Abstract:In this paper, we present our winning modeling approach, DiConfounder, for the Atlantic Causal Inference Conference (ACIC) 2022 Data Science data challenge. Our method ranks 1st in RMSE and 5th in coverage among the 58 submissions. We propose a transformed outcome estimator by connecting the difference-in-difference and conditional average treatment effect estimation problems. Our comprehensive multistage pipeline encompasses feature engineering, missing value imputation, outcome and propensity score modeling, treatment effects modeling, and SATT and uncertainty estimations. Our model achieves remarkably accurate predictions, with an overall RMSE as low as 11 and 84.5% coverage. Further discussions explore various methods for constructing confidence intervals and analyzing the limitations of our approach under different data generating process settings. We provide evidence that the clustered data structure is the key to success. We also release the source code on GitHub for practitioners to adopt and adapt our methods.
{"title":"Estimating Treatment Effects over Time with Causal Forests: An application to the ACIC 2022 Data Challenge","authors":"Shu Wan, Guanghui Zhang","doi":"10.1353/obs.2023.0026","DOIUrl":"https://doi.org/10.1353/obs.2023.0026","url":null,"abstract":"Abstract:In this paper, we present our winning modeling approach, DiConfounder, for the Atlantic Causal Inference Conference (ACIC) 2022 Data Science data challenge. Our method ranks 1st in RMSE and 5th in coverage among the 58 submissions. We propose a transformed outcome estimator by connecting the difference-in-difference and conditional average treatment effect estimation problems. Our comprehensive multistage pipeline encompasses feature engineering, missing value imputation, outcome and propensity score modeling, treatment effects modeling, and SATT and uncertainty estimations. Our model achieves remarkably accurate predictions, with an overall RMSE as low as 11 and 84.5% coverage. Further discussions explore various methods for constructing confidence intervals and analyzing the limitations of our approach under different data generating process settings. We provide evidence that the clustered data structure is the key to success. We also release the source code on GitHub for practitioners to adopt and adapt our methods.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"9 1","pages":"59 - 71"},"PeriodicalIF":0.0,"publicationDate":"2023-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"43810955","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract:In this American Causal Inference Conference (ACIC) 2022 challenge submission, the canonical difference-in-differences (DID) estimator has been used with inverse probability weighting (IPW) and strong simplifying assumptions to produce a benchmark model of the sample average treatment effect on the treated (SATT). Despite the restrictive assumptions and simple model, satisfactory performance in both point estimate and confidence intervals was observed, ranking in the top half of the competition.
{"title":"Inverse Probability Weighting Difference-in-Differences (IPWDID)","authors":"Yuqin Wei, M. Epland, Jingyuan Liu","doi":"10.1353/obs.2023.0027","DOIUrl":"https://doi.org/10.1353/obs.2023.0027","url":null,"abstract":"Abstract:In this American Causal Inference Conference (ACIC) 2022 challenge submission, the canonical difference-in-differences (DID) estimator has been used with inverse probability weighting (IPW) and strong simplifying assumptions to produce a benchmark model of the sample average treatment effect on the treated (SATT). Despite the restrictive assumptions and simple model, satisfactory performance in both point estimate and confidence intervals was observed, ranking in the top half of the competition.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"9 1","pages":"73 - 81"},"PeriodicalIF":0.0,"publicationDate":"2023-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"49451652","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract:We present the lmtp R package for causal inference from longitudinal observational or randomized studies. This package implements the estimators of Díaz et al. (2021) for estimating general non-parametric causal effects based on modified treatment policies. Modified treatment policies generalize static and dynamic interventions, making lmtp and all-purpose package for non-parametric causal inference in observational studies. The methods provided can be applied to both point-treatment and longitudinal settings, and can account for time-varying exposure, covariates, and right censoring thereby providing a very general tool for causal inference. Additionally, two of the provided estimators are based on flexible machine learning regression algorithms, and avoid bias due to parametric model misspecification while maintaining valid statistical inference.
