Article
Statistics & Probability
Yu Luo, Daniel J. Graham, Emma J. McCoy
Summary: Frequentist semiparametric theory has been widely used in the development of doubly robust causal estimation. In this paper, a fully semiparametric Bayesian framework is proposed for DR causal inference. The framework combines nonparametric Bayesian procedures with empirical likelihood via semiparametric linear regression, allowing for consistent parameter estimation even with correct specification of only one model.
JOURNAL OF STATISTICAL PLANNING AND INFERENCE
(2023)
Article
Computer Science, Interdisciplinary Applications
Kecheng Wei, Guoyou Qin, Jiajia Zhang, Xuemei Sui
Summary: This study focuses on estimating the ATE and ATT in causal inference, and addresses the challenges of unbalanced covariates and missing outcomes in observational data. The doubly robust estimators remain consistent under correct specification of the propensity score and selection probability models, or the outcome regression model. The asymptotic normality of the estimators is established under regularity conditions, and simulation studies confirm the finite-sample performance of the proposed methods.
COMPUTATIONAL STATISTICS & DATA ANALYSIS
(2022)
Article
Statistics & Probability
Yifan Cui, Hongming Pu, Xu Shi, Wang Miao, Eric Tchetgen Tchetgen
Summary: This article introduces the framework of proximal causal inference and makes contributions to nonparametric proximal identification of the average treatment effect, semiparametric theory for proximal estimation, and characterization of proximal doubly robust estimators. Identification and efficiency results for the average treatment effect on the treated are also provided.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2023)
Article
Biology
Dingke Tang, Dehan Kong, Wenliang Pan, Linbo Wang
Summary: In this paper, the authors propose a causal ball screening method for confounder selection from ultra-high dimensional data sets. Unlike variable selection for prediction modeling, their method aims to control for confounding while improving efficiency in the causal effect estimate. Theoretical analyses and data analysis demonstrate the superiority of their proposal in realistic settings.
Article
Statistics & Probability
Kosuke Imai, Zhichao Jiang, Anup Malaniam
Summary: This article focuses on the nonparametric identification of complier average direct and spillover effects in two-stage randomized experiments with interference and noncompliance. The proposed methodology offers consistent estimators and establishes relationships with popular two-stage least squares estimators. Results are motivated by a randomized evaluation of India's National Health Insurance Program and can be implemented using open-source software.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2021)
Article
Mathematics
Shaojie Wei, Gaorong Li, Zhongzhan Zhang
Summary: In this paper, a doubly robust method using both propensity score and outcome models is proposed to estimate the causal effect of a treatment. By combining estimation equations, a new estimator is derived that retains robustness and improves efficiency even under misspecified conditional mean working models. Through simulation experiments and real data analysis, the proposed method is shown to be competitive with its competitors, in line with asymptotic theory expectations.
COMMUNICATIONS IN MATHEMATICS AND STATISTICS
(2022)
Article
Statistics & Probability
Sat Yajit Ghosh, Zhiqiang Tan
Summary: The paper introduces a method of regularized calibrated estimation for estimating parameters in two working models with high-dimensional data, providing valid Wald confidence intervals for the parameter of interest. A computationally tractable two-step algorithm is proposed, along with rigorous theoretical analysis of the regularized calibrated estimators, demonstrating sufficiently fast rates of convergence.
Article
Computer Science, Artificial Intelligence
Tao Zhang, Hao-Ran Shan, Max A. Little
Summary: This paper introduces a graph neural network model called Causal GraphSAGE (C-GraphSAGE) that incorporates causal inference into the sampling stage of GraphSAGE to improve classifier robustness. Experimental results demonstrate that C-GraphSAGE outperforms other graph neural network models in terms of classification performance when perturbation is present.
PATTERN RECOGNITION
(2022)
Article
Computer Science, Artificial Intelligence
Ana Rita Nogueira, Andrea Pugnana, Salvatore Ruggieri, Dino Pedreschi, Joao Gama
Summary: This article explores the complexity of causality and its significance in the field of artificial intelligence. Causality research aims at obtaining causal knowledge from observational data and estimating the impact of variable changes on outcomes. The article also provides a practical toolkit for researchers and practitioners, including software, datasets, and examples.
WILEY INTERDISCIPLINARY REVIEWS-DATA MINING AND KNOWLEDGE DISCOVERY
(2022)
Article
Biology
Shuwei Li, Limin Peng
Summary: This article fills the research gap by proposing a nonparametric maximum likelihood estimator for causal treatment effects in time-to-event outcomes subject to interval censoring. A reliable and computationally stable expectation-maximization algorithm is designed, and the asymptotic properties of the proposed estimators are established. Extensive simulation studies and an application to colorectal cancer screening data demonstrate the satisfactory performance and advantages of the proposed method over naive methods.
Article
Statistics & Probability
Xiang Zhou
Summary: This article focuses on how the treatment affects outcomes through multiple causally ordered mediators and proposes a method to identify path-specific effects using Pearl's model, applying two estimators to this model and discussing their respective advantages and disadvantages.
