Share2]_cons) (Share3_cons: [Share3]_cons - [Share2]_cons) (Share4_cons: 0 - [Share2]_cons) (Share1_age: [Share1]age - [Share2]age) (Share3_age: [Share3]age - [Share2]age) (Share4_age: 0 - [Share2]age)

and so on for each variable listed in -membership()-.

[Daniel PerezT]

Dear Hong Il Yoo, thank you so much, I just checked it and it works perfectly.

I also wanted to thank you for the useful command, it really helped me in my research.

Kind regards,
Dani.

[Hong Il Yoo]

Daniel PerezT Great to hear that, and thanks a lot for your kind words