How to get summary statistics by group

85
votes

I'm trying to get multiple summary statistics in R/S-PLUS grouped by categorical column in one shot. I found couple of functions, but all of them do one statistic per call, like aggregate().

data <- c(62, 60, 63, 59, 63, 67, 71, 64, 65, 66, 68, 66, 
          71, 67, 68, 68, 56, 62, 60, 61, 63, 64, 63, 59)
grp <- factor(rep(LETTERS[1:4], c(4,6,6,8)))
df <- data.frame(group=grp, dt=data)
mg <- aggregate(df$dt, by=df$group, FUN=mean)    
mg <- aggregate(df$dt, by=df$group, FUN=sum)

What I'm looking for is to get multiple statistics for the same group like mean, min, max, std, ...etc in one call, is that doable?

12
rs
This one is a pretty basic question with multiple answers. You may not be familiar with RSeek (LINK) and the sos library (LINK) Both are great resources to help you figure out the answers to questions. Ibet with those resources you'll be able to answer your own question in seconds. – Tyler Rinker
There's an extra comma at the end of the data <- c( line. – BenBarnes
I just found a wonderful R package tables. You can tabulate data by as many categories as you desire and calculate multiple statistics for multiple variables - it truly is amazing! But wait, there's more! The package has functions to generate LaTeX code for your tables for easy import to your documents. – StatGrrl

12 Answers

123
votes

1. tapply

I'll put in my two cents for tapply().

tapply(df$dt, df$group, summary)

You could write a custom function with the specific statistics you want or format the results:

tapply(df$dt, df$group,
  function(x) format(summary(x), scientific = TRUE))
$A
       Min.     1st Qu.      Median        Mean     3rd Qu.        Max. 
"5.900e+01" "5.975e+01" "6.100e+01" "6.100e+01" "6.225e+01" "6.300e+01" 

$B
       Min.     1st Qu.      Median        Mean     3rd Qu.        Max. 
"6.300e+01" "6.425e+01" "6.550e+01" "6.600e+01" "6.675e+01" "7.100e+01" 

$C
       Min.     1st Qu.      Median        Mean     3rd Qu.        Max. 
"6.600e+01" "6.725e+01" "6.800e+01" "6.800e+01" "6.800e+01" "7.100e+01" 

$D
       Min.     1st Qu.      Median        Mean     3rd Qu.        Max. 
"5.600e+01" "5.975e+01" "6.150e+01" "6.100e+01" "6.300e+01" "6.400e+01"

2. data.table

The data.table package offers a lot of helpful and fast tools for these types of operation:

library(data.table)
setDT(df)
> df[, as.list(summary(dt)), by = group]
   group Min. 1st Qu. Median Mean 3rd Qu. Max.
1:     A   59   59.75   61.0   61   62.25   63
2:     B   63   64.25   65.5   66   66.75   71
3:     C   66   67.25   68.0   68   68.00   71
4:     D   56   59.75   61.5   61   63.00   64
52
votes

dplyr package could be nice alternative to this problem:

library(dplyr)

df %>% 
  group_by(group) %>% 
  summarize(mean = mean(dt),
            sum = sum(dt))

To get 1st quadrant and 3rd quadrant 

df %>% 
  group_by(group) %>% 
  summarize(q1 = quantile(dt, 0.25),
            q3 = quantile(dt, 0.75))
36
votes

Using Hadley Wickham's purrr package this is quite simple. Use split to split the passed data_frame into groups, then use map to apply the summary function to each group. 

library(purrr)

df %>% split(.$group) %>% map(summary)
19
votes

There's many different ways to go about this, but I'm partial to describeBy in the psych package:

describeBy(df$dt, df$group, mat = TRUE)
12
votes

take a look at the plyr package. Specifically, ddply

ddply(df, .(group), summarise, mean=mean(dt), sum=sum(dt))
9
votes

after 5 long years I'm sure not much attention is going to be received for this answer, But still to make all options complete, here is the one with data.table

library(data.table)
setDT(df)[ , list(mean_gr = mean(dt), sum_gr = sum(dt)) , by = .(group)]
#   group mean_gr sum_gr
#1:     A      61    244
#2:     B      66    396
#3:     C      68    408
#4:     D      61    488
6
votes

Besides describeBy, the doBy package is an another option. It provides much of the functionality of SAS PROC SUMMARY. Details: http://www.statmethods.net/stats/descriptives.html

6
votes

The psych package has a great option for grouped summary stats:

library(psych)
    
describeBy(dt, group="grp")

produces lots of useful stats including mean, median, range, sd, se.

5
votes

While some of the other approaches work, this is pretty close to what you were doing and only uses base r. If you know the aggregate command this may be more intuitive. 

with( df , aggregate( dt , by=list(group) , FUN=summary)  )
2
votes

Not sure why the popular skimr package hasn’t been brought up. Their function skim() was meant to replace the base R summary() and supports dplyr grouping:

library(dplyr)
library(skimr)

starwars %>%
  group_by(gender) %>%
  skim()

#> ── Data Summary ────────────────────────
#>                            Values    
#> Name                       Piped data
#> Number of rows             87        
#> Number of columns          14        
#> _______________________              
#> Column type frequency:               
#>   character                7         
#>   list                     3         
#>   numeric                  3         
#> ________________________             
#> Group variables            gender    
#> 
#> ── Variable type: character ──────────────────────────────────────────────────────
#>    skim_variable gender    n_missing complete_rate   min   max empty n_unique
#>  1 name          feminine          0         1         3    18     0       17
#>  2 name          masculine         0         1         3    21     0       66
#>  3 name          <NA>              0         1         8    14     0        4
#>  4 hair_color    feminine          0         1         4     6     0        6
#>  5 hair_color    masculine         5         0.924     4    13     0        9
#>  6 hair_color    <NA>              0         1         4     7     0        4
#> # [...]
#> 
#> ── Variable type: list ───────────────────────────────────────────────────────────
#>   skim_variable gender    n_missing complete_rate n_unique min_length max_length
#> 1 films         feminine          0             1        9          1          5
#> 2 films         masculine         0             1       24          1          7
#> 3 films         <NA>              0             1        3          1          2
#> 4 vehicles      feminine          0             1        3          0          1
#> 5 vehicles      masculine         0             1        9          0          2
#> 6 vehicles      <NA>              0             1        1          0          0
#> # [...]
#> 
#> ── Variable type: numeric ────────────────────────────────────────────────────────
#>   skim_variable gender    n_missing complete_rate  mean     sd    p0   p25   p50
#> 1 height        feminine          1         0.941 165.   23.6     96 162.  166. 
#> 2 height        masculine         4         0.939 177.   37.6     66 171.  183  
#> 3 height        <NA>              1         0.75  181.    2.89   178 180.  183  
#> # [...]
1
votes

First, it depends on your version of R. If you've passed 2.11, you can use aggreggate with multiple results functions(summary, by instance, or your own function). If not, you can use the answer made by Justin.

0
votes

this may also work,

spl <- split(mtcars, mtcars$cyl)
list.of.summaries <- lapply(spl, function(x) data.frame(apply(x[,3:6], 2, summary)))
list.of.summaries