In rpahl/GroupSeq: Group Sequential Design Probabilities - With Graphical User Interface

library(knitr)
knitr::opts_chunk$set(
  comment = "#",
  prompt = F,
  tidy = FALSE,
  cache = FALSE,
  collapse = T
)

old <- options(width = 100L, digits = 10)

Let's assume you have planned a two-sided 3-stage Pocock design with 80% power.

include_graphics("figures/task4-two-sided-3-stage-Pocock-setup.png")

The resulting design requires a drift of about 3.

include_graphics("figures/task4-two-sided-3-stage-Pocock.png")
include_graphics("figures/task4-two-sided-3-stage-Pocock-graph.png")

During the study you perform the interim analysis as planned when about 67% of all samples were collected and obtain a standardized effect size of 2.5, exceeding the critical bound (2.2894) so that you can abort the study and reject H0.

To compute the confidence interval at this point, open option -4- of GroupSeq and enter the values as shown below.

include_graphics("figures/task4-CI-setup.png")

Hitting CALCULATE yields the following.

include_graphics("figures/task4-CI-result.png")

So for this study, the resulting confidence interval of the standardized effect is (0.30, 5.38).

Depending on the underlying data distribution and applied test statistic this will be "back-calculated" to obtain the confidence interval for the value of interest (e.g. mean difference for some normally distributed data or difference in proportions of some binomially distributed data).

options(old)

rpahl/GroupSeq documentation built on Nov. 12, 2023, 12:25 p.m.