In this episode we work through a paper that attempts to estimate the benefit of attending AA for patients with alcohol use disorder. AA promises that "it works if you work it"--is this true? Precisely how true is it? And how do you figure that out, given that patients are free to choose to attend AA (or not), and typically have deeply individual reasons for attending (or not). We revisit our past discussion about "unobserved confounders" and work through some confusing wording choices made by the authors (the words "randomization" and "group assignment" mean different things). This paper is part of the ABPN MOC. As always, the podcast authors have no knowledge of the ABPN MOC content.