The function gFWE performs different multiple test procedures for controlling
the generalized family-wise error rate (gFWE), i.e. the probability of rejecting at
least u+1 (default: u=0) true hypotheses is less than or equal to alpha=0.05.
[O] = gFWE(Input,n1,samp) returns the number of rejected hypotheses,
the rank (O(:,1)), the indices of the rejected hypotheses (0(:,2)) and the corresponding
p-values (O(:,3)) for the procedure of Troendle (1995) with the significance level
alpha=0.05. Note, the default is a procedure for controlling the family-wise error
rate, i.e. the probability of rejecting at least one true hypotheses is less
than or equal to alpha=0.05.
----------
INPUT
These input arguments are required:
Input: data matrix with the size [n,k]
n1: number of patients in group one (0 < n1 <= n ),
restricted by the kind of samp
samp: kind of sample
single sample 'single' (n1 = n)
paired sample 'paired' (n1 = n/2; n must be even)
independent sample 'indept' (n1 < n)
[...] = gFWE(...,'PARAM1',VAL1,'PARAM2',VAL2,...) specifies additional
parameters and their values. Valid parameters are the following:
Parameter Value
'u' number of accepted type 1 errors;
default: u = 0 (i.e. no type 1 error is accepted!)
u must be in the interval 0 <= u <= k/2
'test' Value for single sample
'ttest' to compute the t-Test
assumption : normal(gaussian) distribution
'wilcox' to compute the Wilcoxen signed rank test
assumption : symmetrical distribution
'sign' (the default) to compute the sign-test
assumption : none
Value for paired sample
'ttest' to compute the t-Test
assumption : normal(gaussian) distribution
'wilcox' to compute the Wilcoxen signed rank test
assumption : symmetrical distribution
'sign' (the default) to compute the sign-test
assumption : none
Value for independent sample
'ttest' to compute the t-Test
assumption : normal(gaussian) distribution
'wilcox' (the default) to compute the Wilcoxen rank test (Wilcoxen-Man-Whitney-Test)
assumption : none
'tail' The alternative hypothesis against which to compute
p-values for testing the hypothesis of no differences.
Choices are:
tail Alternative Hypothesis
'~=' (the default) "there is a significant difference" (two-sided test)
'>' "the values of group 1 are higher than the values of group 2" (one-sided test)
'<' "the values of group 1 are smaller than the values of group 2" (one-sided test)
'proc' Values for u > 0
'Av' (the default) chooses the procedure A
(conservative) of Korn et. al (2004)
'Ae' chooses the procedure A of Korn et al. (2004)
'TL' chooses the procedure of Troendle (1995) and the extention of van der Laan et al.
'HH' chooses the procedure of Hommel and Hoffmann (1987)
'HL' chooses the procedure of Holm and the extention of van der Laan et al.
---
Values for u = 0 (the default value for u)
'Av' (the default) chooses the procedure of Troendle (1995)
'Ho' chooses the procedure of Holm
'B' number of permutations (for procedures with permutation tests: Av; Ae; TL)
default: 500
B must be in the intervall
500 <= B <= 2^n1 for single and paired sample
(for 2^n1 < 500 : B = min(B,2^n1))
500 <= B <= n! / n1!*(n-1) for independent sample
(for n! / n1!*(n-n1)! < 500 : B = min(B,n!/n1!*(n-n1)!))
'alpha' 0.05 (the default) significance level
alpha must be a scalar and in the interval 0 < alpha <= 0.2
OUTPUT
[O] = gFWE(Input,n1,samp) returns the rank (O(:,1)),
the indices of the rejected hypotheses (O(:,2)) and
the adjusted p-values (O(:,3)).
-----------
REFERENCES
[1] Hemmelmann, C., Horn, M., S�e, T., Vollandt, R., Weiss, S. (2005):
New concepts of multiple tests and their use for evaluating
high-dimensional EEG data, Vol 142/2 pp 209-217.
Copyright (C) 2006 by Claudia Hemmelmann <claudia.hemmelmann@mti.uni-jena.de>
Institute of Medical Statistics, Computer Sciences and Documantation
University of Jena
This work was supported by DFG Project VO 683/2-1
This is part of the BIOSIG-toolbox http://biosig.sf.net/