prm_test: Perform a full permutation test

In holub008/nymph: Randomization tests for nonparametric inference

Description

This function should only be used for academic purposes or small sample sizes (time grows factorially with number of observations) If the "parallel" pacakge is installed, use the global mc.cores option to control level of parallelism

Usage

prm_test(data, statistics, group_var, statistic_names = names(statistics),
  max_observations = 10)

Arguments

data	a data frame to draw statistics from
statistics	a function or vector of functions accepting a data frame, returning a scalar statistic
group_var	a character, singular column name in the data representing groups.
statistic_names	the names of statistics that are computed. must be specified if the supplied statistics do not have names
trials	the number of Monte Carlo trials to perform

Author(s)

kholub

Examples

library(dplyr)
data(mpg)
test_data <- mpg %>% 
  filter(class %in% c('suv', 'compact')) %>%
  sample_n(7) # this just for computational feasibility of illustrative example
prm_res <- prm_test(test_data, c(median_difference = function(df) {
                                                                     group_medians <- df %>%
                                                                       group_by(class) %>%
                                                                       summarize(
                                                                         median = median(hwy))
                                                                     as.numeric(group_medians[group_medians$class == 'suv', 'median']) -
                                                                     as.numeric(group_medians[group_medians$class == 'compact', 'median'])
                                                                  }), 
                    'class')

holub008/nymph documentation built on June 4, 2019, 10:47 a.m.