{"title":"lmtp: An R Package for Estimating the Causal Effects of Modified Treatment Policies","authors":"Nicholas T Williams, I. Díaz","doi":"10.1353/obs.2023.0019","DOIUrl":"https://doi.org/10.1353/obs.2023.0019","url":null,"abstract":"Abstract:We present the lmtp R package for causal inference from longitudinal observational or randomized studies. This package implements the estimators of Díaz et al. (2021) for estimating general non-parametric causal effects based on modified treatment policies. Modified treatment policies generalize static and dynamic interventions, making lmtp and all-purpose package for non-parametric causal inference in observational studies. The methods provided can be applied to both point-treatment and longitudinal settings, and can account for time-varying exposure, covariates, and right censoring thereby providing a very general tool for causal inference. Additionally, two of the provided estimators are based on flexible machine learning regression algorithms, and avoid bias due to parametric model misspecification while maintaining valid statistical inference.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"9 1","pages":"103 - 122"},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"47362691","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract:Inverse probability of treatment weighted estimators and doubly robust estimators (including augmented inverse probability of treatment weight and targeted minimum loss estimators) are widely used in causal inference to estimate and draw inference about the average effect of a treatment. As an intermediate step, these estimators require estimation of key nuisance parameters, which are often regression functions. Typically, regressions are estimated using maximum likelihood and parametric models. Confidence intervals and p-values may be computed based on standard asymptotic results, such as the central limit theorem, the delta method, and the nonparametric bootstrap. However, in high-dimensional settings, maximum likelihood estimation often breaks down and standard procedures no longer yield correct inference. Instead, we may rely on adaptive estimators of nuisance parameters to construct flexible regression estimators. However, use of adaptive estimators poses a challenge for performing statistical inference about an estimated treatment effect. While doubly robust estimators facilitate inference when all relevant regression functions are consistently estimated, the same cannot be said when at least one nuisance estimator is inconsistent. drtmle implements doubly robust confidence intervals and hypothesis tests for targeted minimum loss estimates of the average treatment effect, in addition to several other recently proposed estimators of the average treatment effect.
{"title":"Doubly-Robust Inference in R using drtmle","authors":"D. Benkeser, N. Hejazi","doi":"10.1353/obs.2023.0017","DOIUrl":"https://doi.org/10.1353/obs.2023.0017","url":null,"abstract":"Abstract:Inverse probability of treatment weighted estimators and doubly robust estimators (including augmented inverse probability of treatment weight and targeted minimum loss estimators) are widely used in causal inference to estimate and draw inference about the average effect of a treatment. As an intermediate step, these estimators require estimation of key nuisance parameters, which are often regression functions. Typically, regressions are estimated using maximum likelihood and parametric models. Confidence intervals and p-values may be computed based on standard asymptotic results, such as the central limit theorem, the delta method, and the nonparametric bootstrap. However, in high-dimensional settings, maximum likelihood estimation often breaks down and standard procedures no longer yield correct inference. Instead, we may rely on adaptive estimators of nuisance parameters to construct flexible regression estimators. However, use of adaptive estimators poses a challenge for performing statistical inference about an estimated treatment effect. While doubly robust estimators facilitate inference when all relevant regression functions are consistently estimated, the same cannot be said when at least one nuisance estimator is inconsistent. drtmle implements doubly robust confidence intervals and hypothesis tests for targeted minimum loss estimates of the average treatment effect, in addition to several other recently proposed estimators of the average treatment effect.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"9 1","pages":"43 - 78"},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"41508466","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract:Most nutritional epidemiology studies investigating diet-disease trends use unsupervised dimension reduction methods, like principal component regression (PCR) and sparse PCR (SPCR), to create dietary patterns. Supervised methods, such as partial least squares (PLS), sparse PLS (SPLS), and Lasso, offer the possibility of more concisely summarizing the foods most related to a disease. In this study we evaluate these five methods for interpretable reduction of food frequency questionnaire (FFQ) data when analyzing a univariate continuous cardiac-related outcome via a simulation study and data application. We also demonstrate that to control for covariates, various scientific premises require different adjustment approaches when using PLS. To emulate food groups, we generated blocks of normally distributed predictors with varying intra-block covariances; only nine of 24 predictors contributed to the normal response. When block covariances were informed by FFQ data, the only methods that performed variable selection were Lasso and SPLS, which selected two and four irrelevant variables, respectively. SPLS had the lowest prediction error, and both PLS-based methods constructed four patterns, while PCR and SPCR created 24 patterns. These methods were applied to 120 FFQ variables and baseline body mass index (BMI) from the Multi-Ethnic Study of Atherosclerosis, which includes 6814 participants aged 45-84, and we adjusted for age, gender, race/ethnicity, exercise, and total energy intake. From 120 variables, PCR created 17 BMI-related patterns and PLS selected one pattern; SPLS only used five variables to create two patterns. All methods exhibited similar predictive performance. Specifically, SPLS’s first pattern highlighted hamburger and diet soda intake (positive associations with BMI), reflecting a fast food diet. By selecting fewer patterns and foods, SPLS can create interpretable dietary patterns while maintaining predictive ability.
