[R] using weighted.mean with tapply()

[Martyn Plummer]

On Wed, 2005-08-03 at 17:00 -0400, roger bos wrote:
> I am trying to calculate the weighted mean for a of 10 deciles and I
> get an error:
> > decile <- tapply(X=mat$trt1m, INDEX=mat$Rank, FUN=weighted.mean, w=mat$mcap)
> Error in FUN(X[[1]], ...) : 'x' and 'w' must have the same length
> 
> All three of my inputs have the same length, as shown below, and the
> weighted.mean calculation works by itself, just not in tapply()
> 
> > length(mat$Rank)
> [1] 1853
> > length(mat$mcap)
> [1] 1853
> > length(mat$trt1m)
> [1] 1853
> > mean(mat$trt1m)
> [1] -0.04475397
> weighted.mean(mat$trt1m, w=mat$mcap)
> [1] -0.04819243
> > mat$mcap[is.na(mat$mcap)] <- min(mat$mcap, na.rm=TRUE)
> 
> I am probably making a simple error in how I pass the optional
> parameter w.  Any help would be greatly appreciated.

When you use tapply, only the first argument of the function you supply
is split by the index variable.  There is no way for tapply to know if
further arguments should be split or not, so it doesn't split them.  
People are regularly caught out by this.  If you look carefully at the
documentation for split, there is a warning about it.

Here is a very dense solution (due to Peter Dalgaard):

by(mat, mat$Rank, with, weighted.mean(trt1m,  mcap))

Alternatively you can split up the data frame using split, write your
own custom wrapper function for doing the weighted mean, and use lapply
or sapply:

sapply(split(mat,mat$Rank), function(x) {weighted.mean(x$trt1m,x$mcap)})

Martyn



-----------------------------------------------------------------------
This message and its attachments are strictly confidential. ...{{dropped}}