dof bug issue #106

[FellowFish]

edited the code in effsize file(effsize boot function) to account for paired and unpaired cases

[josesho]

Hi @FellowFish ,

Can you include below a copy-pastable example?

e.g.

library(dabestr)

# Performing unpaired (two independent groups) analysis.
unpaired_mean_diff <- dabest(iris, Species, Petal.Width,
                             idx = c("setosa", "versicolor", "virginica"),
                             paired = TRUE)

# Code demonstrating you have fixed the bug goes below.
# 

Thanks!
Joses

[FellowFish]

Sorry for the late reply. Below is the reproducible code.
Using the data given from issue #106:

d <- structure(list(change = c(-3, -3, 3, 3, -3, -3, 25, 25, -8, -8, 2, 2, 24, 
                               24, 2, 2, 10, 10, 24, 24, 0, 0, 37, 37),
                    Timepoint = c("T1", "T2", "T1", "T2", "T1", "T2", "T1", "T2", 
                                  "T1", "T2", "T1", "T2", "T1", "T2", "T1", "T2", 
                                  "T1", "T2", "T1", "T2", "T1", "T2", "T1", "T2"), 
                    Result = c(35, 32, 81, 84, 69, 66, 82, 107, 84, 76, 108, 110,
                               75, 99, 57, 59, 102, 112, 68, 92, 76, 76, 75, 112), 
                    Id = c(1L, 1L, 2L, 2L, 3L, 3L, 4L, 4L, 5L, 5L, 6L, 6L, 7L, 
                           7L, 8L, 8L, 9L, 9L, 10L, 10L, 11L, 11L, 12L, 12L)),
                    row.names = c(NA, -24L), class = c("tbl_df", "tbl", "data.frame"))

d_wide <- structure(list(Id = 1:12, T1 = c(35, 81, 69, 82, 84, 108, 75, 57, 102, 68, 76, 75), 
                         T2 = c(32, 84, 66, 107, 76, 110, 99, 59, 112, 92, 76, 112),
                         change = c(-3, 3, -3, 25, -8, 2, 24, 2, 10, 24, 0, 37)), 
                    row.names = c(NA, -12L), class = c("tbl_df", "tbl", "data.frame"))
dabest.obj <- d %>%
  dabest(Timepoint, Result, 
         idx = c("T1", "T2"), 
         paired = TRUE, id.col = Id)

dabest.obj %>% mean_diff()

Performing Bootstrap we get the confidence interval:

set.seed(12345)
boot.ci(boot(d_wide %>% data.frame, 
             statistic = function(data, indices) 
               return(mean(data[indices, "T2"]) - mean(data[indices, "T1"])), 
             R=10000))

Using Dabestr we get the confidence interval, which is quite different from the confidence interval gotten from bootstrapping:

With the edit we get, with a much closer confidence interval:

The edit takes into account whether the data is paired or not, and computes the bootstrap accordingly.

[josesho]

Update:

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

devtools::install_github("ACCLAB/dabestr",  ref = "v0.3.9999")