Misaligned mean difference line

[johannesbjork]

When plotting proportional data, I found that the mean difference line is not aligned to the mean difference distribution.

Here is an example.

df <- cbind(data.frame(value=sample(1:9, 100, replace= TRUE)/10),data.frame(grp=sample(c("A","B"),100,replace=T)))

plot(dabest(df, x = grp, y = value, idx = c("A","B")))

[josesho]

Sorry for the delay. I'm seeing this on my end as well; definitely a bug.

You can use www.estimationstats.com to produce the correct Gardner-Altman plot.

Will let you know once I've squashed this bug!

PS. We are working on more effective ways to display proportions. (Here, I assume each data point is a proportion already?)

[josesho]

More troubleshooting.

library(tidyverse)

set.seed(12345)
df <- cbind(data.frame(value=sample(1:9, 100, replace = TRUE)/10), 
            data.frame(grp=sample(c("A","B"), 100, replace = T))
            )
df %>% group_by(grp) %>% count()

# A tibble: 2 x 2
# Groups:   grp [2]
  grp       n
  <fct> <int>
1 A        54
2 B        46

Using dabestr to analyse this....

library(dabest)

df_analysed <- dabest(df, x = grp, y = value, idx = c("A", "B"))
df_analysed

produces

DABEST (Data Analysis with Bootstrap Estimation) v0.2.1
============================================================

Variable: value 

Unpaired mean difference of B (n=54) minus A (n=100)
 -0.0127 [95CI  -0.0986; 0.0774]


5000 bootstrap resamples.
All confidence intervals are bias-corrected and accelerated.

Note how the N for A is now 100 instead of 54. The bug likely lies in how the data.frame is being handled....

[josesho]

This is likely to be related to #30 . For some reason, unequal group sizes jams up the munging....

[josesho]

Hi @johannesbjork ,

This bug occurs because grp is used internally. To quickly and simply mitigate this, simply change the x-column to another name.

library(dabest)

set.seed(12345)
df <- cbind(data.frame(value=sample(1:9, 100, replace = TRUE)/10), 
           # we rename the `grp` column to `ggrp`.
            data.frame(ggrp=sample(c("A","B"), 100, replace = T)) 
)


df_analysed <- dabest(df, x = ggrp, y = value, idx = c("A", "B"))
plot(df_analysed)

[josesho]

Pull request #43 fixes this. Run

devtools::install.github('ACCLAB/dabestr')

to install v0.2.1.9000, which is the beta release. This should be on CRAN in a week or so.