hete_split: Split Estimators for Control and Treatment

In wlattner/hete: Heterogeneous Treatment Effect Modeling

Description

Split Estimators for Control and Treatment

Usage

hete_split(x, ...)

## Default S3 method:
hete_split(x, y, tmt, est, tmt_est = est, ctl_est = est,
  ...)

## S3 method for class 'formula'
hete_split(x, data, est, tmt_est = est, ctl_est = est,
  ...)

Arguments

x	a data.frame or matrix with predictor variables measured prior to treatment or unaffected by treatment. Alternatively, this can be a model formula: outcome ~ x1 + x2 | treatment
...	ignored.
y	a vector of outcomes.
tmt	a vector indicating which units received treatment.
est, tmt_est, ctl_est	an estimator to use for modeling the treatment effect. This must be a function which takes two arguments, x and y and returns an object which has an implementation of predict. The predict implementation must return a single vector with the estimated outcome or probability of success in the case of binary outcomes. By default the same estimator, est is used for both models in the meta-estimator. You may optionally select a different estimator for each models. tmt_est and ctl_est are used to estimate the response in the treatment and control groups respectively.
data	a data.frame containing the variables specified in x when using a formula to specify the model.

Value

A hete_split object.

Examples

## Not run: 
library(tidyverse)
data(gotv)

df <- gotv %>%
  filter(treatment %in% c("Control", "Neighbors")) %>%
  mutate(treatment = ifelse(treatment == "Control", 0, 1))

m <- hete_split(voted ~ . | treatment, data = df, est = random_forest)
p <- predict(m, df)

uc <- uplift(df$voted, df$treatment, p)
plot(uc)


## End(Not run)

wlattner/hete documentation built on May 4, 2019, 12:57 a.m.