dplyr sample_n from a single group

Question

I have some data where the summary of the number of observations looks like:

# A tibble: 14 x 3
# Groups:   status [2]
   status  year     n
    <dbl> <dbl> <int>
 1      0  2010  4593
 2      0  2011 10990
 3      0  2012 27711
 4      0  2013 99989
 5      0  2014 95407
 6      0  2015 89010
 7      0  2016 72289
 8      1  2010   584
 9      1  2011   785
10      1  2012   640
11      1  2013   667
12      1  2014   377
13      1  2015   460
14      1  2016   104

Where the class of one group is signficantly higher than the class of another group. How can I randomly sample the class of 0 without doing anything to the class of 1. That is, I would like to keep all class 1 observations and randomly sample the class 0 observations by 4593 (which is the minimum number of observations for that year)

Using group_by(status, year) and then sample_n() doesn't work since the 4593 value is greater than the values in the class 1 group.

Some random sample of my data:

    structure(list(status = c(0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1), 
    year = c(2013, 2014, 2012, 2013, 2016, 2013, 2015, 2014, 
    2013, 2016, 2015, 2016, 2011, 2014, 2016, 2012, 2013, 2012, 
    2014, 2014, 2012, 2012, 2012, 2016, 2016, 2012, 2016, 2015, 
    2013, 2014, 2015, 2013, 2015, 2015, 2014, 2015, 2011, 2014, 
    2013, 2012, 2011, 2016, 2015, 2015, 2015, 2014, 2012, 2013, 
    2015, 2012, 2015, 2016, 2015, 2013, 2014, 2014, 2014, 2013, 
    2013, 2016, 2016, 2013, 2015, 2012, 2014, 2014, 2013, 2015, 
    2014, 2016, 2016, 2014, 2012, 2016, 2013, 2010, 2011, 2014, 
    2016, 2013, 2016, 2014, 2014, 2013, 2013, 2013, 2016, 2016, 
    2012, 2014, 2013, 2015, 2016, 2013, 2013, 2015, 2013, 2014, 
    2013, 2015, 2013, 2013, 2011, 2014, 2016, 2013, 2010, 2012, 
    2014, 2012, 2011, 2011, 2013, 2015, 2014, 2010, 2010, 2013, 
    2010, 2014, 2011, 2011, 2014, 2013, 2014, 2015, 2015, 2013, 
    2014, 2013, 2011, 2013, 2014, 2013, 2011, 2013, 2012, 2015, 
    2012, 2012, 2012, 2010, 2013, 2013, 2011, 2011, 2011, 2012, 
    2016, 2013, 2011, 2011, 2012, 2012, 2014, 2010, 2013, 2014, 
    2011, 2012, 2010, 2012, 2012, 2011, 2015, 2011, 2011, 2013, 
    2015, 2010, 2015, 2011, 2015, 2015, 2012, 2012, 2013, 2012, 
    2014, 2014, 2012, 2012, 2014, 2010, 2011, 2013, 2014, 2012, 
    2013, 2016, 2014, 2012, 2012, 2013, 2010, 2012, 2013, 2014, 
    2014, 2011)), groups = structure(list(status = c(0, 1), .rows = structure(list(
    1:100, 101:200), ptype = integer(0), class = c("vctrs_list_of", 
"vctrs_vctr"))), row.names = c(NA, -2L), class = c("tbl_df", 
"tbl", "data.frame"), .drop = TRUE), row.names = c(NA, -200L), class = c("grouped_df", 
"tbl_df", "tbl", "data.frame"))

what is your desired output? also "randomly sample the class 0 observations by 4593" does that mean you want sample_n() where size = 4593? — EJJ
Apologies, the class refers to the status variable which has 2 clases 0 and 1. I would like to group_by or filter the class 0 variable and then take a random sample of these for each year. Using something like sample_n() with size = 4593 is what I am looking for yes, but this does not work when doing data %>% group_by(status, year) %>% sample_n(size = 4593) since it returns Error: size must be less or equal than 584 (size of data), set replace = TRUE to use sampling with replacement. Setting replace = TRUE doesn't give me the correct output either. — user8959427

www www · Accepted Answer · 2020-02-20T16:51:00

I think this will work. dat is your example data frame. The code below split the data frame by status, and then use imap to evaluate if sampling is needed. If the name of the list element is "0", it will conduct sampling. You can change the size = 1 to the minimum number of your real-world data frame.

library(dplyr)
library(purrr)

dat2 <- dat %>%
  split(f = .$status) %>%
  imap(function(x, y){
    if (y %in% "0"){
      x <- x %>% 
        group_by(status, year) %>%
        sample_n(size = 1) 
    }
    return(x)
  }) %>%
  bind_rows()

dplyr sample_n from a single group

1 Answers