R: Custom randomization test function to test variables in a data frame-CodePudding

I'm doing an assignment in R where I need to take a data frame with multiple variables and create a function() that resamples the absolute mean differences between the two categories in the data frame.

For the sake of my question I'll add an example data frame:

Variable 1	Variable 2	Variable 3	Category
1	2	3	1
4	5	6	1
7	8	9	2
10	11	12	2

The function needs to accept three arguments: a numeric vector, the two categories within the data frame, and nsim (number of times to resample randomly). The output should be a vector of length nsim with the resampled absolute mean differences.

This is the function I've tried, but when testing the output is always "Nan".

setseed(12345)
test<-function(x, category1, category2, nsim){
 resampled<-sample(df, size=length(nrow(df)), replace=F)
 category1.mean<-sum(df$x[resampled=="category1"])/length(df$x[resampled=="category1"])
 category2.mean<-sum(df$x[resampled=="category2"])/length(df$x[resampled=="category2"])
 return(abs(category1.mean-category2.mean)}

I'm not sure if I'm misunderstanding anything based on how function() works or if I'm misunderstanding the question or the data but I've tried a few things to try to fix the Nan output without success.

Can anyone help me out?

CodePudding user response：

The code below uses replicate to run nsim times the resampling and calculations function f.

x<-'Variable1   Variable2   Variable3   Category
1   2   3   1
4   5   6   1
7   8   9   2
10  11  12  2'
df1 <- read.table(textConnection(x), header = TRUE)

test <- function(data, x, category1, category2, nsim){
  f <- function(data, x, category1, category2) {
    i <- sample(nrow(data), replace = TRUE)
    d <- data[i, ]
    j1 <- which(d[["Category"]] == category1)
    j2 <- which(d[["Category"]] == category2)
    v1 <- d[j1, x, drop = TRUE]
    v2 <- d[j2, x, drop = TRUE]
    diff_means <- if(length(v1) == 0 & length(v2) == 0) {
      NaN
    } else if(length(v1) == 0) {
      mean(v2)
    } else if(length(v2) == 0) {
      mean(v1)
    } else mean(v1) - mean(v2)
    abs(diff_means)
  }
  replicate(nsim, f(data, x, category1, category2))
}

set.seed(2022)

# amd: absolute mean differences
amd <- test(df1, "Variable1", 1, 2, nsim = 1e3)
hist(amd)

^{Created on 2022-05-26 by the reprex package (v2.0.1)}