Unlocking IRV in Aspen
October 1, 2010
ASPEN – The mood in Aspen City Hall on May 5, 2009 seemed optimistic as the ballots for what many called that day’s “historic” city election rolled through the electronic scanners. The machines were meant to protect the electoral decisions of the entire community based on the pictures they took of the seemingly innocent sheets of paper – a very important task.
City Councilman Steve Skadron sat at a corner of the roundtable setup of the room being interviewed by former Mayor Helen Klanderud and former Aspen finance director Paul Menter about how Skadron felt the election had been conducted. Their conversation spanned a number of topics, from the conduct of the mayoral and City Council candidates – very professional, Skadron said – to the viability of the new “instant runoff voting,” or IRV, system the city was using to select its new officials.
But nobody discussed how, as city officials implemented IRV, they discovered that the vote-counting technology is actually a lot more complicated than it appeared at first glance, that they were deploying a system that few really understood – or seem to understand now.
There was no talk in that meeting about the previous night, when three private election engineers, hired by the city to count the ballots, reconfigured the tabulation software for the election computers after they discovered an egregious flaw. It had tallied test sample ballots incorrectly, causing the test candidate with the fewest votes to win.
The software was part of a system the company, True Ballot, Inc., or TBI, brings to elections – mostly for private institutions like labor unions, but also for cities and Indian tribal governments – to tabulate them. It was a system able to perform complex election tabulations. But it was not certified, or even reviewed, by any governmental agency that normally does so.
It was essentially a privatization of the May 2009 election. The ballots were “spit out of a black box,” as Councilman Dwayne Romero put it in an interview last month. That phrase is common parlance among election enthusiasts, who like to call themselves “wonks,” to describe a machine that hides the mechanics of the count from the voters.
Officials from TBI said the private model became Aspen’s vote-counting paradigm because only TBI had the ability to tabulate an instant runoff election. And that was true, with the exception of a handcount by the city clerk’s office. The city of Aspen didn’t have the expertise to administer the system it had chosen.
Few Aspenites knew the intricacies of the system the city was about to use. On the surface, everything was peachy.
Since then, however, the confusion inherent to Aspen’s spring election has been illustrated by statements from people who have derided the system and its implementation in Aspen. Recently, Mayor Mick Ireland asked Rob Ittner, who is now a candidate for Pitkin County commissioner, and Marilyn Marks, who ran for mayor in the May election, if they thought the ballots were confusing. Yes, they both said. Some of the critics – whether they’re candidates in the November election or simply individuals interested in election integrity – say they hope city voters will endorse a new voting system in the November election. They also hope an investigation by the city’s election oversight board will put to rest allegations by four local election-watchers that the city engaged in widespread unethical and illegal conduct.
The allegations have been part of the almost year-and-a-half-long road of controversy that has ultimately led the City Council to weigh new voting procedures and has put Aspen on the national map as what Bev Harris, an investigative journalist who specializes in elections, called the “poster child” of what can go wrong with IRV.
In the first half of the decade, after he’d run successfully for Aspen City Council, Jack Johnson had a vision. Having survived a May election and then a June runoff, he had discovered that only a small percentage of Aspen’s electorate showed up at the polls the second time around. He felt that more voters would participate if Aspen were to streamline its voting process, use fewer city resources and not require the voters to return to the polls a month after the first round of voting. (The city’s old system called for a runoff election if no candidate won a majority of votes.)
So in 2004, Johnson and City Clerk Kathryn Koch traveled to San Francisco to observe the hot new “instant runoff voting” method. What they came back with eventually became a proposal asking voters for permission to implement IRV.
It seemed like a good idea. IRV is supposed to work like this: On the ballot, voters rank by number the candidates of their choice. In the case of the 2009 election, it was from one to four for the mayoral candidates, and one to nine for the City Council candidates. If the top vote-getter didn’t receive more than 50 percent of the total, then the candidate in last place would be eliminated, and each of the ballots that put the loser first would go to the person ranked second on that ballot. That process, known as sequential elimination, would be repeated until one candidate achieved a majority.
This system, proponents said, would eliminate the expensive runoff, save the city thousands of dollars and boost confidence in the voting process. Voters approved the concept in 2007.
