Question

Possible Duplicate:
In R, correlation test between two columns, for each of the groups in a third column

My data frame (g) contains 2 columns with continuous data and other columns with categorical data. I want to test for correlations between the 2 continuous variables, in different groups which are defined by a third column.
g (157X3000) look like:

     Geno          GDW         GN        M1     M2      M3
1 SB002XSB012 -17.1597630   52.31961    G/G    C/C     T/T
3 SB002XSB044  -3.6537657   53.81305    G/G    C/G     G/G
4 SB002XSB051  -7.8411596   58.05924    A/G    C/C     G/T
5 SB002XSB067   2.8412103   30.85074    A/G    G/G     G/T
6 SB002XSB073 -16.0789550  -10.09913    A/A    C/G     G/G
7 SB002XSB095   0.1759709   10.28837    A/A    G/G     T/T

I’m looking for the correlations between GDW and GN in each of the groups as defined by each M. I tried :

q<- function (x) {  
    spl<-split(g[,2:3],x)  
    r<-function(x) { 
        if ((nrow(x[[1]][1]))>2)  
          cor.test(x[[1]][1],x[[1]][2],use="pairwise.complete.obs")[3:4] else Na  
    }  
    cor<- sapply(spl,r)  
} 

all.cor<- apply(g[,4:ncol(g)],2,q)

and got:

Error in if ((nrow(x[[1]][1])) > 2) cor.test(x[[1]][1], x[[1]][2], use = "pairwise.complete.obs")[3:4] else Na : 
  argument is of length zero

Answer 1

or if you use the plyr package you could say (copying Antoine’s sample data)

library(plyr)
g <- data.frame(col1=rnorm(100, 1, 1), col2=rnorm(100, 10, 3), col3=c(rep("a", 50), rep("b", 50)))

co <- ddply(g, .(col3), function(adf) cor(adf[,1], adf[,2]))

to give a data.frame looking something like…

    col3         V1
  1    a -0.1697979
  2    b  0.1660783

EDIT: adding looping for multiple columns
*EDIT2: removing ridiculously complicated expression!*

library(plyr)
g <- data.frame(col1=rnorm(100, 1, 1),
            col2=rnorm(100, 10, 3),
            col3=sample(c('a','b','c'), 100, replace=TRUE),
            col4=sample(c('a','b','c'), 100, replace=TRUE),
            col5=sample(c('a','b','c'), 100, replace=TRUE),
            col6=sample(c('a','b','c'), 100, replace=TRUE))
for (i in 3:6) {
  co <- ddply(g, i, function(adf) cor(adf[,1], adf[,2]))
  names(co) <- c('variable',paste('CorCol',i, sep='-'))
  if(exists('odf')) { 
    odf <- merge(odf, co, by='variable', all=TRUE)
  } else {
    odf <- co
  } 
} 

Results are in data.frame odf with a column for each correlation so looks like:

> odf
  variable    CorCol-3    CorCol-4    CorCol-5    CorCol-6
1        a  0.29596471 -0.12278082  0.02184259  0.11972933
2        b -0.11793616  0.08827011  0.11030097 -0.03682823
3        c -0.09552299  0.12951251 -0.03855727 -0.03082486