slipper (version 0.0.0.9001)

slipper_lm: Bootstrap a linear regression model

Description

Takes a data frame, and a model to fit to the data and each bootstrap replicate. Bootstrapping is by default resampling cases, but if you set boot_resid=TRUE then resampling residuals will be performed. If you pass a null model formula that includes a subset of the variables in the full model (i.e. it is a nested model) then the bootstrap statistics will come from the bootstrapped null data and can be used for a hypothesis test.

Usage

slipper_lm(df, formula, null_formula = NULL, B = 100, boot_resid = FALSE)

Arguments

df

A data frame

formula

A bare formula to pass to the lm command

null_formula

(optional) If NULL, standard bootstrapping is performed. If a bare nested null formula is passed the bootstrapped statistics come from the null.

B

the number of bootstrap samples to draw

boot_resid

If TRUE then bootstrapping residuals is performed.

Value

out A data frame with the values, whether they come from the observed data or the bootstrapped data, and the coefficient name.

Examples

# NOT RUN {
# Bootstrap a regression model 
slipper_lm(mtcars,mpg ~ cyl,B=100)

# Bootstrap a regression model with piping
mtcars %>% slipper_lm(mpg ~ cyl,B=100)

# Bootstrap residuals for a regression model
mtcars %>% slipper_lm(mpg ~ cyl,B=100,boot_resid=TRUE)

# Bootsrap confidence intervals
mtcars %>% slipper_lm(mpg ~ cyl,B=100) %>% 
filter(type=="bootstrap",term=="cyl") %>%
 summarize(ci_low = quantile(value,0.025),
           ci_high = quantile(value,0.975))
           
# Bootstrap hypothesis test - here I've added one to the numerator
# and denominator because bootstrap p-values should never be zero.

boot = mtcars %>% slipper_lm(mpg ~ cyl, null_formula = mpg ~ 1,B=1000) %>%
                     filter(term=="cyl") %>%
                     summarize(num = sum(abs(value) >= abs(value[1])),
                               den = n(),
                               pval = num/den)
# }