Reporting significance level in corrplot()

2
votes

I'm currently using corrplot() from the corrplot package in R and I've stumbled across two problems. For simplicity, I'll use the same notation as the help/introduction page for corrplot.

  1. I'd like to inscribe either my p-value or how significant the test was (or both!) in all cells, not just the insignificant ones.

  2. I'd like these inscriptions only in the upper triangular.

To address 2) first, I've been able to use this, but if feels kind of hacky:

corrplot(M, type="upper", p.mat = res1[[1]], insig="p-value", tl.pos="n")
corrplot(M, type="lower", add=T, tl.pos="d", cl.pos="n"))

However I haven't been able to figure out number 1. Any suggestions would be helpful!

1
rr-corrplot
You can add sig.level=0 to the first plot - user20650
While technically that works, corrplot() rounds to two decimal places, so it doesn't show precision. It would be more useful to have some kind of indication for how small the p-value is. - pocketlizard
So you would like to add, for example"<0.0001" or similar? I think this will be easier to do by adding a format to the pvalue matrix and then plotting using text. - user20650
Sure, something similar to that would work. Perhaps asterisks denoting significance or something along those lines. It seems like such a pain to input them manually with the text function... - pocketlizard

1 Answers

4
votes

The quick way is to add sig.level=0 to the first plot, so all p-values are shown (actually, due to numerical precision some p-values will be exactly zero, so you may need to set it to sig.level=-0.1, for example)

require(corrplot)

# Data
M <- mtcars[3:7]
pval <- psych::corr.test(M, adjust="none")$p

# Corrplot
corrplot(cor(M), type="upper", p.mat=pval, insig="p-value", 
                                               tl.pos="n", sig.level=0)
corrplot(cor(M), type="lower", add=T, tl.pos="d", cl.pos="n")

This gives

enter image description here

However, if you want to add more detail to the p values it is probably easier to post-format the plot and add them using a text call

# Plot
corrplot(cor(M), type="upper", tl.pos="n")

# Get positions & plot formatted p-values
pos <- expand.grid(1:ncol(pval), ncol(pval):1)
text(pos, p_format(pval))

# lower tri
corrplot(cor(M), type="lower", add=T, tl.pos="d", cl.pos="n")

To give

enter image description here

Format function

p_format <- function(x, ndp=3)
{
  out <- format(round(as.numeric(x),ndp),ns=ndp,scientific=F,just="none")
  ifelse(out=="0.000","<0.0001", out)
}

My view (fwiw) is that this is adding too much info to the plot