IDEAS home Printed from https://ideas.repec.org/p/arx/papers/2506.00152.html
   My bibliography  Save this paper

Aligning Language Models with Observational Data: Opportunities and Risks from a Causal Perspective

Author

Listed:
  • Erfan Loghmani

Abstract

Large language models are being widely used across industries to generate content that contributes directly to key performance metrics, such as conversion rates. Pretrained models, however, often fall short when it comes to aligning with human preferences or optimizing for business objectives. As a result, fine-tuning with good-quality labeled data is essential to guide models to generate content that achieves better results. Controlled experiments, like A/B tests, can provide such data, but they are often expensive and come with significant engineering and logistical challenges. Meanwhile, companies have access to a vast amount of historical (observational) data that remains underutilized. In this work, we study the challenges and opportunities of fine-tuning LLMs using observational data. We show that while observational outcomes can provide valuable supervision, directly fine-tuning models on such data can lead them to learn spurious correlations. We present empirical evidence of this issue using various real-world datasets and propose DeconfoundLM, a method that explicitly removes the effect of known confounders from reward signals. Using simulation experiments, we demonstrate that DeconfoundLM improves the recovery of causal relationships and mitigates failure modes found in fine-tuning methods that ignore or naively incorporate confounding variables. Our findings highlight that while observational data presents risks, with the right causal corrections, it can be a powerful source of signal for LLM alignment. Please refer to the project page for code and related resources.

Suggested Citation

  • Erfan Loghmani, 2025. "Aligning Language Models with Observational Data: Opportunities and Risks from a Causal Perspective," Papers 2506.00152, arXiv.org.
  • Handle: RePEc:arx:papers:2506.00152
    as

    Download full text from publisher

    File URL: http://arxiv.org/pdf/2506.00152
    File Function: Latest version
    Download Restriction: no
    ---><---

    References listed on IDEAS

    as
    1. Max H. Farrell & Tengyuan Liang & Sanjog Misra, 2021. "Deep Neural Networks for Estimation and Inference," Econometrica, Econometric Society, vol. 89(1), pages 181-213, January.
    2. Victor Chernozhukov & Denis Chetverikov & Mert Demirer & Esther Duflo & Christian Hansen & Whitney Newey & James Robins, 2018. "Double/debiased machine learning for treatment and structural parameters," Econometrics Journal, Royal Economic Society, vol. 21(1), pages 1-68, February.
    3. Ali Goli & Amandeep Singh, 2024. "Frontiers: Can Large Language Models Capture Human Preferences?," Marketing Science, INFORMS, vol. 43(4), pages 709-722, July.
    4. Elea McDonnell Feit & Ron Berman, 2019. "Test & Roll: Profit-Maximizing A/B Tests," Marketing Science, INFORMS, vol. 38(6), pages 1038-1058, November.
    Full references (including those not matched with items on IDEAS)

