[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
jverkuilen <jverkuilen@gc.cuny.edu> |

To |
<statalist@hsphsun2.harvard.edu> |

Subject |
RE: st: Categorical dependent variable models |

Date |
Tue, 21 Apr 2009 11:00:54 -0400 |

Irina, I could be wrong but it sounds like your data don't have information that would identify more complex models, e.g., repeated observations per subject for a mixed model or alternative-specific covariates. I'd suggest seeing if you overlooked some covariates or interactions that might account for the IIA violation. Essentially two of the products sound like imperfect copies of each other, which is a recipe for IIA to fail, though if it's not failing too badly you might just have to hold your nose and dea with it as Maarten said. JV -----Original Message----- From: "Paley, Irina" <Irina.Paley@occ.treas.gov> To: "'statalist@hsphsun2.harvard.edu'" <statalist@hsphsun2.harvard.edu> Sent: 4/20/2009 9:48 PM Subject: RE: st: Categorical dependent variable models Thanks so much for your response. The three choices are three mortgage products. Products 2 and 3 imply significantly lower monthly payments. Product 2 has much higher interest rate risk than product 1 and 3. My goal is to estimate effect of gender on choice of each of the three products. But when I try nested logit, for example-- stata gives me error message saying that gender doesn't vary within the product-- in other words it's not alternative specific but consumer specific. Suppose my tree is: p1 vs p2 and p3 (level 1) p2 vs p3 (level 2) Don't I have to assume that choice at level 1 is affected by product characterisics only? Or can I find a way to let gender affect choice at both levels? Thanks, Irina -----Original Message----- From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2harvard.edu] On Behalf Of jverkuilen Sent: Monday, April 20, 2009 9:31 PM To: statalist@hsphsun2.harvard.edu Subject: RE: st: Categorical dependent variable models Nested logit seems like a good choice, in theory. -help nlogit-. Much depends on the nature of the choices and why IIA fails. If it just fails a little, I would probably still consider it a viable assumption. If it fails badly and there are good reasons why, look for a different model. For instance if the choice set is among three modes of transportation to work (hey, it's tradition!), {Car, Bus, Train}, Bus and Train might be expected to share similarity because both are mass transit whereas car is private. Nested logit wold let you make the choice essentially become first {Car, Mass Transit} and then, assuming Mass Transit is chosen, {Bus, Train}. -----Original Message----- From: "Paley, Irina" <Irina.Paley@occ.treas.gov> To: "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu> Sent: 4/20/2009 6:32 PM Subject: st: Categorical dependent variable models Dear statalist members, Thanks for your previous help in my search for the right model, I am hoping you can prod me a little further. I want to estimate the effect of gender on the choice between 3 types of products, and the data I have only has consumer characterisics (I can stretch it somewhat to get a couple of product characteristics, though). the data fails the iia assumption, so I want to use something more fitting than mlogit. Is there a model, perhaps a variant of mixed logit or multinomial probit, that can allow me to have only consumer specific characteristics, and that would allow me to circumvent the assumption of uncorrelated outcome specific errors? I would really appreciate your suggestions. Thanks, Irina -----Original Message----- From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2harvard.edu] On Behalf Of Martin Weiss Sent: Monday, April 20, 2009 6:04 PM To: statalist@hsphsun2.harvard.edu Subject: st: Re: GMM-estimation of probit models <> I do not think that there is a routine at the moment, yet this problem is very similar (I imagine) to the one that Austin presented in Chicago, July 2008: http://www.stata.com/meeting/snasug08/nichols_gmm.pdf If you can tweak those moment conditions from poisson to probit, you can write the thing yourself... Also note the command -findit ivpois- HTH Martin _______________________ ----- Original Message ----- From: "Carlos Soares" <pipoca2140@gmail.com> To: <statalist@hsphsun2.harvard.edu> Sent: Monday, April 20, 2009 11:54 PM Subject: st: GMM-estimation of probit models > Dear Statalisters, > > Would any of you know of a Stata routine to run GMM-estimation of > probit models with endogenous regressors? > > Thanks, > > Carlos > * > * For searches and help try: > * http://www.stata.com/help.cgi?search > * http://www.stata.com/support/statalist/faq > * http://www.ats.ucla.edu/stat/stata/ > * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**Re: AW: st: AW: beta coefficients for interaction terms** - Next by Date:
**st: renaming using the file name** - Previous by thread:
**RE: st: Categorical dependent variable models** - Next by thread:
**st: Statistical test for trends** - Index(es):

© Copyright 1996–2015 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |