Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: mlogit vs mprobit


From   André Paul <hcpats@hotmail.fr>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: mlogit vs mprobit
Date   Fri, 24 Mar 2006 18:13:51 +0100

Thanks for yoyr answer.
Yes, I used exactly the same data. I did:

xi: mlogit outc nb_in nb_app share_no_x share_no_y_ fwd d_pr_app succes_mean stock_pim_mean nb_ds i.ap_year i.area2 df*, base(0)

xi: mprobit outc nb_in nb_app share_no_x share_no_y_ fwd d_pr_app succes_mean stock_pim_mean nb_ds i.ap_year i.area2 df*, base(0) tech(bfgs)

and then

mfx, predict(outcome(0))
...etc

Note that I used the bfgs algorithm (insteas of the default "nr"). I also tried to use the "dfp" algorithm which gives the same results. The reason is that the "nr" algorith makes 1 iteration every half an hour and the other techniques are much faster, but I don't think that the difference comes from there.

There are 4 categories which differe a lot in terms of size:
outcome 0: 294 obs.
outcome 1: 3277
outcome 2: 176
outcome 3: 2674

As I wrote in the previous message, the marginal effects are not significant when I use mlogit (i.e all the P>|z| values are equal to 1 or very close to 1) whereas some of the coefficiants of the marginal effects are significant with mprobit

André


From: Richard Williams <Richard.A.Williams.5@ND.edu>
Reply-To: statalist@hsphsun2.harvard.edu
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: mlogit vs mprobit
Date: Fri, 24 Mar 2006 11:17:28 -0500

At 10:16 AM 3/24/2006, André Paul wrote:
Dear all,

when I estimate successively a mlogit and a mprobit model, I get, as expected roughly the same coefficients.
However, when I compute the marginal effects, the standard errors (of the marginal effects) are much lower with mprobit. Actually, when I use mlogit, none of the marginal effects are significant, whereas, most of them become significant when I use mprobit.

Could someone give me the reason of this?

Thanks,
André
I just tried an example, and both the marginal effects and standard errors were very similar for both mlogit and mprobit. Are you sure something wasn't different between the two runs, e.g. were the samples and variables the same throughout? Did you use the -mfx- command in both cases, or a different command? Are some of the categories extremely thin, are the models having trouble converging? Perhaps you could post exactly what you did.

As a sidelight, Tamas Bartus's -margeff- command will quickly estimate the marginal effects after mlogit. Alas, it doesn't support mprobit. Note that, by default, it estimates the marginal effects a little differently than -mfx- does. If you want it to clone -mfx-'s behavior, give the command

margeff, at(mean)


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
FAX: (574)288-4373
HOME: (574)289-5227
EMAIL: Richard.A.Williams.5@ND.Edu
WWW (personal): http://www.nd.edu/~rwilliam
WWW (department): http://www.nd.edu/~soc


*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
_________________________________________________________________
MSN Hotmail : créez votre adresse e-mail gratuite & à vie ! http://www.imagine-msn.com/hotmail/default.aspx?locale=fr-fr

*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/




© Copyright 1996–2020 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index