Statalist


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

st: Correcting stanadard error in 2SLS


From   "Eva Poen" <eva.poen@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: Correcting stanadard error in 2SLS
Date   Fri, 29 Feb 2008 13:11:48 +0000

A friend of mine is having problems posting to statalist. I am sending
this on her behalf.

Eva

*************************************
From: Bidisha Sayema [lexshb@nottingham.ac.uk]

Dear Statalisters

I posted a mail few weeks before but havent got any reply. In
connection to that problem (pls see below) I am writing once again as
I am using bootstrap as a method of solution.

I am using 2 sample 2 stage least square method (Angrist & Krueger
1990). I would like to draw a bootstrap sample of my 1st sample (where
explanatory variable X and instruments Zs are present) and from that I
estimate the parametres needed to estimate the predicted value of X.
Then I want to draw the 2nd sample (which has our dependent variable Y
and instruments Z) and use that sample (and combine the parametres
from the 1st estimation) to generate a predicted value of X. With this
2nd sample I want to run the regression of Y on predicted value of X
and estimate the coefficients. I want to repeat all these steps 1000
times and want to estimate the SEs of the coefficients of our final
regression by the standard deviation of the bootstrap estimates.

My question is (very basic though) if I just run 2 separate bootstrap
regression of 1000 reps in both of the stages, will that give me the
result that I want to have (as outlined in last para)? My question is
if I do separate bootstrap regressions then will STATA  be able to
correct the bias in SE generated by using predicted value in the 2nd
stage regression?

Many thanks in advance.

Best Regards
Bidisha

From: Bidisha Sayema
Sent: 13 February 2008 00:12
To: statalist@hsphsun2.harvard.edu
Cc: Bidisha Sayema
Subject: FW: st: Correcting stanadard error in 2SLS


Thanks for your reply but although the method is intuitively an IV,
due to the nature of data I am not (infact I cannot) using ivreg. Both
stages involve OLS but in the 2nd stage as I am using predicted value
as a regresor I need to correct the SEs. I am wondering whether
bootstrapping the SEs would suffice the purpose and would produce
correct SEs. Any suggestion is appreciated.

Regards
Bidisha



--------------------------------------------------------------------------------
From: rraciborski@gmail.com [mailto:rraciborski@gmail.com] On Behalf
Of Rafal Raciborski
Sent: 12 February 2008 20:07
To: lexshb@nottingham.ac.uk
Subject: Re: st: Correcting stanadard error in 2SLS


use ivreg in stata 9 or ivregress in stata 10, the standard errors are
automatically calculated the correct way, no need to do it manually.
rafal


On Feb 12, 2008 2:27 PM, Bidisha Sayema <lexshb@nottingham.ac.uk> wrote:

Dear Satalisters,

I am using a two sample two stage least square method which essentially
involves getting predicted value of one of the explanatory variables
from one sample and then using that predicted value in another sample to
estimate the final regression. Therefore in my 1st sample I have the
explanatory variable X and instruments Zs and in 2nd sample I don't have
X but I have the dependent variable Y and same instrument set Z. In my
1st stage I estimate model of X while regressing X on Zs. In the 2nd
stage I use the coefficients of Zs obtained from the 1st stage
regression with the values of Zs available in the 2nd sample to get a
predicted value of Xhat. This Xhat is used as a regressor in my final
regression of Y. However as I am using predicted value of Xhat in my
final regression, I need to correct the standard errors manually. It
would be greatly appreciated if someone inform me how to do it in STATA.

Regards
Bidisha
*
*   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