The R Function of the Day series will focus on describing in plain language how certain R functions work, focusing on simple examples that you can apply to gain insight into your own data.
Today, I will discuss the tapply function.
What situation is tapply useful in?
In statistics, one of the most basic activities we do is computing summaries of variables. These summaries might be as simple as an average, or more complex. Let’s look at some simple examples.
When you read the results of a medical trial, you will see things such as “The average age of subjects in this trial was 55 years in the treatment group, and 54 years in the control group.”
As another example, let’s look at one from the world of baseball.
Batting Leaders per Team
Team | Player | Batting Average |
---|---|---|
Minnesota Twins | Joe Mauer | .374 |
Seattle Mariners | Ichiro Suzuki | .355 |
Boston Red Sox | Kevin Youkilis | .309 |
… | … | … |
These two examples have a lot in common, even if they don’t appear to when first reading. In the first example, we have a dataset from a medical trial. We want to break up the dataset into two groups, treatment and control, and then compute the sample average for age within each group.
In the second example, we want to break up the dataset into 30 groups, one for each MLB team, and then compute the maximum batting average within each group.
So what is in common?
In each case we have
- A dataset that can be broken up into groups
- We want to break it up into groups
- Within each group, we want to apply a function
The following table summarizes the situation.
Example | Group Variable | Summary Variable | Function |
---|---|---|---|
Medical Example | Treatment | age | mean |
Baseball Example | Team | batting average | max |
The tapply function can solve both of these problems for us!
How do I use tapply?
The tapply function is simple to use. First, we will generate some data.
> ## generate data for medical example > medical.example <- data.frame(patient = 1:100, age = rnorm(100, mean = 60, sd = 12), treatment = gl(2, 50, labels = c("Treatment", "Control"))) > summary(medical.example) patient age treatment Min. : 1.00 Min. : 29.40 Treatment:50 1st Qu.: 25.75 1st Qu.: 54.31 Control :50 Median : 50.50 Median : 61.24 Mean : 50.50 Mean : 61.29 3rd Qu.: 75.25 3rd Qu.: 66.22 Max. :100.00 Max. :102.47 > ## generate data for baseball example > ## 5 teams with 5 players per team > > baseball.example <- data.frame(team = gl(5, 5, labels = paste("Team", LETTERS[1:5])), player = sample(letters, 25), batting.average = runif(25, .200, .400)) > summary(baseball.example) team player batting.average Team A:5 a : 1 Min. :0.2172 Team B:5 c : 1 1st Qu.:0.2553 Team C:5 d : 1 Median :0.2854 Team D:5 e : 1 Mean :0.2887 Team E:5 f : 1 3rd Qu.:0.3013 g : 1 Max. :0.3859 (Other):19
Now we have some sample data. Using tapply is now straightforward. In general, the call to the function will look like the example in the first comment. Then, actual calls to the function using the data we defined above are shown.
> ## Generic Example > ## tapply(Summary Variable, Group Variable, Function) > > ## Medical Example > tapply(medical.example$age, medical.example$treatment, mean) Treatment Control 62.26883 60.30371 > ## Baseball Example > tapply(baseball.example$batting.average, baseball.example$team, max) Team A Team B Team C Team D Team E 0.3784396 0.3012680 0.3488655 0.2962828 0.3858841
Summary of tapply
The tapply function is useful when we need to break up a vector into groups defined by some classifying factor, compute a function on the subsets, and return the results in a convenient form. You can even specify multiple factors as the grouping variable, for example treatment and sex, or team and handedness.
[…] in this case the longest run of heads is 9 and the longest run of tails is 8. The tapply function was discussed in a previous R Function of the Day […]
I would also suggest looking at the plyr and reshape packages on CRAN.
> library(plyr)
> ddply(baseball.example, .(team), summarise, max = max(batting.average))
team max
1 Team A 0.3850099
2 Team B 0.3899625
3 Team C 0.3784054
4 Team D 0.3616533
5 Team E 0.3554805
> library(reshape)
> recast(baseball.example, . ~team, max)
Using team, player as id variables
value Team A Team B Team C Team D Team E
1 (all) 0.3850099 0.3899625 0.3784054 0.3616533 0.3554805
That’s great! I did take a look at Hadley’s plyr package when it was in its infancy, maybe about a year ago. I think it’s come a long way since then, and it is definitely on my list of things to check out.
By the way, Hadley seems to have a ton of interesting projects on his github page at http://github.com/hadley , I think a lot of them are not very mature yet, but he has some really interesting ideas for extending R!
Baseball is the most interesting game on the planet. I’m excited about the new season. Should be fantastic.
OK, so how do you do two grouping variables, not just one?
Simply pass a list as the second argument. For example,
tapply(df$var1, list(df$var2, df$var3), mean)
You might also be interested in the plyr package for breaking up data into subsets and summarizing them.