*
* Example from page 275, section 8.8
* Chow test
*
cal 1970
open data table8-9.prn
data(format=prn,org=columns) 1970:1 1995:1
table
*
* Chow test for sample split at 1981:1
*
linreg savings 1970:1 1981:1
# constant income
compute ndf1=%ndf,rss1=%rss
linreg savings 1982:1 1995:1
# constant income
compute ndf2=%ndf,rss2=%rss
linreg savings 1970:1 1995:1
# constant income
compute ndfr=%ndf,rssr=%rss
compute rssu=rss1+rss2,ndfu=ndf1+ndf2
compute fchow=((rssr-rssu)/%nreg)/(rssu/ndfu)
*
cdf(title="Chow Test for Split at 1981:1") ftest fchow %nreg ndfu
*
* Test for constant variance. Note that, unlike most F tests, this is a
* two-tailed test: either a small value or a large value should lead to
* rejection. The cdf instruction produces a significance level (p-value) for the
* more typical one-tailed test. To get a two-tailed significance (for H0: same
* variance vs H1: variances aren't the same), multiply small p-values by 2, and
* transform large ones to 2x(1-stated pvalue). For instance if you exchange the
* roles of samples 2 and 1 in the formulas below, you would get a p-value of
* .98972. The two-tailed test significance would be 2x(1-.98972), i.e. you would
* reject same variances at the .05 level.
*
cdf(title="Test of equal variances") ftest (rss2/ndf2)/(rss1/ndf1) ndf2 ndf1