Professional Documents
Culture Documents
Tukey Kramer
Tukey Kramer
Tukey Kramer
i j
L L
12-6
Example: Widths of 95% Acceptance
Intervals Increasing with K in TukeyKramer
Test (n
1
=n
2
==n
K
=10)
12-7
Example of TukeyKramer Test
(Example 12.2 in ISSO)
Goal: With K = 4, test null hypothesis L(u
1
) = L(u
2
) =
L(u
3
) = L(u
4
) based on 10 measurements at each u
i
All (six) differences o
ij
must lie in acceptance
intervals [1.23, 1.23]
Find that o
34
= 1.72
Have o
34
e [1.23, 1.23]
Since at least one o
ij
is not in acceptance interval, reject
null hypothesis
Conclude at least one u
i
likely better than others
Further analysis required to find u
i
that is better
i j
L L
12-8
Multiple Comparisons Against One Candidate
Assume prior information suggests one of K points is
optimal, say u
m
Reduces number of comparisons from K(K1)/2 differences
o
ij
= to only K1 differences o
mj
Under null hypothesis, L(u
m
) > L(u
j
) for all j
Aim to reject null hypothesis
Implies that L(u
m
) < L(u
j
) for at least some j
Tests based on critical values < 0 for observed
differences o
mj
To show that L(u
m
) < L(u
j
) for all j requires additional
analysis
i j
L L
o
mj
12-9
Example of Many-to-One Test with
Known Variances (Example 12.3 in ISSO)
Suppose K = 4, m = 2 Need to compute 3 critical
values , , and for acceptance regions
Valid to take
Under Bonferroni/Chebyshev:
Under Bonferroni/normal noise:
Under Slepian/normal noise:
Note tighter (smaller) acceptance regions when assuming
normal noise
21
o
23
o
24
o
24 23 21
o = o = o o
96 . 3 = o
10 . 1 = o
09 . 1 = o
} {
2 2 j j
o > o
Widths of 95% Acceptance Intervals (< 0)
for Tukey-Kramer and Many-to-One Tests
(n
1
=n
2
==n
K
=10)
12-11
Ranking and Selection:
Indifference Zone Methods
Consider usual problem of determining best of K possible
options, represented u
1
, u
2
,, u
K
Have noisy loss measurements y
k
(u
i
)
Suppose analyst is willing to accept any u
i
such that L(u
i
)
is in indifference zone [L(u
-
), L(u
-
) + o)
Analyst can specify o such that
P(correct selection of u = u
-
) > 1 o
whenever L(u
i
) L(u
-
) > o for all u
i
= u
-
Can use independent sampling or common random
numbers (see Section 14.5 of ISSO)