Why We Must Question Our Elections
Monday 17 January 2005
I am a statistician. When I testified about electoral tampering in Martin County, Florida, in November 2000, I focused
exclusively on the fact that the number of disputed ballots would have changed the outcome. That was shortsighted. As
U.S. newspapers have written about the Ukraine, an election's outcome may be less important than how it was conducted.
Democratic elections must be verifiably fair.
Before November 2, several U.S. newspapers pursued concerns about election integrity. They reported on the vulnerability
of electronic voting to simple errors and malicious hacking, and the unnecessary dangers posed by non-verifiable
touch-screen voting. They exposed obstructionist maneuvers, such as Ohio Secretary of State Kenneth Blackwell's telling
election commissioners to reject voter registrations submitted on thin paper.
Post-election, though, the mainstream media has largely ignored or ridiculed concerns about U.S. electoral errors and
For example, a front page article by Tom Zeller, Jr. in November 12's New York Times told us that the "elections
department in Cleveland set off a round of Web-blog hysteria when it posted turnout figures on its site that seemed to
show more votes being cast in some communities than there were registered voters." But, the figures did show over 93,000
more votes than voters. Why is it hysterical to be disturbed by this?
On the evening of November 2, Election Day ("exit") poll data showed comfortable margins for Kerry. These figures
disappeared shortly after midnight, replaced by numbers very close to the final tallies. In Ohio, for example, as late
as midnight the Bush/Kerry split was 47.9%/52.1%. Later, it was 50.9%/48.6%. The official vote tally is 51.0/48.5. The
press has published speculation about how the early numbers might have come to be wrong, yet, two months after the
election, there is still no convincing explanation of the discrepancy.
Polls and the official tabulations each measure the voters' collective intent. Each is imperfect. Differences between
poll results and final tallies raise legitimate questions about the accuracy of the tallies.
In Florida 2000, the polls identified more Gore voters than Bush ones, while the official tally declared Bush, just
barely, the winner. Which was right? Recall that Palm Beach County's infamous butterfly ballot nullified thousands of
ballots intended for Gore. Statewide, 3% of ballots were "spoiled" - that is, not counted. While only 11% of Florida
voters, African-Americans contributed 54% of the 180,000 spoiled votes. Since 87% of African-Americans in Florida vs.
45% of whites voted for Gore over Bush, these ballots alone likely represent a Gore advantage of approximately 60,000
votes - more than 1% of the statewide total. 
If more Florida voters cast their votes for Gore, but more Gore ballots were invalidated, then the polls and the
tallies would each have captured a different truth.
Although "Bush Would Have Won" headlines blanketed the country in April 2001, the articles belied their headlines. Bush
would have won under the counting standards advocated by Gore, while Gore would have, under other counting rules -
including the one proposed by Bush! The media chose reassurance (the system worked!) over the more important message in
Florida 2000: referees, not voters, determined the outcome. When different counting rules produce different Presidents,
reform is desperately needed.
Statistics can flag anomalies and provide efficient methods for studying them. For example, some analysts questioned
this year's vote tallies in Florida's Broward and Palm Beach counties. Well-conducted audits could resolve such
questions, using probability sampling to examine only a small fraction of the votes cast in districts with the most
reason for concern. Unfortunately, about 30% of the votes in the last election - and all in Broward and Palm Beach -
used non-verifiable, touch-screen voting.
Snohomish County, Washington, also used non-verifiable touch screen voting in all precincts (polling locations) on
election day 2004. Among about 100,000 touch screen votes in the famously close governor's race, Republican candidate
Rossi had an 8,000 vote advantage; while among about 200,000 paper (absentee) ballots, Democrat Gregoire had 2,000 more
Some voters spoke of the touch screen machine changing their vote. Countywide, there were 19 formally reported
instances of machine switching; every one favored the Republican. Could these be random errors, equally likely to go
either way, like tosses of a fair coin? Well, fair coins just don't come up the same 19 times in a row! 
However, Rossi did not have a "touch screen advantage" everywhere. Among 90 precincts with no reported machine problems
(vote switching, machine freeze-ups, or repairs within two weeks of the election), 44 had touch screen vote counts more
favorable to Rossi than paper ballots, while 46 had a touch screen advantage for Gregoire. However, among the precincts
with reported voting machine problems, Rossi had a touch screen advantage in 56 out of 58 (96.6%). Such differences
demand an explanation. 
While reported e-voting problems have mostly disadvantaged Democrats, Republicans have also been harmed by prematurely
adopting non-auditable technology. In Fairfax County, Virginia, a Republican-endorsed school board member, Rita
Thompson, appears to have lost because her name appeared on the edge of the screen and was difficult to select. In many
races the margin of victory was smaller than the number of spoiled, lost or discounted votes. Improved electoral quality
requires investigating all important sources of errors, not just those that may have produced a different winner.
Voting machines are less secure than slot machines; voter registration lists, less reliable than many other government
and commercial lists. Why have we - democracy's customers - not demanded better? Sadly, our media chooses to deride the
doubts rather than report the problems.
Electoral audits must not be ad hoc enterprises assembled only reluctantly and only when the outcome of an already-held
election might plausibly be overturned. Improvement requires looking systematically at the current reality, acting to
fix problems, and continuing to monitor and act until we get to where we need to be. The media does us no favors by
pretending that the current situation is not messy.
I accept that about half the nation's voters chose George Bush in each of the last two elections. Not every anomaly is
an error, and not every error is malicious. Still, I look for the mainstream media to join my statistical colleagues and
myself in insisting on meaningful election oversight.
With fair electoral administration and credible audits, only the fringe will doubt winners of close races; without this,
any reasonable person should. When will Americans demand elections that truly deserve our trust?
1. Voting Irregularities in Florida During the 2000 Presidential Election, http://www.usccr.gov/
3. More precisely, the chance of a fair coin toss coming up the same 19 times in a row equals 1/2 to the 18th power,
approximately 4 in a million.
4. Using a contingency table analysis: If there were no systematic association between having a polling problem and the
Republican having a touch screen advantage, the probability that we would see an association this strong is less than 1
in a million (Chi-square = 36.6, 1 df).
Arlene S. Ash, Ph.D. is a Fellow of the American Statistical Association and a Research Professor at Boston
University's Schools of Medicine and Public Health. Contact Dr. Ash.