[R] using tapply with multiple variables

Jim Lemon jim at bitwrit.com.au
Sun May 1 14:29:51 CEST 2011


On 05/01/2011 05:28 AM, Kevin Burnham wrote:
> HI All,
>
> I have a long data file generated from a minimal pair test that I gave to
> learners of Arabic before and after a phonetic training regime.  For each of
> thirty some subjects there are 800 rows of data, from each of 400 items at
> pre and posttest.  For each item the subject got correct, there is a 'C' in
> the column 'Correct'.  The line:
>
> tapply(ALLDATA$Correct, ALLDATA$Subject, function(x)sum(x=="C"))
>
> gives me the sum of correct answers for each subject.
>
> However, I would like to have that sum separated by Time (pre or post).  Is
> there a simple way to do that?
>
>
> What if I further wish to separate by Group (T or C)?
>
Hi Kevin,
When I looked at this, I immediately thought of the brkdnNest function 
(which uses tapply internally). In order to get the counts with the 
current function, I had to create a new variable (newcorrect). However, 
the idea so attracted me that I programmed it into the code (thanks).
Here is a way to get your summary by Subject and Time:

ALLDATA<-data.frame(Subject=rep(1:30,each=800),
  Occasion=factor(rep(c("pre","post"),2400),levels=c("pre","post")),
  Correct=sample(c("C","I"),2400,TRUE))
tapply(ALLDATA$Correct,list(ALLDATA$Subject,ALLDATA$Occasion),
  function(x) sum(x=="C"))
library(plotrix)
brkdnNest(Correct~Subject+Occasion,ALLDATA,FUN="propbrk",trueval="C")
ALLDATA$newcorrect<-ALLDATA$Correct=="C"
brkdnNest(newcorrect~Subject+Occasion,ALLDATA,FUN="sum")

To get the three level breakdown, add another factor:

ALLDATA$Group<-rep(c("T","C"),each=1200)
brkdnNest(newcorrect~Group+Subject+Occasion,ALLDATA,FUN="sum")

Notice that this gives you all of the subjects for each Group, even if 
they weren't in that Group. I'll work on that one, for I have just 
switched to using "tapply" for this breakdown, as it doesn't discard NA 
values (the cause of the minor bug in barNest)

Jim



More information about the R-help mailing list