design randomized experiments

In adaptive experiments, data-driven design adjustments invalidate conventional statistical inference; existing methods suffer from narrow applicability and strong assumptions. This paper proposes a selective randomization inference framework: it models the data-generating process via a directed acyclic graph (DAG) and implements conditional-on-selection inference within randomization tests. It is the first systematic integration of this principle into randomization-based inference—requiring neither i.i.d. nor parametric modeling assumptions, and accommodating arbitrary adaptive experimental designs. To address disconnected confidence intervals, we innovatively introduce a holdout-unit method. Theoretically and empirically, our approach strictly controls selective Type-I error and constructs valid confidence intervals for homogeneous treatment effects. It substantially outperforms conventional methods in both robustness and generality.

This paper addresses identification and inference for the average treatment effect (ATE) and average treatment effect on the treated (ATT) under covariate-adaptive randomization (CAR) with noncompliance. First, it precisely characterizes the sharp identification sets for ATE and ATT within the CAR framework—where sampling yields non-i.i.d. data—using extremal identification theory. Second, it proposes boundary estimators and confidence interval constructions that are both consistent and asymptotically efficient. Third, it develops a hybrid weighting strategy that jointly leverages empirical sampling frequencies and known randomization probabilities; it establishes that ATE estimation achieves asymptotic efficiency using only empirical frequencies, whereas ATT estimation requires the numerator to employ true compliance probabilities and the denominator empirical frequencies. These results provide both theoretical foundations and practical guidelines for robust causal inference in CAR-based randomized controlled trials.

Traditional randomized controlled trials (RCTs) suffer from cross-group spillovers and interference in multi-population interaction settings—e.g., buyer-seller or creator-subscriber systems—leading to biased estimation of the average treatment effect (ATE). This paper introduces the first systematic multidimensional randomization design framework, relaxing the conventional single-layer randomization assumption. By integrating hierarchical randomization, cross-group assignment, and potential outcomes modeling, it jointly identifies both the ATE and cross-group interference effects. We establish theoretical guarantees: the proposed unbiased estimator is consistent and asymptotically normal. Simulation results demonstrate substantially higher statistical power compared to standard RCTs. This work extends the scope of causal questions addressable through experimental design and provides a rigorous foundation for causal inference in complex intervention environments—particularly platform economies—where interdependent user behaviors induce non-negligible interference.

This paper investigates the performance of the difference-in-means estimator in two-arm randomized experiments under continuous, binary, proportion, and survival endpoints, considering both equal and unequal allocation within Neyman and superpopulation modeling frameworks. Methodologically, it integrates randomization inference, minimax analysis, asymptotic statistics, and NP-hardness arguments, complemented by Monte Carlo simulations. The study establishes, for the first time, that Fisher’s blocked design is asymptotically optimal under Kapelner’s tail criterion. It systematically characterizes the theoretical boundaries between complete randomization (Neyman model) and deterministic perfect balance (superpopulation model), identifying blocked design as the optimal compromise. Both theoretical analysis and simulation results consistently demonstrate that blocking substantially reduces estimation error and robustly outperforms complete randomization and pairwise matching across all endpoint types and allocation ratios.

This paper addresses complex randomized experiments subject to interference between units—such as social network interventions—where standard causal inference assumptions fail. Method: We develop a design-based theoretical framework for estimating treatment effects, introducing a family of design-compatible estimators and a scalar, interpretable measure of “experimental complexity.” We establish its theoretical connection to design variance, derive the asymptotic variance lower bound for unbiased estimation under arbitrary designs, and propose a consistent variance estimator. Contributions/Results: Through interference modeling, design-based inference foundations, and network experiment simulations, we validate our approach on real-world social network data from an insurance adoption study. Our estimators achieve significantly improved estimation accuracy and consistent variance estimation compared to existing methods, providing a theoretically rigorous yet practically implementable analytical framework for complex experimental designs.

This work addresses the challenge of finite-sample inference for individual regression coefficients in fixed-design linear models when errors exhibit dependence or heteroskedasticity. The authors propose a unified randomization testing framework based on group permutations, which rigorously controls Type I error under exchangeable errors and enhances power through design-dependent geometric separation. The approach is further extended to non-exchangeable settings, establishing quantitative robustness for approximately symmetric errors. The study proves that the resulting Type I error bound of level $2\alpha$ is tight and, by integrating a constructive algorithm, sub-Gaussian analysis, and conformal inference, achieves substantially improved power under heavy-tailed designs while preserving finite-sample validity.

This work addresses a key limitation of the standard debiasing approach for the Randomized Response (RR) mechanism under local differential privacy, which often yields negative estimates that lack interpretability as histogram probabilities. The paper presents the first closed-form maximum likelihood estimator (MLE) for the true underlying distribution under RR, circumventing the high computational cost of iterative Bayesian updating (IBU) algorithms. The proposed estimator is theoretically elegant, computationally efficient, and guarantees non-negative, accurate reconstruction of the original histogram. Empirical evaluations demonstrate that this analytical solution significantly outperforms existing correction methods in both estimation accuracy and computational efficiency, offering practitioners a reliable and scalable tool for privacy-preserving frequency estimation.

Classical randomized experiments struggle to identify causal effects in market platforms featuring cross-group strategic interactions and complex spillovers. To address this, we propose a novel multi-stage randomization design and the first finite-sample valid inferential framework that explicitly accounts for interference. We define composite causal parameters—including average direct, primary, and multiple types of spillover effects—that are both identifiable and substantively interpretable. Our method integrates graph-based randomization, hierarchical–clustered randomization, inverse-probability weighting, and Hájek-type bias correction, and establishes a finite-sample central limit theorem. We prove that all estimators achieve √n-consistency and asymptotic normality, ensuring statistical validity while substantially improving estimation precision for spillover effects. The framework is scalable and directly applicable to large-scale online marketplace experiments.

This paper addresses finite-sample inference for regression-adjusted average treatment effect (ATE) estimation in high-dimensional randomized experiments where $p > n$. We propose a design-based non-asymptotic analytical framework—distinct from existing approaches relying on correct model specification or large-sample approximations. For the first time, we establish a non-asymptotic theory for high-dimensional regression-adjusted estimators, leveraging Stein’s exchangeable pair technique, Doob martingales, and Freedman’s inequality to characterize how the geometric structure of covariates governs estimation precision. The resulting confidence intervals are instance-adaptive and data-driven, with explicitly computable widths that remain informative even when $p > n$. This work provides the first rigorous, practical, design-based theoretical foundation for robust causal inference in small-sample, high-dimensional settings.

This study addresses the challenge of unbiased comparison between two pre-specified matching mechanisms in finite populations, where interference from one mechanism can contaminate the evaluation of the other. The authors propose an alternating-path randomization design based on a decomposition of the disagreement set, which requires no assumptions about outcome or behavioral models. By leveraging unique alternating paths and cycle decompositions from graph theory—such as augmenting paths and Eulerian cycle decompositions—the method constructs a controlled interference structure and sets the optimal randomization probability to √2−1 to minimize worst-case variance. The approach integrates Horvitz–Thompson estimation with minimax optimization, establishing unbiasedness of the estimator, proving a finite-population central limit theorem applicable to complex path structures, and extending successfully to many-to-one matching settings with capacity constraints.