How to generate bootstrapping samples in R?

Question

Here is my original dataset from a voting activity. Each participant voted for one option (A,B,C,D,E,F) listed below. The numerical value is the total number of participants who voted for each option.

I want to bootstrap the voting for 1000 times (sample with replacement) and make a comparison between the pre-event and post-event voting for each category using independent sample t-test.

However, I don't know how to start with generating the 1000 bootstrapping samples...I have the "Boot" packages installed on R. Could anyone give me a hint what kind of steps I should take in order to generate 1000 bootstrap samples based on the original data I provide above?

Answer 1

Creating samples

You can create samples using sample and replicate functions. I create smp with the first given row.

> smp <- c(rep('A', 8), rep('B', 4), rep('C', 2), rep('D', 11), rep('E', 21), rep('F', 0))
> table(smp)
smp
 A  B  C  D  E 
 8  4  2 11 21 
> 
> samples <- replicate(1000, sample(smp)) # creating 1000 samples
> dim(samples) # 1000 samples of 46 votes
[1]   46 1000
> 
> samples[,1:3]
      [,1] [,2] [,3]
 [1,] "E"  "A"  "E" 
 [2,] "D"  "E"  "E" 
 [3,] "E"  "D"  "C" 
 [4,] "D"  "E"  "E" 
 [5,] "A"  "D"  "D" 
 [6,] "B"  "E"  "E" 
 [7,] "E"  "A"  "D" 
 [8,] "B"  "C"  "D" 
 [9,] "B"  "A"  "E" 
[10,] "E"  "E"  "E" 
[11,] "A"  "E"  "E" 
[12,] "D"  "E"  "D" 
[13,] "D"  "E"  "B" 
[14,] "D"  "B"  "E" 
[15,] "E"  "E"  "D" 
[16,] "E"  "E"  "E" 
[17,] "E"  "B"  "B" 
[18,] "A"  "D"  "E" 
[19,] "E"  "E"  "E" 
[20,] "E"  "E"  "C" 
[21,] "E"  "E"  "E" 
[22,] "E"  "D"  "A" 
[23,] "D"  "D"  "B" 
[24,] "E"  "E"  "E" 
[25,] "D"  "D"  "A" 
[26,] "E"  "E"  "D" 
[27,] "C"  "A"  "E" 
[28,] "D"  "E"  "D" 
[29,] "D"  "E"  "E" 
[30,] "D"  "A"  "E" 
[31,] "A"  "D"  "E" 
[32,] "E"  "E"  "A" 
[33,] "E"  "E"  "D" 
[34,] "E"  "E"  "A" 
[35,] "A"  "E"  "E" 
[36,] "D"  "C"  "A" 
[37,] "A"  "D"  "A" 
[38,] "E"  "D"  "D" 
[39,] "E"  "D"  "E" 
[40,] "A"  "E"  "A" 
[41,] "B"  "A"  "A" 
[42,] "C"  "A"  "D" 
[43,] "A"  "A"  "D" 
[44,] "E"  "B"  "E" 
[45,] "E"  "D"  "B" 
[46,] "E"  "B"  "E" 

If all you need are samples, sample solves your problem.

Bootstraping

To bootstrap you need to compute a statistic. For example, I compute the weighted mean for a table of votes. I'm using boot::boot to bootsprap. You must pass the original sample and a handler which receives the original sample s and a vector with the indexes shuffled (idx). This function returns a boot object that shows the bootstrap statistics.

> boot.smp <- boot::boot(smp, function(s, idx) {
+ tt <- table(s[idx])
+ weighted.mean(tt, as.numeric(factor(names(tt), labels=1:6, levels=LETTERS[1:6])))
+ }, 1000)
> boot.smp

CASE RESAMPLING BOOTSTRAP FOR CENSORED DATA


Call:
boot::boot(data = smp, statistic = function(d, w) {
    tt <- table(d[w])
    weighted.mean(tt, as.numeric(factor(names(tt), labels = 1:6, 
        levels = LETTERS[1:6])))
}, R = 1000)


Bootstrap Statistics :
    original    bias    std. error
t1*     11.4 0.4126077    1.274827

The boot object has other methods to compute confidence interval and return much more information.