Comment: Self-Experimentation for Causal Effects

Self-experimentation what docs that haw to do with the academic field of statistics? It has a lot to do with it because thinking carefully about the issues raised by the type of self-cxpcrimcntarion Seth Roberts discusses leads to a better understanding of the foundations of causal inference, Let's begin by considering what the causal effect is for one "unit" that is, one object let's say Seth Roberts at one point in time, I lc wants to reduce his acne and is considering whether to usc a pill or a cream that evening. "What will my acne be like tomorrow morning if I take the pill?""What will my acne be like if I usc the cream?"lfhe could have answers to both of these questions, he would USl' the product that would lead to less acne in the morning, But he GInnot get answers to both questions; the best he can do is to choose one product and observe the result. The causal effect, however, is the comparison of the observed result under the chosen treatment with the unobserved result under the unchosen treatment. How docs Seth learn about the causal effect, which involves the comparisons of two "potential outcomes" from the observation of only one? ;rhe answer replication, more units. Now in statistics replication usually means more objects, as in a big randomized experiment with half the people assigned to one treatment and half to another, Two problems face Seth. first, he cares most about what works on him and not on others, although he'd certainly like to know what appears to work on others because it would suggest an answer for him, Second, he cant con-