In September 2008, the City Council appointed a number of community members to a group charged with choosing the actual method to tabulate the new ballots. But the group, called the IRV task force, quickly discovered a problem: The City Council race involved two open seats, instead of just one.
Rob Richie, the executive director of FairVote, a Maryland non-profit that advocates on a national level for IRV, said at the time that a two-seat race does not lend itself to IRV because of a wide range of logistical complications. And the task force meetings became confusing, to the point that the board lost three members and brought in city staff members to replace them, without approval from the City Council. City attorney John Worcester, special counsel Jim True, Koch and Mayor Ireland now sat on the board.
“People who were going to be subject to the election system were designing it,” said Menter, who quit working for the city in December 2008 and, after the 2009 election, became an outspoken critic of the election.
And the more fundamental, mathematical problem of using IRV to choose two candidates remained. The group would either have to commission two counts of the ballots because, on an IRV ballot, a voter can’t rank two people in the same place. Or the task force members would have to come up with a creative counting method. And that’s what they did.
The method chosen in February 2009 by half the members of the task force – Worcester, True, then-Councilman Johnson, Ireland, Koch, Assistant City Manager Barry Crook and Pitkin County Clerk Janice Vos-Caudill, each of them local government employees – was a sequential elimination system that would count the ballots a second time after the winner of the first seat was established.
The rest of the task force, all private citizens, did not vote. Su Lum, a member of the task force (Lum is also a columnist and advertising account manager for The Aspen Times) had expressed confusion in a Feb. 25 e-mail message to Worcester about which task force members could vote. Were city officials planning the method that might keep some of them in office, or was it citizens?
Called out by election enthusiasts and an editorial in The Aspen Times that asked the government officials on the board to recuse themselves, the task force reformed, eliminating the government employees on the board and bringing in several independent citizens. And then the balance of opinion changed. The task force finally chose a counting procedure called the “True Method,” named after True, who created it.
The True Method uses a “batch elimination” process to select winning candidates. In the event that no candidate receives a majority, the True Method would automatically eliminate five of the nine City Council candidates who received the fewest first-place votes. After that batch elimination, the normal IRV process would kick in with the remaining candidates.
But some warned that this method had dangerous implications.
Richie, of FairVote, said at the time that the True Method wouldn’t work. “While the proposal is a creative effort to apply [IRV] for a multi-seat race, it violates one of the key principles of an [IRV] election – ranking a lesser choice should not help defeat your higher choice,” he wrote in a letter dated Feb. 4 to Johnson. Colorado law explicitly prohibits a system that allows a person with less support to gain the advantage.
“They knowingly violated that statute,” said Millard Zimet, an Aspen lawyer who has complained about the conduct of the election.
Even Caleb Kleppner, an analyst with TBI, said in an e-mail message to Koch on Dec. 8, 2008 that a two-seat count in an IRV race was unprecedented. “In terms of what system to use, there is no standard two-seat IRV system, so whatever you come up with is going to be somewhat homegrown,” Kleppner wrote.
Richie has shifted his position since, saying the True Method is “defensible” because it was the most familiar way to bring IRV to Aspen voters: “I think they’re right. I think that Jim True’s system more approximates what Aspen had in the past.”
But Ittner, a restaurateur and 2010 county commissioner candidate, asks himself, did the True Method do what an election system is supposed to do? In his mind, the answer is a resounding no.
Ittner, who calls himself a “math guy,” feels the method could actually alter the vote count. According to his analysis, which he presented to the task force before the election, it would be possible for a person in the True Method’s automatic throwaway pile to have received more votes than a winning candidate.
“It was more about a quick election than about finding out what [voters] wanted,” said Harvie Branscomb, an El Jebel resident who has been an active critic of the May 2009 election.
With all the controversy flying around about the complexity of the proposed methods, the city knew it had to find a very accurate way to count the ballots.
Months before the election, Aspen hired Kleppner, Nick Koumoutseas and John Seibel of True Ballot, Inc., a private, election-tallying enterprise from Maryland. TBI was the one of the only outfits in the country with the ability to do an instant runoff for Aspen’s two-seat election.
TBI’s system is a unique vote-counting program that relies on Microsoft software, a completely different system than the vote-counting machines commonly used by cities and counties around the country. TBI scans the ballots into its computers, and the company’s special software counts it, relying on rules specific to each election.
