mt.sample.teststat {multtest}R Documentation

Permutation distribution of test statistics and raw (unadjusted) p-values

Description

These functions provide tools to investigate the permutation distribution of test statistics, raw (unadjusted) p-values, and class labels.

Usage

mt.sample.teststat(V,classlabel,test="t",fixed.seed.sampling="y",B=10000,na=.mt.naNUM,nonpara="n")
mt.sample.rawp(V,classlabel,test="t",side="abs",fixed.seed.sampling="y",B=10000,na=.mt.naNUM,nonpara="n")
mt.sample.label(classlabel,test="t",fixed.seed.sampling="y",B=10000)

Arguments

V A numeric vector containing the data for one of the variables (genes).
classlabel A vector of integers corresponding to observation (column) class labels. For k classes, the labels must be integers between 0 and k-1. For the blockf test option, observations may be divided into n/k blocks of k observations each. The observations are ordered by block, and within each block, they are labeled using the integers 0 to k-1.
test A character string specifying the statistic to be used to test the null hypothesis of no association between the variables and the class labels.
If test="t", the tests are based on two-sample Welch t-statistics (unequal variances).
If test="t.equalvar", the tests are based on two-sample t-statistics with equal variance for the two samples. The square of the t-statistic is equal to an F-statistic for k=2.
If test="wilcoxon", the tests are based on standardized rank sum Wilcoxon statistics.
If test="f", the tests are based on F-statistics.
If test="pairt", the tests are based on paired t-statistics. The square of the paired t-statistic is equal to a block F-statistic for k=2.
If test="blockf", the tests are based on F-statistics which adjust for block differences (cf. two-way analysis of variance).
side A character string specifying the type of rejection region.
If side="abs", two-tailed tests, the null hypothesis is rejected for large absolute values of the test statistic.
If side="upper", one-tailed tests, the null hypothesis is rejected for large values of the test statistic.
If side="lower", one-tailed tests, the null hypothesis is rejected for small values of the test statistic.
fixed.seed.sampling If fixed.seed.sampling="y", a fixed seed sampling procedure is used, which may double the computing time, but will not use extra memory to store the permutations. If fixed.seed.sampling="n", permutations will be stored in memory. For the blockf test, the option n was not implemented as it requires too much memory.
B The number of permutations. For a complete enumeration, B should be 0 (zero) or any number not less than the total number of permutations.
na Code for missing values (the default is .mt.naNUM=--93074815.62). Entries with missing values will be ignored in the computation, i.e., test statistics will be based on a smaller sample size. This feature has not yet fully implemented.
nonpara If nonpara="y", nonparametric test statistics are computed based on ranked data.
If nonpara="n", the original data are used.

Value

For mt.sample.teststat, a vector containing B permutation test statistics.

For mt.sample.rawp, a vector containing B permutation unadjusted p-values.

For mt.sample.label, a matrix containing B sets of permuted class labels. Each row corresponds to one permutation.

Author(s)

Yongchao Ge, yongchao.ge@mssm.edu,
Sandrine Dudoit, http://www.stat.berkeley.edu/~sandrine.

See Also

mt.maxT, mt.minP, golub.

Examples


# Gene expression data from Golub et al. (1999)
data(golub)

mt.sample.label(golub.cl,B=10)

permt<-mt.sample.teststat(golub[1,],golub.cl,B=1000)
qqnorm(permt)
qqline(permt)

permt<-mt.sample.teststat(golub[50,],golub.cl,B=1000)
qqnorm(permt)
qqline(permt)

permp<-mt.sample.rawp(golub[1,],golub.cl,B=1000)
hist(permp)

[Package multtest version 1.8.0 Index]