make_full_rank: Make a design matrix full rank

In WeightIt: Weighting for Covariate Balance in Observational Studies

Make a design matrix full rank

Description

When writing user-defined methods for use with weightit(), it may be necessary to take the potentially non-full rank covs data frame and make it full rank for use in a downstream function. This function performs that operation.

Usage

make_full_rank(mat, with.intercept = TRUE)

Arguments

mat	a numeric matrix or data frame to be transformed. Typically this contains covariates. NAs are not allowed.
with.intercept	whether an intercept (i.e., a vector of 1s) should be added to mat before making it full rank. If TRUE, the intercept will be used in determining whether a column is linearly dependent on others. Regardless, no intercept will be included in the output.

Details

make_full_rank() calls qr() to find the rank and linearly independent columns of mat, which are retained while others are dropped. If with.intercept is set to TRUE, an intercept column is added to the matrix before calling qr(). Note that dependent columns that appear later in mat will be dropped first.

See example at method_user.

Value

An object of the same type as mat containing only linearly independent columns.

Note

Older versions would drop all columns that only had one value. With with.intercept = FALSE, if only one column has only one value, it will not be removed, and it will function as though there was an intercept present; if more than only column has only one value, only the first one will remain.

See Also

method_user, model.matrix()

Examples

set.seed(1000)
c1 <- rbinom(10, 1, .4)
c2 <- 1-c1
c3 <- rnorm(10)
c4 <- 10*c3
mat <- data.frame(c1, c2, c3, c4)

make_full_rank(mat) #leaves c2 and c4

make_full_rank(mat, with.intercept = FALSE) #leaves c1, c2, and c4

WeightIt documentation built on April 4, 2025, 2:05 a.m.