[Limdep Nlogit List] Latent Class Ordered Probit with Random Parameters?

[Steven Lavrenz]

Hi Bill,

I appreciate the response. Understood on LCRPLogit and binary/multinomial…the data in question has 4 levels (no injury, minor injury, major injury, fatal injury) so that’s why we were initially looking at the OP models.

What I’m trying to develop a better sense of is the tradeoffs with LCRPLogit and LCFPOrderedProbit (to use my own term)…it seems the major thing we’d be losing out on with LCRPLogit is the information found in the ordering of the injury severities. But I also know that in past years, Mannering and others have done work to suggest that this ordering can in some cases lead to erroneous inferences.

From a basic modeling standpoint, is there validity to estimating both models, and then [assuming no weird parameter estimates] choosing the one with the better log likelihood?

Best,
Steve


From: William Greene <wgreene at stern.nyu.edu>
Sent: Wednesday, July 10, 2024 10:25
To: Limdep and Nlogit Mailing List <limdep at mailman.sydney.edu.au>; Steven Lavrenz <slavrenz at wayne.edu>
Subject: Re: [Limdep Nlogit List] Latent Class Ordered Probit with Random Parameters?

[EXTERNAL]
Steve. A latent class random parameters ordered probit model is not possible in NLOGIT.
Also, the LCRPLogit model is not a binary logit model; it is a variant of the multinomial
logit model.  LCRPLogit is the only model form in NLOGIT that accommodates random
parameters with latent classes.  The ordered probit/logit model seems like the right model
for injury severity.  I'm not sure what else to suggest (or what the profession would suggest).
I've seen work by Kara Kockelman and by Chandra Bhat, but maybe they're out of date.
Regards
Bill Greene

On Tue, Jul 9, 2024 at 4:36 PM Steven Lavrenz via Limdep <limdep at mailman.sydney.edu.au<mailto:limdep at mailman.sydney.edu.au>> wrote:
Is this type of model possible in NLOGIT? I can estimate latent class ordered probit, ordered probit with random parameters, and latent class logit with random parameters. But I can't seem to get latent class OP with random parameters to work.

I have tried adding the 'RPM/fcn' commands, as well as trying the 'rpl' command mentioned in the documentation. All that happens is a fixed parameter latent class ordered probit model estimates.

If this is not possible, should I consider a LCRPLOGIT model instead, versus a fixed parameter LC ordered probit one? The dependent variable in question is crash injury severity, and I know the profession has cooled in recent years on the use of OP models in this space.

Thanks,
Steve




_______________________________________________
Limdep site list
Limdep at mailman.sydney.edu.au<mailto:Limdep at mailman.sydney.edu.au>
http://limdep.itls.usyd.edu.au


--
William Greene
Department of Economics
University of South Florida, Tampa
Department of Economics, emeritus
Stern School of Business, New York University
URL: https://url.au.m.mimecastprotect.com/s/kYORCVARKgC0vJEvRfGe1tO?domain=people.stern.nyu.edu
Email: wgreene at stern.nyu.edu<mailto:wgreene at stern.nyu.edu>
Editor in Chief: Foundations and Trends in Econometrics
Associate Editor: Economics Letters