Читать книгу New Horizons in Modeling and Simulation for Social Epidemiology and Public Health - Daniel Kim - Страница 17

Randomized experiments are the gold standard of study designs to establish cause‐and‐effect relationships. Yet, it is often neither feasible nor ethical to conduct experiments that randomly assign people or places to different levels of social determinants of health. As a result, evidence on the impacts of the social determinants of health has been largely based on observational studies, i.e. ecological, cohort, case–control, and cross‐sectional studies. Within such observational studies, traditional epidemiological approaches for studying the impacts of social determinants of health include multivariate analysis, which controls for factors that predict both the social determinants and health outcomes, i.e. so‐called potential “confounders.”

In addition, studies have explored these relationships by testing for single or multiple factors as potential mediators of the population health impacts of social determinants that could lend plausibility to the presence of causal associations. Because such social determinants are often contextual or area‐based factors (e.g. factors at the neighborhood or regional level), multilevel models that incorporate the hierarchical structure of data—such as individuals living within neighborhoods or states—are used to account for similarities and statistical nonindependence of individuals living within the same geographical areas (Goldstein et al. 2002).