ciCalibrate

ciCalibrate is an R package for computing support intervals for unknown univariate parameters. A support interval can either be computed based on a parameter estimate and standard error or based on a confidence interval for the respective parameter. The main function for doing so is ciCalibrate, see the documentation with ?ciCalibrate for the available options. Theoretical background on support intervals is provided in the accompanying paper Pawel et al. (2023) and also Wagenmakers et al. (2020).

Installation

## development version from GitHub (requires remotes package)
## remotes::install_github(repo = "SamCH93/ciCalibrate")

## from CRAN
install.packages(pkgs = "ciCalibrate")

Usage

library("ciCalibrate")

## data from RECOVERY trial
logHR <- -0.19 # estimate
se <- 0.05 # standard error of estimate
ci95 <- logHR + c(-1, 1) * qnorm(p = 0.975) * se # 95% Wald-CI

## default normal prior for logHR under the alternative H1
pm <- 0 # center around value of no effect
psd <- 2 # unit-information standard deviation for a logHR

## compute a support interval with support level = 10
si10 <- ciCalibrate(estimate = logHR, se = se, siLevel = 10, method = "SI-normal",
                    priorMean = pm, priorSD = psd)

## compute instead with confidence interval as input
si10 <- ciCalibrate(ci = ci95, ciLevel = 0.95, siLevel = 10, method = "SI-normal",
                    priorMean = pm, priorSD = psd)
si10

#> Point Estimate [95% Confidence Interval] 
#> -0.19 [-0.29,-0.092]
#> 
#> Calibration Method
#> Normal prior for parameter under alternative
#> with mean m = 0 and standard deviation sd = 2
#> 
#> k = 10 Support Interval
#> [-0.27,-0.11]

## plot Bayes factor function and support interval
plot(si10)

Output of the command plot(si10): the Bayes factor function and the 10 support interval

References

Pawel, S., Ly, A., and Wagenmakers, E.-J. (2023). Evidential Calibration of Confidence Intervals. The American Statistician. doi:10.1080/00031305.2023.2216239