Bootstrap sampling distributions ("what if we ran the experiment many times...")

Estimate the sampling distribution of a sample statistic or regression coefficient from a linear model (lm)

what_if(
  data,
  variable = NULL,
  model = NULL,
  stat = NULL,
  group = NULL,
  focus = NULL,
  experiments = 100L,
  sample_size = 100L,
  plot = TRUE,
  p.values = FALSE,
  seed = sample.int(.Machine$integer.max, 1),
  ...
)

Arguments

data	data of class data.frame
variable	a continuous variable from the data frame. Only relevant if using `stat` (to boostrap a sample statistic)
model	a model or formula, e.g., `y ~ x1 + x2`. Use this to boostrap a linear model
stat	an R summary statistic function (e.g., `mean()`, `median()`, `max()`, `sd()`) entered as a string (e.g., "mean", "median", "max", "sd")
group	categorical grouping variable from data (defaults to NULL). If activated each simulation evenly samples data according to group
focus	an independent variable from model (defaults to FALSE). Only required if model is a multiple linear regression
experiments	number of experiments, i.e. runs of the boostrap sampler (defaults to 100)
sample_size	sample size of each experiment (defaults to 100)
plot	plot the sampling distribution of the regression coefficient (defaults to TRUE)
p.values	plot the cumulative distribution of p-values and report the number of significant results (defaults to TRUE)
seed	random number generator seed
...	additional arguments to adjust the plots

Value

ggplot object if plot = TRUE, otherwise a data.frame

Author

Lawrence R. De Geest

Examples

# using the nhanes data
data("nhanes", package = "SuffolkEcon")

# sampling distribution of the mean of height
what_if(data = nhanes, variable = height, stat = 'mean')

# sampling distribution and p-values for lm(formula = weight ~ height, data = nhanes)
what_if(data = nhanes, model = weight ~ height, sample_size = 30, experiments = 100, p.values = TRUE)

# increase the number of experiments
what_if(data = nhanes, model = weight ~ height, sample_size = 30, experiments = 1000, p.values = TRUE)