Conformal prediction intervals for the individual treatment effect

TL;DR

This paper introduces conformal prediction intervals for the individual treatment effect that provide coverage guarantees in a flexible non-parametric setting, demonstrating their effectiveness through extensive simulations.

Contribution

It develops new conformal prediction interval procedures for the individual treatment effect with coverage guarantees in complex, non-linear, heteroskedastic, and non-Gaussian settings.

Findings

Neural networks can produce narrower intervals with sufficient data.

The proposed methods achieve reliable coverage in various simulation scenarios.

Complex algorithms outperform simple ones in interval precision when data is ample.

Abstract

We propose several prediction intervals procedures for the individual treatment effect with either finite-sample or asymptotic coverage guarantee in a non-parametric regression setting, where non-linear regression functions, heteroskedasticity and non-Gaussianity are allowed. The construct the prediction intervals we use the conformal method of Vovk et al. (2005). In extensive simulations, we compare the coverage probability and interval length of our prediction interval procedures. We demonstrate that complex learning algorithms, such as neural networks, can lead to narrower prediction intervals than simple algorithms, such as linear regression, if the sample size is large enough.