Y603
Lectures Online
Lecture #8
1. Nonorthogonal Planned and
Posteriori Comparisons
|
Other
restrictions
|
Testing
(p-1) contrasts with a control-group
mean
|
Testing
C contrasts
|
|
none
|
Dunnett's
test
|
Dunn, Dunn-
test
Holmís
test
|
|
Unequal n's or
heterogeneous
variances
|
Dunnett's test
with
modifications
|
Dunn-
test with Welch df
Holmís test with
Welch df
|
|
Other
restrictions
|
Testing
all pairwise contrasts
|
Testing
pairwise contrasts of ordered
means
|
|
None
|
Tukey's test
Fisher-Hayter
test
|
Newman-Keuls
Test
|
2. Holmís sequentially
rejective Bonferroni Test
Is a step-down version of
the Dunn procedure
First it rank-orders the
magnitude of the test statistic 
(p. 143) based on Dunnís procedure
Tests the largest 
with a critical t value at the a
FW/C level. In other words, tests the largest
mean difference with Dunn-Sidakís MSD
where
is obtained with C=total number of
comparisons.
Tests the second largest
with a critical t value at the a
FW/(C-1) level. Or alternatively, tests the
second largest mean difference with Dunn-Sidakís
MSD where
is obtained with C-1 comparisons.
Continues this process until no
significant mean difference is found or the smallest mean
difference is tested with the old-fashioned
Studentís t with df=
and C=1.
Controls the familywise type-I
error rate at a level less than  .
This procedure has not been
implemented by SAS yet. So you are back to the old,
by-hand method!
3. Dunnettís Multiple
Comparison Test (Section 4.3 in Kirk)
Always tests (p-1)
pairwise comparisons
Comparisons are formed between
(p-1) group means against a control
Controls the type-I error rate
at the familywise level
Requires all sample sizes be
equal (a harmonic average of different nís may be
substituted)
Declares a pair-wise comparison
to be significant if it exceeds the following critical
difference:
MSD=
, where 
critical value is obtained from Table E.7 (p.
810-811)
This test can be carried out as
a two-tailed test as well as a one-tailed test by
SAS.
SAS commands for carrying
out the two-tailed Dunnettís
Procedure:
PROC GLM;
CLASS treat;
MODEL
score=treat;
MEANS treat/DUNNETT
('4') ALPHA=.10;
SAS commands for carrying
out the one-tailed Dunnettís
Procedure:
PROC GLM;
CLASS treat;
MODEL
score=treat;
MEANS treat/DUNNETTL
('4') ALPHA=.10;
or
MEANS treat/DUNNETTU
('4') ALPHA=.10;
4. Assignments:
(1) Review Sections 4.4
and 4.5 in Kirk.
(2) Apply Holmís
procedure to questions 2,3,4,5,7, and 8 from Chapter 5 in
Kirk, assume C=number of all pairwise
comparisons.
(3) Apply the Dunnett procedure
to questions 2,3,4,5,7, and 8 from Chapter 5 in Kirk. For
questions 3, 5, 7, and 8, conduct a two-tailed Dunnett
test using the first group as the control group. For
questions 2 and 4, assume the control group is the last
group and carry out a one-tailed test.
(4) Preview Sections 4.5, 4.7,
and 4.8 in Kirk.
|
