Select with shortcuts {cherry} | R Documentation |

Calculates confidence limits for the number of true hypotheses on the basis of the closed testing procedure using specific local tests that allow shortcuts.

```
pickFisher (p, select = seq_along(p), alpha=0.05, silent=FALSE)
curveFisher (p, select = seq_along(p), order, alpha=0.05, plot = TRUE)
pickSimes (hommel, select, alpha=0.05, silent=FALSE)
curveSimes (hommel, select, order, alpha=0.05, plot = TRUE)
pickMeinshausen (p, PM, select = seq_along(p), alpha=0.05, silent=FALSE)
curveMeinshausen (p, PM, select = seq_along(p), order, alpha=0.05, plot = TRUE)
```

`p` |
The vector of p-values for all tested hypotheses. |

`hommel` |
The hommel object, obtained from the hommelFast function. |

`PM` |
A matrix of permutation p-values. Rows are hypotheses; columns are permutations. |

`select` |
The indexing vector of the p-values of the hypotheses to be selected. May be any type of appropriate indexing vector (integers, logical, or character). |

`order` |
The indexing vector specifying the order in which p-values of the hypotheses are to be rejected. May be integer or character. |

`alpha` |
The significance level of the test procedure. |

`silent` |
If |

`plot` |
If |

The results of the `pickFisher`

, `pickSimes`

and `pickMeinshausen`

functions are identical to applying `closed`

and `pick`

, for specific choices of the local test, but are computationally more efficient. `pickFisher`

uses local tests based on Fisher combinations. This local test is only valid if p-values of true hypotheses are independent. `pickSimes`

uses a local test based on Simes' inequality. It is valid if p-values of true hypotheses are independent but also under some forms of positive correlations. The Hommel variant of the Simes local test is valid under any dependence structure of the p-values. `pickMeinshausen`

is a permutation-based variant of `pickSimes`

. See the reference below.

In the curve functions, the user may specify either `select`

or `order`

. Specifying `order`

fixes the precise order in which hypotheses are selected, whereas specifying `select`

only specifies which hypotheses are candidates for selection, leaving the order to be chosen by the function to maximize the number of correct rejections.

For `pickFisher`

, `pickSimes`

and `pickMeinshausen`

, the function returns the lower confidence limit for the number of false hypotheses (correct rejection) among the set `reject`

. The upper confidence limit is always equal to the number of rejections made. `curveFisher`

and `curveSimes`

return the same confidence limit, but for selecting only the first 1,2,3,... hypotheses from the selected set.

Jelle Goeman: j.j.goeman@lumc.nl; Aldo Solari

Goeman and Solari (2011) Statistical Science 26 (4) 584-597.

Meinshausen (2006) Scandinavian Journal of Statistics 33 (2), 227-237.

```
# Fisher's method
data(NAEP)
pickFisher(NAEP, c("NH", "NC", "IA"))
pickFisher(NAEP, 1:7)
curveFisher(NAEP)
curveFisher(NAEP, order=7:1)
# Simes method
hom <- hommelFast(NAEP)
pickSimes(hom, c("NH", "NC", "IA"))
pickSimes(hom, 1:7)
curveSimes(hom)
curveSimes(hom, select=1:7)
# Meinshausen's method
# This example uses data from the multtest package on bioconductor
if(require("multtest")) {
data(golub)
smallglb <- golub[1:500,]
TM<-sapply(1:nrow(smallglb), function(i) {
mt.sample.teststat(smallglb[i,], golub.cl,test="t.equalvar",B=500)
})
PM<-2*(1-pt(abs(TM),df=length(golub.cl)-2)) # permutation matrix
# p-values
p<-apply(smallglb,1, function(z) t.test(z[golub.cl==0],z[golub.cl==1],var.equal=TRUE)$p.value)
pickMeinshausen(p, PM, select=1:100)
pickMeinshausen(p, PM, select=sort.list(p)[1:100])
curveMeinshausen(p,PM, select=1:200)
curveMeinshausen(p,PM, order=1:200)
}
```

[Package *cherry* version 0.6-14 Index]