Still looking for health outcomes in all the wrong places? Misinterpreted observational evidence, medication adherence promotion, and value-based insurance design.

Above all, the right question needs to be framed.”1 To those who would argue that sophisticated statistical modeling obviates the need to consider confounding factors and alternative explanations when interpreting observational associations, Freedman’s answer was simple: “The technology is relatively easy to use. ... However, the appearance of methodological rigor can be deceptive.”1 More contemporary guidelines for research reporting acknowledge the same understanding: when interpreting associations, multivariate modeling is often appropriate and valuable, but is still potentially vulnerable to the effects of confounding and bias and is therefore no substitute for a randomized design.2,3,4