bootstrap by group in tibble

Question

Suppose I have a tibble tbl_

tbl_ <- tibble(id = c(1,1,2,2,3,3), dta = 1:6)
tbl_
# A tibble: 6 x 2
     id   dta
  <dbl> <int>
1     1     1
2     1     2
3     2     3
4     2     4
5     3     5
6     3     6

There are 3 id groups. I want to resample entire id groups 3 times with replacement. For example the resulting tibble can be:

     id   dta
  <dbl> <int>
1     1     1
2     1     2
3     1     1
4     1     2
5     3     5
6     3     6

but not

     id   dta
  <dbl> <int>
1     1     1
2     1     2
3     1     1
4     2     4
5     3     5
6     3     6

or

     id   dta
  <dbl> <int>
1     1     1
2     1     1
3     2     3
4     2     4
5     3     5
6     3     6

Why `id = 1,1,2,2,3,3` case is not a valid option? – MKR Apr 29 '18 at 18:42 — MKR, Apr 29 '18 at 18:42
Because second row is 1 1, when it has to be 1 2. – stollenm Apr 29 '18 at 18:46 — stollenm, Apr 29 '18 at 18:46

score 2 · Accepted Answer · answered Apr 30 '18 at 00:26

Here is one option with sample_n and distinct

library(tidyverse)
distinct(tbl_, id) %>% 
    sample_n(nrow(.), replace = TRUE) %>% 
    pull(id) %>% 
    map_df( ~ tbl_ %>% 
         filter(id == .x)) %>%
    arrange(id)
# A tibble: 6 x 2
#     id   dta
#  <dbl> <int>
#1  1.00     1
#2  1.00     2
#3  1.00     1
#4  1.00     2
#5  3.00     5
#6  3.00     6

MKR · Answer 2 · 2018-04-29T19:10:46.030

An option can be to get the minimum row number for each id. That row number will be used to generate random samples from wiht replace = TRUE.

library(dplyr)

tbl_ %>% mutate(rn = row_number()) %>%
  group_by(id) %>%
  summarise(minrow = min(rn)) ->min_row

indx <- rep(sample(min_row$minrow, nrow(min_row), replace = TRUE), each = 2) + 
        rep(c(0,1), 3)

tbl_[indx,]
# # A tibble: 6 x 2
#    id     dta
#   <dbl>  <int>
# 1  1.00     1
# 2  1.00     2
# 3  3.00     5
# 4  3.00     6
# 5  2.00     3
# 6  2.00     4

Note: In the above answer the number of rows for each id has been assumed as 2 but this answer can tackle any number of IDs. The hard-coded each=2 and c(0,1) needs to be modified in order to scale it up to handle more than 2 rows for each id

bootstrap by group in tibble

2 Answers2