    Most related items

    These are the items that most often cite the same works as this one and are cited by the same works as this one.
    1. Kyle Colangelo & Ying-Ying Lee, 2019. "Double debiased machine learning nonparametric inference with continuous treatments," CeMMAP working papers CWP72/19, Centre for Microdata Methods and Practice, Institute for Fiscal Studies.
    2. Jelena Bradic & Weijie Ji & Yuqian Zhang, 2021. "High-dimensional Inference for Dynamic Treatment Effects," Papers 2110.04924, arXiv.org, revised May 2023.
    3. Davide Viviano & Jelena Bradic, 2019. "Synthetic learner: model-free inference on treatments over time," Papers 1904.01490, arXiv.org, revised Aug 2022.
    4. Shi, Chengchun & Zhou, Yunzhe & Li, Lexin, 2024. "Testing directed acyclic graph via structural, supervised and generative adversarial learning," LSE Research Online Documents on Economics 119446, London School of Economics and Political Science, LSE Library.
    5. Combes, Pierre-Philippe & Gobillon, Laurent & Zylberberg, Yanos, 2022. "Urban economics in a historical perspective: Recovering data with machine learning," Regional Science and Urban Economics, Elsevier, vol. 94(C).
    6. Luo, Shanshan & Zhang, Yechi & Li, Wei & Geng, Zhi, 2025. "Multiply robust estimation of causal effects using linked data," Computational Statistics & Data Analysis, Elsevier, vol. 209(C).
    7. Semenova, Vira, 2023. "Debiased machine learning of set-identified linear models," Journal of Econometrics, Elsevier, vol. 235(2), pages 1725-1746.
    8. Kyle Colangelo & Ying-Ying Lee, 2020. "Double Debiased Machine Learning Nonparametric Inference with Continuous Treatments," Papers 2004.03036, arXiv.org, revised Sep 2023.
    9. Nan Liu & Yanbo Liu & Yuya Sasaki, 2024. "Estimation and Inference for Causal Functions with Multiway Clustered Data," Papers 2409.06654, arXiv.org.
    10. Phillip Heiler & Michael C. Knaus, 2021. "Effect or Treatment Heterogeneity? Policy Evaluation with Aggregated and Disaggregated Treatments," Papers 2110.01427, arXiv.org, revised Aug 2023.
    11. Yikun Zhang & Yen-Chi Chen, 2025. "Doubly Robust Inference on Causal Derivative Effects for Continuous Treatments," Papers 2501.06969, arXiv.org, revised Apr 2025.
    12. Huber, Martin, 2019. "An introduction to flexible methods for policy evaluation," FSES Working Papers 504, Faculty of Economics and Social Sciences, University of Freiburg/Fribourg Switzerland.
    13. Daniel Jacob, 2021. "CATE meets ML," Digital Finance, Springer, vol. 3(2), pages 99-148, June.
    14. Chenyu Hou, 2023. "Learning and Subjective Expectation Formation: A Recurrent Neural Network Approach," Discussion Papers dp23-13, Department of Economics, Simon Fraser University.
    15. Achim Ahrens & Christian B. Hansen & Mark E. Schaffer & Thomas Wiemann, 2024. "ddml: Double/debiased machine learning in Stata," Stata Journal, StataCorp LLC, vol. 24(1), pages 3-45, March.
    16. Chen, Jiafeng & Ritzwoller, David M., 2023. "Semiparametric estimation of long-term treatment effects," Journal of Econometrics, Elsevier, vol. 237(2).
    17. Achim Ahrens & Christian B. Hansen & Mark E. Schaffer & Thomas Wiemann, 2025. "Model Averaging and Double Machine Learning," Journal of Applied Econometrics, John Wiley & Sons, Ltd., vol. 40(3), pages 249-269, April.
    18. Stijn Vansteelandt & Oliver Dukes, 2022. "Authors' reply to the Discussion of ‘Assumption‐lean inference for generalised linear model parameters’ by Vansteelandt and Dukes," Journal of the Royal Statistical Society Series B, Royal Statistical Society, vol. 84(3), pages 729-739, July.
    19. Jason Poulos & Shuxi Zeng, 2021. "RNN‐based counterfactual prediction, with an application to homestead policy and public schooling," Journal of the Royal Statistical Society Series C, Royal Statistical Society, vol. 70(4), pages 1124-1139, August.
    20. Alena Skolkova, 2023. "Instrumental Variable Estimation with Many Instruments Using Elastic-Net IV," CERGE-EI Working Papers wp759, The Center for Economic Research and Graduate Education - Economics Institute, Prague.

    More about this item

    NEP fields

    This paper has been announced in the following NEP Reports:

    Statistics

    Access and download statistics

    Corrections

    All material on this site has been provided by the respective publishers and authors. You can help correct errors and omissions. When requesting a correction, please mention this item's handle: RePEc:arx:papers:2506.00152. See general information about how to correct material in RePEc.

    If you have authored this item and are not yet registered with RePEc, we encourage you to do it here. This allows to link your profile to this item. It also allows you to accept potential citations to this item that we are uncertain about.

    If CitEc recognized a bibliographic reference but did not link an item in RePEc to it, you can help with this form .

    If you know of missing items citing this one, you can help us creating those links by adding the relevant references in the same way as above, for each refering item. If you are a registered author of this item, you may also want to check the "citations" tab in your RePEc Author Service profile, as there may be some citations waiting for confirmation.

    For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: arXiv administrators (email available below). General contact details of provider: http://arxiv.org/ .

    Please note that corrections may take a couple of weeks to filter through the various RePEc services.

    IDEAS is a RePEc service. RePEc uses bibliographic data supplied by the respective publishers.