Abstract:The estimation of causal effects in observational studies is usually limited by the lack of randomization, which can result in different treatment or exposure groups differing systematically with respect to characteristics that influence outcomes. To remove such systematic differences, methods to ’’balance” subjects on observed covariates across treatment or exposure levels have been developed over the past three decades. These methods have been primarily developed in settings with binary treatment or exposures. However, in many observational studies, the exposures are continuous instead of being binary or discrete, and are usually considered as doses of treatment. In this manuscript we consider estimating the causal effect of early childhood lead exposure on youth academic achievement, where the exposure variable blood lead concentration can take any values that are greater than or equal to 0, using three balancing methods: propensity score analysis, non-bipartite matching, and Bayesian regression trees. We find some evidence that the standard logistic regression analysis controlling for age and socioeconomic confounders used in previous analyses (Zhang et al. (2013)) overstates the effect of lead exposure on performance on standardized mathematics and reading examinations; however, significant declines remain, including at doses currently below the recommended exposure levels.
摘要：观察性研究中因果效应的估计通常受到缺乏随机化的限制，这可能导致不同的治疗或暴露组在影响结果的特征方面存在系统性差异。为了消除这种系统性差异，在过去三十年中，已经开发出了在不同治疗或暴露水平的观察到的协变量上“平衡”受试者的方法。这些方法主要是在二元治疗或暴露的环境中开发的。然而，在许多观察性研究中，暴露是连续的，而不是二元或离散的，通常被视为治疗剂量。在这篇手稿中，我们考虑使用三种平衡方法来估计儿童早期铅暴露对青少年学业成绩的因果影响，其中暴露变量血铅浓度可以取大于或等于0的任何值：倾向得分分析、非二分匹配和贝叶斯回归树。我们发现一些证据表明，先前分析中使用的控制年龄和社会经济混杂因素的标准逻辑回归分析（Zhang et al.（2013））夸大了铅暴露对标准化数学和阅读考试成绩的影响；然而，仍有显著下降，包括目前低于建议暴露水平的剂量。
{"title":"Application of Propensity Scores to a Continuous Exposure: Effect of Lead Exposure in Early Childhood on Reading and Mathematics Scores","authors":"M. Elliott, Nanhua Zhang, Dylan S. Small","doi":"10.1353/obs.2015.0002","DOIUrl":"https://doi.org/10.1353/obs.2015.0002","url":null,"abstract":"Abstract:The estimation of causal effects in observational studies is usually limited by the lack of randomization, which can result in different treatment or exposure groups differing systematically with respect to characteristics that influence outcomes. To remove such systematic differences, methods to ’’balance” subjects on observed covariates across treatment or exposure levels have been developed over the past three decades. These methods have been primarily developed in settings with binary treatment or exposures. However, in many observational studies, the exposures are continuous instead of being binary or discrete, and are usually considered as doses of treatment. In this manuscript we consider estimating the causal effect of early childhood lead exposure on youth academic achievement, where the exposure variable blood lead concentration can take any values that are greater than or equal to 0, using three balancing methods: propensity score analysis, non-bipartite matching, and Bayesian regression trees. We find some evidence that the standard logistic regression analysis controlling for age and socioeconomic confounders used in previous analyses (Zhang et al. (2013)) overstates the effect of lead exposure on performance on standardized mathematics and reading examinations; however, significant declines remain, including at doses currently below the recommended exposure levels.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/obs.2015.0002","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"43901491","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}
The economist Paul Samuelson said, “My belief is that nothing that can be expressed by mathematics cannot be expressed by careful use of literary words.” Paul Rosenbaum brings this perspective to causal inference in his new book Observation and Experiment: An Introduction to Causal Inference (Harvard University Press, 2017). The book is a luminous presentation of concepts and strategies for causal inference with a minimum of technical material. An example of how Rosenbaum explains causal inference in a literary way is his use of a passage from Robert Frost’s poem “The Road Not Taken” to illuminate how causal questions involve comparing potential outcomes under two or more treatments where we can only see one potential outcome:
{"title":"Book review of “Observation and Experiment: An Introduction to Causal Inference” by Paul R. Rosenbaum","authors":"Dylan S. Small","doi":"10.1353/obs.2017.0008","DOIUrl":"https://doi.org/10.1353/obs.2017.0008","url":null,"abstract":"The economist Paul Samuelson said, “My belief is that nothing that can be expressed by mathematics cannot be expressed by careful use of literary words.” Paul Rosenbaum brings this perspective to causal inference in his new book Observation and Experiment: An Introduction to Causal Inference (Harvard University Press, 2017). The book is a luminous presentation of concepts and strategies for causal inference with a minimum of technical material. An example of how Rosenbaum explains causal inference in a literary way is his use of a passage from Robert Frost’s poem “The Road Not Taken” to illuminate how causal questions involve comparing potential outcomes under two or more treatments where we can only see one potential outcome:","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/obs.2017.0008","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"41412544","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}
