Article
Biochemical Research Methods
Qian Gao, Yu Zhang, Hongwei Sun, Tong Wang
Summary: This paper reviews the methods for estimating causal effects in observational studies and evaluates their performance in high-dimensional settings. The simulation experiments show that GLiDeR and hdCBPS approaches perform well in terms of estimation accuracy, but further studies are needed for constructing valid confidence intervals.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Biology
P. R. Rosenbaum, D. B. Rubin
Summary: This article emphasizes the importance of design activities prior to examining outcome variables in experimental or observational studies. Balancing the propensity scores is an aspect of the design of observational studies, which can be achieved through matching or balancing on the propensity score. Controlling for observed covariates is a crucial step from association to causation.
Article
Education & Educational Research
Wendy Chan
Summary: As causal inference research progresses, propensity score methods have played an important role in estimation of causal impacts and generalization. Under certain assumptions, propensity score methods can reduce bias and provide valid inferences.
ASIA PACIFIC EDUCATION REVIEW
(2023)
Article
Mathematical & Computational Biology
Siyun Yang, Elizabeth Lorenzi, Georgia Papadogeorgou, Daniel M. Wojdyla, Fan Li, Laine E. Thomas
Summary: This article introduces analytical methods and visualization tools for causal subgroup analysis, including subgroup weighted average treatment effect and overlap weighting method. The proposed methods aim to achieve balance within subgroups and to address the bias-variance tradeoff in SGA. The Connect-S plot is designed for visualizing subgroup covariate balance.
STATISTICS IN MEDICINE
(2021)
Article
Automation & Control Systems
Laura Forastiere, Fabrizia Mealli, Albert Wu, Edoardo M. Airoldi
Summary: In this study, a new covariate-adjustment estimator is proposed to estimate the direct treatment and spillover effects in observational studies on networks. Under assumptions of neighborhood interference and unconfoundedness of individual and neighborhood treatment, the estimator balances individual and neighborhood covariates using a generalized propensity score and conducts adjustment using penalized spline regression. The Bayesian inference strategy accounts for uncertainty in propensity score estimation and incorporates random effects and community detection algorithm to model the correlation among connected units. A simulation study is conducted to evaluate the performance of the proposed estimator on different network topologies.
JOURNAL OF MACHINE LEARNING RESEARCH
(2022)
Article
Health Care Sciences & Services
Stephen P. Fortin, Stephen S. Johnston, Martijn J. Schuemie
Summary: The study compared large-scale CM (LS-CM) and large-scale PSM (LS-PSM) in terms of post-match sample size, covariate balance, and residual confounding. Results showed that LS-CM could find the largest matched sample and achieve covariate balance in small sample studies.
BMC MEDICAL RESEARCH METHODOLOGY
(2021)
Article
Engineering, Multidisciplinary
Hening Huang
Summary: This paper proposes a propensity-based framework for measurement uncertainty analysis. The measurand is regarded as a random variable characterized by central tendency and dispersion. The state of propensity is described by a probability density function (PDF). The framework encodes the state of propensity of the measurand based on all available information about influence quantities.
Article
Computer Science, Artificial Intelligence
Weina Zhang, Xingming Zhang, Dongpei Chen
Summary: Implicit feedback data has various forms of interaction, such as clicking, collection, and play count, posing a challenge to recommendation systems. This paper introduces a Causal Neural Fuzzy Inference algorithm to address missing data in implicit recommendations through joint learning, demonstrating effectiveness and advancement in experiments on realistic datasets.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Mathematics, Interdisciplinary Applications
David A. Stephens, Widemberg S. Nobre, Erica E. M. Moodie, Alexandra M. Schmidt
Summary: This paper studies Bayesian approaches to causal inference using propensity score regression. It emphasizes the importance of considering model mis-specification in causal inference and proposes fully Bayesian inference methods based on decision-theoretic arguments. The authors also propose a computational approach based on the Bayesian bootstrap for inference.
Article
Economics
F. Blasques, P. Gorgi, S. J. Koopman
Summary: The existing methods for handling missing observations in time-varying parameter observation-driven models lead to inconsistent inference. To address this issue, a novel estimation procedure based on indirect inference is proposed, which has been shown to deliver consistent inference in both theoretical and empirical studies.
JOURNAL OF ECONOMETRICS
(2021)
Article
Mathematical & Computational Biology
Youfei Yu, Min Zhang, Xu Shi, Megan E. V. Caram, Roderick J. A. Little, Bhramar Mukherjee
Summary: This article focuses on comparing multiple treatments and binary outcomes using propensity score-based methods, evaluating their relative performance through simulation studies. The methods are applied to assess the effects of four common therapies for castration-resistant advanced-stage prostate cancer. The data consists of medical and pharmacy claims from a large national private health insurance network, with the adverse outcome being admission to the emergency room within a short time window of treatment initiation.
STATISTICS IN MEDICINE
(2021)
Review
Multidisciplinary Sciences
Fan Li, Peng Ding, Fabrizia Mealli
Summary: This paper critically reviews the Bayesian perspective of causal inference based on the potential outcomes framework. It discusses the key elements of Bayesian inference in causal effect estimation, including the assignment mechanism, the general structure of Bayesian inference, and sensitivity analysis. It also highlights unique issues in Bayesian causal inference, such as the role of propensity score, the definition of identifiability, and the choice of priors in different scenarios. Additionally, it emphasizes the importance of covariate overlap and the design stage in Bayesian causal inference, and extends the discussion to complex assignment mechanisms like instrumental variables and time-varying treatments. The strengths and weaknesses of the Bayesian approach to causal inference are identified, and key concepts are illustrated through examples.
PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES
(2023)
Article
Biology
A. Giffin, B. J. Reich, S. Yang, A. G. Rappold
Summary: This article presents a new causal framework for causal inference in the presence of spatial interference. The framework allows for estimation of direct and spill-over effects, taking into account the influence of exposure at nearby locations. The study demonstrates the efficacy of a generalized propensity score in removing measured confounding.
Article
Public, Environmental & Occupational Health
Richard Wyss, Sebastian Schneeweiss, Kueiyu Joshua Lin, David P. Miller, Linda Kalilani, Jessica M. Franklin
Summary: The propensity score is widely used in healthcare database studies to control for a large number of variables. However, there is little research on comparing large-scale propensity score analyses using different methods for confounder selection and adjustment. In this article, a framework is proposed to supplement balance diagnostics and use synthetically generated control studies to screen analyses that show bias caused by measured confounding. This framework utilizes a model for treatment assignment to create pseudo-treatment groups and generate partially simulated datasets that approximate the study population.
Article
Public, Environmental & Occupational Health
Eric R. Cohn, Jose R. Zubizarreta
Summary: This paper introduces a multivariate matching method called profile matching for randomized experiments and observational studies. It finds balanced samples across treatment groups based on covariate profiles, facilitating generalization and personalization of causal inferences. The method achieves covariate balance without requiring a specific matching ratio, making it suitable for diverse treatment categories.