Koumoutseas hailed the system as one that can conform to any election, no matter the type of ballot. It was the best fit for the True Method, to which it conformed. And while he said IRV has its problems – most importantly a reported propensity for complex mathematical anomalies that can allow lower rankings to count against higher rankings – the TBI count essentially eliminates them, he said. Also, it allows for an easy audit of the election, Koumoutseas said.
Sometimes on a ballot, voters will indicate their preferred candidate with a check mark instead of filling in the entire box next to that candidate’s name. If some voters are in a hurry, they might mark completely outside of the box. Or, if there is space for a write-in candidate, voters might write the name of the person they prefer and not fill in the checkbox at all. Or, as Koumoutseas pointed out, a voter might write in large letters across the boxes indicating the person they want in office: “I ONLY WANT MICK IRELAND” is a possibility. The lines that person makes on the ballot would inevitably cross the boxes of some of the candidates they didn’t want and the computerized scanner would count that vote for another person, even though the intent of the voter was clear on the ballot. The list of similar problems, which are not unique to IRV, and are reminiscent of the “hanging chads” in the 2000 presidential election in Florida, goes on and on.
The TBI scanners create an electronic image that allows the analysts to electronically manipulate the ballot to reflect the voter’s intent. The analyst simply clicks on the name of the candidate the voter clearly wanted in office, and the system highlights it.
“Hopefully the system picks it up and we have ways to correct the system to pick those up,” Koumoutseas said.
The information on the ballots is then transferred into a Microsoft spreadsheet in lines that are called “ballot strings,” each of which represents one voter in an election. In the case of IRV, the ranked choices are indicated in lines across the page.
Branscomb said the system is one of the best out there.
But it remains a mystery to many observers, who have referred to the TBI system as a “black box.” And the city’s contract with the company allowed TBI to do whatever it wanted with the election data, including the images of the scanned ballots. There was no clause saying, for example, that TBI could not sell the information.
Elizabeth Milias, who became a vehement critic of city politics after she was “disappeared” (as she puts it) from the Election Commission after a tense dispute last fall with the city, says the system was not fully vetted by public officials. In essence the system circumvented the normal public election process, she claims. Normally, vote counting technology, like the Accuvote machines used in most city elections and all county elections in Colorado, is certified by the Secretary of State’s office. But that process is constantly in flux and can take up to three years to get a new technology approved, Branscomb said, which poses a number of logistical problems to vote-counting innovators.
TBI’s system endured two process tests for the Aspen election and has been tested in private elections, but it went without a comprehensive analysis in Aspen by pretty much everyone but TBI.
Two tests of the system, called logic and accuracy tests, or LATs, were conducted by TBI and city officials leading up to the election. These tests are performed before every election, to ensure the counting technology is working properly. City residents, candidates and election followers gathered in the basement of City Hall on April 21 to fill out mock-ups of the ballots that had been approved for the election. Koch randomly handed 86 of them to the people in the room, they filled them out and handed them right back to Koch, sources said. The testers didn’t tally their votes, and the pile was counted by city officials and sent via Federal Express to TBI by Milias. This test found no problems.
Normally, Koch would have handed out 100 of the test ballots and the attendees would have tallied them, per state rules for LATs. But True and Worcester said in a recent statement that the meeting attendees got bored with the process and didn’t want to continue. “Unfortunately, the audience lost patience and only prepared 76. … No one objected to proceeding with the 76 ballots,” the statement says. And none of the public counters could tell for sure if the count was right because the attendees did not record the tally. In the same statement, True and Worcester said Koch offered to have audience members count the ballots, but they declined.
The second LAT, conducted on the eve of the election, found that the company’s system was counting the results backward. “It reminded me of the Bible,” Milias said recently. “The last shall be first, and the first shall be last.” But Kleppner reconfigured the software on the spot – one of the great things the TBI system allows, Koumoutseas said.
After the polls closed at 7 p.m. on May 5, ballots started flowing in from polling locations to City Council chambers, where they entered TBI’s machines. Skadron, Klanderud and Menter talked about the election, and observers sat in the chairs across the room watching the ballots flash across the plasma screens that line the walls. The display was meant to let Aspen voters watch the public process of ballot tabulation. Throughout the count, Kleppner’s computer kept making a strange sound – “Uh-oh, Uh-oh, Uh-oh” – an indication that ballots were being rejected from the system.