{"title":"Comparison of dimension reduction methods for the identification of heart-healthy dietary patterns","authors":"Natalie C. Gasca, R. McClelland","doi":"10.1353/obs.2023.0020","DOIUrl":"https://doi.org/10.1353/obs.2023.0020","url":null,"abstract":"Abstract:Most nutritional epidemiology studies investigating diet-disease trends use unsupervised dimension reduction methods, like principal component regression (PCR) and sparse PCR (SPCR), to create dietary patterns. Supervised methods, such as partial least squares (PLS), sparse PLS (SPLS), and Lasso, offer the possibility of more concisely summarizing the foods most related to a disease. In this study we evaluate these five methods for interpretable reduction of food frequency questionnaire (FFQ) data when analyzing a univariate continuous cardiac-related outcome via a simulation study and data application. We also demonstrate that to control for covariates, various scientific premises require different adjustment approaches when using PLS. To emulate food groups, we generated blocks of normally distributed predictors with varying intra-block covariances; only nine of 24 predictors contributed to the normal response. When block covariances were informed by FFQ data, the only methods that performed variable selection were Lasso and SPLS, which selected two and four irrelevant variables, respectively. SPLS had the lowest prediction error, and both PLS-based methods constructed four patterns, while PCR and SPCR created 24 patterns. These methods were applied to 120 FFQ variables and baseline body mass index (BMI) from the Multi-Ethnic Study of Atherosclerosis, which includes 6814 participants aged 45-84, and we adjusted for age, gender, race/ethnicity, exercise, and total energy intake. From 120 variables, PCR created 17 BMI-related patterns and PLS selected one pattern; SPLS only used five variables to create two patterns. All methods exhibited similar predictive performance. Specifically, SPLS’s first pattern highlighted hamburger and diet soda intake (positive associations with BMI), reflecting a fast food diet. By selecting fewer patterns and foods, SPLS can create interpretable dietary patterns while maintaining predictive ability.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"9 1","pages":"123 - 156"},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"49570747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract:Instrumental variable (IV) strategies are widely used to estimate causal effects in economics, political science, epidemiology, sociology, psychology, and other fields. When there is unobserved heterogeneity in causal effects, standard linear IV estimators only represent effects for complier subpopulations (Imbens and Angrist, 1994). Marginal treatment effect (MTE) methods (Heckman and Vytlacil, 1999, 2005) allow researchers to use additional assumptions to extrapolate beyond complier subpopulations. We discuss a flexible framework for MTE methods based on linear regression and the generalized method of moments. We show how to implement the framework using the ivmte package for R.
{"title":"ivmte: An R Package for Extrapolating Instrumental Variable Estimates Away From Compliers*","authors":"Joshua Shea, Alexander Torgovitsky","doi":"10.1353/obs.2023.0016","DOIUrl":"https://doi.org/10.1353/obs.2023.0016","url":null,"abstract":"Abstract:Instrumental variable (IV) strategies are widely used to estimate causal effects in economics, political science, epidemiology, sociology, psychology, and other fields. When there is unobserved heterogeneity in causal effects, standard linear IV estimators only represent effects for complier subpopulations (Imbens and Angrist, 1994). Marginal treatment effect (MTE) methods (Heckman and Vytlacil, 1999, 2005) allow researchers to use additional assumptions to extrapolate beyond complier subpopulations. We discuss a flexible framework for MTE methods based on linear regression and the generalized method of moments. We show how to implement the framework using the ivmte package for R.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"9 1","pages":"1 - 42"},"PeriodicalIF":0.0,"publicationDate":"2023-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"45602939","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract:In this commentary, I provide a personal perspective on how the propensity score has become important to epidemiology.
