HK_decomposition: Heiler & Knaus decomposition of effect heterogeneity under...

In MCKnaus/causalDML: Causal Double Machine Learning

Heiler & Knaus decomposition of effect heterogeneity under non-homogeneous treatments

Description

This function builds on an APO_dml object to calculate the decomposition parameters nATE(Z) = rATE(Z) + Delta(Z) described in Heiler and Knaus (2021).

Usage

HK_decomposition(
  APO_dml,
  z = NULL,
  intercept = TRUE,
  spline = FALSE,
  subset = NULL,
  ...
)

Arguments

APO_dml	APO_dml object containing the effective treatment specific doubly robust score and nuisance parameters.
z	Vector or matrix containing the low-dimensional heterogeneity variables to be considered. If NULL, the unconditional decomposition parameters are calculated.
intercept	Adds an intercept to the variables in z. Set FALSE e.g. to get subgroup effects if one-hot coded z.
spline	If TRUE, B-spline regression based on crs is conducted.
subset	Optional logical vector if decomposition should be run for subsample of original sample.
...	Pass option for crs.

Value

HK_decomposition object with a list of nATE, rATE and Delta list containing each

score	The parameter specific Neyman orthogonal score that was used as pseudo-outcome.
list	The lm object run to get the least squares coefficients (not with corrected standard errors).
vcov	The adjusted variance-covariance matrix derived in Heiler and Knaus (2021) for rATE and Delta. For nATE standard heteroskedasticity robust variance-covariance matrix.
results	A coeftest object with the coefficients and the appropriate inference results.

References

• Heiler, P., Knaus, M. C. (2021). Effect or Treatment Heterogeneity? Policy Evaluation with Aggregated and Disaggregated Treatments arXiv preprint arXiv:2110.01427. https://arxiv.org/abs/2110.01427

MCKnaus/causalDML documentation built on June 11, 2025, 12:30 a.m.