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 at the end of May, and its replacement, statalist.org is already up and running.


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

st: RE: ivreg2 weak-id statistic and quadratic terms


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: RE: ivreg2 weak-id statistic and quadratic terms
Date   Mon, 20 Feb 2012 22:35:34 -0000

Hi Miroslav, hi all.

I've checked this with the toy auto dataset.  I can replicate this
behaviour.

Miroslav - either before or after rescaling your covariates, do the
estimated coefficients vary hugely in scale?

In my toy auto dataset example, I am pretty sure that it is driven by
scaling problems.  For example, after

sysuse auto, clear
gen double weight2=weight^2
ivreg2 price (mpg=turn) weight weight2

gives a large weak ID stat of 11.5.  But there are big scaling problems
in the first-stage and main estimations, with coeffs that are something
like a factor of 10^8 different in magnitude.

If I estimate and just partial out the constant,

ivreg2 price (mpg=turn) weight weight2, partial(_cons)

the ill-conditioning is less pronounced and I get a weak ID stat of
0.73.

If I partial out all the exogenous covariates,

ivreg2 price (mpg=turn) weight weight2, partial(weight weight2)

the ill-conditioning is gone and I again get a weak ID stat of 0.73.

I will investigate further and will report back to the list if I find
anything more.  It may be that -ivreg2- could handle these cases more
robustly.

--Mark (ivreg2 coauthor)

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Miros Lav
> Sent: 20 February 2012 21:25
> To: statalist@hsphsun2.harvard.edu
> Subject: st: ivreg2 weak-id statistic and quadratic terms
> 
> Dear all,
> 
> I am using ivreg2 to estimate a model where a control 
> variable enters with a quadratic term. A simplified version 
> of the command is as follows
> 
> ivreg2 y   (a=instrument)  x x^2, r cluster(id).
> 
> Estimating this model results in a very large 
> Kleinbergen-Paap weak-id F statistic.
> 
> However, generating z=x/1000 and z^2=z*z and estimating the model
> 
> ivreg2 y   (a=instrument)  z z^2, r cluster(id)
> 
> results in a very low Kleinbergen-Paap weak-id F statistic.
> 
> (The z-statistics and significance levels in the first and 
> second stage regressions are the same as in the previous model.)
> 
> Does anyone have an idea why these two equivalent models 
> result in very different Kleinbergen-Paap weak-id F statistic?
> 
> Thanks for your help!
> 
> Miroslav
> *
> *   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/


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