Changing numbers call state's election accuracy to question

By Eric Pryne

Seattle Times 20 February 2005

Idea of closer scrutiny met with mixed reaction

As she watched hundreds of votes appear and disappear in the dramatic recounts of last year's governor's race, Michelle Anderson wondered why the numbers were changing so much.

"It surprised me," the independent voter from Lake Stevens said.

"Look at our banking system. I can go anywhere in the world and withdraw money at an ATM, and it balances out to the penny."

The shifting recount totals have taken a back seat in the titanic struggle over the governorship as Democrats and Republicans wrangle in a Wenatchee courtroom over provisional ballots, dead voters and felons.

Still, for many people the numbers were the first indication that the accuracy of Washington elections ? and maybe elections in general ? is not what they had imagined. Doubts persist, as state lawmakers debate election-reform proposals and King County officials review their own procedures.

Why weren't all the votes counted right the first time? Have elections always been this inaccurate?

The answer depends on how you define accuracy.

Secretary of State Sam Reed assured voters before the recounts began in November that past recounts had shown Washington elections were 99.99 percent accurate.

His office measures accuracy by looking at how much the margin between candidates changes between an initial vote count and recounts. By that standard, this election was no less accurate than earlier ones, even though the outcome changed.

But by another measure ? total changes in raw numbers of votes ? the initial count in this election was much further off than in statewide recounts of the not-so-distant past.
  
  
During the first recount of the governor's race, when the ballots were tabulated a second time by machines, election workers added and subtracted more than 2,600 votes to the totals for Democrat Christine Gregoire, Republican Dino Rossi and Libertarian Ruth Bennett. During machine recounts in 1990 and 1991, workers added and subtracted just 450 and 691, respectively.

Some find that trend disturbing. "The initial count should be more accurate," says Don Whiting, former assistant secretary of state.

State and county elections officials say the numbers were up in the 2004 election because workers did a better job of detecting and fixing mistakes voters made on their ballots ? mistakes that prevented those votes from being read properly by machines.

In a sense, they suggest, they've traded one kind of accuracy for another.

"We in effect bend over backward for the voter," said John Pearson, deputy state-elections superintendent. "There is a trade-off in terms of the numbers not balancing every time [in recounts]. I think the trade-off is worth it."

Chads and ovals
Vote totals change in recounts for two reasons: Vote-counters catch mistakes voters made that weren't caught the first time through. And vote-counters catch mistakes they've made themselves.

During the first recount last November, for instance, Cowlitz County discovered it had counted 99 ballots twice in the initial count. Snohomish County found 224 uncounted ballots in a locker.

Processing errors like those are nothing new: In 2000, Douglas County discovered during recounts for the U.S. Senate and secretary of state that in the first tally it had accidentally double-counted more than 700 ballots ? 6 percent of the total county vote.

Proportionately, it was a bigger mistake than any county made in 2004. But it didn't attract a lot of attention because it didn't make a difference. The races weren't as tight.

Pearson and other election officials say processing errors weren't responsible for most of the thousands of votes that were added and subtracted during the recounts last year.

Technology was.

Since the early 1990s, "optical-scan" ballots, which voters mark with a pencil or pen, have replaced punch-card ballots in most Washington counties. It's part of a national trend that has been accelerated by the 2000 presidential-election debacle in Florida. Congress concluded in 2002 that punch cards are more error-prone than other voting systems, and ordered all states to eliminate them by 2006.

But, while optical-scan ballots are considered more reliable, vote totals in Washington recounts began to fluctuate much more dramatically after the new systems had largely supplanted punch cards.

Fourteen counties still used punch cards last fall. On average, their vote totals changed much less between the initial count and the recounts than in the counties that used optical scan.

When Spokane County used punch cards in 2000, its vote totals in the U.S. Senate recount increased by just 28. When it used optical-scan ballots last year, the first recount produced 261 more votes.

Elections officials say the numbers jumped because people interact differently with the two types of ballots.

Voters don't have many options when they mark punch-card ballots, says Snohomish County Auditor Bob Terwilliger, and election workers don't have much to look for when they inspect them.

"You either have that chad punched out or you don't," he says, referring to the tiny pieces of paper voters detach from punch cards to mark their choices.