JOURNAL OF THE ROYAL STATISTICAL SOCIETY SERIES B-STATISTICAL METHODOLOGY
(2022)
Article
Public, Environmental & Occupational Health
Kieran S. O'Brien, Ahmed M. Arzika, Ramatou Maliki, Abdou Amza, Farouk Manzo, Alio Karamba Mankara, Elodie Lebas, Catherine Cook, Catherine E. Oldenburg, Travis C. Porco, Benjamin F. Arnold, Stefano Bertozzi, Jeremy D. Keenan, Thomas M. Lietman
Summary: A randomized controlled trial in Niger found that biannual azithromycin distribution to children aged 1-59 months reduced all-cause mortality. Analysis of compliance-related subgroups supported the implementation of this intervention in programmatic settings.
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
(2022)
Article
Economics
Arthur Lewbel, Jin Young Choi, Zhuzhu Zhou
Summary: Consider two parametric models, where at least one is correctly specified but the correct one is unknown. Both models share a common vector of parameters. A consistent estimator for this common parameter vector, regardless of the correct model, is referred to as Doubly Robust (DR). We propose a general technique, called Over-identified Doubly Robust (ODR), for constructing DR estimators assuming the models are over-identified. ODR is a simple extension of the Generalized Method of Moments, and we demonstrate its application in various models, particularly in instrumental variables estimation where one of two instrument vectors may be invalid.
JOURNAL OF ECONOMETRICS
(2023)
Article
Mathematical & Computational Biology
Denis Agniel, Boris P. Hejblum, Rodolphe Thiebaut, Layla Parast
Summary: When evaluating the effectiveness of a treatment, policy, or intervention, it can be difficult to measure the desired outcome. In such cases, finding a surrogate outcome that is easier, quicker, or cheaper to measure can be a useful alternative.
Article
Statistics & Probability
Benjamin Gochanour, Sixia Chen, Laura Beebe, David Haziza
Summary: In this study, a semiparametric multiply robust multiple imputation method is developed for evaluating the impact of non-randomized treatment on various health outcomes in observational studies. The method combines information from multiple propensity score models and outcome regression models, and is robust against model misspecifications. The simulation study demonstrates the advantages of the proposed method in terms of balancing efficiency, bias, and coverage probability compared to existing methods.
Article
Statistics & Probability
Aaron Fisher, Edward H. Kennedy
Summary: Estimators based on influence functions (IFs) are effective in estimating specific targets, but can be difficult to understand and therefore underutilized. Through visual examples, a better understanding and trust in IF-based estimators can be fostered.
AMERICAN STATISTICIAN
(2021)
Article
Public, Environmental & Occupational Health
Ashley Naimi, Jacqueline E. Rudolph, Edward H. Kennedy, Abigail Cartus, Sharon Kirkpatrick, David M. Haas, Hyagriv Simhan, Lisa M. Bodnar
Summary: The study introduces the incremental propensity score approach to quantify effects more aligned with assumptions, compared to average treatment effects. Analyzing the relationship between vegetable intake and preeclampsia risk, it is found that incremental PS effects are more useful in addressing public health questions with fewer assumptions.
Article
Statistics & Probability
Matteo Bonvini, Edward H. Kennedy
Summary: This article introduces a novel sensitivity analysis approach, with the sensitivity parameter set as the proportion of unmeasured confounding, to investigate the impact of unmeasured confounders on research outcomes. By making different probability assumptions, sharp boundaries on the average treatment effect are derived, and nonparametric estimators are proposed for flexible covariate adjustment. Additionally, a one-number summary is introduced to assess the robustness of the study to the number of confounded units.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2022)
Article
Immunology
Natalia L. Oliveira, Edward H. Kennedy, Ryan Tibshirani, Andrew Levine, Eileen Martin, Cynthia Munro, Ann B. Ragin, Leah H. Rubin, Ned Sacktor, Eric C. Seaberg, Andrea Weinstein, James T. Becker
Summary: The study longitudinally examined cognitive impairment in HIV patients and identified clinical AIDS diagnosis, hepatitis B or C infection as strong predictors of future impairment. The relative importance of an AIDS diagnosis diminished over time, suggesting a critical change in the epidemic landscape.
Article
Public, Environmental & Occupational Health
Yongqi Zhong, Edward H. Kennedy, Lisa M. Bodnar, Ashley Naimi
Summary: The AIPW package is able to implement AIPW estimation method, including cross-fitting and flexible covariate adjustment. Through simulated RCT and simulation studies, we found that the performance of the AIPW package is comparable to other doubly robust estimation methods, and cross-fitting can significantly reduce bias and improve confidence interval coverage.
