Select covariates that predict one variable

Runs a single LASSO regression of one variable on a pool of candidate covariates and returns the covariates that survive. This is the narrow, treatment-free counterpart to lasso_select_covariates: it knows nothing about arms, assignment, or estimands, and answers only "which of these candidates predict this column".

Usage

lasso_select_one(
  response,
  covariates,
  data,
  subset = NULL,
  lambda_rule = c("lambda.1se", "lambda.min"),
  seed = 999
)

Arguments

response: Column to be predicted, as a character scalar or a bare column name.
covariates: A one-sided formula of candidate covariates (e.g. ~ X1 + X2), a character vector of column names, or a bare expression such as X1 + X2.
data: A data frame.
subset: Optional logical vector, the same length as nrow(data), restricting the rows selection runs on. NA is treated as FALSE. Use it for equations estimated on part of the sample, such as an outcome equation among respondents who answered.
lambda_rule: Either "lambda.1se" (default, the one-standard-error rule) or "lambda.min" (the cross-validation minimum).
seed: Integer seed for the cross-validation fold assignment, so selection is reproducible. The global random number state is restored on exit, so calling this function does not disturb the stream of the script that called it.

Value

A character vector of selected covariate names, in the order they were supplied. Empty when nothing was selected or selection could not be run, which are not distinguished: both mean "no covariates to use".

Details

It exists because some procedures need selection equations that lasso_select_covariates cannot express. That function takes outcome ~ treatment and selects arm by arm and on each arm indicator, which is the right thing for regression adjustment of an experiment. A diagnostic for differential attrition, by contrast, needs two equations with two different left-hand sides and no treatment in either: the outcome among respondents who answered, and an indicator for having dropped out among all those assigned. Calling this function twice expresses that directly.

Relation to the other selector

lasso_select_covariates returns a one-sided formula, because its answer is consumed by lm_lin. This function returns a character vector, because its answer is usually combined with another selection before any model is fit. Wrap the result in reformulate if a formula is what you need.

Unusable candidates

Candidates that are absent from data, entirely missing, or constant are dropped before the complete-case filter rather than after, and the check is applied over the subset actually used. Dropping them afterwards is a trap: a single all-NA candidate empties the complete-case set, and selection then silently returns nothing while appearing to have run. A covariate that is constant among the rows being used is equally unusable even when it varies in the full data, which is why the check follows subset.

Non-numeric candidates are expanded through model.matrix and mapped back to their source covariate, so a factor is returned whole rather than as a subset of its indicator columns.

Examples

set.seed(1)
n <- 400
dat <- data.frame(X1 = rnorm(n), X2 = rnorm(n), X3 = rnorm(n))
dat$Y <- 1.5 * dat$X1 + rnorm(n)

# X1 predicts Y; X2 and X3 are noise
lasso_select_one("Y", ~ X1 + X2 + X3, data = dat)
#> [1] "X1"

# Selection restricted to half the sample
lasso_select_one("Y", ~ X1 + X2 + X3, data = dat,
                 subset = seq_len(n) <= n / 2)
#> [1] "X1"