Estimation APITidy functions for performing an analysis of heterogeneous treatment effects. Once a configuration has been defined (like by the Recipe API), these functions are the workhorses that perform all estimation. 


Attach an 

Define splits for crossfitting 

Estimate models of nuisance functions 

Construct Pseudooutcomes 

Estimate Quantities of Interest 

Recipe APITidy functions for configuring a “recipe” for how to estimate heterogeneous treatment effects. This is the easiest way to get started with setting up a configuration for an HTE analysis. 

Create a basic config for HTE estimation 

Add an additional model to the propensity score ensemble 

Uses a known propensity score 

Add an additional diagnostic to the propensity score 

Add an additional model to the outcome ensemble 

Add an additional diagnostic to the outcome model 

Add an additional model to the joint effect ensemble 

Add an additional diagnostic to the effect model 

Adds moderators to the configuration 

Adds variable importance information 

Removes variable importance information 

Model ConfigurationClasses to define the configuration of models to be used in the eventual HTE analysis. These are the classes which define the underlying configurations in the Recipe API. They’re most useful for advanced users who want the most granular control over their analysis, but most users will be best served by the Recipe API. 

Base Class 

Base Class of Model Configurations 

Models for displaying resultsThese model configurations include valid standard errors and are well suited for providing usable output to be plotted or returned for formal inference. 

Configuration for a Stratification Estimator 

Configuration for a Kernel Smoother 

Models for nuisance functionsThese models are most useful for estimating nuisance functions in the course of HTE estimation. In particular, 

Configuration of Known Model 

Configuration of a Constant Estimator 

Configuration for a SuperLearner Ensemble 

Configuration of SuperLearner Submodel 

Analysis ConfigurationThese classes configure the overall shape of the HTE analysis: essentially, how all the various components and models should fit together. They explain what models should be estimated and how those models should be combined into relevant quantities of interest. These, too, underlie the Recipe API, and should rarely need to be used directly. 

Configuration of Model Diagnostics 

Configuration of Quantities of Interest 

Configuration of Quantities of Interest 

Configuration of Variable Importance 

Configuration of HTE estimandsThere are two configurations for HTE estimands. 

Configuration of Marginal CATEs 

Configuration of Partial CATEs 

Internal FunctionsThe remaining functions are not really useful for endusers. Documentation is provided in order to provide additional details about the internal workings of the methods. 

Elastic net regression with pairwise interactions 

Prediction for an SL.glmnet object 

Calculates a SATE and a PATE using AIPW 

Calculate diagnostics 

Regression ROC Curve calculation 

Calculate Linear Variable Importance of HTEs 

Calculate Variable Importance of HTEs 

Partition the data into folds 

R6 class to represent data to be used in estimating a model 

Checks that a dataframe has an attached configuration for HTEs 

Checks that an appropriate identifier has been provided 

Checks that nuisance models have been estimated and exist in the supplied dataset. 

Checks that splits have been properly created. 

Checks that an appropriate weighting variable has been provided 

Fits a plugin model using the appropriate settings 

Fits a propensity score model using the appropriate settings 

Fits a Tlearner using the appropriate settings 

Fits a treatment effect model using the appropriate settings 

Removes rows which have missing data on any of the supplied columns. 