Define a covariate for a causal DGP

Constructs a fully-specified covariate descriptor for use in causalsim_dgp(). Every covariate has a distribution family, optional distribution parameters, and one or more causal roles that control how it enters the automatically-generated treatment and outcome models.

Usage

covar(dist = "normal", role = "confounder", ...)

Arguments

dist

Character. Distribution family. One of "normal", "binary", "uniform".

role

Character vector. Causal role(s). One or more of:

"confounder"

Enters both the propensity model and the outcome baseline — creates confounding bias in naive estimators.

"instrument"

Enters the propensity model only (exclusion restriction holds).

"effect_modifier"

Available for use in the effect function; excluded from the propensity model unless also a "confounder".

"noise"

Independent of both treatment and outcome; adds variance without confounding.

Roles are not mutually exclusive: role = c("confounder", "effect_modifier") specifies a variable that both confounds and moderates the effect.

...

Distribution parameters.

• "normal": mean (default 0), sd (default 1)

• "binary": prob (default 0.5)

• "uniform": min (default 0), max (default 1)

Value

An S3 object of class causalsim_covar.

Examples

# Standard normal confounder
covar("normal", role = "confounder", mean = 0, sd = 1)
#> <causalsim_covar>  dist: normal(mean=0, sd=1)  role: [confounder]

# Binary effect modifier
covar("binary", role = "effect_modifier", prob = 0.4)
#> <causalsim_covar>  dist: binary(prob=0.4)      role: [effect_modifier]

# Variable that both confounds and moderates the effect
covar("normal", role = c("confounder", "effect_modifier"))
#> <causalsim_covar>  dist: normal                role: [confounder, effect_modifier]