uplift: Uplift Curve

In wlattner/hete: Heterogeneous Treatment Effect Modeling

Description

The uplift curve is analgous to the ROC curve used to evaluate the performance of binary classification models. We order the observations by the predicted treatment effect and then compare the cumulative lift against the observed treatment effect. The observed treatment effect is the lift we would achieve using random targeting or selection.

Usage

uplift(y, tmt, pred_te, bins = 10)

Arguments

y	a vector of outcomes.
tmt	a vector indicating which units received treatment.
pred_te	a vector of predicted treatment effects.
bins	the number of bins to use for building the uplift curve. More bins will result in a smoother curve, but this is limited by the number of distinct values pred_te as well as the number of treatment and control units falling in each bin.

Value

a hete_uplift object with:

uplift_curve	A data.frame with the points of the uplift curve.
q	The q/qini score of the model.
ate_observed	The observed average treatment effect.
ate_predicted	The mean predicted treatment effect.

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