hardhat (version 0.2.0)

model_matrix: Construct a design matrix

Description

model_matrix() is a stricter version of stats::model.matrix(). Notably, model_matrix() will never drop rows, and the result will be a tibble.

Usage

model_matrix(terms, data)

Arguments

terms

A terms object to construct a model matrix with. This is typically the terms object returned from the corresponding call to model_frame().

data

A tibble to construct the design matrix with. This is typically the tibble returned from the corresponding call to model_frame().

Value

A tibble containing the design matrix.

Details

The following explains the rationale for some of the difference in arguments compared to stats::model.matrix():

• contrasts.arg: Set the contrasts argument, options("contrasts") globally, or assign a contrast to the factor of interest directly using stats::contrasts(). See the examples section.

• xlev: Not allowed because model.frame() is never called, so it is unnecessary.

• ...: Not allowed because the default method of model.matrix() does not use it, and the lm method uses it to pass potential offsets and weights through, which are handled differently in hardhat.

Examples

# NOT RUN {
# ---------------------------------------------------------------------------
# Example usage

framed <- model_frame(Sepal.Width ~ Species, iris)

model_matrix(framed$terms, framed$data)

# ---------------------------------------------------------------------------
# Missing values never result in dropped rows

iris2 <- iris
iris2$Species[1] <- NA

framed2 <- model_frame(Sepal.Width ~ Species, iris2)

model_matrix(framed2$terms, framed2$data)

# ---------------------------------------------------------------------------
# Contrasts

# Default contrasts
y <- factor(c("a", "b"))
x <- data.frame(y = y)
framed <- model_frame(~y, x)

# Setting contrasts directly
y_with_contrast <- y
contrasts(y_with_contrast) <- contr.sum(2)
x2 <- data.frame(y = y_with_contrast)
framed2 <- model_frame(~y, x2)

# Compare!
model_matrix(framed$terms, framed$data)
model_matrix(framed2$terms, framed2$data)

# Also, can set the contrasts globally
global_override <- c(unordered = "contr.sum", ordered = "contr.poly")

rlang::with_options(
  .expr = {
    model_matrix(framed$terms, framed$data)
  },
  contrasts = global_override
)
# }