ESHS/HSS Annual Meeting

Statistical Individuals

• In Event: Epidemiology Unbound 1

English Abstract

One of the achievements of proponents of the spread of statistical methods is the now-widespread belief that they are not just applicable in the aggregate but can be useful for determining an individual’s risk of disease or the answer to a standalone research question. At the center of this achievement has been the field of epidemiology, which is in some ways ironic given its status as a numerical but not statistical discipline for the first half of the century. This paper focuses on the development of two statistical methods, both of which started outside of epidemiology, were refined within the field, and then took on lives of their own outside of it. First, the formalization of the statistics of research synthesis (meta-analysis, procedures for pooling ratios and test scores, etc.) as a way of making judgments about evidence from multiple sources. And second, the development of logistic regression as a technique for replacing formal ideas of causality with data-driven assessment of causal influences. In both cases, US and UK biostatisticians between the 1930s and the 1980s drove these developments, which in turn lay the groundwork for more recent controversies over estimating average treatment effects, n-of-1 trials, and other tools of precision research in medicine and the human sciences. This paper engages directly with the themes of the symposium, showing how to “unbound” the history of epidemiology and connect it with modern economics, statistics, and the social sciences more broadly; it also engages directly with the themes of the conference, in particular showing how statistics has come to play a central role in adjudicating the scientific contestation that often arises from evidentiary plurality. Epidemiologists and biostatisticians played key role in setting up statistical methods as a technology of consensus, enabling researchers to navigate complex issues of causation and conflicting data.

Author

• Christopher J Phillips, Carnegie Mellon University