CODEWITHSUNDEEP
rfunctiondataframedplyrsummary
jonas
jonas

Reputation: 13969

Bootstrap CI for several variables of column in dataframe

I would like to bootstrap confidence intervals for a proportion from a data.frame. I would like to get the results for the variables in one of my columns. I have managed to perform the bootstrap for a vector but do not know how to scale it up to a data.frame from here. A simplified example setting a threshold value of 10 and looking at the proportion less than 10 in the data.

Vector solution:

library(boot)

vec <- abs(rnorm(1000)*10) #generate example vector

data_to_tb <- vec

tb <- function(data) {
  sum(data < 10, na.rm = FALSE)/length(data) #function for generating the proportion
}

tb(data_to_tb)

boot.out <- boot(data = data_to_tb, function(u,i) tb(u[i]),  R = 999)
quantile(boot.out$t, c(.025,.975))

And from here I would like to do the same for a data.frame containing two columns. I would like to return the result in a "summarized" data.frame if possible, with columns (x, sample, proportion, CI) :

x    n   proportion  CI

A    xx  xx          xx
B    xx  xx          xx
C    xx  xx          xx

Would be extra good if dplyr package could be used. Here is a simplified example of my data:

Example:

dataframe <- data.frame(x = sample(c("A","B","C"),100,replace = TRUE), vec =abs(rnorm(100)*10))

head(dataframe)
##   x        vec
## 1 B 0.06735163
## 2 C 0.48612358
## 3 B 2.34190635
## 4 C 0.36393262
## 5 A 7.99762969
## 6 B 1.43293330

Upvotes: 1

Views: 1446

Answers (1)

shadow
shadow

Reputation: 22323

You can use group_by and summarise from dplyr to achieve the desired result. See below for the code. 

# load required package
require(dplyr)
# function to calculate the confidence interval
CIfun <- function(v, probs = c(.025, .975)) {
  quantile(boot(data = v, function(u,i) tb(u[i]),  R = 999)$t, probs)
}
# using summarise from dplyr
dataframe %>% group_by(x) %>%
  summarise(n = n(), 
            proportion = tb(vec), 
            `2.5%` = CIfun(vec, .025), 
            `97.5%`= CIfun(vec, .975))

Upvotes: 5

Related Questions