modernBoot (version 0.1.1)

wild_boot_lm: Wild Bootstrap for Linear Model Coefficients

Description

Performs wild bootstrap resampling for linear regression models to handle heteroscedasticity. Supports Rademacher and Mammen weight schemes.

Usage

wild_boot_lm(fit, R = 2000, type = c("rademacher", "mammen"))

Value

A list with two elements:

coef

numeric vector of original fitted model coefficients, including intercept if present. Names preserved from original model.

boot

numeric matrix of dimensions (p x R) where p is number of coefficients. Each column contains one bootstrap replicate of coefficients. Row names are coefficient names; column names are bootstrap iteration numbers.

Arguments

fit

an object of class 'lm' from lm. Should be fitted linear regression model. Heteroscedasticity is allowed and expected - this method is specifically designed to handle non-constant variance in residuals. Model should have at least 1 predictor.

R

integer number of bootstrap replicates (default 2000). Larger values (5000-10000) recommended for confidence intervals in publications. Must be >= 1 and a whole number.

type

character string specifying weight distribution scheme. Options: "rademacher" (default, faster, robust) generates weights as +1/-1 with equal probability. "mammen" (asymptotically optimal) uses empirically calibrated Mammen distribution with golden ratio. Choice has minimal practical impact on results.

Details

The wild bootstrap works by resampling residuals with random signs/weights while keeping predictors fixed. This is particularly useful for heteroscedastic data.

Examples

set.seed(42)
x <- rnorm(50)
y <- 2 + 1.5 * x + rnorm(50, sd = abs(x))  # heteroscedastic
fit <- lm(y ~ x)
result <- wild_boot_lm(fit, R = 500, type = "rademacher")
head(result$boot, 3)