EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

2325. Estimating continuous treatment effects from observational data: An empirical evaluation of methods, scenarios, and challenges

Invited abstract in session MC-31: Causal Machine Learning, stream Analytics.

Monday, 12:30-14:00
Room: 046 (building: 208)

Authors (first author is the speaker)

1. Christopher Bockel-Rickermann
Faculty of Economics & Business, KU Leuven
2. Tim Verdonck
University of Antwerp
3. Wouter Verbeke
Faculty of Economics and Business, KU Leuven

Abstract

Our study investigates the use of causal machine learning for estimating the effects of continuous treatments from observational data. This is a challenging task, as in such data treatments are often not randomly assigned, leading to "confounding", a systematic difference between units that were assigned different treatments. Additionally, many modern decision-making problems are high-dimensional and complex, hindering the application of traditional statistical causal inference techniques, and calling for specialized data-driven methodologies. We review established causal machine learning methods for continuous treatment effect estimation, as well as previously presented benchmarking datasets. The goal of our study is to derive sensitivities of model performance to different data-generating processes and accompanying challenges, such as the strength of confounding, the amount of training data, and the complexity of relationships between variables of interest. To further aid practitioners and researchers, we also test several state-of-the-art methodologies from supervised learning in the capability to estimate treatment effects, aiming to further aid the decision when to adopt targeted methodologies for continuous treatment effect estimation.

Keywords

Status: accepted

Back to the list of papers