An optical-scan ballot, in contrast, "opens up all kinds of opportunities for voters to get creative in marking their ballots," Pearson says.

Some check or underline or circle the name of a candidate or party on the ballot, rather than filling in the oval or other space the scanner scans. Some don't fill in the oval darkly enough. Some fill in two ovals ? one for a candidate, one for a write-in ? then print the name of that same candidate on the write-in line.

Machines won't count any of those votes.

Adams County's scanner reads only marks made in pencil; auditor Nancy McBroom says some voters last fall ? especially those voting by mail ? filled out their ballots in ink. Others changed their votes, but didn't erase their first marks completely enough to satisfy the machine.

Many of those mistakes were corrected before the ballots were counted the first time ? McBroom says her staff duplicated or re-marked about one in every 10 ballots.

But other errors slipped through undetected, or weren't fixed completely before the first count. "There are so many more things to look for" with optical scan, Pearson says.

The beauty of the system is that if there's a recount, it's often fairly easy to discern a voter's intent and count the vote. One consequence: the numbers change more.

Voters may have made even more mistakes with punch cards ? mistakes that didn't turn up in recounts because they were more difficult to detect, let alone fix.

Ballots without votes
After the 2000 election, researchers at the California Institute of Technology and the Massachusetts Institute of Technology explored how different voting systems might enable or interfere with voters' ability to state their preferences. Their measuring stick: the number of ballots in presidential elections that didn't record a vote for any presidential candidate.

Numbers varied widely across the nation by ballot type, researchers found ? not something you'd expect if all such "undervotes" were intentional. In 2000, 3 percent of the punch-card ballots, but just 1.2 percent of the optical-scan ballots, didn't contain a vote for the nation's top office.

The same pattern emerged in Washington last year. On average, punch-card counties recorded more undervotes for president than did optical-scan counties.

Yakima County used punch cards in 2000, then optical-scan and electronic voting last year. The county's undervote for president ped from nearly 4.7 percent to less than 1.2 percent.

"A punch-card ballot is much less user-friendly," says Thurston County Auditor Kim Wyman, who supervises elections in one of the larger counties that still uses them.

That's become increasingly evident as more people vote by mail, many elections officials say. At polling places, it's easy to punch out the right chad; there's a template that aligns candidates' names with the corresponding chads.

At home, it's more complex: In all but one Washington county ? Franklin County ? the cards don't contain candidates' names. Voters must look in a separate, printed guide to learn which chad is a vote for which candidate.

Sometimes they get it wrong.

Sometimes voters poke holes in punch cards above, below or beside chads. When election workers inspect those ballots, "you don't know what their intent was," says Yakima County assistant auditor Lynda Sissom.

With optical-scan ballots, elections officials agree, determining intent isn't such a mystery.

Paul Miller, director of election information in the Secretary of State's office, tells of confused punch-card voters mistakenly punching out, say, Chad 98 when they intended to vote for or against Initiative 98.

"The system's never going to pick that up," he says.

"The voters may make a mistake," adds Pam Floyd, assistant elections director for voter services, "but it's harder to know that they've made a mistake."

Anderson, the Lake Stevens voter, wonders whether ballots that aren't filled out correctly should be counted at all. Some states do reject them, Pearson says.

"But we are a 'voter-intent' state," he adds. "If you can figure out what the voter intended to do, you count that vote."

Is 100% accuracy possible?
Voters and vote-counters have always made mistakes. They always will, Pearson says: If there were another recount, the numbers could change again.

"It's a massive organizational challenge," says University of Washington communications professor Philip Howard.

After the November elections, Howard led a team that used the Caltech-MIT data to calculate the reliability of voting systems used by different states. The team then compared that number with the vote difference between candidates in high-profile races.

Researchers found that the margin of error for the voting systems was greater than the winners' victory margins in seven close contests: the presidential races in Iowa, New Hampshire and New Mexico, Senate contests in Florida, Kentucky and South Dakota ? and Washington's governor's race.

In each case, Howard says, the certified winner theoretically could be the loser. Runoff elections may be the best solution when the outcome is that close, he suggests.

"It doesn't matter how many times you count," Howard says. "You always get a different number."