Difference between tilde and "by" while using aggregate function in R

Question

Every time I do an aggregate on a data.frame I default to using the "by = list(...)" parameter. But I do see solutions on stackoverflow and elsewhere where tilde (~) is used in the "formula" parameter. I kinda see the "by" parameter as the "pivot" around these variables.

In some cases, the output is exactly the same. For example:

aggregate(cbind(df$A, df$B, df$C), FUN = sum, by = list("x" = df$D, "y" = df$E))

AND

aggregate(cbind(df$A, df$B, df$C) ~ df$E, FUN = sum)

What is the difference between the two and when do you use which?

Answer 1

I would not entirely disagree that it doesn't really matter which approach you use, however, it is important to note that they do behave differently.

I'll illustrate with a small example.

Here's some sample data:

set.seed(1)
mydf <- data.frame(A = c(1, 1, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4),
                   B = LETTERS[c(1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2)],
                   matrix(sample(100, 36, replace = TRUE), nrow = 12))
mydf[3:5] <- lapply(mydf[3:5], function(x) {
  x[sample(nrow(mydf), 1)] <- NA
  x
})
mydf
#    A B X1  X2 X3
# 1  1 A 27  69 27
# 2  1 A 38  NA 39
# 3  1 A 58  77  2
# 4  2 A 91  50 39
# 5  2 A 21  72 87
# 6  3 B 90 100 35
# 7  3 B 95  39 49
# 8  3 B 67  78 60
# 9  3 B 63  94 NA
# 10 4 B NA  22 19
# 11 4 B 21  66 83
# 12 4 B 18  13 67

First, the formula interface. The following three commands will all yield the same output.

aggregate(cbind(X1, X2, X3) ~ A + B, mydf, sum)
aggregate(cbind(X1, X2, X3) ~ ., mydf, sum)
aggregate(. ~ A + B, mydf, sum)
#   A B  X1  X2  X3
# 1 1 A  85 146  29
# 2 2 A 112 122 126
# 3 3 B 252 217 144
# 4 4 B  39  79 150

Here's a related command for the "by" interface. Pretty cumbersome to type (but that can be addressed by using with, if required).

aggregate(cbind(mydf$X1, mydf$X2, mydf$X3), 
          by = list(mydf$A, mydf$B), sum)
  Group.1 Group.2  V1  V2  V3
1       1       A 123  NA  68
2       2       A 112 122 126
3       3       B 315 311  NA
4       4       B  NA 101 169

Now, stop and make note of any differences.

The two that pop into my mind are:

1. The formula method does a nicer job of preserving names but it doesn't let you control the names directly in your command, which you can do in the data.frame method:

   aggregate(cbind(NewX1 = mydf$X1, NewX2 = mydf$X2, NewX3 = mydf$X3), 
             by = list(NewA = mydf$A, NewB = mydf$B), sum)
   

2. The formula method and the data.frame method treat NA values differently. To get the same result with the formula method as you do with the data.frame method, you need to use na.action = na.pass.

   aggregate(. ~ A + B, mydf, sum, na.action=na.pass)
   

Again, it is not entirely wrong to say "I don't think it really matters", and I'm not going to state my preference here since that's not really what Stack Overflow is about, but it is important to always read the function documentation carefully before making such decisions.

Answer 2

From the help page,

aggregate.formula is a standard formula interface to aggregate.data.frame

So I don't think it really matters. Use whichever approach you're comfortable with, or which fits existing variables and formulas in your workspace.