Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Biometrika Trust

A Comparison of Two Modified Bonferroni Procedures


Author(s): Gerhard Hommel
Source: Biometrika, Vol. 76, No. 3 (Sep., 1989), pp. 624-625
Published by: Biometrika Trust
Stable URL: http://www.jstor.org/stable/2336131 .
Accessed: 17/06/2014 21:38

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Biometrika Trust is collaborating with JSTOR to digitize, preserve and extend access to Biometrika.

http://www.jstor.org

This content downloaded from 188.72.126.181 on Tue, 17 Jun 2014 21:38:34 PM


All use subject to JSTOR Terms and Conditions
Biometrika (1989), 76, 3, pp. 624-5
Printed in Great Britain

A comparisonof two modified Bonferroni procedures


BY GERHARD HOMMEL
Institut fur Medizinische Statistik und Dokumentation, University of Mainz, D-6500 Mainz,
Federal Republic of Germany

SUMMARY

It is shown that Hommel's (1988) modified Bonferroni procedure is at least as powerful as


Hochberg's (1988) procedure, and, in general, more powerful.
Some key words: Familywise error rate; Modified Bonferroni procedure; Multiple test procedure; Strong
control.

Let there be n null hypotheses H1,..., H, with corresponding p-values PI, .-. , P". A multiple
test procedure is said to control strongly the familywise error rate a, if the probability of a false
rejection of any Hi (i = 1, . . ., n) is at most a, irrespective of which and how many of the Hi are
true (Hochberg & Tamhane, 1987, pp. 3, 7).
Let P(O), . . . , P(n) be the ordered p-values. If I is any nonempty subset of {1, . . ., n} with II =r,
we denote by P(i,), . . ., P(i,) the ordered p-values Pi, ic L The Simes test of HI =(n {Hi: i c I} is
defined by the rule 'reject HI, if P(i,) <ka/r for at least one k (1 < ks r)'. Simes (1986) proved
that this test is a level a test if the p-values Pi, i c I, are independent, and he suggested by
simulation that it is a level a test if the p-values are based on a large variety of multivariate
distributions.
We consider the following two procedures.
Procedure 1 (Hochberg, 1988). Reject Hi if there exists aj (1 S j S n) such that P(j) S a/(n -j + 1)
and Pi - P(j).
Procedure 2 (Hommel, 1988). Let
J = {i' c {1, . .. , n}: P(n-i'+k) > ka/ i'; k = 1, ..., i'}.
If J is nonempty, reject Hi whenever Pi -c a/j' with j' = max J. If J is empty, reject all
Hi (i = 1, ... ., n)-
In the above papers it has been shown that both procedures control strongly the familywise
error rate a provided each Simes test of HI is a level a test. The advantage of Procedure 1 is
that the computation is simpler than that of Procedure 2. Except for simplicity, however, we show
that Procedure 2 should be preferred.
LEMMA.If a hypothesis Hi is rejectedby Procedure 1, it is also rejectedby Procedure2. For n > 2,
there are situations where Procedure 2 rejects and Procedure 1 does not.
Proof. If Hi is rejected by Procedure 1, there is aj with Pi <P(J)-a/(n-j+ 1). If j=n, all
Hi (i = 1, .. ., n) are rejected by both procedures. Assume that j < n.
If i' and k are such that nB i'- n-j+1 and k=j+i'-n, then 1z ks i' and
P(n-i'+k) = aljs x(n -j + 1) -, kal (k - 1 +n -j + 1) = kal i'.
Therefore j' = max J - n -j, and Hi is also rejected by Procedure 2 because Pi < a/(n -j).
To show the second pa;t, we assume that the p-values fulfill:
a(n-j + 1)<P(j)1a, (j ., n-2);

This content downloaded from 188.72.126.181 on Tue, 17 Jun 2014 21:38:34 PM


All use subject to JSTOR Terms and Conditions
Miscellanea 625
Then, by Procedure 1 no Hi is rejected, whereas Procedure 2 rejects all Hi corresponding to
P() 5... ., P(n_2), because of j' = 2. 0I
For both procedures the strong control of the familywise error rate a is ensured even if the
free combinations condition (Holm, 1979) is not satisfied. This can be shown by means of the
'modified closure principle' described by Hommel (1988).

REFERENCES
HOCHBERG, Y. (1988). A sharper Bonferroni procedure for multiple tests of significance. Biometrika 75,
800-2.
HOCHBERG, Y. & TAMHANE, A. C. (1987). Multiple Comparison Procedures.New York: Wiley.
HOLM, S. (1979). A simple sequentially rejective multiple test procedure. Scand. J. Statist. 6, 65-70.
HOMMEL, G. (1988). A stagewise rejective multiple test procedure based on a modified Bonferroni test.
Biometrika 75, 383-6.
SIMES, R. J. (1986). An improved Bonferroni procedure for multiple tests of significance. Biometrika73, 751-4.

[Received January 1989. Revised March 1989]

This content downloaded from 188.72.126.181 on Tue, 17 Jun 2014 21:38:34 PM


All use subject to JSTOR Terms and Conditions

You might also like