*
* Example from section 7.1.2 starting page 169
*
open data cps_small.dat
data(format=free,org=columns) 1 1000 wage educ exper female black white midwest south west
*
set black_female = black*female
*
linreg wage
# constant educ black female black_female
exclude(title="Test of Demographics")
# black female black_female
*
linreg wage
# constant educ black female black_female south midwest west
*
exclude(title="Test of Regional Differences")
# south midwest west
*
* Sample break down
*
set educ_south = educ*south
set black_south = black*south
set female_south = female*south
set black_female_south = black*female*south
*
linreg wage
# constant educ black female black_female south educ_south black_south female_south black_female_south
exclude(title="Chow Test Using Dummies")
# south educ_south black_south female_south black_female_south
*
* The SMPL option restricts the regression to the set of observations where the
* value of the expression is non-zero. The expression .not.south will be 0
* ("false") where SOUTH is non-zero, and 1 ("true") where SOUTH is zero.
*
*
linreg(smpl=.not.south) wage
# constant educ black female black_female
*
* %NDF and %RSS are two variables defined by LINREG. %NDF is the degrees of
* freedom, %RSS is the sum of squared residuals. We need to keep these values for
* later use. To do this, we use the instruction COMPUTE. COMPUTE does general
* calculations with single values or matrices. Here compute ndf1=%ndf takes the
* current value of %ndf and puts it into a new variable called ndf1. As you can
* see, you can do more than one such assignment on a single COMPUTE.
*
compute ndf1=%ndf,rss1=%rss
*
linreg(smpl=south) wage
# constant educ black female black_female
compute ndf2=%ndf,rss2=%rss
*
linreg wage
# constant educ black female black_female
*
* CDF gives access to tables for four standard distributions: normal, chisquared,
* t (called "ttest") and F (called "ftest"). Given an input value for the test
* statistic and (if necessary) the degrees of freedom, it formats up a display of
* the test with its significance level. There are also functions which return the
* signficance values for tests using these statistics (and some others) in case
* you want to format the output differently.
*
cdf(title="Chow Test by Subsample Regression") ftest $
  ((%rss-(rss1+rss2))/%nreg)/((rss1+rss2)/(ndf1+ndf2)) %nreg ndf1+ndf2