摘要:在这篇评论中，我从个人角度阐述了倾向评分对流行病学的重要性。
{"title":"The central role of the propensity score in epidemiology","authors":"Brian K. Lee","doi":"10.1353/obs.2023.0004","DOIUrl":"https://doi.org/10.1353/obs.2023.0004","url":null,"abstract":"Abstract:In this commentary, I provide a personal perspective on how the propensity score has become important to epidemiology.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"19 1","pages":"55 - 57"},"PeriodicalIF":0.0,"publicationDate":"2023-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"66460632","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract:Rosenbaum and Rubin’s paper is highly cited because the basic idea is simple and insightful, and it has applications to important practical problems in treatment comparisons with observational data, and selection bias and nonresponse in surveys. I discuss several issues related to the method, including use of the propensity score for weighting or prediction, and two robust methods that use the propensity score as a covariate and can be more efficient that weighting when the weights are highly variable, namely Penalized Spline of Propensity Prediction (PSPP) and Penalized Spline of Propensity for Treatment Comparisons (PENCOMP). Approaches to addressing highly variable weights are discussed, including omitting variables in the propensity model that are unrelated to outcomes, and redefining the estimand.
{"title":"Some Reflections on Rosenbaum and Rubin’s Propensity Score Paper","authors":"R. Little","doi":"10.1353/obs.2023.0006","DOIUrl":"https://doi.org/10.1353/obs.2023.0006","url":null,"abstract":"Abstract:Rosenbaum and Rubin’s paper is highly cited because the basic idea is simple and insightful, and it has applications to important practical problems in treatment comparisons with observational data, and selection bias and nonresponse in surveys. I discuss several issues related to the method, including use of the propensity score for weighting or prediction, and two robust methods that use the propensity score as a covariate and can be more efficient that weighting when the weights are highly variable, namely Penalized Spline of Propensity Prediction (PSPP) and Penalized Spline of Propensity for Treatment Comparisons (PENCOMP). Approaches to addressing highly variable weights are discussed, including omitting variables in the propensity model that are unrelated to outcomes, and redefining the estimand.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"9 1","pages":"69 - 75"},"PeriodicalIF":0.0,"publicationDate":"2023-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"49646766","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract:The foundational propensity score paper by Rosenbaum and Rubin (1983a) laid the foundation for a set of methods widely used in the design of non-experimental studies. This commentary reflects on the theoretical contributions of that paper –especially the idea of the propensity score as a balancing score –as well as on the wide variety of contexts in which the general idea of a balancing score has since been applied. Areas in which the fundamental ideas of a balancing score –which can help equate two groups on the basis of a set of covariates –have been extended include mediation analysis and generalizability. The commentary also touches on common misperceptions regarding propensity scores, and on the key role of the “other” Rosenbaum and Rubin (1983b) paper, which laid out a method for assessing the sensitivity of study results to violation of the key assumption underlying most uses of propensity scores –that of no unmeasured confounding. All together, this body of work has changed how many fields conduct non-experimental studies, and other related types of studies, and with many applications and extensions yet to come.
{"title":"What is a propensity score? Applications and extensions of balancing score methods","authors":"E. Stuart","doi":"10.1353/obs.2023.0011","DOIUrl":"https://doi.org/10.1353/obs.2023.0011","url":null,"abstract":"Abstract:The foundational propensity score paper by Rosenbaum and Rubin (1983a) laid the foundation for a set of methods widely used in the design of non-experimental studies. This commentary reflects on the theoretical contributions of that paper –especially the idea of the propensity score as a balancing score –as well as on the wide variety of contexts in which the general idea of a balancing score has since been applied. Areas in which the fundamental ideas of a balancing score –which can help equate two groups on the basis of a set of covariates –have been extended include mediation analysis and generalizability. The commentary also touches on common misperceptions regarding propensity scores, and on the key role of the “other” Rosenbaum and Rubin (1983b) paper, which laid out a method for assessing the sensitivity of study results to violation of the key assumption underlying most uses of propensity scores –that of no unmeasured confounding. All together, this body of work has changed how many fields conduct non-experimental studies, and other related types of studies, and with many applications and extensions yet to come.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":"9 1","pages":"113 - 117"},"PeriodicalIF":0.0,"publicationDate":"2023-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"45664986","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：