Statistics lost one of our giants when Larry Brown passed away in February, 2018 at the age of 77. Many lost a friend, a mentor and a teacher. I had the good fortune to be Larry’s colleague for the past 16 years. While I wasn’t as close to him as many others, Larry was my friend and I looked up to him. Larry loved thinking. When Larry got interested in something and scratched his head, I could see his enjoyment. Steam seemed to come out as Larry was thinking and then he happily shared his thoughts. Larry’s classes and talks were full of insights. After joining the faculty at Wharton, I attended Larry’s linear models first year PhD course even though I had seen the material before. I was glad I took the time to do so as I learned a lot, and seeing how Larry thought deeply through things from different perspectives (e.g., he often presented both a geometric and a statistical perspective) was memorable and inspiring. Larry often mentioned questions he had about methods or results, and directions of research he thought could be expanded upon, which I think motivated students to see statistics as a field full of open questions and research opportunities rather than a dead field. Larry was generous with his time. Whenever I had a student for whom it was unclear which other faculty members had the expertise to serve on their dissertation committee, I suggested asking Larry because I knew he would be willing to spend time talking with the student, read the dissertation seriously and have something thoughtful to say. A few days before Larry’s passing, when he knew his time was short, he was writing recommendation letters for students. Larry spent much time on public service, and he encouraged me about its value to society even though one may not get recognition for it. Larry worked hard. He was active in research, teaching and mentoring students and public service until his passing. The large number of Larry’s former students who traveled to his funeral from places far away at short notice (Hong Kong even!) was a testament to Larry’s impact on their lives. Larry also made good time for family and friends. Besides the much time spent together with his wife Linda and their family, Larry made the time for trips over the summer alone with his sons. Larry treated people with respect and decency regardless of their status. At a time when I was the postdoctoral coordinator for our department, a PhD student from a little known university in India contacted Larry about a post doc opening in our department and
{"title":"Larry Brown: Remembrance and Connections of His Work to Observational Studies","authors":"Dylan S. Small","doi":"10.1353/obs.2018.0011","DOIUrl":"https://doi.org/10.1353/obs.2018.0011","url":null,"abstract":"Statistics lost one of our giants when Larry Brown passed away in February, 2018 at the age of 77. Many lost a friend, a mentor and a teacher. I had the good fortune to be Larry’s colleague for the past 16 years. While I wasn’t as close to him as many others, Larry was my friend and I looked up to him. Larry loved thinking. When Larry got interested in something and scratched his head, I could see his enjoyment. Steam seemed to come out as Larry was thinking and then he happily shared his thoughts. Larry’s classes and talks were full of insights. After joining the faculty at Wharton, I attended Larry’s linear models first year PhD course even though I had seen the material before. I was glad I took the time to do so as I learned a lot, and seeing how Larry thought deeply through things from different perspectives (e.g., he often presented both a geometric and a statistical perspective) was memorable and inspiring. Larry often mentioned questions he had about methods or results, and directions of research he thought could be expanded upon, which I think motivated students to see statistics as a field full of open questions and research opportunities rather than a dead field. Larry was generous with his time. Whenever I had a student for whom it was unclear which other faculty members had the expertise to serve on their dissertation committee, I suggested asking Larry because I knew he would be willing to spend time talking with the student, read the dissertation seriously and have something thoughtful to say. A few days before Larry’s passing, when he knew his time was short, he was writing recommendation letters for students. Larry spent much time on public service, and he encouraged me about its value to society even though one may not get recognition for it. Larry worked hard. He was active in research, teaching and mentoring students and public service until his passing. The large number of Larry’s former students who traveled to his funeral from places far away at short notice (Hong Kong even!) was a testament to Larry’s impact on their lives. Larry also made good time for family and friends. Besides the much time spent together with his wife Linda and their family, Larry made the time for trips over the summer alone with his sons. Larry treated people with respect and decency regardless of their status. At a time when I was the postdoctoral coordinator for our department, a PhD student from a little known university in India contacted Larry about a post doc opening in our department and","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/obs.2018.0011","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"42764860","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 demonstrate a selective inferential approach for discovering and making confident conclusions about treatment effect heterogeneity. Our method consists of two stages. First, we use Robinson’s transformation to eliminate confounding in the observational study. Next we select a simple model for effect modification using lasso-regularized regression and then use recently developed tools in selective inference to make valid statistical inference for the discovered effect modifiers. We analyze the Mindset Study data-set provided by the workshop organizers and compare our approach with other benchmark methods.
