Report p-values as <0.001 (or similar)

[garthtarr]

It would be great to have an option to report small p-values at some arbitrary threshold, e.g. "p < 0.001" or "p < 0.0001" . Looks like some efforts along this path exist in stat_cor() but it is hardcoded and at machine level precision.

I'm thinking that instead of this:

ggscatter(mtcars, x = "wt", y = "mpg", add = "reg.line") +
  stat_cor(
    aes(label = paste(..rr.label.., ..p.label.., sep = "~`,`~")),
    label.x = 3
  )

You might be able to implement something like this:

ggscatter(mtcars, x = "wt", y = "mpg", add = "reg.line") +
  stat_cor(
    aes(label = paste(..rr.label.., ..p.rounded.., sep = "~`,`~")),
    label.x = 3, accuracy = 0.001
  )

where accuracy defines how far the p-value rounding went, much like the pvalue_format() function from the scales package does. I.e.

p.vals = c(0.123, 0.015, 0.0015, 0.00015, 1e-10)
accuracy = 0.01
scales::pvalue(p.vals, accuracy = accuracy, add_p = TRUE)
# [1] "p=0.12" "p=0.02" "p<0.01" "p<0.01" "p<0.01"
accuracy = 0.001
scales::pvalue(p.vals, accuracy = accuracy, add_p = TRUE)
# [1] "p=0.123" "p=0.015" "p=0.002" "p<0.001" "p<0.001"

I note that you've used spaces around the = and < in the ggpubr package, so that could be added back in:

p.vals = c(0.123, 0.015, 0.0015, 0.00015, 1e-10)
accuracy = 0.01
pfmt = scales::pvalue(p.vals, accuracy = accuracy, add_p = TRUE)
gsub(pattern = "(=|<)", replacement = " \\1 ", x = pfmt)
# [1] "p = 0.123" "p = 0.015" "p = 0.002" "p < 0.001" "p < 0.001"

[kassambara]

good idea and thank you for the well documented examples. I will implement this as soon as possible

[kassambara]

Hi,

With the current version, you can pass any pvalue formatting function you want to stat_cor().

For example:

# Custom formatting fucntion
format_pval <- function(pval){
  pval <- scales::pvalue(pval, accuracy= 0.0001, add_p = TRUE)
  gsub(pattern = "(=|<)", replacement = " \\1 ", x = pval)
}
# Display formatted p-value
ggscatter(mtcars, x = "wt", y = "mpg", add = "reg.line") +
  stat_cor(
    aes(label = paste(..rr.label.., format_pval(..p..), sep = "~`,`~")),
    label.x = 3
  )

[garthtarr]

Thanks!

[cigan77]

Is it possible to avoid the space between the R2 value and comma?

[garthtarr]

You can use * in place of ~ to prevent a space being generated:

library(ggpubr)
# Custom formatting fucntion
format_pval <- function(pval){
  pval <- scales::pvalue(pval, accuracy= 0.0001, add_p = TRUE)
  gsub(pattern = "(=|<)", replacement = " \\1 ", x = pval)
}
# Display formatted p-value
ggscatter(mtcars, x = "wt", y = "mpg", add = "reg.line") +
  stat_cor(
    aes(label = paste(..rr.label.., format_pval(..p..), sep = "*`,`~")),
    label.x = 3
  )

[kassambara]

With the latest dev version, you can now specify the argument p.accuracy and r.accuracy:

suppressPackageStartupMessages(library(ggpubr))
# Default plot
sp <- ggscatter(mtcars, x = "wt", y = "mpg", add = "reg.line")
sp + stat_cor(label.x = 3)
#> `geom_smooth()` using formula 'y ~ x'

# Specify the number of decimal places of precision for 
# the p-value and the correlation coefficient r
sp + stat_cor(label.x = 3, p.accuracy = 0.001, r.accuracy = 0.01)
#> `geom_smooth()` using formula 'y ~ x'

^{Created on 2020-05-10 by the reprex package (v0.3.0.9001)}

[johannesjuliusm]

This is fantastic, but is it true that this functionality is not available for stat_compare_means()? It would be fantastic to be able to do the same p-value formatting throughout statistical tests and procedures in ggpubr.

[dchiu911]

Any updates on adding p.accuracy to stat_compare_means()?