90

I have 92 set of data of same type.

I want to make a correlation matrix for any two combinations possible.

i.e., I want a matrix of 92x92.

such that element (ci,cj) should be correlation between ci and cj.

How do I do that?

AndrewGB
  • 16,126
  • 5
  • 18
  • 49
Swapnil 'Tux' Takle
  • 1,187
  • 2
  • 9
  • 9

5 Answers5

106

An example,

 d &lt- data.frame(x1=rnorm(10),
                 x2=rnorm(10),
                 x3=rnorm(10))
cor(d) # get correlations (returns matrix)
bright-star
  • 6,016
  • 6
  • 42
  • 81
Manuel Ramón
  • 2,490
  • 2
  • 18
  • 23
76

You could use 'corrplot' package. 

d <- data.frame(x1=rnorm(10),
                 x2=rnorm(10),
                 x3=rnorm(10))
M <- cor(d) # get correlations

library('corrplot') #package corrplot
corrplot(M, method = "circle") #plot matrix

enter image description here

More information here: http://cran.r-project.org/web/packages/corrplot/vignettes/corrplot-intro.html

Jot eN
  • 6,120
  • 4
  • 40
  • 59
  • 1
    Is it possible to obtain the graph similar to these ones http://cran.r-project.org/web/packages/corrplot/vignettes/corrplot-intro.html, or a simple matrix, but with the R-squared instead of pearson, kendall, or spearman correlation? – FraNut Jan 22 '15 at 12:13
  • 1
    R2 equals the square of the Pearson correlation coefficient. So all you need is multiply M by M (multiply correlation matrix by itself), before creating the plot. – Jot eN Jan 23 '15 at 11:42
18

The cor function will use the columns of the matrix in the calculation of correlation. So, the number of rows must be the same between your matrix x and matrix y. Ex.:

set.seed(1)
x <- matrix(rnorm(20), nrow=5, ncol=4)
y <- matrix(rnorm(15), nrow=5, ncol=3)
COR <- cor(x,y)
COR
image(x=seq(dim(x)[2]), y=seq(dim(y)[2]), z=COR, xlab="x column", ylab="y column")
text(expand.grid(x=seq(dim(x)[2]), y=seq(dim(y)[2])), labels=round(c(COR),2))

enter image description here

Edit:

Here is an example of custom row and column labels on a correlation matrix calculated with a single matrix:

png("corplot.png", width=5, height=5, units="in", res=200)
op <- par(mar=c(6,6,1,1), ps=10)
COR <- cor(iris[,1:4])
image(x=seq(nrow(COR)), y=seq(ncol(COR)), z=cor(iris[,1:4]), axes=F, xlab="", ylab="")
text(expand.grid(x=seq(dim(COR)[1]), y=seq(dim(COR)[2])), labels=round(c(COR),2))
box()
axis(1, at=seq(nrow(COR)), labels = rownames(COR), las=2)
axis(2, at=seq(ncol(COR)), labels = colnames(COR), las=1)
par(op)
dev.off()

enter image description here

Community
  • 1
  • 1
Marc in the box
  • 11,769
  • 4
  • 47
  • 97
  • 1
    @Manuel Ramón 's example is probably best for your case (a single matrix) - organize your data sets as columns. – Marc in the box May 21 '12 at 08:08
  • 1
    in the image above, how can one 'invert' the colors, the is red one correlation is close to -1 or 1 and white when close to 0? – tagoma Dec 04 '12 at 18:29
  • 1
    `image(x=seq(dim(x)[2]), y=seq(dim(y)[2]), z=COR, col=rev(heat.colors(20)), xlab="x column", ylab="y column")` – Marc in the box Dec 05 '12 at 06:47
  • 1
    @Marcinthebox how would you add variable labels to x and y axis (instead of numbers)? Thanks – Agustín Indaco Apr 10 '17 at 22:57
  • 1
    @AgustínIndaco - I have updated my answer with a further example. The `image` function doesn't automatically take the row and column names, so this must be added. – Marc in the box Apr 11 '17 at 16:36
16

Have a look at qtlcharts. It allows you to create interactive correlation matrices:

library(qtlcharts)
data(iris)
iris$Species <- NULL
iplotCorr(iris, reorder=TRUE)

enter image description here

It's more impressive when you correlate more variables, like in the package's vignette: enter image description here

epo3
  • 2,991
  • 2
  • 33
  • 60
2

There are other ways to achieve this here: (Plot correlation matrix into a graph), but I like your version with the correlations in the boxes. Is there a way to add the variable names to the x and y column instead of just those index numbers? For me, that would make this a perfect solution. Thanks!

edit: I was trying to comment on the post by [Marc in the box], but I clearly don't know what I'm doing. However, I did manage to answer this question for myself.

if d is the matrix (or the original data frame) and the column names are what you want, then the following works:

axis(1, 1:dim(d)[2], colnames(d), las=2)
axis(2, 1:dim(d)[2], colnames(d), las=2)

las=0 would flip the names back to their normal position, mine were long, so I used las=2 to make them perpendicular to the axis.

edit2: to suppress the image() function printing numbers on the grid (otherwise they overlap your variable labels), add xaxt='n', e.g.:

image(x=seq(dim(x)[2]), y=seq(dim(y)[2]), z=COR, col=rev(heat.colors(20)), xlab="x column", ylab="y column", xaxt='n')
Community
  • 1
  • 1
TSeymour
  • 729
  • 3
  • 17