Statalist


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

Re: st: Is there a kind of stochasticity in the execution of xthtaylor?


From   Hewan Belay <hewan_belay@yahoo.com>
To   Stata List <statalist@hsphsun2.harvard.edu>
Subject   Re: st: Is there a kind of stochasticity in the execution of xthtaylor?
Date   Wed, 16 Jul 2008 14:23:13 -0700 (PDT)

Dear Partha,
Not to be excluded the possibility that it's only near-perfect 
collinearity. But how would that explain why xthtaylor would give 
different results in repeated results of the same command? The extent of
collinearity would explain why we see some particular result, but not why 
we see varying results, no?
Hewan

---------------------------------------------------------------------------From   Partha Deb <partha.deb@hunter.cuny.edu> 
To   statalist@hsphsun2.harvard.edu 
Subject   Re: st: Is there a kind of stochasticity in the execution of xthtaylor? 
Date   Wed, 16 Jul 2008 16:56:30 -0400 

---------------------------------------------------------------------------
My quick peek into the ado file for xthtaylor shows that it does check for collinearity. Is it possible that yours is not perfect (but near perfect)?

Partha


Hewan Belay wrote:


Dear Partha,

Yes, it clearly has something to do with the presence of collinearity. But it is strange that Stata (at least for xthtaylor) stochastically chooses which collinear variable to drop (and doesn't "silently" drop it as in other cases of perfect collinearity, but in my demonstration regressions "sets" one of the collinear variables' coefficients to zero, and on another run does same but with another of the collinear variable's coefficient). Wonder if this is due to some mild bug in the xthtaylor programme... In any case, dropping one of the collinear time dummies myself, prior to running the command, takes care of the proximate problem (though leaves above question open).

Thanks much,
Hewan

---------------------------------------------------------------------------



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