aggregate.formula implicitly removes rows containing NA

The documentation for `aggregate` makes it sound like  
aggregate.formula should behave identically to aggregate.data.frame  
(apart from the way the parameters are passed).  But it looks like  
aggregate.formula is quietly removing rows where any of the "output"  
variables (those on the LHS of the formula) are NA.  This differs  
from how aggregate.data.frame works.  Is this expected behavior?

Here are a couple of examples:

d <- data.frame(a=rep(1:2, each=2),

+                 b=c(1,2,NA,3))

aggregate(d["b"], d["a"], mean)

 a   b
1 1 1.5
2 2  NA

aggregate(b ~ a, d, mean)

 a   b
1 1 1.5
2 2 3.0

It's removing whole rows even if just one of the columns is NA, i.e.:

d <- data.frame(a=rep(1:2, each=2),

+                 b=c(1,2,NA,3),
+                 c=c(NA,2,3,NA))

aggregate(cbind(b,c) ~ a, d, mean)

 a b c
1 1 2 2

The documentation for `aggregate` makes it sound like aggregate.formula should behave identically to aggregate.data.frame (apart from the way the parameters are passed).  But it looks like aggregate.formula is quietly removing rows where any of the "output" variables (those on the LHS of the formula) are NA.  This differs from how aggregate.data.frame works.  Is this expected behavior?

Here are a couple of examples:

d<- data.frame(a=rep(1:2, each=2),

+                 b=c(1,2,NA,3))

aggregate(d["b"], d["a"], mean)

   a   b
1 1 1.5
2 2  NA

aggregate(b ~ a, d, mean)

   a   b
1 1 1.5
2 2 3.0

It's removing whole rows even if just one of the columns is NA, i.e.:

d<- data.frame(a=rep(1:2, each=2),

+                 b=c(1,2,NA,3),
+                 c=c(NA,2,3,NA))

aggregate(cbind(b,c) ~ a, d, mean)

   a b c
1 1 2 2

Daniel

On Jan 11, 2011, at 5:41 PM, Dickison, Daniel wrote:

The documentation for `aggregate` makes it sound like  
aggregate.formula should behave identically to aggregate.data.frame  
(apart from the way the parameters are passed).  But it looks like  
aggregate.formula is quietly removing rows where any of the "output"  
variables (those on the LHS of the formula) are NA.  This differs  
from how aggregate.data.frame works.  Is this expected behavior?

Here are a couple of examples:

d <- data.frame(a=rep(1:2, each=2),

+                 b=c(1,2,NA,3))

aggregate(d["b"], d["a"], mean)

a   b
1 1 1.5
2 2  NA

aggregate(b ~ a, d, mean)

a   b
1 1 1.5
2 2 3.0

It's removing whole rows even if just one of the columns is NA, i.e.:

d <- data.frame(a=rep(1:2, each=2),

+                 b=c(1,2,NA,3),
+                 c=c(NA,2,3,NA))

aggregate(cbind(b,c) ~ a, d, mean)

a b c
1 1 2 2

The help page for aggregate gives the calling defaults for  
aggregate.formula as:
## S3 method for class 'formula' aggregate(formula, data, FUN, ...,  
subset, na.action = na.omit)
So the description you give seems to be adhering to what I would have  
expected (had I initially read the help page.)
-- 
David Winsemius, MD
West Hartford, CT