I would like to create a custom code that creates a compact letter display from a pairwise test I have performed.
I have done this with pairwise t-tests with success (packages for this exist), and I am also familiar with the package library(multcomp)
when I run linear models and the function cld()
to get the compact letter displays, but they will not work for my specific case here.
I work with kaplan meier survival data often, and after I run the pairwise_survdiff()
function to see if any statistical differences exist between groups (found in the packages library(survival)
and library(survminer)
, I am easily able to extract a table to display all pairwise comparisons and their corresponding p-values. I have included an example for you here today. (see df
below)
When their are many comparisons to do by hand, this becomes a mess to found out which groups are different / similar, and it's prone to human error when many levels exist, and up to now, I've always done it by hand. I would like to change this.
Could someone help me with a code that helps do this automatically?
Here is a mock dataframe df
with 10 treatments (named treatment-1....treatment-10), and the rows are filled with p-values. Let's assume anything below p<0.05 as significant. However, it would be very cool to have a code that would allow a more conservative approach, and say set the desired cut off for statistical significance (say anything below p<0.01 as significant for example).
Thanks for your help, and again, here is a play datatframe
df <- read.table("https://pastebin.com/raw/ZAKDBjVs", header = T)
While reflecting on this, I believe I found an answer on my own, with the library(mulcompView)
and library(rcompanion)
Nonetheless, I think it's important, since I have seen / heard this question multiple times. Here is how I solved my problem
library(rcompanion)
library(multcompView)
df <- read.table("https://pastebin.com/raw/ZAKDBjVs", header=T)
PT1 = fullPTable(df)
multcompLetters(PT1,
compare="<",
threshold=0.05,
Letters=letters,
reversed = FALSE)
This gives me the desired output with the compact letter displays between groups. Additionally, one could edit the statistical threshold to be either more/less conservative by changing the threshold=
Very happy with the result. This has bothered me for a while. I hope it is useful to other members