{"title":"Selective Inference for Effect Modification: An Empirical Investigation","authors":"Qingyuan Zhao, Snigdha Panigrahi","doi":"10.1353/obs.2019.0007","DOIUrl":"https://doi.org/10.1353/obs.2019.0007","url":null,"abstract":"Abstract:We demonstrate a selective inferential approach for discovering and making confident conclusions about treatment effect heterogeneity. Our method consists of two stages. First, we use Robinson’s transformation to eliminate confounding in the observational study. Next we select a simple model for effect modification using lasso-regularized regression and then use recently developed tools in selective inference to make valid statistical inference for the discovered effect modifiers. We analyze the Mindset Study data-set provided by the workshop organizers and compare our approach with other benchmark methods.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/obs.2019.0007","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"48461945","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}
{"title":"Standing on the shoulders of Austin Bradford Hill: The refinement of “specificity” as a consideration in causal inference","authors":"N. Weiss","doi":"10.1353/OBS.2020.0009","DOIUrl":"https://doi.org/10.1353/OBS.2020.0009","url":null,"abstract":"","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/OBS.2020.0009","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"49002003","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}
{"title":"The Inheritance bequeathed to William G. Cochran that he willed forward and left for others to will forward again: The Limits of Observational Studies that seek to Mimic Randomized Experiments","authors":"Thomas D. Cook","doi":"10.1353/OBS.2015.0012","DOIUrl":"https://doi.org/10.1353/OBS.2015.0012","url":null,"abstract":"","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/OBS.2015.0012","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"44922735","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 fundamental obstacle of observational studies is that of unmeasured confounding. If all potential confounders are measured within the data, and treatment occurs at but a single time-point, conventional regression adjustment methods provide consistent estimates and allow for valid hypothesis testing in a relatively straightforward manner. In situations for which treatment occurs at several successive timepoints, as in many longitudinal studies, another type of confounding is also problematic: even if all confounders are known and measured in the data, time-dependent confounding may bias estimates and invalidate testing due to collider-stratification. While “causal inference methods” can adequately adjust for time-dependent confounding, these methods require strong and unverifiable assumptions. Alternatively, instrumental variable analysis can be used. By means of a simple illustrative scenario and simulation studies, this paper sheds light on the issues involved when considering the relative merits of these two approaches for the purpose of hypothesis testing in the presence of time-dependent confounding.
{"title":"The Validity and Efficiency of Hypothesis Testing in Observational Studies with Time-Varying Exposures","authors":"Harlan Campbell, P. Gustafson","doi":"10.1353/obs.2018.0010","DOIUrl":"https://doi.org/10.1353/obs.2018.0010","url":null,"abstract":"Abstract:The fundamental obstacle of observational studies is that of unmeasured confounding. If all potential confounders are measured within the data, and treatment occurs at but a single time-point, conventional regression adjustment methods provide consistent estimates and allow for valid hypothesis testing in a relatively straightforward manner. In situations for which treatment occurs at several successive timepoints, as in many longitudinal studies, another type of confounding is also problematic: even if all confounders are known and measured in the data, time-dependent confounding may bias estimates and invalidate testing due to collider-stratification. While “causal inference methods” can adequately adjust for time-dependent confounding, these methods require strong and unverifiable assumptions. Alternatively, instrumental variable analysis can be used. By means of a simple illustrative scenario and simulation studies, this paper sheds light on the issues involved when considering the relative merits of these two approaches for the purpose of hypothesis testing in the presence of time-dependent confounding.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/obs.2018.0010","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"48954253","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:Initial data analyses (IDA) are often performed as part of studies with primary-data collection, where data are obtained to address a predefined set of research questions, and with a clear plan of the intended statistical analyses. An informal or unstructured approach may have a large and non-transparent impact on results and conclusions presented in publications. Key principles for IDA are to avoid analyses that are part of the research question, and full documentation and transparency.We develop a framework for IDA from the perspective of a study with primary-data collection and define and discuss six steps of IDA: (1) Metadata setup to properly conduct all following IDA steps, (2) Data cleaning to identify and correct data errors, (3) Data screening that consists of understanding the properties of the data, (4) Initial data reporting that informs all potential collaborators working with the data about insights, (5) Refining and updating the analysis plan to incorporate the relevant findings, (6) Reporting of IDA in research papers to document steps that impact the interpretation of results. We describe basic principles to be applied in each step and illustrate them by example.Initial data analysis needs to be recognized as an important part and independent element of the research process. Lack of resources or organizational barriers can be obstacles to IDA. Further methodological developments are needed for IDA dealing with multi-purpose studies or increasingly complex data sets.
