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 on April 23, and its replacement, statalist.org is already up and running.


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

Re: Re: st: Comparison of Coefficients in Conditional Logistic regression


From   Maarten Buis <maartenlbuis@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: Re: st: Comparison of Coefficients in Conditional Logistic regression
Date   Wed, 20 Jun 2012 15:12:48 +0200

On Wed, Jun 20, 2012 at 2:43 PM, A. Karvounidis wrote:
> thanks for the help. Could you please be more specific with what do you mean by "derive good starting values
> from the separately estimated models(*), and use those."? Moreover you said that if I want to use coefficients from the separetly regressions I need to do some computations. What computations can I do in order to compare them? thats the point...

Just to be sure we are on the same page: you have to use the model
with interaction effects. There is no other way. The challenge is to
estimate it.

Have you tried estimating your model with interactions after
appropriately centering your variable? This is the first thing you
must try, as all other "solutions" will require lots of work for you
and there is not much we can do to help you.

Assuming you tried to estimate the interaction model with
appropriately centered variables and it failed to converge. Than the
next step would be to seriously seriously seriously reconsider the
model, as in all likelihood you are either making a mistake with the
specification of the model or your model just too complex to be
meaningfully estimated with your data.

Only than can we start to think about starting values. In that case
estimate the separate models. Than just write down the separate models
and the model with interaction effects and derive the relationship
between the two. I would start with sticking to the additive form of
the model, i.e. the effects on the log odds of success. The math is
typically simple, usually no more than some adding and subtracting,
but there are many different ways in which you can parameterize both
models, and the results will critically depend on that. So you need to
do this yourself, as only you know how you parameterized your model.
After that you use those results to derive the "values" for the
interaction model from the separate models, store those initial values
in a matrix, assign the appropriate column names to it and feed it to
-clogit- using the -from()- option.

-- Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany


http://www.maartenbuis.nl
--------------------------
*
*   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/


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