Statalist


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

st: Testing Equality of Coefficients from two regressions


From   "Kemal Turkcan" <kturkcan@akdeniz.edu.tr>
To   statalist@hsphsun2.harvard.edu
Subject   st: Testing Equality of Coefficients from two regressions
Date   Wed, 23 Apr 2008 12:53:01 +0300

Dear Statalisters,
I have the following problem but couldnt solved it quite sometime. I have a 
panel data set consists of 30 countries and 17 years. I have performed 
several tests (such as Hausman test, Breush-Pagan test, autocorrelation 
test, and and heteroskedasticity test) that indicate XTGLS is correct 
procedure in my case. But I have realized that XTGLS postestimation can not 
produce e(RSS) which is necessary to test (Chow's F test) the equality of 
coefficients across two different groups (in my case: developed and 
developing countries).  XTGLS postestimation can produce e(b) and e(V) which 
can be used to calculate the WALD test,asymtotically very similar to Chow's 
test. Something like that:
	W =  (RB-r)'(R*V*R')(Rb-r) 
The problem is how I can create a restriction matrix "R" when I try to 
compare the sets of all coefficients from two regressions.

The model I am estimating is:
xtgls LIIT LAVGDP LDIFGDP LAVGDPP LDIFGDPP LDIST LMFDIOUT LEXCH, panels
(hetero) corr(ar1) igls force

I tried to use xtreg,re procedure to calculate Chow test as follows:
xtreg LIIT LAVGDP LDIFGDP LAVGDPP LDIFGDPP LDIST LMFDIOUT LEXCH if 
core==1,fe vce(robust)
est store core
gen residualcore=e(rss)
gen ncore=e(N)

xtreg LIIT LAVGDP LDIFGDP LAVGDPP LDIFGDPP LDIST LMFDIOUT LEXCH if 
periphery==1,fe vce(robust)
est store periphery
gen residualperiphery=e(rss)
gen nperiphery=e(N)

xtreg LIIT LAVGDP LDIFGDP LAVGDPP LDIFGDPP LDIST LMFDIOUT LEXCH,fe vce
(robust)
est store total
gen residualtotal=e(rss)

gen k=7

gen F1=residualtotal-(residualcore+residualperiphery)/k
gen F2=(residualcore+residualperiphery)/((ncore+nperiphery-2)*k)

But this procedure is not correct when you have autocorrelation and 
heteroskedasticity problem. Can I use REG procedure to calculate FGLS with 
panel data? If I can, then I can easily compare coefficients using SUEST.
Any help wilp be appreaciated,
Kemal Turkcan

*
*   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