Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

Re: st: mlogit problem


From   Richard Williams <richardwilliams.ndu@gmail.com>
To   statalist@hsphsun2.harvard.edu, statalist@hsphsun2.harvard.edu
Subject   Re: st: mlogit problem
Date   Sun, 17 Feb 2013 19:23:27 -0500

At 10:07 AM 2/17/2013, JVerkuilen (Gmail) wrote:
On Sun, Feb 17, 2013 at 6:46 AM, saqlain raza <bhatti_sb@yahoo.com> wrote:
> Hello,
> I am trying to fit -mlogit- with aroung 60 covariates (discrete and continous) initially to see the significnat variables. My dependent variable has four alternate choices. Upto 44 covariates it converges. But, if I add one more variable, iteration process starts and at the end, the result is not converged. Any help will be highly acknowleged.
>
> Saqlain
> PhD Researcher

You need to indicate your N but you are running a probably unfittable
model. With 60 covariates in an mlogit model you have 180 parameters
and the chances of a non-concave likelihood, collinearity problems or
perfect prediction become higher and higher. This is simply going to
fail. I'd really go back to rethink your problem as looking at
statistical significance is usually not a very good way to do variable
selection anyway.

I am always struck by these problems that involve estimating hundreds, even thousands of parameters. It is very different from what I do in my world. I agree that the problem probably needs to be simplified; the data are likely being spread far too thin.

Having said that, if I was bound and determined to get all 60 of those covariates in, I might

(a) add the -difficult- option -- occasionally it works miracles; or

(b) consider combining some categories, maybe even dichotomize the response variable, if it did not seem too unreasonable to do so.

You could also do stepwise -- which I think would probably be a terrible thing to do, but just putting 60 variables in at once and seeing what works probably isn't any better.


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

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index