I have a very large data set and am attempting to find the correlation between a lot of different (and random) combinations of the data. For instance, I may want the correlation between the 3rd column with the 12th-15th columns, or the correlation between the 20th column with the 1st-4th columns, etc...
I am currently using pairs.panels() function from the psych library, but am unable to pinpoint the specific pairing of columns I would like.
Here is df, a dummy data.frame with 26 columns, each containing random values, so the correlations should be reasonably low for any pair of columns.
cols = lapply(1:26, function(dummy) runif(30))
df = do.call(data.frame, cols)
names(df) = LETTERS
If you want the correlation between column "X" and columns "A", "C", and "E", try sapply with the cor function.
sapply(df[c("A","C","E")], cor, df["X"])
Or use column numbers:
sapply(df[c(1,3,5)], cor, df[24])
If you want all the permutative combinations of correlations between two groups of columns, try:
firstGroup <- c(1,3,5,20)
secondGroup <- c(14,20,25)
combos <- expand.grid(firstGroup, secondGroup)
result <- mapply(cor, df[combos$Var1], df[combos$Var2])
resultAsMatrix <- matrix(result, nrow = length(firstGroup), dimnames = list(firstGroup, secondGroup))
To get:
> resultAsMatrix
14 20 25
1 -0.22949844 -0.1527876 -0.11877405
3 0.23174965 0.0311125 0.33570756
5 0.01491815 -0.1263007 -0.16688800
20 0.18007802 1.0000000 0.04638838
EDIT:
@user20650 pointed out that the cor function has the capacity to compare two matrices built in. So:
cor(df[firstGroup], df[secondGroup])
yields the matrix I created manually, above:
N T Y
A -0.22949844 -0.1527876 -0.11877405
C 0.23174965 0.0311125 0.33570756
E 0.01491815 -0.1263007 -0.16688800
T 0.18007802 1.0000000 0.04638838