By Arlene Ash and Mary Batcher
With increasingly heated allegations of "rigged elections," things have probably not gotten better since a September 29 poll concluded that "more than 15 million voters may stay home on Election Day" over concerns about cyber-security. Equally problematic would be doubts about who won following November 8. A vibrant democracy requires trusted elections. Paper validation of ballots cast and meaningful audits of those ballots are important - and neglected - tools for bolstering trust.
As statisticians working in healthcare and business, we frequently help researchers, patients, and business executives think about the probability and severity of potential risks. Based on the news coverage it receives, you might think that the problem of people who are not entitled to vote showing up at polling places is rampant. It is not. A comprehensive study of all American elections between 2000 and 2014 identified only 31 possible cases out of a billion votes cast. That is, only 0.000003 percent of votes might have been due to the kind of fraud that Voter ID laws could possibly prevent!
In contrast, electoral malpractice, intentional or not - including confusing ballot designs, computer security breaches and malfunctions, long lines, partisan administration, misleading information about where and how to vote, poorly maintained voting lists, and overly aggressive voter list purges - plague every American election.
Many of us woke up to these issues in 2000, as wrangling over the vote for president in Florida introduced a whole new vocabulary ("butterfly ballots," "hanging chads") for what can go wrong. There, the "margin of sloppiness" far exceeded the slender official 537-vote "margin of victory." Despite much subsequent research, we will never really know which candidate received more votes.
To be fair, conducting high-integrity elections is hard, and perfection is not possible. However, if "we the people" want elections worthy of our trust, our tax dollars and laws should be used to adopt best practices, many of which are summarized on the U.S. Election Assistance Commission website.
One key best practice is an impartial, meaningful audit of voter-verified ballots. Unfortunately, after Florida we turned to computers to side-step the messy problem of interpreting paper ballots, forgetting the need for auditable records. The voter-verified ballot is the record of the underlying "truth" that an official vote count must reflect. No electronic recount of machine ballots can tell how often the tabulated count fails to reflect the voter's choice.
Can you be confident that your vote will be counted accurately?
American elections - now almost entirely computerized - provide opportunities for both simple errors and hacking. Typically, in 1 to 3% of ballots in top-of-the-ticket races (e.g., for U.S. Senator, not "dog-catcher") no vote is counted. With confusing ballot designs, such "undercounts" can exceed 7% or even 10%. Although some undercounts are real (voters choose not to choose), many legitimate votes are lost. But the vast majority of American electoral contests are not audited. No one can honestly claim to know how often votes are miscounted.
Although some states require so-called post-election audits, these rarely compare the vote tally to voter-verified records. They might, for example, involve re-running ballots through vote-reading machines. According to Verified Voting only 12 states have even minimally reasonable post-election audits.
So what makes a post-election audit acceptable? It requires: a voter-verified paper record with a secure chain of custody and random selection of ballots for audit. The ballots audited must also be sufficiently comprehensive to limit the risk of a wrong outcome, providing confidence "that the winner won." Audit protocols should also be efficient. If the election is not close, we need only to confirm that huge errors are unlikely. More generally, a "risk-limiting audit" adapts to what it finds - continuing to examine ballots until it reaches an acceptable level of certainty about who won. A very close race could require examining every ballot.
Although only 27 states require a paper audit trail, just 5 states have no paper records whatsoever. Thus, most states could conduct proper audits - but few do. Some states require audits of a fixed number (or fraction) of ballots. This is ineffective for finding out who won in a close race, and inefficient (requires more work than needed) when examining a lopsided win.
Confidence that elections are fair is a cornerstone of democracy. Given increasing skepticism about the accuracy of the vote count, resources invested in risk-limiting electoral audits is money well spent.
Arlene Ash is Professor in the Department of Quantitative Health Sciences at the University of Massachusetts Medical School. She has worked on electoral integrity issues with the American Statistical Association and other professional and governmental groups since 2000.
Mary Batcher is a partner in BDS Data Analytics following her retirement from Ernst & Young in 2014. She has been involved in post election auditing since 2000.
Both are members of the American Statistical Association's Election Auditing Resource Team