*
* Specification Tests for Klein's Model I from pp 414-415
*
open data tablef15-1[1].txt
calendar 1921
data(format=prn,org=columns) 1921:1 1942:1 year c p wp in k1 x wg g tax
set totwage = wp+wg
set trend   = (t-1932:1)
*
* Specification tests based upon 2SLS. There is already a specification test
* (called the J test) which is included in the regression output, which will
* generally be similar numerically to the Hausman-Wu test.
*
* Note that the degrees of freedom for the test in all equations should be 4 (8
* instruments less 4 regressors in all cases), not the 2 or 3 cited in the book.
* The 2 or 3 would be the degrees of freedom for a Hausman/Wu test.
*
instruments constant trend wg g tax p{1} x{1} k1
*
linreg(inst) c
# constant p{0 1} totwage
linreg(noprint) %resids
# constant trend wg g tax p{1} x{1} k1
cdf(title="Overidentification Test") chisqr %trsq 4
*
linreg(inst) in
# constant p{0 1} k1
linreg(noprint) %resids
# constant trend wg g tax p{1} x{1} k1
cdf(title="Overidentification Test") chisqr %trsq 4
*
linreg(inst) wp
# constant x x{1} trend
linreg(noprint) %resids
# constant trend wg g tax p{1} x{1} k1
cdf(title="Overidentification Test") chisqr %trsq 4
*
* LIML. The specification test is included in the output from the LIML procedure.
*
@liml c
# constant p{0 1} totwage
@liml in
# constant p{0 1} k1
@liml wp
# constant x x{1} trend