Summary statistics for multiple variables with statistics as rows and variables as columns?

2
votes

I'm trying to use dplyr::summarize() and dplyr::across() to obtain a tibble with several summary statistics in the rows and the variables in the columns. I was only able to achieve this result by using dplyr::bind_rows(), but I'm wondering if there's a more elegant way to get the same output.

> library(tidyverse)
── Attaching packages ────────────────────────────────────────────── tidyverse 1.3.1 ──
✔ ggplot2 3.3.3     ✔ purrr   0.3.4
✔ tibble  3.1.1     ✔ dplyr   1.0.6
✔ tidyr   1.1.3     ✔ stringr 1.4.0
✔ readr   1.4.0     ✔ forcats 0.5.1
── Conflicts ───────────────────────────────────────────────── tidyverse_conflicts() ──
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag()    masks stats::lag()
> 
> bind_rows(min = summarize(starwars, across(where(is.numeric), min, 
+       na.rm = TRUE)), 
+   median = summarize(starwars, across(where(is.numeric), median, 
+       na.rm = TRUE)), 
+   mean = summarize(starwars, across(where(is.numeric), mean, na.rm = TRUE)), 
+   max = summarize(starwars, across(where(is.numeric), max, na.rm = TRUE)), 
+   sd = summarize(starwars, across(where(is.numeric), sd, na.rm = TRUE)), 
+   .id = "statistic")
# A tibble: 5 x 4
  statistic height   mass birth_year
  <chr>      <dbl>  <dbl>      <dbl>
1 min         66     15          8  
2 median     180     79         52  
3 mean       174.    97.3       87.6
4 max        264   1358        896  
5 sd          34.8  169.       155.

Why can't one do it with summarize directly? Seems more elegant than using a list of functions, as suggested by the colwise vignette. Does this violate the principles of a tidy data frame? (It seems to me that staking a bunch of data frames besides one another is far less tidy.)

4
rdplyrtidyversesummarizeacross
Each row in your desired output is defined differently. Row 1 is the min, row 2 is the median, etc. So while this may be convenient to work with, you wouldn't operate on entire columns of that output (e.g., you wouldn't sum over height). So I am not sure that output is considered tidy. The way summarize does it of giving you the wide output is probably more "tidy", but I understand why you want to work with it that way. A lot is philosophy and just understanding what you are trying to do with the data. – Adam
That is an excellent point. Do you think it'd be tidier if I had variables in the rows and statistics on the columns? That'd also be fine for the purposes of presentation. – Lucas De Abreu Maia
You want to store and work with data in a tidy format. For presentation, do what communicates it the best. What you have is probably fine for presentation. I wouldn't stress too much about having your presentation tables "tidy". – Adam

4 Answers

6
votes

Here is a way using purrr to iterate over a list of functions. This is effectively what you were doing with bind_rows(), but in less code.

library(dplyr)
library(purrr)

funs <- lst(min, median, mean, max, sd)

map_dfr(funs,
        ~ summarize(starwars, across(where(is.numeric), .x, na.rm = TRUE)),
        .id = "statistic")

# # A tibble: 5 x 4
#   statistic height   mass birth_year
#   <chr>      <dbl>  <dbl>      <dbl>
# 1 min         66     15          8  
# 2 median     180     79         52  
# 3 mean       174.    97.3       87.6
# 4 max        264   1358        896  
# 5 sd          34.8  169.       155.
2
votes

This resolves in the output you want, but it's not that fancy.

starwars %>% 
  summarise(across(
    where(is.numeric),
    .fns = list(
       min = min,
       median = median, 
       mean = mean, 
       max = max, 
       sd = sd
    ), 
    na.rm = TRUE, 
    .names = "{.col}_{.fn}")) %>% 
  pivot_longer(cols = everything()) %>% 
  mutate(statistic = str_match(name, pattern = ".+_(.+)")[,2],
         name = str_match(name, pattern = "(.+)_.+")[,2]) %>% 
  pivot_wider(names_from = name, values_from = value)
1
votes

You could use gtsummary to summarize the data. Below I subset to numeric columns (although gtsummary handles many different data types. Then I tell the type argument to put my summary stats on different rows and finally tell the statistics argument which summaries I want to display.

library(tidyverse)
library(gtsummary)

starwars[sapply(starwars, is.numeric)] %>% 
    tbl_summary(type = all_continuous() ~ "continuous2",
                statistic = all_continuous() ~ c("{median} ({p25}, {p75})",
                                                 "{min}, {max}",
                                                 "{mean},{sd}"))
1
votes

I would do it this way:

starwars %>%
    summarise(across(where(is.numeric), stat_funs,
        na.rm = TRUE, .names = "{.col}__{.fn}")) %>%
    pivot_longer(everything()) %>%
    separate(name, c('v', 'f'), sep = '__') %>%
    pivot_wider(names_from = v, values_from = value)

#  f      height   mass birth_year
#   <chr>   <dbl>  <dbl>      <dbl>
# 1 min      66     15          8  
# 2 median  180     79         52  
# 3 mean    174.    97.3       87.6
# 4 max     264   1358        896  
# 5 sd       34.8  169.       155.