PlotCorr: Plot a Correlation Matrix

Description

This function produces a graphical display of a correlation matrix. The cells of the matrix can be shaded or colored to show the correlation value.

Usage

PlotCorr(x, cols = colorRampPalette(c(Pal()[2], "white",
                                      Pal()[1]), space = "rgb")(20),
         breaks = seq(-1, 1, length = length(cols) + 1),
         border = "grey", lwd = 1,
         args.colorlegend = NULL, xaxt = par("xaxt"), yaxt = par("yaxt"),
         cex.axis = 0.8, las = 2, mar = c(3, 8, 8, 8), mincor = 0,
         main = "", clust = FALSE, ...)

Value

no values returned.

Arguments

x: x is a correlation matrix to be visualized.
cols: the colors for shading the matrix. Uses the package's option "col1" and "col2" as default.
breaks: a set of breakpoints for the colours: must give one more breakpoint than colour. These are passed to image() function. If breaks is specified then the algorithm used follows cut, so intervals are closed on the right and open on the left except for the lowest interval.
border: color for borders. The default is grey. Set this argument to NA if borders should be omitted.
lwd: line width for borders. Default is 1.
args.colorlegend: list of arguments for the ColorLegend. Use NA if no color legend should be painted.
xaxt: parameter to define, whether to draw an x-axis, defaults to "n".
yaxt: parameter to define, whether to draw an y-axis, defaults to "n".
cex.axis: character extension for the axis labels.
las: the style of axis labels.
mar: sets the margins, defaults to mar = c(3, 8, 8, 8) as we need a bit more room on the right.
mincor: numeric value between 0 and 1, defining the smallest correlation that is to be displayed. If this is >0 then all correlations with a lower value are suppressed.
main: character, the main title.
clust: logical. If set to TRUE, the correlations will be clustered in order to aggregate similar values.
...: the dots are passed to the function image, which produces the plot.

Author

Andri Signorell <andri@signorell.net>

Examples

Run this code

m <- cor(d.pizza[,sapply(d.pizza, IsNumeric, na.rm=TRUE)], use="pairwise.complete.obs")

PlotCorr(m, cols=colorRampPalette(c("red", "black", "green"), space = "rgb")(20))
PlotCorr(m, cols=colorRampPalette(c("red", "black", "green"), space = "rgb")(20),
         args.colorlegend=NA)

m <- PairApply(d.diamonds[, sapply(d.diamonds, is.factor)], CramerV, symmetric=TRUE)
PlotCorr(m, cols = colorRampPalette(c("white", "steelblue"), space = "rgb")(20),
         breaks=seq(0, 1, length=21), border="black",
         args.colorlegend = list(labels=sprintf("%.1f", seq(0, 1, length = 11)), frame=TRUE)
)
title(main="Cramer's V", line=2)
text(x=rep(1:ncol(m),ncol(m)), y=rep(1:ncol(m),each=ncol(m)),
     label=sprintf("%0.2f", m[,ncol(m):1]), cex=0.8, xpd=TRUE)

# Spearman correlation on ordinal factors
csp <- cor(data.frame(lapply(d.diamonds[,c("carat", "clarity", "cut", "polish",
                      "symmetry", "price")], as.numeric)), method="spearman")
PlotCorr(csp)

m <- cor(mtcars)
PlotCorr(m, col=Pal("RedWhiteBlue1", 100), border="grey",
         args.colorlegend=list(labels=Format(seq(-1,1,.25), digits=2), frame="grey"))

# display only correlation with a value > 0.7
PlotCorr(m, mincor = 0.7)
x <- matrix(rep(1:ncol(m),each=ncol(m)), ncol=ncol(m))
y <- matrix(rep(ncol(m):1,ncol(m)), ncol=ncol(m))
txt <- Format(m, digits=3, ldigits=0)
idx <- upper.tri(matrix(x, ncol=ncol(m)), diag=FALSE)

# place the text on the upper triagonal matrix
text(x=x[idx], y=y[idx], label=txt[idx], cex=0.8, xpd=TRUE)

# or let's get rid of all non significant correlations
p <- PairApply(mtcars,  function(x, y) cor.test(x, y)$p.value, symmetric=TRUE)
# or somewhat more complex with outer
p0 <- outer(1:ncol(m),  1:ncol(m),
           function(a, b)
             mapply(
               function(x, y) cor.test(mtcars[, x], mtcars[, y])$p.value,
               a, b))
# ok, got all the p-values, now replace > 0.05 with NAs
m[p > 0.05] <- NA
PlotCorr(m)

# the text
n <- ncol(m)
text(x=rep(seq(n), times=n),
     y=rep(rev(seq(n)), rep.int(n, n)),
     labels=Format(m, digits=2, na.form=""),
     cex=0.8, xpd=TRUE)
# the text could also be set with outer, but this function returns an error,
# based on the fact that text() does not return some kind of result
# outer(X = 1:nrow(m),  Y = ncol(m):1,
#  FUN = "text", labels = Format(m, digits=2, na.form = ""),
#  cex=0.8, xpd=TRUE)


# put similiar correlations together
PlotCorr(m, clust=TRUE)

# same as
idx <- order.dendrogram(as.dendrogram(
          hclust(dist(m), method = "mcquitty")
       ))
PlotCorr(m[idx, idx])


# plot only upper triangular matrix and move legend to bottom
m <- cor(mtcars)
m[lower.tri(m, diag=TRUE)] <- NA

p <- PairApply(mtcars,  function(x, y) cor.test(x, y)$p.value, symmetric=TRUE)
m[p > 0.05] <- NA

PlotCorr(m, mar=c(8,8,8,8), yaxt="n",
         args.colorlegend = list(x="bottom", inset=-.15, horiz=TRUE, 
                                 height=abs(LineToUser(line = 2.5, side = 1)), 
                                 width=ncol(m)))
mtext(text = rev(rownames(m)), side = 4, at=1:ncol(m), las=1, line = -5, cex=0.8)

text(1:ncol(m), ncol(m):1, colnames(m), xpd=NA, cex=0.8, font=2)

n <- ncol(m)
text(x=rep(seq(n), times=n),
     y=rep(rev(seq(n)), rep.int(n, n)),
     labels=Format(t(m), digits=2, na.form=""),
     cex=0.8, xpd=TRUE)