Aug 2026· Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2· pp. 174-185· 0 citations· 34 references
Abstract
Standard supervised learning algorithms prioritize predictive performance over causal inference, making them ill-suited for conditional average dose response (CADR) estimation. Although specialized CADR estimators have been proposed to address this, the field lacks clarity on which dataset characteristics truly challenge model performance, largely because current benchmarking practices fail to isolate different sources of estimation error. In this work, we analyze these current benchmarking practices and introduce a novel decomposition framework that disentangles the contribution of distinct data-generating components, such as confounding, dose distribution non-uniformity, and response surface complexity, to estimator performance. Applying this scheme to several established benchmarks, we uncover that widely used datasets do not primarily test for confounding robustness, as often assumed, but are instead dominated by challenges arising from non-uniform dose distributions. We further propose a new benchmark dataset with high CADR heterogeneity, where confounding does have a substantial impact. Our results call for a rethinking of current evaluation practices and advocate for more diagnostic, data-centric benchmarks to advance the development of robust CADR estimation methods.
This work analyzes current benchmarking practices and introduces a novel decomposition framework that disentangles the contribution of distinct data-generating components, such as confounding, dose distribution non-uniformity, and response surface complexity, to estimator performance.
Christopher Bockel-Rickermann, Daan Caljon, Toon Vanderschueren et al.· Proceedings of the 32nd ACM...· 2 citations
Causal effect estimation is fundamental to personalized decision-making and policy evaluation, with applications spanning healthcare, economics, and social sciences. However, observational data often suffer from selection bias and the absence of counterfactual outcomes, posing significant challenges to inference accuracy. While recent representation learning-based approaches have shown promise, they fail to fully exploit the rich self-supervised information and causal prior knowledge embedded in the data. To address these limitations, we propose Self-supervised Causal Effects Estimation (SCEE), a novel framework that integrates causal priors with self-supervised learning to construct balanced and predictive representations for causal effects estimation. Experimental results on widely used real-world, semi-synthetic, and synthetic benchmarks demonstrate that SCEE consistently outperforms state-of-the-art methods. To further enhance its effectiveness, we investigate different contrastive sample selection strategies, maximizing the potential of contrastive learning in causal inference. Additionally, we analyze the impact of sample reweighting and show that SCEE inherently mitigates distributional discrepancies between treatment and control groups, eliminating the need for explicit reweighting mechanisms.
Xinshu Li, Shiyi Yang, Venus Haghighi et al.· ACM Transactions on Intellig...· 0 citations
We develop a sensitivity-analysis workflow for causal panel estimators, covering synthetic difference-in-differences, matrix completion, fixed-effect imputation, and group-time average treatment effects. The workflow combines Riesz-representation omitted-variable-bias bounds with partial-$R^2$ robustness values and separates two reporting routes. Route A gives a direct sensitivity profile for additive or projected confounding summarized by outcome-side and Riesz-side partial $R^2$ values. Route B treats observed-covariate benchmarks as auxiliary data only when benchmark-count, alpha-side alignment, model-check, dependence, and dominance diagnostics are credible; otherwise its main role is demotion. We derive estimator-specific Riesz diagnostics and clarify which are fixed-weight, target-level, or first-stage-conditional rather than full derivatives of regularized training maps. Monte Carlo stress tests distinguish calibrated benchmark settings from dominance failure, coarse alpha-side benchmarks, benchmark dependence, noisy covariates, and concentrated SDID weights. In the California tobacco-control panel, the SDID estimate is $-15.60$ packs per capita; corrected finite-donor placebo inference gives standard error 9.49 and add-one $p=0.051$. A refit-weight finite-difference audit changes the Route A nullification robustness value from 0.054 to 0.045, leaving the low-single-digit conclusion unchanged. A county-level minimum-wage application applies the same profile to a multi-cohort staggered panel.
Post-click conversion rate (CVR) is a key metric in various scenarios including e-commerce and advertising, reflecting the efficiency and user experience in the second stage of the conversion process. Estimating the causal effect on CVR is therefore of great practical importance. However, directly applying existing causal inference methods to clicked samples introduces sample selection bias and increased variance due to the exclusion of non-click data. Recent studies on CVR prediction introduce"ideal loss", which optimizes model parameters using an unbiased estimate of the loss over the full sample. Nevertheless, there is no guarantee that unbiasedness of the loss implies unbiasedness of the final estimator. We revisit this challenge from the perspective of semiparametric theory. Specifically, we develop a new doubly robust causal effect estimator for chain-structured outcomes such as CVR, and derive its theoretical properties in detail. It achieves a faster convergence rate compared to nuisance parameters estimation and is therefore more robust when using flexible nonparametric estimators, including neural networks. Based on these theoretical findings, we further design a framework based on targeted regularization to improve numerical stability and practical applicability. Extensive experiments on synthetic and real-world data demonstrate the effectiveness and robustness of our method. In addition, we find that naively combining loss debiasing with standard causal estimators underperforms our method, highlighting the necessity of developing the new estimator tailored to this CVR-style objective with solid theoretical guarantees.
Estimating the average treatment effect (ATE) remains a fundamental challenge in observational studies in the presence of poor or limited covariate overlap. Although the inverse probability weighting (IPW) estimator is a widely used approach for estimating the ATE, its performance can deteriorate substantially when overlap is limited, often resulting in increased finite sample bias and unreliable confidence intervals. One common strategy is to shift attention from the original target estimand, the ATE, to alternative estimands that are less sensitive to extreme propensity scores; however, doing so changes the scientific question of interest. In this manuscript, we propose a novel ATE estimator that preserves the original target estimand, the ATE, while improving robustness to limited overlap. A key idea is that a class of estimands can be expressed by a polynomial function of a hyperparameter characterizing the estimands. Exploiting this structure, the proposed method computes IPW estimators for a sequence of such estimands, models these estimates using a polynomial function, and extrapolates to recover the ATE. We show that the estimator has consistency and asymptotic normality under weaker overlap conditions than required for the standard IPW estimator. Simulation studies demonstrate that the proposed method improves estimation accuracy and interval performance in settings with limited overlap. In addition to its theoretical and empirical advantages, the proposed approach has a clear interpretation and is easy to implement using standard statistical software.
Shunichiro Orihara, S. Komukai, Fan Li· 0 citations
In real-world applications, data are often error-contaminated; naively applying conventional methods without accommodating the measurement error effects often yields inconsistent estimates. Biased results can be further exacerbated by the ultrahigh-dimensionality of covariates. Focusing on the widely used function-on-scalar linear regression model, this article develops new methods for simultaneous parameter estimation and variable selection with error-prone covariates that can be ultrahigh-dimensional. The proposed framework provides flexibility to handle different types of measurement error models. We rigorously establish asymptotic properties of the proposed estimators under mild conditions. Notably, the convergence rates and limiting distributions of the proposed estimators depend on the nature of measurement error. Our findings highlight the significant differences of settings with ultrahigh dimensions compared to scenarios with finite dimensions, as well as the drastically different influence of different measurement error processes. For efficient computation, we design algorithms with data-driven tuning. We evaluate the finite sample performance of the proposed method through simulation studies and a real data application, demonstrating its effectiveness in addressing the challenges posed by error-contaminated and ultrahigh-dimensional of covariates.