Adel Daoud, Richard Johansson, Connor T. Jerzak
arXiv 30 Dec 2025 · Econometrics
arXiv:2601.02400 · PDF · DOI · OpenAlex · Extracted main text
Text-based causal inference increasingly employs textual data as proxies for unobserved confounders, yet this approach introduces a previously undertheorized source of bias: treatment leakage. Treatment leakage occurs when text intended to capture confounding information also contains signals predictive of treatment status, thereby inducing post-treatment bias in causal estimates. Critically, this problem can arise even when documents precede treatment assignment, as authors may employ future-referencing language that anticipates subsequent interventions. Despite growing recognition of this issue, no systematic methods exist for identifying and mitigating treatment leakage in text-as-confounder applications. This paper addresses this gap through three contributions. First, we provide formal statistical and set-theoretic definitions of treatment leakage that clarify when and why bias occurs. Second, we propose four text distillation methods -- similarity-based passage removal, distant supervision classification, salient feature removal, and iterative nullspace projection -- designed to eliminate treatment-predictive content while preserving confounder information. Third, we validate these methods through simulations using synthetic text and an empirical application examining International Monetary Fund structural adjustment programs and child mortality. Our findings indicate that moderate distillation optimally balances bias reduction against confounder retention, whereas overly stringent approaches degrade estimate precision.
appendix boundary found by appendix_titled_section at “Appendix” · 96% of the source is main text. Read the extracted text to check this.
The works this paper leans on most, across its whole bibliography — not restricted to papers in our corpus. Ranked by composite intensity, which combines how often a work is mentioned, how many sections mention it, and how much of that falls in the main text rather than the appendix.
| Reference | Intensity | Mentions | Sections | Main text | |
|---|---|---|---|---|---|
| 1 | Reagan Mozer, Luke Miratrix, Aaron Russell Kaufman, and Jason L. Ana… (2020) Matching with text data: An experimental evaluation of methods for matching documents and of measuring match quality | 1.000 | 5 | 3 | 100% |
| 2 | Shauli Ravfogel, Yanai Elazar, Hila Gonen, Michael Twiton, and Yoav… (2020) Null it out: Guarding protected attributes by iterative nullspace projection | 0.928 | 4 | 3 | 100% |
| 3 | David M. Blei, Andrew Y. Ng, and Michael I. Jordan (2003) Latent Dirichlet allocation | 0.843 | 3 | 3 | 100% |
| 4 | Adel Daoud, Bernhard Reinsberg, Alexander E. Kentikelenis, Thomas H.… (2019) The International Monetary Fund's interventions in food and agriculture: An analysis of loans and conditions self | 0.843 | 3 | 3 | 100% |
| 5 | Margaret E. Roberts, Brandon M. Stewart, and Richard A. Nielsen (2020) Adjusting for confounding with text matching | 0.843 | 3 | 3 | 100% |
| 6 | Axel Dreher (2009) IMF conditionality: Theory and evidence | 0.811 | 4 | 2 | 100% |
| 7 | Thomas Stubbs, Alexander Kentikelenis, Bernhard Reinsberg, and Lawre… (2020) Whatever it takes? the global financial safety net, COVID-19, and developing countries | 0.811 | 4 | 2 | 100% |
| 8 | Adel Daoud, Elias Nosrati, Bernhard Reinsberg, Alexander E. Kentikel… (2017) Impact of International Monetary Fund programs on child health self | 0.737 | 3 | 2 | 100% |
| 9 | Zach Wood-Doughty, Ilya Shpitser, and Mark Dredze (2021) Generating synthetic text data to evaluate causal inference methods | 0.737 | 3 | 2 | 100% |
| 10 | Adel Daoud, Connor Jerzak, and Richard Johansson (2022) Conceptualizing treatment leakage in text-based causal inference self | 0.644 | 4 | 1 | 100% |
Showing the top 10 of 49 scored citations.