{"title":"A Contemporary Conceptual Framework for Initial Data Analysis","authors":"M. Huebner, S. le Cessie, C. O. Schmidt, W. Vach","doi":"10.1353/obs.2018.0014","DOIUrl":"https://doi.org/10.1353/obs.2018.0014","url":null,"abstract":"Abstract:Initial data analyses (IDA) are often performed as part of studies with primary-data collection, where data are obtained to address a predefined set of research questions, and with a clear plan of the intended statistical analyses. An informal or unstructured approach may have a large and non-transparent impact on results and conclusions presented in publications. Key principles for IDA are to avoid analyses that are part of the research question, and full documentation and transparency.We develop a framework for IDA from the perspective of a study with primary-data collection and define and discuss six steps of IDA: (1) Metadata setup to properly conduct all following IDA steps, (2) Data cleaning to identify and correct data errors, (3) Data screening that consists of understanding the properties of the data, (4) Initial data reporting that informs all potential collaborators working with the data about insights, (5) Refining and updating the analysis plan to incorporate the relevant findings, (6) Reporting of IDA in research papers to document steps that impact the interpretation of results. We describe basic principles to be applied in each step and illustrate them by example.Initial data analysis needs to be recognized as an important part and independent element of the research process. Lack of resources or organizational barriers can be obstacles to IDA. Further methodological developments are needed for IDA dealing with multi-purpose studies or increasingly complex data sets.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/obs.2018.0014","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"43137526","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:Thistlethwaite and Campbell (1960) (TC) introduced the Regression Discontinuity Design (RDD) as a strategy for learning about the causal effects of interventions in 1960. Their introduction highlights the most important strengths and weaknesses of the RDD. The main points of the original paper have held up well to more formal scrutiny. However, TC did not address “manipulation of assignment scores” as an important validity threat to the design. The insight that manipulation is a central validity threat is the most important conceptual advance in the methodological literature since its introduction. Although most modern RDD analyses include density tests for assessing manipulation, results are most convincing when diagnostic probes are used to address specific, plausible threats to validity. In this paper, we examine validity threats to two common RD designs used to evaluate the effects of No Child Left Behind and state pre-kindergarten programs.
