Cross-fitted partially linear DML (ATE/ATT) on OTIS

Wraps a Frisch-Waugh-Lovell partialling-out estimator with n_folds cross-fitting on the OLS nuisance functions \(E[Y|X]\) and \(E[D|X]\), then regresses outcome residuals on treatment residuals for the ATE; heteroskedasticity-robust standard errors. ATT is the ATE divided by the treated share (a simple weighting approximation; for the production-grade DML use DoubleML).

Usage

morie_otis_otdml(
  df,
  outcome = "Y",
  treatment = "D",
  covariates = NULL,
  n_folds = 3L,
  seed = 123L
)

Arguments

df

data.frame.

outcome, treatment

Column names.

covariates

Character vector of covariate column names. If NULL, defaults to the standard OTIS set.

n_folds

Integer fold count (default 3L).

seed

Integer RNG seed.

Value

morie_otis_result.

Details

Categorical covariates are dummy-coded with model.matrix.

References

Chernozhukov, V. et al. (2018). Double/debiased machine learning for treatment and structural parameters. Econometrics Journal, 21(1), C1-C68.