AMERICAN JOURNAL OF EPIDEMIOLOGY
(2021)
Article
Public, Environmental & Occupational Health
Gabriel Conzuelo Rodriguez, Lisa M. Bodnar, Maria M. Brooks, Abdus Wahed, Edward H. Kennedy, Enrique Schisterman, Ashley Naimi
Summary: This study compared the performance of correctly specified parametric models and nonparametric models in evaluating effect modification. The findings suggest that generalized linear models have the highest power for detecting effect modification with binary exposures, while the DR-learner is comparable to flexible parametric models for continuous modifiers, especially in capturing quadratic and nonlinear monotonic functions.
AMERICAN JOURNAL OF EPIDEMIOLOGY
(2022)
Editorial Material
Public, Environmental & Occupational Health
Ashley Naimi, Alan E. Mishler, Edward H. Kennedy
AMERICAN JOURNAL OF EPIDEMIOLOGY
(2023)
Article
Mathematical & Computational Biology
Youjin Lee, Edward H. Kennedy, Nandita Mitra
Summary: This article proposes nonparametric estimators for addressing censored survival outcomes, providing estimates for the local average treatment effect on survival probabilities. The proposed estimators possess double-robustness properties and can easily incorporate nonparametric estimation using machine learning tools. Simulation studies demonstrate the flexibility and double robustness of the proposed estimators.
Article
Mathematical & Computational Biology
Asma Bahamyirou, Mireille E. Schnitzer, Edward H. Kennedy, Lucie Blais, Yi Yang
Summary: This paper proposes a two-stage procedure to automatically select effect modifying variables in a Marginal Structural Model (MSM), and verifies its performance through simulation studies and analysis of pregnancy data.
INTERNATIONAL JOURNAL OF BIOSTATISTICS
(2022)
Article
Public, Environmental & Occupational Health
Lisa M. Bodnar, Abigail R. Cartus, Edward H. Kennedy, Sharon Kirkpatrick, Sara M. Parisi, Katherine P. Himes, Corette B. Parker, William A. Grobman, Hyagriv N. Simhan, Robert M. Silver, Deborah A. Wing, Samuel Perry, Ashley Naimi
Summary: This study found that dietary patterns with a high periconceptional density of fruits and vegetables appear to be more protective against preeclampsia for women with higher BMI than for leaner women.
AMERICAN JOURNAL OF EPIDEMIOLOGY
(2022)
Article
Medicine, General & Internal
Yongqi Zhong, Maria M. Brooks, Edward H. Kennedy, Lisa M. Bodnar, Ashley Naimi
Summary: This study demonstrates the use of ensemble machine learning with augmented inverse probability weighting for per-protocol effect estimation in randomized clinical trials. The study evaluates the per-protocol effect of aspirin on pregnancy. The findings suggest that adherence to a low-dose aspirin protocol is associated with an increased probability of hCG-detected pregnancy.
Article
Public, Environmental & Occupational Health
Jacqueline E. Rudolph, David Benkeser, Edward H. Kennedy, Enrique F. Schisterman, Ashley Naimi
Summary: Estimating and identifying average causal effects in longitudinal data with time-varying exposures poses important challenges, particularly in meeting the positivity condition. Using the example of the EAGeR Trial, we found that violating positivity assumption limits our ability to make causal interpretations. More flexible estimation approaches can mitigate the effects of nonpositivity, although it may result in greater uncertainty. When facing nonpositivity, using a flexible and transparent approach, applying parametric methods to fill gaps in the data, or targeting estimands less vulnerable to positivity violations are viable options.
AMERICAN JOURNAL OF EPIDEMIOLOGY
(2022)
Article
Statistics & Probability
Matteo Bonvini, Edward H. Kennedy, Valerie Ventura, Larry Wasserman
Summary: This paper develops statistical methods for causal inference in epidemics and estimates the effect of social mobility on deaths in the first year of the Covid-19 pandemic. By proposing a marginal structural model and conducting several types of sensitivity analyses, the study finds that the data support the idea that reduced mobility causes reduced deaths. However, there is sensitivity to model misspecification and unmeasured confounding, implying caution in interpreting the size of the causal effect. The work highlights the challenges in drawing causal inferences from pandemic data.
ANNALS OF APPLIED STATISTICS
(2022)
Article
Public, Environmental & Occupational Health
Jacqueline E. Rudolph, Kwangho Kim, Edward H. Kennedy, Ashley I. Naimi
Summary: This study aims to assess the incremental effects of time-varying exposure, using data from an experiment on the impact of aspirin on pregnancy outcomes. The results suggest that increasing women's probability of taking aspirin has little impact on pregnancy rates.
Article
Mathematics, Interdisciplinary Applications
Kwangho Kim, Edward H. Kennedy, Ashley Naimi
Summary: This study addresses the issues of dropout and positivity violations in modern longitudinal studies by generalizing the effects of recent incremental interventions. Efficient nonparametric estimators are proposed, showing fast convergence rates and uniform inferential guarantees. In an infinite time horizon setting, incremental intervention effects demonstrate near-exponential gains in statistical precision compared to conventional deterministic effects.
JOURNAL OF CAUSAL INFERENCE
(2021)