It is 1999, and hospital report cards have arrived: newspapers publish mortality league tables, patients choose surgeons by them, and administrators live in fear of them. You have mortality rates across 300 hospitals with wildly different volumes — a rural hospital with twelve cardiac cases sits in the same table as an urban center with four thousand. Rank them without crowning small-sample flukes: raw rates make the tiny hospitals both the best and worst in the nation by pure noise. Partial pooling — shrinking each hospital's estimate toward the ensemble in proportion to its ignorance — is the principled answer, and it must be defended to journalists who will call it fudging. Rank naively and a good small hospital is destroyed by three unlucky deaths.
Abebe's field — mechanism design and algorithms for social good, co-founding the MD4SG community in the late 2010s — studies how allocation and measurement systems interact with inequality, and a published hospital ranking is such a system: her framework would ask who bears the cost when a rural hospital is destroyed by three unlucky deaths, and how patients' choices reallocate under the table. Her technical work on income shocks and subsidy allocation shows real modeling depth about institutions serving the vulnerable. But the problem's core is the estimator, and partial pooling, variance components, and ranking uncertainty are not her tools — her methods are combinatorial and game-theoretic, not small-sample inferential — and her career postdates the 1999 setting by two decades. Vital critic of the table's consequences; not its statistician. Low-middle.
Pearl's Bayesian networks (1980s) made conditional-independence structure computable, and by 1999 his do-calculus had formalized when observational comparisons license causal claims — directly relevant, because a mortality league table is an implicit causal assertion that the hospital, not its case mix, produced the deaths. He would correctly demand the confounding be drawn before any ranking is trusted. But the problem's stated core — partial pooling, shrinkage in proportion to sample size, ranking under binomial noise — is estimation, and Pearl has spent a career insisting that estimation is precisely what his contribution is not: no variance components, no empirical Bayes, no small-sample craft in his toolkit. Belief propagation computes posteriors in given networks but does not tell you to shrink the twelve-case hospital. Right neighborhood, wrong house; middle-low.
Battle #48 · 8/9/2026, 8:37:54 PM · this result is deterministic: the same two personas on this problem always resolve the same way.