April/May . 2018 IMS Bulletin . 11

OBITUARY: David L. Wallace 1928–2017 Univ. of Chicago Special Collections Univ. Even though David Wallace was not 77 Federalist papers, John Jay had written courtesy of Donald Rocker, by Photo well known across the broad landscape of fve (and no others), David Wallace in 1978 statistics, among academic statisticians he had written 43, and had was widely considered to be one of the most written 14. Tat left 12 where there was a part, on a 1958 foundational paper on insightful statisticians of his generation. dispute (Hamilton vs. Madison) and three asymptotic expansions. To students, and David was not a prolifc publisher, but he joint papers where the relative contributions others around him, David provided a strong was a penetrating thinker, and a ferce and of the two were in doubt. To solve the voice supporting the importance of statisti- inspirational oral commentator; when he problem, Mosteller and Wallace provided cal theory when tied to problems arising in did write up his work, his publications were the frst large-scale computer-based analysis data analysis, and he imparted a sense that gems. of text, using Bayes classifers built on data analysis was a deep subject worthy of Best known was his landmark study, data-driven priors in hierarchical models. serious intellectual pursuit. with Frederick Mosteller, of disputed Teir work required important technical David was born in Homestead, PA, authorship among the Federalist Papers, innovations (including the application of and went to Carnegie Tech (now Carnegie the series of political tracts that laid the Laplace’s method to Bayesian computation), Mellon) for Bachelor’s and Master’s degrees foundation for the U.S. Constitution. as well as labor-intensive coding procedures, (1948 and 1949), then to Princeton for When Mosteller and Wallace published and it was a model of painstaking, thorough a PhD, where his thesis supervisor was their work, in a 1963 JASA paper and a analysis in reaching defnitive conclusions. . He received his PhD in 1953 1964 book, they provided a compelling When the book was published it garnered and then held a post-doctoral position at solution to a 175-year-old problem: Which headlines in the national press: “Computer MIT (where he shared an ofce with John of these famous Federalist Papers had been Scans Federalist Papers,” NY Times (Front Nash, whom he had known as a student at written by each of the potential authors? Page); “IBM Machine Picks Federalist Princeton). In 1954 he accepted appoint- It had been generally agreed that, of the Papers’ Author,” NY Herald Tribune; and “A ment as an Assistant Professor of Statistics Computer Makes History, Spots Federalist at the University of Chicago, and remained Papers’ Author,” Chicago Sun Times. there until he retired in 1995. Tere, David In the 1960s David also helped develop played a vital role in developing the curricu- modern methods for real-time forecasting of lum and setting the intellectual and collegial elections. John Tukey and political scientist tone of the Department of Statistics, and Richard Scammon assembled a team for served as its chair from 1977–1980. Lacking NBC, on which David played a key role, in a feasible set of statistical programs for a public competition to be the frst network instruction, in the 1970s he wrote the sta- to announce results during the evening tistical package SNAP, which was used with of election day, and to do so accurately. success until it was superseded by larger and Te methods developed were regarded broader-based packages. as proprietary, and were not published, He was an inspiring teacher, and his but from later descriptions we know the image remains vivid, with the white lab coat team used Bayesian hierarchical models he wore to protect his suit from the clouds based on early, incomplete counts to make of chalk stirred up by his sometimes-impas- projections across precincts, and to evaluate sioned lectures. Te Department ofers an uncertainty. Both this and the authorship annual David Wallace Prize in his honor to work anticipated methods that would much the best statistical application by a graduate later become standard in statistics and student. machine learning. Robert E. Kass, Carnegie Mellon University David’s reputation was also based, in and Stephen M. Stigler, University of Chicago