powergrid

The powergrid package is intended to allow users to easily evaluate a function across a grid of input parameters. The utilities in the package are aimed at performing analyses of power and sample size, allowing for easy search of minimum n (or min/max of any other parameter) to achieve a minimal level of power (or a maximum of any value). Also, plotting functions are included that present the dependency of n and power in relation to further parameters.

Note that the package is currently in a late stage of development. You are encouraged to use the package released here on SCTO’s github. For replicability, make sure you explicitly refer to the current release when loading the package in your code, using argument tag to install_github.

devtools::install_github("SwissClinicalTrialOrganisation/powergrid",
                         build_vignette = TRUE)
library(powergrid)

Powergrid is a package mainly developed by Gilles Dutilh, partly funded by a Statistical programming grant from the SCTO. Valuable contributions have been made by Richard Allen of the statistics team at DKF basel.

Example usage

Define a grid of parameters to evaluate a function across:

sse_pars <- list( # a simple list
  n = seq(from = 10, to = 60, by = 5), # sample size
  sd = seq(from = 0.1, to = 1, by = 0.1) # standard deviation
)

Define a function to evaluate the parameters across the grid. The function should take the parameters as input and return a single value (e.g., power, sample size, etc.). For example, we can use the power.t.test function from the stats package to calculate power for a t-test:

PowFun <- function(n, sd){
  ptt = power.t.test(n = n/2,
                     delta = .6,
                     sd = sd,
                     sig.level = 0.05)
  return(ptt$power)
}

Evaluate the function at each grid node:

power <- PowerGrid(pars = sse_pars, fun = PowFun)

Display the results:

PowerPlot(power)

See the vignette for more details on how to use the package and its functions.

Please don’t hesitate making an issue above or contributing through a pull request. You may also contact Gilles by email.