Statalist


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

Re: st: clogit prediction in the presence of collinearity


From   sjsamuels@gmail.com
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: clogit prediction in the presence of collinearity
Date   Tue, 16 Feb 2010 18:48:44 -1000

Actually my statements:
"In -clogit-, there is an intercept for each pair, it but does not
appear in the prediction.   -logit- has only a singleintercept, which
is part of the xb prediction."

are irrelevant for the algorithm Laura gave,  as any intercept will cancel out.

However the estimated coefficients from the two commands will differ,
and -logit- will accept some predictors that -clogit- rejects because
they are constant within group.

Steve


On Tue, Feb 16, 2010 at 11:51 AM,  <sjsamuels@gmail.com> wrote:
> Laura, why do you want to manually construct the predictions?  Which
> predictions? What problem are you trying to solve?  Please be specific
> and illustrated with code and results if you have them.
>
> In any case, to answer your specific questions:
>
> 1. Your found "algorithm" is incorrect: The command "clogit x y z" is
> not legal syntax; there is no group variable.  If you mean "logit x y
> z", then the algorithm you found  does not produce the -pc1-
> prediction.  In -clogit-, there is an intercept for each pair, it but
> does not appear in the prediction.   -logit- has only a single
> intercept, which is part of the xb prediction. Also, -logit- knows
> nothing about the group variable, whereas -clogit- measures
> associations within groups.
>
>
> 2.  -predict- after -clogit- works the same way with collinear
> variables as it does without.  Collinearity  does not affect
> predictions much, if at all, in general.  What is your reason for
> thinking otherwise?
>
>
>
> On Mon, Feb 15, 2010 at 10:30 AM, Laura Zoratto
> <laura.zoratto@graduateinstitute.ch> wrote:
>> dear all,
>>
>> would anyone know how Stata predicts the probability of a positive outcome
>> after estimating a clogit, in the presence of collinearity (and the colinear
>> variable does not get dropped)? I found somewhere that the pc1 command in
>> Stata is equivalent to:
>>
>> clogit x y z
>> predict xb, xb
>> gen top=exp(xb)
>> by countrypair, sort: egen bot=total(exp(xb))
>> gen pc1=top/bot
>>
>> But I dont know what it does when there is a collinear variable affecting
>> the results . I need to calculate manually the predicted value...
>>
>> thank you,
>> Laura
>>
>> *
>> *   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/
>
>
>
> --
> Steven Samuels
> sjsamuels@gmail.com
> 18 Cantine's Island
> Saugerties NY 12477
> USA
> 845-246-0774
>



-- 
Steven Samuels
sjsamuels@gmail.com
18 Cantine's Island
Saugerties NY 12477
USA
845-246-0774

*
*   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   |   What's new   |   Site index