Overt Bias in Observational Studies

Abstract

An observational study is biased if the treated and control groups differ prior to treatment in ways that matter for the outcomes under study. An overt bias is one that can be seen in the data at hand for instance, prior to treatment, treated subjects are observed to have lower incomes than controls. A hidden bias is similar but cannot be seen because the required information was not observed or recorded. Overt biases are controlled using adjustments, such as matching or stratification. In other words, treated and control subjects may be seen to differ in terms of certain observed covariates, but these visible differences may be removed by comparing treated and control subjects with the same values of the observed covariates, that is, subjects in the same matched set or stratum defined by the observed covariates. It is natural to ask when the standard methods for randomized experiments may be applied to matched or stratified data from an observational study. This chapter discusses a model for an observational study in which there is overt bias but no hidden bias. The model is, at best, one of many plausible models, but it does clarify when methods for randomized experiments may be used in observational studies, and so it becomes the starting point for thinking about hidden biases. Dealing with hidden bias is the focus of Chapters 4 through 8.KeywordsObservational StudyPropensity ScoreTreatment AssignmentTreated SubjectPublic High SchoolThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.