ABSTRACT
Observational studies have the advantage of realistic exposure conditions. This chapter describes strategies for identifying and accounting for confounding variables in the analysis of observational data. It defines confounding and introduces the general methodological strategy of “statistical control.” The chapter addresses the following question: How does the data analyst identify a set of variables for statistical control of confounding in a regional observational study? It discusses three strategies for identifying confounding variables that may require statistical control. A propensity score is said to be a “balancing score,” a combination of the confounding variables such that when the propensity score is held relatively constant, the stressor of interest is approximately independent of the confounding variables. Finally, the chapter describes statistical methods to control or mitigate the influence of identified confounders in data analysis.
