# permp

##### Exact permutation p-values

Calculates exact p-values for permutation tests with permutations sampled with replacement.

- Keywords
- htest

##### Usage

`permp(x, nperm, n1, n2, total.nperm=NULL, method="auto", twosided=TRUE)`

##### Arguments

- x
- numeric vector or array giving the number of permutations that yield test statistics at least as extreme as that observed.
- nperm
- number of permutations performed.
- n1
- sample size of group 1. Not required if
`total.nperm`

is supplied. - n2
- sample size of group 2. Not required if
`total.nperm`

is supplied. - total.nperm
- total number of permutations allowable from the design of the experiment.
- method
- computation method, possible values are
`"exact"`

,`"approximate"`

or`"auto"`

. - twosided
- logical, is the test two-sided so that the test statistic is symmetric between the two groups?

##### Details

This function can be used for calculating exact p-values for permutation tests where permutations are sampled with replacement, using theory and methods developed by Phipson and Smyth (2010).
`total.nperm`

is the total number of distinct values of the test statistic that are possible.
This is generally equal to the number of possible permutations, unless a two-sided test is conducted with equal sample sizes, in which case `total.nperm`

is half the number of permutations, because the test statistic must then be symmetric in the two groups.
`total.nperm`

can be supplied directly by the user, alternatively it will be computed from `n1`

and `n2`

.
When `method="exact"`

, the p-values are computed to full machine precision by evaluating a summation.
When `method="approximate"`

, an approximation is used that is faster and uses less memory.
If `method="auto"`

, the exact calculation is used when `total.nperm`

is less than or equal to 10,000, otherwise the approximation is used.

##### Value

- vector or array of p-values, of same dimensions as
`x`

##### References

Phipson, B and Smyth, GK (2010). Exact permutation p-values when permutations are randomly drawn. Technical Report, Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia.

##### Examples

```
# Consider a two-sided permutation test with 99 permutations,
# 5 reaching the statistically significant threshold.
# Assume a two group experiment with 6 in each group.
# Input total.nperm=462
permp(x=5, nperm=99, total.nperm=462)
# Input n1=6 and n2=6
permp(x=5, nperm=99, n1=6, n2=6)
# Suppose we have a vector of p-values taking on the values 0 to 10 for the same experiment described above.
x<-0:10
permp(x=x, nperm=99, total.nperm=462)
```

*Documentation reproduced from package statmod, version 1.4.14, License: LGPL (>= 2)*