Event Date
Speaker: Philip B. Stark, Distinguished Professor, Dept Statistics UC Berkeley
Title: "Evidence-Based Elections and Nonparametric Sequential Tests"
Abstract: Elections are conducted using electronic hardware, software, and complex logistics with many opportunities for bugs, hacking, misconfiguration, human error, and fraud. How can we tell whether the reported winners really won? Hand-marked paper ballots, verifiable physical chain of custody, compliance audits, and risk-limiting audits (RLAs) make it possible to provide strong evidence that outcomes are correct when they are correct, and to rectify tabulation errors that altered reported outcomes. RLAs pose election integrity as a statistical hypothesis test. For a broad spectrum of social choice functions--including plurality, multi-winner plurality, super-majority, ranked-choice voting, D'Hondt and other proportional representation systems, approval voting, and Borda count--the hypothesis that the reported outcome is incorrect can be expressed as a set of assertions about finite lists of non-negative numbers: the mean of one or more of the lists is less than or equal to 1/2. Martingale methods can be used to test those hypotheses practically using a broad variety of sampling methods, including sampling individual ballots or clusters of ballots, with or without replacement (or using Bernoulli sampling), with or without stratification, with or without weights. RLAs have been endorsed by the National Academies, the ASA, and others. RLAs are now in law in half a dozen states and are in pending federal legislation. RLAs have been piloted in more than 10 U.S. states and in Denmark.
COVID-19 will remain a risk in the 2020 presidential election; in-person voting involves voters and pollworkers congregating in small spaces and touching the same equipment. Senators Klobuchar and Wyden have introduced a politically controversial bill that requires states to implement universal vote-by-mail to minimize transmission of COVID-19 while providing every registered voter the opportunity to cast a ballot. Would that support or undermine evidence-based elections? Is it better or worse than requiring voters to cast ballots in person? What are the tradeoffs?
This is the annual UC Berkeley / UC Davis Joint Colloquium, and will be presented this year via Zoom.