Valid Inference for Treatment Effects under Multimodal Confounding

Martin Spindler, Philipp Bach, Victor Chernozhukov, Sven Klaassen, Jan Teichert-Kluge, Suhas Vijaykumar
Proceedings of the Fifth Conference on Causal Learning and Reasoning, PMLR 323:222-247, 2026.

Abstract

This paper provides methods for the valid estimation and inference of treatment effects in the presence of unstructured, multimodal data, namely text and images, as confounders. We develop a neural network architecture that is adapted to the double machine learning (DML) framework, specifically the partially linear model. An additional contribution of our paper is a new method to generate a semi-synthetic dataset which can be used to evaluate the performance of causal effect estimation in the presence of text and images as confounders. The proposed methods and architectures are evaluated on a correspondingly generated semi-synthetic dataset and compared to standard approaches, highlighting the potential benefit of using text and images directly in causal studies and of our approach. In the experiments, our methods performs well and achieves the nominal coverage. Our findings might be valuable for researchers and practitioners in economics, marketing, finance, medicine and data science in general who are interested in estimating causal quantities using non-traditional data.

Cite this Paper

BibTeX
@InProceedings{pmlr-v323-spindler26a, title = {Valid Inference for Treatment Effects under Multimodal Confounding}, author = {Spindler, Martin and Bach, Philipp and Chernozhukov, Victor and Klaassen, Sven and Teichert-Kluge, Jan and Vijaykumar, Suhas}, booktitle = {Proceedings of the Fifth Conference on Causal Learning and Reasoning}, pages = {222--247}, year = {2026}, editor = {Mazaheri, Bijan and Hanson, Niels Richard}, volume = {323}, series = {Proceedings of Machine Learning Research}, month = {06--08 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v323/main/assets/spindler26a/spindler26a.pdf}, url = {https://proceedings.mlr.press/v323/spindler26a.html}, abstract = {This paper provides methods for the valid estimation and inference of treatment effects in the presence of unstructured, multimodal data, namely text and images, as confounders. We develop a neural network architecture that is adapted to the double machine learning (DML) framework, specifically the partially linear model. An additional contribution of our paper is a new method to generate a semi-synthetic dataset which can be used to evaluate the performance of causal effect estimation in the presence of text and images as confounders. The proposed methods and architectures are evaluated on a correspondingly generated semi-synthetic dataset and compared to standard approaches, highlighting the potential benefit of using text and images directly in causal studies and of our approach. In the experiments, our methods performs well and achieves the nominal coverage. Our findings might be valuable for researchers and practitioners in economics, marketing, finance, medicine and data science in general who are interested in estimating causal quantities using non-traditional data.} }
Endnote
%0 Conference Paper %T Valid Inference for Treatment Effects under Multimodal Confounding %A Martin Spindler %A Philipp Bach %A Victor Chernozhukov %A Sven Klaassen %A Jan Teichert-Kluge %A Suhas Vijaykumar %B Proceedings of the Fifth Conference on Causal Learning and Reasoning %C Proceedings of Machine Learning Research %D 2026 %E Bijan Mazaheri %E Niels Richard Hanson %F pmlr-v323-spindler26a %I PMLR %P 222--247 %U https://proceedings.mlr.press/v323/spindler26a.html %V 323 %X This paper provides methods for the valid estimation and inference of treatment effects in the presence of unstructured, multimodal data, namely text and images, as confounders. We develop a neural network architecture that is adapted to the double machine learning (DML) framework, specifically the partially linear model. An additional contribution of our paper is a new method to generate a semi-synthetic dataset which can be used to evaluate the performance of causal effect estimation in the presence of text and images as confounders. The proposed methods and architectures are evaluated on a correspondingly generated semi-synthetic dataset and compared to standard approaches, highlighting the potential benefit of using text and images directly in causal studies and of our approach. In the experiments, our methods performs well and achieves the nominal coverage. Our findings might be valuable for researchers and practitioners in economics, marketing, finance, medicine and data science in general who are interested in estimating causal quantities using non-traditional data.
APA
Spindler, M., Bach, P., Chernozhukov, V., Klaassen, S., Teichert-Kluge, J. & Vijaykumar, S.. (2026). Valid Inference for Treatment Effects under Multimodal Confounding. Proceedings of the Fifth Conference on Causal Learning and Reasoning, in Proceedings of Machine Learning Research 323:222-247 Available from https://proceedings.mlr.press/v323/spindler26a.html.

Related Material