Skip to contents

Wild bootstrap for linear regression with heteroskedasticity

Source: R/bootstrap_methods.R
wild_bootstrap.Rd

Multiplies the residuals by random weights (Rademacher or Mammen) and refits OLS. sandwich::vcovBS implements the standard wild bootstrap variance-covariance and fwildclusterboot::boottest adds cluster-wild p-values; both are cross-referenced here. The inline implementation is retained because rmorie's API returns the resampled coefficient distribution (not just a vcov), which is what downstream MRM analyses consume.

Usage

wild_bootstrap(
  y,
  X,
  statistic_idx = 2L,
  n_boot = 999L,
  ci_level = 0.95,
  weight_distribution = "rademacher",
  seed = 42L
)

Arguments

y

Numeric response vector.

X

Numeric design matrix (include an intercept column).

statistic_idx

Column index of the coefficient of interest (1-based).

n_boot

Number of replicates.

ci_level

Confidence level.

weight_distribution

"rademacher" or "mammen".

seed

Random seed.

Value

A morie_bootstrap_result.

See also

sandwich::vcovBS, fwildclusterboot::boottest, morie_did_wild_cluster_bootstrap().

Examples

set.seed(1)
X <- cbind(1, rnorm(50)); y <- drop(X %*% c(1, 2)) + rnorm(50)
str(wild_bootstrap(y, X, n_boot = 199L), max.level = 1)
#> List of 11
#>  $ estimate         : num 1.95
#>  $ se               : num 0.172
#>  $ ci_lower         : num 1.67
#>  $ ci_upper         : num 2.31
#>  $ bias             : num 0.0126
#>  $ n_boot           : int 199
#>  $ method           : chr "wild"
#>  $ ci_method        : chr "percentile"
#>  $ boot_distribution: num [1:199] 1.94 1.6 1.91 1.91 2.01 ...
#>  $ original_estimate: num 1.95
#>  $ acceleration     : num 0
#>  - attr(*, "class")= chr [1:2] "morie_bootstrap_result" "list"