Pairwise partial correlation of a matrix, controlling by one variable

问题

I have a 100-column table for which I would like to run pairwise partial correlations, controlling by the 100th column's variable using the pcor.test function from the ppcor package. Is there any partial correlation function in R that I can use the returns something like rcorr, taking the pairwise correlations of the whole matrix but only controlling by one variable?

回答1:

It sounds like for an n-column matrix you want to output a (n-1) x (n-1) matrix of the pairwise correlations of the first n-1 columns, controlling for the last (using the pcor.test function from the ppcor package).

You could do this with the sapply function, looping through each column and computing its correlation to all other columns with pcor.test:

# Sample dataset with 5 columns
set.seed(144)
dat <- matrix(rnorm(1000), ncol=5)

# Compute the 4x4 correlation matrix, controlling for the fifth column
library(ppcor)
sapply(1:(ncol(dat)-1), function(x) sapply(1:(ncol(dat)-1), function(y) {
  if (x == y) 1
  else pcor.test(dat[,x], dat[,y], dat[,ncol(dat)])$estimate
}))
#              [,1]        [,2]       [,3]        [,4]
# [1,]  1.000000000 -0.01885158 0.06037621 0.004032437
# [2,] -0.018851576  1.00000000 0.09560611 0.097152907
# [3,]  0.060376208  0.09560611 1.00000000 0.105123093
# [4,]  0.004032437  0.09715291 0.10512309 1.000000000

来源：https://stackoverflow.com/questions/30198968/pairwise-partial-correlation-of-a-matrix-controlling-by-one-variable