Skip to contents

Function to compute estimates on external dataset

Source: R/apply_OTR.R
compute_estimates_external.Rd

Function to compute estimates on external dataset

Usage

compute_estimates_external(
  df,
  Y_name,
  A_name,
  W_list,
  Z_list,
  CATE_preds = NULL,
  d_pred = NULL,
  threshold = 0,
  nuisance_models = NULL,
  sl.library.outcome = NULL,
  sl.library.treatment = NULL,
  sl.library.missingness = NULL,
  k_folds = 5,
  ps_trunc_level = 0.01,
  outcome_type = "gaussian",
  id_name = NULL
)

Arguments

df

dataframe containing external dataset to apply rule(s) to; must contain Z_list variables that are same as pre-trained rule(s)

Y_name

name of outcome variable in df

A_name

name of treatment variable in df

W_list

character vector containing names of covariates in the dataframe to be used for fitting nuisance models

Z_list

character vector containing names of variables in df used to fit CATE model (variables used in treatment rule; must be same names as used in pre-fit CATE model(s))

CATE_preds

dataframe with CATE predictions for each CATE model & truncation flags for extreme predictions

d_pred

vector of binary treatment decisions (if have existing decisions and just want to estimate effects)

threshold

character vector of decision thresholds for CATE to determine OTR. Values should be positive if `Y_name` is desirable outcome, negative if `Y_name` is undesirable outcome. If threshold is 0, use +0 for desirable, -0 for undesirable.

nuisance_models

list of objects of class `Nuisance` containing outcome, treatment, and missingness SuperLearner models (only include if using pre-fit nuisance models)

sl.library.outcome

character vector of SuperLearner libraries to use to fit the outcome models (if fitting nuisance internally)

sl.library.treatment

character vector of SuperLearner libraries to use to fit the treatment models (if fitting nuisance internally)

sl.library.missingness

character vector of SuperLearner libraries to use to fit the missingness models (if fitting nuisance internally)

k_folds

integer number of folds to use for nuisance model cross-validation (must specify if fitting nuisance models here)

ps_trunc_level

numeric level below which propensity scores will be truncated (to avoid errors in computing AIPTW)

outcome_type

outcome_type specifying continuous (outcome_type = "gaussian") or binary (outcome_type = "binomial") outcome Y (if not providing pre-fit nuisance models)

id_name

name of participant ID variable if present in data

Value

List of "Results" objects for each threshold. Each results object contains:
aggregated_results

Results averaged over each nuisance model

k_fold_results

Results for each nuisance model

decision_df

Dataframe with CATE prediction from every individual CATE_model, flag for CATE truncation, average CATE predition used for treatment decision, treatment decision

k_non_na

equal to number of k_folds used for nuisance models. Ignore– placeholder to keep same format as Results object from internal compute estimates