Abstract: Studies of health behaviors using observational data are prone to bias from selection in behavior choices. How important are these biases? Are they dynamic - that is, are they influenced by the recommendations we make? Are there formal assumptions under which we can use information we have about selection on observed variables to learn…