[Limdep Nlogit List] Finding confidence intervals of WTA after LCRPLOGIT

[Kolady, Deepthi Elizabeth P]

Hello Dr. Greene,

Thank you very much for catching the error in my code. I will correct it and re-run the model. I have used a small number of Halton draws to reduce the convergence time so that I can share the results with you. I plan to use at least 500 Halton draws in the final model.

Yes, I have estimated the WTA using ratios of parameters (-(Coeff attribute/coefficient price)). For this purpose, I have kept the price fixed and attributes random. I am interested in calculating the 95% confidence intervals for class-wise WTA calculated as ratios of parameters using the Krinsky and Robb method. I see that we can use the Wald and Krinsky and Robb (K&R) command after Logit, but I'm not sure how to do it after LCRPLOGIT.

Also, to plot class-wise k-density, I assume I calculate individual WTA values, use class probabilities to assign each respondent to a class, and then use those class-wise individual WTA values to plot K-density plots. Am I right on this?

Thanks and regards,
Deepthi



-----Original Message-----
From: Limdep <limdep-bounces at mailman.sydney.edu.au> On Behalf Of William Greene via Limdep
Sent: Wednesday, March 27, 2024 9:18 AM
To: Limdep and Nlogit Mailing List <limdep at mailman.sydney.edu.au>
Cc: William Greene <wgreene at stern.nyu.edu>
Subject: Re: [Limdep Nlogit List] Finding confidence intervals of WTA after LCRPLOGIT

[You don't often get email from limdep at mailman.sydney.edu.au. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]

Deepthi: Your results did not show how you computed class-wise WTA values.
I assume they are computed using ratios of parameters.  But, since the parameters are random, what you are computing is a distribution of WTA values. A "confidence interval" is ambiguous.  The variation in an interval has two meanings: (1) sampling variability based on the estimated parameters and (2) the estimated variation due to the randomness of the specified parameters.  I'd note, ratios of random parameters are problematic as well.  So,.... before I can suggest how to do the computation you asked about, I need a suggestion of what kind of interval you are looking for.
   I note a subtle fluke in your results that I would guess you did not intend. Notice in the RP model results, the third and fourth estimated standard errors are identical.
That happened because the variables CONVTOCT and CONVTONT have the same first 6 characters.  The first 6 characters in the name of the variable with the RP coefficient are used to create the name for the standard deviation.  The constructed name is Ns******, which will end up the same both times.  When parameter names are the same, NLOGIT takes that to mean you want to constrain them to be equal, which I assume you did not expect.  To fix this, just change one of the variable names.
You'll then have to adjust the RHS and FCN lists.
One other small point.   You have a lot of random parameters.  The model
estimated OK,
but only 20 Halton draws is probably a little low.  You might get better results is you raise that to 50 or 100.
Sincerely,
B,. Greene

On Wed, Mar 27, 2024 at 9:31 AM Kolady, Deepthi Elizabeth P via Limdep < limdep at mailman.sydney.edu.au> wrote:

> Hello Dr.Greene,
>
> I use discrete choice data to estimate latent class-wise willingness
> to accept (WTA) values. I am using the LCRPLOGIT command because I
> would like to combine the random parameter logit model with the latent
> class model to account for heterogeneity. I am using the LCM command
> to use a variable of interest to determine class probability (in my
> case, the climate change concern scale).
>
> I have attached my model estimation and results. I used this result to
> estimate class-wise WTA values. Can I estimate the confidence interval
> for WTA values using Krinsky and Robb's method after LCRPLOGIT estimation?
> I can generate a K-density plot and show the distribution of
> class-wise or individual WTA, but as a reviewer asked, I need help
> figuring out how to get confidence intervals to report.
>
> Thank you in advance for your help.
>
> Sincerely,
> Deepthi
>
>
>
> _______________________________________________
> Limdep site list
> Limdep at mailman.sydney.edu.au
> http://limde/
> p.itls.usyd.edu.au%2F&data=05%7C02%7Cdeepthi.kolady%40okstate.edu%7Ce4
> 735b99bc12465968ec08dc4e68dbae%7C2a69c91de8494e34a230cdf8b27e1964%7C0%
> 7C0%7C638471459491925062%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiL
> CJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=cyhgE
> UgmZdG6q11cLbCVwuMXGAj33Yqj%2Fw1%2FyAx87co%3D&reserved=0
>


--
William Greene
Department of Economics, emeritus
Stern School of Business, New York University
URL: https://url.au.m.mimecastprotect.com/s/rgHqCmO5glu1jgKVKFGqX7c?domain=people.stern.nyu.edu
Email: wgreene at stern.nyu.edu
Editor in Chief: Foundations and Trends in Econometrics Associate Editor: Economics Letters Associate Editor: Journal of Business and Economic Statistics _______________________________________________
Limdep site list
Limdep at mailman.sydney.edu.au
http://limdep.itls.usyd.edu.au/