Econometric approaches to epidemiologic data: Relating endogeneity and unobserved heterogeneity to confounding