How to display letters to pairwise comparison plot?

Question

How can I display letters on a ggstatsplot package plot for Kruskal Walllis test? This is a reproducible example based on this question.

set.seed(123)

# Create vector for number of cases per month
cases_per_month <- c(10, 25, 20, 20, 25, 20, 19, 5)

# Create vector for months (April to November)
months <- c("April", "May", "June", "July", "August", "September", "October", "November")

# Create empty vectors for final dataset
dataset <- data.frame(mean_severity = numeric(), month = character())

# Generate dataset
dat <- list()

for (i in 1:length(months)) {
  month <- rep(months[i], cases_per_month[i])
  severity <- sample.int(n = 10, size = cases_per_month[i], replace = TRUE)
  
  # generate some differences in the sample
  if (i %in% c(1, 4, 7)){
    severity <- severity^2
  }
  
  temp_data <- data.frame(mean_severity = severity, month = month)
  dat[[i]] <- rbind(dataset, temp_data)
}

# Using rbind to combine rows
dat <- do. Call(rbind, dat)

Currently, I have bars showing p-values. I want letters instead of bars showing p values. The question has been answered here. Apparently, AddLetters function should show letters instead of p values as shown below in his example, but it runs indefinitely without displaying any letters in my case. Is there any other way of displaying letters?

Example plot where letters are shown instead of bars here

I'm not sure that using the `AddLetters` function will work, partly because the package is hard to use. The keyword to search on is `"CLD"` or "compact letter display". There are some solutions using `geom_text` here: https://stackoverflow.com/questions/69309101/automatically-adding-letters-of-significance-to-a-ggplot-barplot-using-output-fr and https://community.rstudio.com/t/add-significance-letters-to-compare-treatments-for-each-variety-in-boxplot-with-tukey-test/139042/3 — David, Jun 30 '23 at 03:45
@David Thanks. I was able to replicate the above example for anova, but not Kruskal-Wallis test. Got an error `Error in x$terms %||% attr(x, "terms") %||% stop("no terms component nor attribute") : no terms component nor attribute` — Ahsk, Jun 30 '23 at 13:26

score 2 · Accepted Answer · answered Jun 30 '23 at 19:14

We have to change the input to match multcompLetters.

Here is how to do it.

library(Matrix)
library(PMCMRplus)
library(ggstatsplot)

#this does the plot in your post
p <- ggbetweenstats(data = dat, y = mean_severity, x = month, type = "nonparametric")

#this does the Dunn pairwise tests
#kwAllPairsDunnTest(mean_severity ~ month, data=dat, p.adjust.method = "holm")

#now we have to format the p-value matrix into a symmetric matrix for multcompLetters
pval.matrix <- kwAllPairsDunnTest(x = dat$mean_severity, 
               g = as.factor(dat$month), p.adjust.method = "holm")

> pval.matrix$p.value
                April      August        July         June         May   November      October
August    0.004092534          NA          NA           NA          NA         NA           NA
July      1.000000000 0.001097334          NA           NA          NA         NA           NA
June      0.004907027 1.000000000 0.001911186           NA          NA         NA           NA
May       0.029411147 1.000000000 0.015500796 1.0000000000          NA         NA           NA
November  0.136459815 1.000000000 0.167977744 1.0000000000 1.000000000         NA           NA
October   1.000000000 0.000118815 1.000000000 0.0002455998 0.002648157 0.07101786           NA
September 0.009390306 1.000000000 0.004164969 1.0000000000 1.000000000 1.00000000 0.0006461642

We use forceSymmetric from the Matrix package after manipulating the pairwise p-value matrix:

#square and diagonalize the p-value matrix
new.pval.matrix <- rbind(1,pval.matrix$p.value)
new.pval.matrix <- cbind(new.pval.matrix, 1)
diag(new.pval.matrix) <- 1

new.pval.matrix <- as.matrix(forceSymmetric(new.pval.matrix, "L"))

#Add September to the row and column names
rownames(new.pval.matrix)[dim(pval.matrix$p.value)+1] <- 
                    rownames(pval.matrix$p.value)[dim(pval.matrix$p.value)[1]]
colnames(new.pval.matrix)[dim(pval.matrix$p.value)+1] <- 
                    rownames(pval.matrix$p.value)[dim(pval.matrix$p.value)[1]]

> new.pval.matrix
                April      August        July         June         May   November      October    September
April     1.000000000 0.004092534 1.000000000 0.0049070268 0.029411147 0.13645981 1.0000000000 0.0093903064
August    0.004092534 1.000000000 0.001097334 1.0000000000 1.000000000 1.00000000 0.0001188150 1.0000000000
July      1.000000000 0.001097334 1.000000000 0.0019111864 0.015500796 0.16797774 1.0000000000 0.0041649687
June      0.004907027 1.000000000 0.001911186 1.0000000000 1.000000000 1.00000000 0.0002455998 1.0000000000
May       0.029411147 1.000000000 0.015500796 1.0000000000 1.000000000 1.00000000 0.0026481566 1.0000000000
November  0.136459815 1.000000000 0.167977744 1.0000000000 1.000000000 1.00000000 0.0710178590 1.0000000000
October   1.000000000 0.000118815 1.000000000 0.0002455998 0.002648157 0.07101786 1.0000000000 0.0006461642
September 0.009390306 1.000000000 0.004164969 1.0000000000 1.000000000 1.00000000 0.0006461642 1.0000000000

Now multcompLetters works:

> multcompLetters(new.pval.matrix)
    April    August      July      June       May  November   October September 
      "a"       "b"       "a"       "b"       "b"      "ab"       "a"       "b"

We can follow your link on how to prepare the CLD:

data.summary <- group_by(dat, month) %>% 
  summarise(mean=mean(mean_severity), sd=sd(mean_severity)) %>% 
  arrange(desc(mean))
#match ordering of the factors [month]
data.summary <- data.summary[order(data.summary$month),]

CLD <- multcompLetters(new.pval.matrix)
data.summary$CLD <- CLD$Letters

#you'll likely need to change these graphics options for your purposes
p + geom_text(data = data.summary, aes(label=CLD,x=month, y=mean), 
    position=position_dodge2(0.75), hjust = 3)

How to display letters to pairwise comparison plot?

1 Answers1