innocuous

An R package that (quietly) overrides lm() from base R, and implements it incorrectly. Used as an example of why R package validation is required, and namespacing is useful.

Example

Set up some data with a known relationship:

set.seed(123)
x <- rnorm(20)
y <- 3*x + 1 + rnorm(20, sd = 0.01)

Fit using lm() from {stats} in base R:

lm(y ~ x)

Call:
lm(formula = y ~ x)

Coefficients:
(Intercept)            x  
     0.9996       2.9992  

Fit using lm() from {innocuous}:

library(innocuous)
lm(y ~ x)

Call:
stats::lm(formula = formula, method = "qr", model = TRUE, x = FALSE, 
    y = FALSE, qr = TRUE, singular.ok = TRUE, contrasts = NULL)

Coefficients:
(Intercept)            x  
      4.000        2.999 

Look at the intercept coefficient! Note that when loading the package with library(), no masking warnings were delivered to the user. The description of the Call in the output also suggests it was fitted using stats::lm().