calculate_power_curves function

Calculate Power Curves

Calculate and optionally plot power curves for different effect sizes and trial counts. This function takes a

calculate_power_curves(
  trials,
  effectsize = 1,
  candidateset = NULL,
  model = NULL,
  alpha = 0.05,
  gen_args = list(),
  eval_function = "eval_design",
  eval_args = list(),
  random_seed = 123,
  iterate_seed = FALSE,
  plot_results = TRUE,
  auto_scale = TRUE,
  x_breaks = NULL,
  y_breaks = seq(0, 1, by = 0.1),
  ggplot_elements = list()
)

Arguments

• trials: A numeric vector indicating the trial(s) used when computing the power curve. If a single value, this will be fixed and only effectsize will be varied.
• effectsize: Default 1. A numeric vector indicating the effect size(s) used when computing the power curve. If a single value, this will be fixed and only trials will be varied. If using a length-2 effect size with eval_design_mc() (such as a binomial probability interval), the effect size pairs can be input as entries in a list.
• candidateset: Default NULL. The candidate set (see gen_design() documentation for more information). Provided to aid code completion: can also be provided in gen_args.
• model: Default NULL. The model (see gen_design() and eval_design() documentation for more information). Provided to aid code completion: can also be provided in gen_args/eval_args.
• alpha: Default 0.05. The allowable Type-I error rate (see eval_design() documentation for more information). Provided to aid code completion: can also be provided in eval_args.
• gen_args: Default list(). A list of argument/value pairs to specify the design generation parameters for gen_design().
• eval_function: Default "eval_design". A string (or function) specifying the skpr power evaluation function. Can also be "eval_design_mc", "eval_design_survival_mc", and "eval_design_custom_mc".
• eval_args: Default list(). A list of argument/value pairs to specify the design power evaluation parameters for eval_function.
• random_seed: Default 123. The random seed used to generate and then evaluate the design. The seed is set right before design generation.
• iterate_seed: Default FALSE. This will iterate the random seed with each new design. Set this to TRUE to add more variability to the design generation process.
• plot_results: Default TRUE. Whether to print out a plot of the power curves in addition to the data frame of results. Requires ggplot2.
• auto_scale: Default TRUE. Whether to automatically scale the y-axis to 0 and 1.
• x_breaks: Default NULL, automaticly generated by ggplot2.
• y_breaks: Default seq(0,1,by=0.1). Y-axis breaks.
• ggplot_elements: Default list(). Extra ggplot2 elements to customize the plot, passed in as elements in a list.

Returns

A data.frame of power values with design generation information.

Examples

if(skpr:::run_documentation()) {
cand_set = expand.grid(brew_temp = c(80, 85, 90),
                      altitude = c(0, 2000, 4000),
                      bean_sun = c("low", "partial", "high"))
#Plot power for a linear model with all interactions
calculate_power_curves(trials=seq(10,60,by=5),
                      candidateset = cand_set,
                      model = ~.*.,
                      alpha = 0.05,
                      effectsize = 1,
                      eval_function = "eval_design") |>
 head(30)

}
if(skpr:::run_documentation()) {
#Add multiple effect sizes
calculate_power_curves(trials=seq(10,60,by=1),
                      candidateset = cand_set,
                      model = ~.*.,
                      alpha = 0.05,
                      effectsize = c(1,2),
                      eval_function = "eval_design") |>
 head(30)
}
if(skpr:::run_documentation()) {
#Generate power curve for a binomial model
calculate_power_curves(trials=seq(50,150,by=10),
                      candidateset = cand_set,
                      model = ~.,
                      effectsize = c(0.6,0.9),
                      eval_function = "eval_design_mc",
                      eval_args = list(nsim = 100, glmfamily = "binomial")) |>
 head(30)
}
if(skpr:::run_documentation()) {
#Generate power curve for a binomial model and multiple effect sizes
calculate_power_curves(trials=seq(50,150,by=10),
                      candidateset = cand_set,
                      model = ~.,
                      effectsize = list(c(0.5,0.9),c(0.6,0.9)),
                      eval_function = "eval_design_mc",
                      eval_args = list(nsim = 100, glmfamily = "binomial")) |>
 head(30)
}