- logit PðXÞ¼aþb 1 Fþb 2 BASEþg 1 POST
þg 2 PFþg 3 OCCþg 4 AGEþg 5 GEN
þd 1 FBASEþd 2 FPOST
þd 3 FPFþd 4 FOCCþd 5 FAGE
þd 6 FGENþd 7 BASEPOST
þd 8 BASEPFþd 9 BASEOCC
þd 10 BASEAGEþd 11 BASEGEN
- F 3 POST:EV
F 3 BASE:EE
POST 3 BASE:EV
PF 3 POST:Neither
BASE 3 PF:EV
- Choicescanddare reasonable.
Choiceais not reasonable because the two
nonsignificant chunk tests do not imply that the
overall chunk test is nonsignificant; in fact, the overall
chunk test was significant.
Choicebis not reasonable because the corresponding
chunk test wasmoresignificant than the chunk test
for interaction terms involving BASE.
- Using the model of question 2 ,
H 0 :d 2 ¼d 3 ¼d 4 ¼d 5 ¼d 6 ¼0.
LR¼2lnLReduced(2lnLFull)approxw^25
underH 0 ,
where the Full Model is the model of question 2 , and
the Reduced Model does not contain the product
termsF 3 POST, F 3 PF, F 3 OCC, F 3 AGE, and
FGEN.
- PF, AGE,andGENare eligible to be dropped from the
model. These variables areVvariables that are not
components of the three product terms found
significant from interaction assessment.
- No. The variablesPOSTandOCCare lower order
components of the product termsF 3 POSTand
BASE 3 OCCfound significant, and are therefore to
be retained in the model (from the hierarchy
principle). Tests of significance forPOSTandOCC
will depend on the coding of the variables in the
model, so such tests are inappropriate, since they
should be independent of coding.
- a. logit PðXÞ¼aþb 1 Fþb 2 BASEþg 1 POSTþg 2 PF
þg 3 OCCþg 4 AGEþg 5 GEN
þd 1 FBASEþd 2 FPOST
þd 9 BASEOCC
b. X*¼(3, 1, POST, PF, OCC, AGE, GEN), whereas
X¼(2, 0, POST, PF, OCC, AGE, GEN),
so
OR¼exp½b 1 þb 2 þ 3 d 1 þd 2 POSTþd 9 OCC
Chapter 8 677
