Dear listers,
I am exploring currently the solution for Chow test in SPSS given in
http://listserv.uga.edu/cgi-bin/wa?A2=ind0110&L=spssx-l&P=R11527 .
There are 3 methods of computing F-statistics for Chow test suggested:
"pen and paper" (after running pooled and separate regressions), REGRESSION
command with interaction, and UNIANOVA. I reproduce these solutions below
after simulating some data for illustration.
My concern is that "pen and paper" and REGRESSION solutions give similar
F-statistics
(and p-values) for Chow test (0.419), whereas UNIANOVA solution gives a bit
different p-value (0.483).
Well, if I drop the 'group' effect from the REGRESSION solution (leaving
only 'grpx' - interaction effect),
the p-value for Chow test will be the same as in UNIANOVA (0.483).
My question than is: which approach for Chow test is correct? Or, if they
are all correct, which p-value one would choose?
Thanks in advance,
Anton
*-----------------Simulate some data.
INPUT PROGRAM.
SET SEED=123.
LOOP i=1 to 150.
COMPUTE y=rv.norm(5,5).
COMPUTE x=y*2+5+rv.norm(0,10).
COMPUTE group=RND(rv.uniform(0,1)).
END CASE.
END LOOP.
END FILE.
END INPUT PROGRAM.
EXE.
*-----------------'Pen and paper' approach for Chow test.
REGRESSION
/MISSING LISTWISE
/STATISTICS COEFF OUTS R ANOVA
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT y
/METHOD=ENTER x .
temporary.
select if group=0.
REGRESSION
/MISSING LISTWISE
/STATISTICS COEFF OUTS R ANOVA
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT y
/METHOD=ENTER x .
temporary.
select if group=1.
REGRESSION
/MISSING LISTWISE
/STATISTICS COEFF OUTS R ANOVA
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT y
/METHOD=ENTER x .
COMPUTE pval = SIG.F(0.876297,2,146) .
EXECUTE .
*-----------------REGRESSION approach for Chow test.
COMPUTE grpx = x * group .
REGRESSION
/DESCRIPTIVES MEAN STDDEV CORR SIG N
/MISSING LISTWISE
/STATISTICS COEFF OUTS CI R ANOVA END CHA
/CRITERIA=PIN(.05) POUT(.10)
/NOORIGIN
/DEPENDENT y
/METHOD=ENTER x
/METHOD=ENTER group grpx.
*-----------------UNIANOVA approach for Chow test.
UNIANOVA
y BY group WITH x
/METHOD = SSTYPE(3)
/INTERCEPT = INCLUDE
/CRITERIA = ALPHA(.05)
/DESIGN = x group*x.
Locked
