HSCI 330 Lecture Notes - Lecture 8: A Priori And A Posteriori, Confounding, Relative Risk
Document Summary
Learning objective #1: apply data-based criteria for confounding to determine if a variable is a confounder. Data-based criteria: a variable is a confounder if crude effect estimate is meaningfully different than the effect estimate after adjustment for the variable. Sometimes people will define a percentage difference for example, if the adjusted effect estimate is different from the crude estimate by 15% or more. There are two ways to look at it: data-based criteria and a priori criteria. We need to decide if a variable will be a confounder. In the above example, i would say that this is a meaningful difference: when designing a study or building a model for analysis of data, when we are judging whether a study is valid (when reading it). The crude estimate tells us that drinking coffee will increase your risk of lung cancer by 42%.