Y603
Lectures Online
Lecture
#9
1. Nonorthogonal Posteriori
Comparisons
|
Other
restrictions
|
Testing
all pairwise contrasts
|
Testing
pairwise contrasts of ordered
means
|
|
None
|
Tukey's
test
Fisher-Hayter
test
|
Newman-Keuls
Test
|
2. Fisher-Hayter
Procedure
It is a two-step
procedure; suitable for all pairwise contrasts
Controls the familywise
type-I error rate at or below 
Uses the same distribution as
the Tukey procedure, i.e., Table E.6 on pages
808-809
Requires the sample sizes be
equal (a harmonic average of different n's may be
substituted)
At Step One, the
null-hypothesis of equal means should be
rejected
At Step Two, a
pairwise comparison is judged to be significant if it
exceeds the MSD:
MSD = 
3. Newman-Keuls
Procedure
Pairwise comparisons by
ordered means
Equal n required (a harmonic
average of different nís may be
substituted)
Uses the same distribution as
the Tukey procedure, i.e., Table E.6 on pages
808-809
When the number of groups is
greater than 3, the familywise type I error rate can
exceed the nominal level of .05 (or the value
designated by a researcher).
It is generally not
recommended due to its fuzzy definition of the alpha
level.
SAS commands for
carrying out the Newman-Keuls
Procedure:
PROC GLM;
CLASS
treat;
MODEL
score=treat;
MEANS TREAT/SNK
ALPHA=.10;
4. Assignments:
(1) Review Chapter 4 in
Kirk
(2) Read the article by
Seaman, Levin, and Serlin (1991) in Psychological
Bulletin: "New development in pairwise multiple
comparisons: Some powerful and practicable
procedures." 110(3), 577-586. [on reserve
in the Ed. Library]
(3) Do questions 13, 14, 15,
16, 19(a), 19(b), 19(c), 19(d), apply the Newman-Keuls
procedure to the same data, and 19(f) of Chapter 4 in
Kirk.
(4) Preview Sections 9.1-9.4
in Kirk
