kfwe: Controlling the Generalized Familywise Error Rate

In someKfwer: Controlling the Generalized Familywise Error Rate

Description

This library collects some procedures controlling the Generalized Familywise Error Rate: Lehmannn and Romano (2005), Guo and Romano (2007) (single step and stepdown), Finos and Farcomeni (2009).

Usage

kfweLR(p, k = 1, alpha = 0.01, disp = TRUE)
kfweGR(p, k = 1, alpha = 0.01, disp = TRUE,
SD=TRUE, const = 10, alpha.prime = getAlpha(k = k, s = length(p), 
                    alpha = alpha, const = const))
kfweOrd(p, k = 1, alpha = 0.01, ord = NULL, 
    alpha.prime = alpha, J = qnbinom(alpha,k,alpha.prime), 
    disp = TRUE, GD=FALSE)
                                                  
getAlpha (s, k = 1, alpha = 0.01, const = 10)

Arguments

p	vector of p-values of length s
s	number of p-values (i.e. hypotheses)
k	number of allowed errors in kFWE controls
alpha	global significance level
ord	the vector of values based on which the p-values have to be ordered
const	Bigger is better (more precise but slower)
J	number of allowed jumps befor stopping
disp	diplay output? TRUE/FALSE
SD	Step-down version of the procedure? (TRUE/FALSE) the step-down version is uniformly more powerful than the single step one.
alpha.prime	univariate alpha for single step Guo and Romano procedure
GD	Logic value. Should the correction for general dependence be applied? (See reference below for further details)

Value

kfweOrd, kfweLR, kfweGR, kfweGR.SD return a vector of kFWE-adjusted p-values. It respect the order of input vector of p-values p.

getAlpha returns the alpha for Guo and Romano procedure.

Author(s)

L. Finos and A. Farcomeni

References

For Lehmann and Romano procedure see:

Lehmann and Romano (2005) Generalizations of the Familywise Error Rate, Annals of Statistics, 33, 1138-1154.

For Guo and Romano procedure see:

Guo and Romano (2007) A Generalized Sidak-Holm procedure and control of genralized error rates under independence, Statistical Applications in Genetics and Molecular Biology, 6, 3.

For Ordinal procedure see:

Finos and Farcomeni (2010) k-FWER control without multiplicity correction, with application to detection of genetic determinants of multiple sclerosis in Italian twins. Biometrics (Articles online in advance of print: DOI 10.1111/j.1541-0420.2010.01443.x)

Examples

set.seed(13)
y=matrix(rnorm(3000),3,1000)+2                      #create toy data
p=apply(y,2,function(y) t.test(y)$p.value)          #compute p-values
M2=apply(y^2,2,mean)                                #compute ordering criterion

kord=kfweOrd(p,k=5,ord=M2)                          #ordinal procedure
kgr=kfweGR(p,k=5)                                   #Guo and Romano

kord=kfweOrd(p,k=5,ord=M2,GD=TRUE)                  #ordinal procedure (any dependence)
klr=kfweLR(p,k=5)                                   #Lehaman and Romano (any dependence)

Example output

Ordered k-FWER procedure
 1000 tests, k=5, alpha=0.01, individual alpha threshold=0.01
 125 jumps allowed
 32 rejections

Guo and Romano k-FWER Step Down procedure
 1000 tests, k=5, alpha=0.01
 0.0013057 individual alpha threshold
 23 rejections

Ordered k-FWER procedure
 1000 tests, k=5, alpha=0.01, individual alpha threshold=0.0003846
 125 jumps allowed
 0 rejections

Lehmann e Romano k-FWER Step Down procedure
 1000 tests, k=5, alpha=0.01
 1 rejections

someKfwer documentation built on May 1, 2019, 10:19 p.m.