Showing effect sizes of paired comparisons

[peafischer]

Hello,

I adapted the code from https://cran.r-project.org/web/packages/dabestr/vignettes/using-dabestr.html
to show how the effect size should become much more precise (i.e. the distribution should become much more narrow) when using a paired comparison with my artificial data.

library(dplyr)
library(tidyverse)

set.seed(54321) # setting the seed to a specific number will result in reproducible random numbers

N = 40
# First, we create a set of vectors of length N=40 and fill them with numbers sampled from a normal distribution using rnorm():
c1 <- rnorm(N, mean = 100, sd = 25)  # we will use these numbers as our control group 1
#g1 <- rnorm(N, mean = 120, sd = 25)  # we will use these numbers as our experimental group 1
g1 <- c1 + 2  + rnorm(N, mean = 0.1, sd = 0.01)
gender <- c(rep('Male', N/2), rep('Female', N/2))
id <- 1: N

# Here we create a data frame, which includes all vectors (like a table) : 
wide_data <- 
  tibble(
    Control1 = c1, 
    Group1 = g1, 
    Gender = gender, ID = id)

View(wide_data)


# Next we reformat the data, so that the data points are stored as one long column vector and the 
# group assignment and the gender is stored in a long column vector as well.
# This is the format we need for using the dabest() function (Note: dabest is an abbreviation for 
# Data Analysis with Bootstrap-coupled ESTimation )

my_data   <- 
  wide_data %>%
  gather(key = Group, value = Measurement, -ID, -Gender)

View(my_data)

library(dabestr)

two.groups.unpaired <- 
  dabest(my_data, Group, Measurement, 
         idx = c("Control1", "Group1"), 
         paired = FALSE)
 

# plot(mean_diff(two.groups.unpaired), color.column = Gender)


plot(cohens_d(two.groups.unpaired), color.column = Gender)  # Cohen's d and Hedge's g (see next
# line) are standardized effect sizes. They normalize the mean difference by the 
# pooled standard deviation.


# Create paired plots:
two.groups.paired <- 
  dabest(my_data, Group, Measurement, 
         idx = c("Control1", "Group1"), 
         paired = TRUE, id.col = ID)  # the ID provides the information about which samples are paired

# plot(mean_diff(two.groups.paired), color.column = Gender)

two.groups.paired %>% 
  cohens_d() %>% 
  plot(color.column = Gender)

However, the visualization of the effect size does not change despite I set "paired" to TRUE. Am I missing a trick?

Many thanks,
Petra

[josesho]

Hi @peafischer

The dev version v0.3.9999 fixes this known bug, and can be installed with

devtools::install_github("ACCLAB/dabestr", ref = "v0.3.9999", force = TRUE)
cf #99

[peafischer]

Thanks a lot, Jose.
If I install the version, I suddenly get issues with multiplots.
E.g. when using:
multi.group <-
my_data %>%
dabest(Group, Measurement,
idx = list(c("Control1", "Group1", "Group3"),
c("Control2", "Group2", "Group4")),
paired = FALSE
)
multi.group.mean_diff <- multi.group %>% mean_diff()

The scatter points turn into straight lines.

If I simply update to the latest developer version
devtools::install_github("ACCLAB/dabestr"), then the multi-plot looks fine again, but the CIs for the paired comparison are again wrong.

Is there any way to update the version so that both functionalities work?

[ZHANGROU-99]

Hi @peafischer
Thank you for bringing this up. We will fix this multiplots bug in v0.3.9999.