{"title":"The Regression Discontinuity Design and the Social Corruption of Quantitative Indicators","authors":"Vivian C. Wong, Coady Wing","doi":"10.1353/obs.2017.0006","DOIUrl":"https://doi.org/10.1353/obs.2017.0006","url":null,"abstract":"Abstract:Thistlethwaite and Campbell (1960) (TC) introduced the Regression Discontinuity Design (RDD) as a strategy for learning about the causal effects of interventions in 1960. Their introduction highlights the most important strengths and weaknesses of the RDD. The main points of the original paper have held up well to more formal scrutiny. However, TC did not address “manipulation of assignment scores” as an important validity threat to the design. The insight that manipulation is a central validity threat is the most important conceptual advance in the methodological literature since its introduction. Although most modern RDD analyses include density tests for assessing manipulation, results are most convincing when diagnostic probes are used to address specific, plausible threats to validity. In this paper, we examine validity threats to two common RD designs used to evaluate the effects of No Child Left Behind and state pre-kindergarten programs.","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/obs.2017.0006","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"48521501","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}
To students of causality, the writings of William Cochran provide an excellent and intriguing vantage point for studying how statistics, lacking the necessary mathematical tools, managed nevertheless to cope with increasing demands for policy evaluation from observational studies. Cochran met this challenge in the years 1955-1980, when statistics was preparing for a profound, albeit tortuous transition from a science of data, to a science of data generating processes. The former, governed by Fisher’s dictum (Fisher, 1922) “the object of statistical methods is the reduction of data” was served well by the traditional language of probability theory. The latter, on the other hand, seeking causal effects and policy recommendations, required an extension of probability theory to facilitate mathematical representations of generating processes. No such representation was allowed into respectable statistical circles in the 1950-60s, when Cochran started looking into the social effects of public housing in Baltimore. While data showed improvement in health and well-being of families that moved from slums to public housing, it soon became obvious that the estimated improvement was strongly biased; Cochran reasoned that in order to become eligible for public housing the parent of a family may have to possess both initiative and some determination in dealing with the bureaucracy, thus making their families more likely to obtain better healthcare than non-eligible families. 1 This led him to suggest “adjustment for covariates” for the explicit purpose of reducing this causal effect bias. While there were others before Cochran who applied adjustment for various purposes, Cochran is credited for introducing this technique to statistics (Salsburg, 2002) primarily because he popularized the method and taxonomized it by purpose of usage. Unlike most of his contemporaries, who considered cause-effect relationships “ill-defined” outside the confines of Fisherian experiments, Cochran had no qualm admitting that he sought such relationships in observational studies. He in fact went as far as dening the objective of an observational study: “to elucidate causal-and-effect relationships” in situations where controlled experiments are infeasible (Cochran, 1965). Indeed, in the paper before us, the word “cause” is used fairly freely, and other causal terms such as “effect,” “influence,” and “explanation” are almost as frequent as “regression” or “variance.” Still, Cochran was well aware that he was dealing with unchartered extra-statistical territory and cautioned us: “Claim of proof of cause and effect must carry with it an explanation of the mechanism by which this effect is produced.”
{"title":"Causal Thinking in the Twilight Zone","authors":"J. Pearl","doi":"10.1353/obs.2015.0020","DOIUrl":"https://doi.org/10.1353/obs.2015.0020","url":null,"abstract":"To students of causality, the writings of William Cochran provide an excellent and intriguing vantage point for studying how statistics, lacking the necessary mathematical tools, managed nevertheless to cope with increasing demands for policy evaluation from observational studies. Cochran met this challenge in the years 1955-1980, when statistics was preparing for a profound, albeit tortuous transition from a science of data, to a science of data generating processes. The former, governed by Fisher’s dictum (Fisher, 1922) “the object of statistical methods is the reduction of data” was served well by the traditional language of probability theory. The latter, on the other hand, seeking causal effects and policy recommendations, required an extension of probability theory to facilitate mathematical representations of generating processes. No such representation was allowed into respectable statistical circles in the 1950-60s, when Cochran started looking into the social effects of public housing in Baltimore. While data showed improvement in health and well-being of families that moved from slums to public housing, it soon became obvious that the estimated improvement was strongly biased; Cochran reasoned that in order to become eligible for public housing the parent of a family may have to possess both initiative and some determination in dealing with the bureaucracy, thus making their families more likely to obtain better healthcare than non-eligible families. 1 This led him to suggest “adjustment for covariates” for the explicit purpose of reducing this causal effect bias. While there were others before Cochran who applied adjustment for various purposes, Cochran is credited for introducing this technique to statistics (Salsburg, 2002) primarily because he popularized the method and taxonomized it by purpose of usage. Unlike most of his contemporaries, who considered cause-effect relationships “ill-defined” outside the confines of Fisherian experiments, Cochran had no qualm admitting that he sought such relationships in observational studies. He in fact went as far as dening the objective of an observational study: “to elucidate causal-and-effect relationships” in situations where controlled experiments are infeasible (Cochran, 1965). Indeed, in the paper before us, the word “cause” is used fairly freely, and other causal terms such as “effect,” “influence,” and “explanation” are almost as frequent as “regression” or “variance.” Still, Cochran was well aware that he was dealing with unchartered extra-statistical territory and cautioned us: “Claim of proof of cause and effect must carry with it an explanation of the mechanism by which this effect is produced.”","PeriodicalId":74335,"journal":{"name":"Observational studies","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2021-06-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1353/obs.2015.0020","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"48533165","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：