Title
Simulation-based model selection for causal effect estimation
Abstract
The causal effect of a binary treatment is typically defined as the difference in expected outcomes under treatment and no treatment. However, due to the fundamental problem of causal inference, individual-level effects are never observed, and the true effect is unknown in real-world data. As a result, selecting among multiple candidate estimators is inherently challenging. A broad range of methods has been proposed, including meta-learners, matching and weighting using, e.g., propensity scores, targeted maximum likelihood estimation, and double machine learning. Empirical studies show that no single method performs best across all scenarios, making model selection crucial, yet difficult in practice.
One promising strategy is simulation-based model selection, where a generative model is learned from observational data and used to simulate datasets with known ground truth, enabling performance comparison of candidate estimators. However, the reliability of such approaches depends on how well the learned simulations reflect the underlying data-generating process.
We propose a comprehensive framework to evaluate simulation-based model selection across a range of data-generating processes that preserve real-world complexity. This setup further allows us to assess robustness to model misspecification, structural complexity, and unobserved confounding, and may be used to provide a principled, dataset-specific strategy for model selection in causal effect estimation.
Bio
Postdoctoral Fellow Milena Pavlovic, Department of Informatics, University of Oslo
Speaker is presenting in person.
Practical
The Tuesdays seminars series are devoted to various topics relevant for the Integreat´s research focus. Presenters from the Integreat community and beyond have 40 minutes to present, followed by a group discussion.
Seminars are open for attendance for everybody.
For those unable to attend in person: https://uio.zoom.us/j/62535744013
Point of contacts:
- Organiser and moderator: Pooya Zakeri, University of Oslo
- Integreat Administrative Leader Maria Dikova, University of Oslo