It is a moment every policy analyst dreads: a program has already been rolled out, city by city, chosen for political convenience rather than by any coin you controlled — and now you must say whether it worked. Randomization, the one clean guarantee, was never on the table. Estimate the program's causal effect anyway, and state with total clarity exactly which assumptions are doing the work that randomization normally would: what you must believe about how cities were selected, which confounders you must rule out. The assumptions ARE the analysis. Get them wrong and a program that did nothing gets scaled nationwide on your say-so, or a program that worked gets killed — and no one will see the flawed assumption buried under the confident estimate.
Chose Partial pooling with a hierarchical model — right call.
Gelman is a working political-science and policy statistician, which is precisely this problem's habitat. His hierarchical Bayesian modeling (Stan, from 2012; multilevel modeling texts from the 1990s–2000s) is built for city-by-city structure: partial pooling across cities, varying treatment effects, and explicit modeling of how units differ. He co-wrote extensively on causal inference with observational data, including the applied-regression textbook treatment of matching, ignorability, and their fragility, and he is publicly relentless about the failure mode the problem names — confident estimates resting on unexamined assumptions. His posterior predictive checking gives a concrete workflow for probing whether the selection model is believable. He is not the originator of the identification theory, so he sits below Rubin, Pearl, and Neyman, but his toolkit applies almost end to end.
The professor has taught causal inference — the potential-outcomes lecture, the do-calculus lecture, the obligatory slide where correlation and causation get divorced — and that is exactly the problem: he is standing in a room with the people the slides are ABOUT. Rubin invented the framework; Santerre once summarized it before lunch. Pearl drew the graphs; Santerre drew them slightly wrong on a whiteboard in 2019 and a student politely fixed the arrow. Neyman wrote the 1923 paper; Santerre assigned it as optional reading, which no one, including him, finished. Consulting has taught him to say 'the assumptions ARE the analysis' with tremendous conviction, and teaching has taught him which specialist to call — which, here, is everyone else on this list. The generalist's fate: fluent in every method, decisive in none. Zero.
Battle #26 · 8/9/2026, 5:19:26 PM · this result is deterministic: the same two personas on this problem always resolve the same way.