0

I have the following dataframe:

set.seed(1)
y <- data.frame(a1 = rnorm(5) , b1 = rnorm(5), c1 = rnorm(5),  a2 = rnorm(5), b2 = rnorm(5), c2 = rnorm(5))

I would like to obtain the correlations of the pairs of columns: cor(a1,a2), cor(b1,b2), cor(c1,c2)

I tried the following but NA's appear as output:

apply(y,2,function(x) cor(x[1],x[3]))

I would like to get the result equivalent to

cor(y[,1],y[,4])
cor(y[,2],y[,5])
cor(y[,3],y[,6])

In my actual data frame, I have many more pairs of columns.

Any ideas?

Thanks for your support.

rwn1v
  • 777
  • 3
  • 10
  • 23

2 Answers2

4
num.vars <- length(y)
var1 <- head(names(y), num.vars / 2)
var2 <- tail(names(y), num.vars / 2)
mapply(cor, y[var1], y[var2])
#         a1         b1         c1 
#  0.2491625 -0.5313192  0.5594564
flodel
  • 87,577
  • 21
  • 185
  • 223
  • classy! what if you if num.vars is odd? – agstudy Feb 20 '14 at 05:06
  • It shouldn't be the case given how the OP posed the problem. But I suppose how `var1` and `var2` are created could use different approaches (e.g. a regex). I would leave the `mapply` alone though. – flodel Feb 20 '14 at 05:08
0

Another approach using variable regular expression on names. This works also if variable names are in arbitrary order.

nn <- 
unique(sub('([0-9]+)','',names(y )))

sapply(nn,function(x){
    xy = y[,grep(x,names(y))]
    cor(xy[,1],xy[,2])})
         a          b          c 
-0.7615458  0.5683647  0.5594564
agstudy
  • 119,832
  • 17
  • 199
  • 261