cate: CATE
In matteobonvini/drl.cate: What the Package Does (One Line, Title Case)

cate	R Documentation

CATE

Description

This function estimates heterogeneous treatment effects (HTEs) defined as E(Y^1 - Y^0 | V = v0).

Usage

cate(
  data,
  learner,
  x_names,
  y_name,
  a_name,
  v_names,
  v0,
  mu1.x,
  mu0.x,
  pi.x,
  drl.v,
  drl.x,
  nsplits = 5,
  foldid = NULL,
  univariate_reg = FALSE,
  partial_dependence = FALSE,
  partially_linear = FALSE,
  additive_approx = FALSE,
  variable_importance = FALSE,
  vimp_num_splits = 1,
  bw.stage2 = NULL,
  sample.split.cond.dens = FALSE,
  cond.dens = NULL,
  cate.w = NULL,
  cate.not.j = NULL,
  reg.basis.not.j = NULL,
  pl.dfs = NULL
)

Arguments

`data`	A data frame containing the dataset.
`learner`	A character string specifying which learner to use (e.g., "dr").
`x_names`	A character vector specifying the names of the confouding variables.
`y_name`	A character string specifying the outcome variable.
`a_name`	A character string specifying the treatment variable.
`v_names`	A character vector specifying the names of the effect modifiers.
`v0`	A matrix of evaluation points, i.e., values of V for which the CATE is estimated (E(Y^1 - Y^0 \| V = v0)).
`mu1.x`	A function taking arguments (y, a, x, new.x). It trains a model estimating E(Y \| A = 1, X) and returns a list of 3 elements: res, model and fit. res is a vector of predictions of the model evaluated at new.x, model is the model object used to estimate E(Y \| A = 1, X) and fit is a function with argument new.x that returns the predictions of the model. See examples.
`mu0.x`	A function taking arguments (y, a, x, new.x). It trains a model estimating E(Y \| A = 0, X) and returns a list of 3 elements: res, model and fit. res is a vector of predictions of the model evaluated at new.x, model is the model object used to estimate E(Y \| A = 0, X) and fit is a function with argument new.x that returns the predictions of the model. See examples.
`pi.x`	A function taking arguments (a, x, new.x). It trains a model estimating P(A = 1 \| X) and returns a list of 3 elements: res, model and fit. res is a vector of predictions of the model evaluated at new.x, model is the model object used to estimate P(A = 1 \| X) and fit is a function with argument new.x that returns the predictions of the model. See examples.
`drl.v`	A function taking arguments (pseudo, v, new.v). It trains a model estimating E(Y^1 - Y^0 \| V) by regressing a pseudo-outcome pseudo onto v and returns a list of 3 elements: res, model and fit. res is a vector of predictions of the model evaluated at new.v, model is the model object used to estimate E(Y^1 - Y^0 \| V) (after possibly model selection) and fit is a function with argument new.v that returns the predictions of the model. See examples. #' @param drl.x A function taking arguments (pseudo, x, new.x). It trains a model estimating E(Y^1 - Y^0 \| X) by regressing a pseudo-outcome pseudo onto x and returns a list of 3 elements: res, model and fit. res is a vector of predictions of the model evaluated at new.x, model is the model object used to estimate E(Y^1 - Y^0 \| X) (after possibly model selection) and fit is a function with argument new.v that returns the predictions of the model. See examples.
`nsplits`	An integer indicating the number of splits used for cross-validation. Ignored if foldid is specified.
`foldid`	An optional vector specifying fold assignments for cross-validation.
`univariate_reg`	A logical indicating whether to perform univariate regression for estimating the CATE as a function of each effect modifier separately (default: FALSE).
`partial_dependence`	A logical indicating whether to compute partial dependence plots (default: FALSE).
`partially_linear`	A logical indicating whether to compute partially linear approximations via Robinson's transformation (default: FALSE).
`additive_approx`	A logical indicating whether to compute the CATE assuming an additive structure (default: FALSE).
`variable_importance`	A logical indicating whether to compute variable importance measures (default: FALSE).
`vimp_num_splits`	An integer specifying the number of splits for variable importance computation (default: 1).
`bw.stage2`	A list of length equal to the number of effect modifiers considered, where each element if a vector of candidate bandwidths for second-stage regression of the pseudo-outcome onto the effect modifier that calculates either the univariate CATE or the Partial Dependence measure (default: NULL). It needs to be provided if univariate_reg or partial_dependence is set to TRUE.
`sample.split.cond.dens`	A logical indicating whether to do sample-splitting for conditional density estimation (default: FALSE).
`cond.dens`	A function
`cate.w`	A function
`cate.not.j`	A function
`reg.basis.not.j`	A function
`pl.dfs`	A list of length equal to the number of effect modifiers considered, where each element is a vector of candidate number of basis elements for the partially linear approximation computed via Robinson trick.