Bootstrap confidence intervals for more than one statistics using boot.ci function
I want to get the loading confidence intervals for more than one statistic using a function boot.ci
. Here is my MWE. I have two statistics in out
and want to find bootstrap confidence intervals for these two statistics. However, the function only boot.ci
provides bootstrap confidence intervals for the first statistic (t1 *) and not for the second statistic (t2 *). Any help would be much appreciated. Thanks to
set.seed(12345)
df <- rnorm(n=10, mean = 0, sd = 1)
Boot.fun <-
function(data, idx) {
data1 <- sample(data[idx], replace=TRUE)
m1 <- mean(data1)
sd1 <- sd(data1)
out <- cbind(m1, sd1)
return(out)
}
Boot.fun(data = df)
library(boot)
boot.out <- boot(df, Boot.fun, R = 20)
boot.out
RDINARY NONPARAMETRIC BOOTSTRAP
Call:
boot(data = df, statistic = Boot.fun, R = 20)
Bootstrap Statistics :
original bias std. error
t1* -0.4815861 0.3190424 0.2309631
t2* 0.9189246 -0.1998455 0.2499412
boot.ci(boot.out=boot.out, conf = 0.95, type = c("norm", "basic", "perc", "bca"))
BOOTSTRAP CONFIDENCE INTERVAL CALCULATIONS
Based on 20 bootstrap replicates
CALL :
boot.ci(boot.out = boot.out, conf = 0.95, type = c("norm", "basic",
"perc", "bca"))
Intervals :
Level Normal Basic
95% (-1.2533, -0.3479 ) (-1.1547, -0.4790 )
Level Percentile BCa
95% (-0.4842, 0.1916 ) (-0.4842, -0.4629 )
Calculations and Intervals on Original Scale
Warning : Basic Intervals used Extreme Quantiles
Some basic intervals may be unstable
Warning : Percentile Intervals used Extreme Quantiles
Some percentile intervals may be unstable
Warning : BCa Intervals used Extreme Quantiles
Some BCa intervals may be unstable
Warning messages:
1: In norm.inter(t, (1 + c(conf, -conf))/2) :
extreme order statistics used as endpoints
2: In norm.inter(t, alpha) : extreme order statistics used as endpoints
3: In norm.inter(t, adj.alpha) :
extreme order statistics used as endpoints
source to share
The package boot
is (IMO) a little awkward for regular use. The short answer is that you need to specify index
(the default is 1) on boot.ci
eg. boot.ci(boot.out,index=2)
... The long answer is that it would be very convenient to get the initial CIs for all bootstrap statistics!
Get the entire CI for a given result slot:
getCI <- function(x,w) {
b1 <- boot.ci(x,index=w)
## extract info for all CI types
tab <- t(sapply(b1[-(1:3)],function(x) tail(c(x),2)))
## combine with metadata: CI method, index
tab <- cbind(w,rownames(tab),as.data.frame(tab))
colnames(tab) <- c("index","method","lwr","upr")
tab
}
## do it for both parameters
do.call(rbind,lapply(1:2,getCI,x=boot.out))
Results (maybe not what you want, but easy to change):
index method lwr upr
normal 1 normal -1.2533079 -0.3479490
basic 1 basic -1.1547310 -0.4789996
percent 1 percent -0.4841726 0.1915588
bca 1 bca -0.4841726 -0.4628899
normal1 2 normal 0.6288945 1.6086459
basic1 2 basic 0.5727462 1.4789105
percent1 2 percent 0.3589388 1.2651031
bca1 2 bca 0.6819394 1.2651031
Alternatively, if you can work with one bootstrap method, my version of the package broom
on Github has this capability (I submitted a pull request)
## devtools::install_github("bbolker/broom")
library(broom)
tidy(boot.out,conf.int=TRUE,conf.method="perc")
source to share