Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Structural equations, latent variables and pathanalysis


From   David Greenberg <dg4@nyu.edu>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Structural equations, latent variables and pathanalysis
Date   Wed, 13 Apr 2005 21:57:24 -0400

It is well-known in OLS regression that the inclusion of variables that
do not actually contribute to the outcome does not bias the parameter
estimates, buy increases standard errors. I believe that this would also
be true for 2SLS. Leaving variables with non-significant coefficients in
should be harmless, particularly if you have enough cases that the gain
in degrees of freedom from omitting them is small. Moreover, if the
variables do contribute (which is possible even when the coefficients
are not significant), you would be introducing some bias from omitting
them. This suggests that you should simply report the results you
already have. David Greenberg, Sociology Department, New York University. 

----- Original Message -----
From: shyamalshyamalshyamal <shyamalshyamalshyamal@yahoo.com>
Date: Wednesday, April 13, 2005 10:43 am
Subject: Re: st: Structural equations, latent variables and path analysis

> 
> Hi:
> I feel IV method for SEM is much easier than some other methods. I
> have a quick question though. After I run 2SLS for non-recursive
> model, what happens if the coefficients are non-significant? Do I
> have to remove the non-significant variable(s), and then start
> reestimating the regression from the beginning? Or, the non-
> significance paths do not matter in this case?
> 
> I would appreciate if someone explains the steps of doing 2SLS to
> estimate paths; also, please include your feedback on non-significant
> paths as well.
> 
> Thanks.
> Shyamal
> --- In statalist@yahoogroups.com, "Nick Cox" <n.j.cox@d...> wrote:
> > Next week would be fine.
> >
> > Joe Newton and Nick Cox
> > editors@s...
> > Editors, Stata Journal
> >
> > Stas Kolenikov
> >
> > > extra two cents: many latent variable structural equation models
> are
> > > estimable by instrumental variable methods. Unlike in economics,
> where
> > > the instruments are usually pulled out of thin air, one can
> derive the
> > > rigorous ways to pick model-implied instruments. IV methods are
> less
> > > efficient that MLE (although I have not seen efficiency losses
> greater
> > > than 30% in those applications), but they are more robust to model
> > > misspecification. Moreover, one can test certain misspecifications
> > > such as omitted paths or measurement error correlations with
> Hausman
> > > test on instruments. Ken Bollen has been the main contributor to
> this
> > > topic (Bollen, KA (1996) 'An Alternative Two Stage Least Squares
> > > (2SLS) Estimator for Latent Variable Equations.' Psychometrika,
> 61:
> > > 109-121; Bollen, K A, and Bauer, D J (2004), Automating the
> Selection
> > > of Model-Implied Instrumental Variables, Soc. Methods & Research,
> 32
> > > (4), 425-452). Being his student, someday I should write a paper
> to
> > > Stata Journal on how to do all that by -ivreg-...
> >
> > *
> > *   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/
> 
> 
> 
> *
> *   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/
> 
*
*   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 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index