st: RE: Weighted 2SLS

["Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>]

Dan,

To answer your last question first, -xtivreg2- (like -ivreg2- and
Stata's official IV commands -ivregress-, -xtivreg-, etc.) doesn't
implement 2SLS literally in 2 stages.  The 2SLS estimator is implemented
as the computationally equivalent (but computationally faster and more
convenient) single-equation IV estimator.

Weighting means weighting observations in this calculation.  So the
answer is no, there is no way for -xtivreg2- (or the other commands
above) to weight one part of the calculation differently from any other
part.

As for whether it makes sense to weight the equivalent of the first
stage differently from the equivalent of the second stage, I guess it's
up to you to work out the econometrics of this.  There are estimators
which treat the first stage differently from the second stage - see, for
example, Brian Poi's paper on Jackknife IV estimation in the SJ, Vol. 6,
No. 3, 2006: http://www.stata-journal.com/sjpdf.html?articlenum=st0108.
I'm not aware, though, of any papers on IV estimators that discuss
whether or why different weighting of the first and second stage would
make sense.

HTH,
Mark


> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> Daniel Brown
> Sent: 28 February 2012 12:55
> To: statalist@hsphsun2.harvard.edu
> Subject: st: Weighted 2SLS
> 
> Dear Statalist,
> 
> I have a question about the use of weights in 2SLS in a panel 
> context, which I believe the xtivreg2 command supports.
> 
> I am employing a 2SLS instrumental variables analysis to a 
> fixed effects regression where both my dependent variable and 
> my instrument are themselves calculated as an average across 
> underlying individuals - such that each observation for my 
> dependent variable and instrument are of different 
> reliability (where I know exactly how many observations have 
> been used to calculate them, and so the weight I would like 
> to attach to each observation - using the aweights command). 
> Intuitively, if I were to attempt to employ fixed effects 
> 2SLS, I thought that I would want to weight my first and 
> second stage regression differently - my first stage 
> regression taking into account the different reliability of 
> observations for my instrument, and the second stage taking 
> into account the different reliability of observations of my 
> dependent variable. Econometrically, I was not sure whether 
> applying weights in this way would undermine the role of the 
> first stage in dividing up exogenous from pot!
>  entially endogenous variation in my independent variable, or 
> whether this approach is acceptable. I am also unsure as to 
> whether the xtivreg2 command allows the use of a different 
> weight for each stage, and if not then whether the weight 
> command in xtivreg2 applies that weight to both stages of 2SLS.
> 
> Thanks,
> Dan Brown
> *
> *   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/
> 


-- 
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.

Heriot-Watt University is the Sunday Times
Scottish University of the Year 2011-2012



*
*   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/