Direct and cross elasticities and margins after lclogit

[Travis Wang]

Hi statalist,

I have run a latent class model with some discrete choice experiment (DCE) data, using lclogit. lclogit is a user-written package by Pacifico and Yoo (http://www.stata-journal.com/article.html?artic).
The DCE asked Respondents to choose between 2 modes of shipping(Liner or Train) based on several attributes (time, cost, delay, etc.) and some membership attributes also are involved. the code of the model estimate is:

lclogit value cost time service_frequency delay ,group( choice_scenario ) id ( ID ) nclasses(2) membership( number_emp frequency_shipment middle_shipment high_shipment  value_shipment) iterate(1000) seed(1234567890)

According to the lclogit command manual I have performed a series of analysis on the data I collected, including parameter estimation and posterior probabilities.
But now I want to perform further analysis on the model, such as Direct and cross elasticities, marginal effects, probability prediction, etc., but I don't know what commands I should use to perform such analysis... So, some help is greatly appreciated ...
Thank you
Best regards,
Travis

[Hong Il Yoo]

Travis Wang:

You can use the postestimation command -lclogitpr- to generate predicted choice probabilities. There's no postestimation command for cross elasticities and marginal effects. You may want to have a look at -nlcom- and -predictnl- commands that will help you write your own command lines to evaluate the cross elasticities and marginal effects.

You may be also interested in the -lclogit2- package which I consider as an improved version of -lclogit: https://journals.sagepub.com/doi/pdf...36867X20931003. It doesn't automate the calculation of cross elasticities and marginal effects, but it runs faster than -lclogit- and offers a standalone -ml maximize- command.

[Travis Wang]

Hong Il Yoo :
I will go and learn these commands to see if I can get more estimates, thanks for the suggestion. Also, I would like to know how to interpret the membership variable parameter.

The above is a parameter estimation I made using the public dataset, whose membership variable is income and in share1 its parameter is 249.91. Is it correct to interpret this estimation as meaning that members of the first class are more likely to have higher income than members of the second class?

Best regards,
Travis

[Hong Il Yoo]

Travis Wang Yes, you can interpret the share equation coefficients in the same way as you'd interpret -mlogit- coefficients. So your results suggest that higher income earners are more likely to be in class 1 than class 2. Having said that, the missing standard error suggests that your model may not have converged properly. I'd like to suggest that you double-check the convergence of your model by playing around with alternative starting values, number of iterations, and maximization techniques.