resample (version 0.2)

bootstrap: One and two sample bootstrap sampling and permutation tests.

Description

Basic resampling. Supply the data and statistic to resample.

Usage

bootstrap(data, statistic, B = 1000,
          args.stat = NULL, seed = NULL, sampler = samp.bootstrap,
          label = NULL, statisticNames = NULL, block.size = 100,
          trace = FALSE)
bootstrap2(data, statistic, treatment, data2 = NULL, B = 1000,
          ratio = FALSE,
          args.stat = NULL, seed = NULL, sampler = samp.bootstrap,
          label = NULL, statisticNames = NULL, block.size = 100,
          trace = FALSE)
permutationTest(data, statistic, B = 999,
          alternative = "two.sided", resampleColumns = NULL,
          args.stat = NULL, seed = NULL, sampler = samp.permute,
          label = NULL, statisticNames = NULL, block.size = 100,
          trace = FALSE)
permutationTest2(data, statistic, treatment, data2 = NULL, B = 999,
          alternative = "two.sided", ratio = FALSE, paired = FALSE,
          args.stat = NULL, seed = NULL, sampler = samp.permute,
          label = NULL, statisticNames = NULL, block.size = 100,
          trace = FALSE)

Arguments

data

vector, matrix, or data frame.

statistic

a function, or expression (e.g. mean(myData, trim = .2).

B

number of replicates (bootstrap samples or permutation resamples).

treatment

a vector with two unique values. For two-sample applications, suppy either treatment or data2.

data2

an object like data; the second sample.

alternative

one of "two.sided", "greater", or "less". If statistic returns a vector, this may be a vector of the same length.

ratio

logical, if FALSE then statistics for two samples are combined using statistic1 - statistic2 (the statistics from the two samples). If TRUE, it uses statistic1 / statistic2.

resampleColumns

integer, or character (a subset of the column names of data); if supplied then only these columns of the data are permuted. For example, for a permutation test of the correlation of x and y, only one of the variables should be per

args.stat

a list of additional arguments to pass to statistic, if it is a function.

paired

logical, if TRUE then observations in data and data2 are paired, and permutations are done within each pair. Not yet implemented.

seed

old value of .Random.seed, or argument to set.seed.

sampler

a function for resampling, see help(samp.bootstrap).

label

used for labeling plots (in a future version).

statisticNames

a character vector the same length as the vector returned by statistic, giving names.

block.size

integer. The B replicates are done this many at a time.

trace

logical, if TRUE an indication of progress is printed.

Value

• a list with class that matches the name of the function (e.g. "bootstrap2"), that inherits from "resample", with components:
• observed
  the value of the statistic for the original data.
• replicates
  a matrix with B rows and p columns.
• n
  number of observations in the original data, or vector of length 2 in two-sample problems.
• p
  length(observed).
• B
  number of replications.
• seed
  the value of the seed at the start of sampling.
• call
  the matched call.
• statistics
  a data frame with p rows, with columns "observed", "mean" (the mean of the replicates), and other columns appropriate to resampling; e.g. the bootstrap objects have columns "SE" and "Bias", while the permutation test objects have "Alternative" and "PValue".
• There are functions for printing and plotting these objects, in particular print, plot, hist, qqnorm, quantile.

code

data2

itemize

• statistic = colMeans(data)

item

• statistic = mean(myData$x)
• statistic = mean(myData[, "x"])

Details

There is considerable flexibility in how you specify the data and statistic.

For the statistic, you may supply a function, or an expression. For example, if data = x, you may specify any of

• statistic = mean

statistic = mean(x) statistic = mean(data)

See Also

resample-package, samp.bootstrap, limits.percentile, limits.t.

Examples

mydata <- data.frame(a = runif(40), b = rnorm(40))
boot1 <- bootstrap(mydata, colMeans)
boot1
par(mfrow = c(2,1))
plot(boot1)
qqnorm(boot1)
par(mfrow = c(1,1))

boot2 <- bootstrap(mydata, mean(a))
boot2