True and Worcester say only a small portion of the ballots were shown that night. And they say most of the ones that appeared only did so partially. But they issued sworn court statements saying that every ballot was shown on the screen. Kleppner also printed a guest opinion in The Aspen Times, saying the same: “… in plain view of the public and cable viewers, True Ballot, Inc. scanned every ballot using commercial imaging scanners, processed the data on the images, and publicly reviewed every ballot twice to make sure the computer interpretation of the ballots matched human interpretation of voter intent.”
But even if all ballots were shown, the 99 ovals that sit next to the 13 names, three write-ins and one policy question are difficult to count in the one or two seconds that they appeared on the screens. And the Grassroots TV broadcast that showed on some of Aspen’s television sets was not long enough to show all the ballots.
“IRV tallies were conducted and announced well after the end of the public TV show and during a process which few were able to observe or understand,” Branscomb said in a written response to Kleppner’s opinion piece.
All this came after city officials made promises during the vetting and approval of IRV that all the ballots would be shown, and the public would be able to verify the count.
The ballots were fed through the machines, and TBI counted the vote. Final numbers for the mayor’s race showed Mick Ireland had won with 1,273 of the 2,544 votes cast – or exactly one more than half. In the City Council race, Derek Johnson won the first seat, and Torre won the second, both with the same number of votes as Ireland.
“In comes Caleb Kleppner, and out comes the winner,” Milias said of the process earlier this year.
And all those numbers remain as the final tally on the city website, despite a publicly disclosed discovery of a counting error in the mayor’s race and several data analyses showing the count in the City Council race was also wrong.
No one had asked any questions about the “Uh-ohs” coming from Kleppner’s laptop. But TBI conducted a review of the mayor’s race that showed the count was off by 37 votes because of the last-minute changes that were made between the second LAT and the election. TBI had set its system to stop counting votes for the winning candidate after that person reached a 50 percent or more majority. Ireland had really received 1,310 compared with Marilyn Marks’ 1,124 votes.
TBI notified the city of the error on May 19, three days before the May 22 deadline for candidates to formally contest the election. But the city waited until May 28 to tell the public about the error; a press release touted the fact the city was committed to disclosing such errors. Worcester and True said the nine days passed because Koch, who they said normally handles such matters, was on vacation. City spokeswoman Sally Spaulding, True and Worcester worked with TBI on official wording for the release – the attorneys didn’t think TBI’s original wording “anomaly” accurately portrayed the mistake, for example – in e-mail messages on May 27 and 28.
The mistake didn’t change the results of the election; in fact, it reinforced them. But the audit was done privately by TBI, without any oversight by Aspen government officials or its residents. And, to the dismay of Marks and a group of election activists who want proof the count was correct, the city has adamantly refused to release any of the election records outside of the complex ballot string.
But what Rob Ittner finds more interesting than the TBI mistake was the True Method’s results in the City Council race. In the first round of tabulation, the batch elimination, the method eliminated Adam Frisch, who had the highest number of votes among the batch-eliminated group of candidates. But, had the batch elimination only taken four people out of the election, and Frisch had stayed in the race, then Torre would not have beaten Frisch. According to Ittner’s in-depth analyses of the ballot strings, Frisch would have outpolled Torre. The battle would have continued with Jack Johnson, Derek Johnson and Frisch, with Derek Johnson coming out on top, and Frisch in second place.
The city has never agreed that Frisch would be in Torre’s place if sequential elimination had been used but, asked in an interview if a different counting method would have resulted in a different tally, True said, “Absolutely.”
While the likelihood of some vast, fraudulent voting scheme to keep Frisch out of office and get Torre on the City Council is almost non-existent because of the extreme difficulty of intentionally gaming IRV, Ittner says the system did an injustice to the election.
“There were a lot more people that voted that preferred Adam before Torre,” Ittner said. “Did the system work? No, in my opinion.”
“It was a Rashomon election,” Zimet said, referencing the acclaimed 1950 Japanese film in which a convicted rapist’s fate depends on four conflicting testimonies by witnesses.
When Kleppner, Koumoutseas and Seibel returned to their TBI headquarters in Maryland the day after the election, they didn’t leave empty-handed. Koumoutseas took at least one copy of the ballot images on a CD with him.
Worcester and True said they were unaware that Koumoutseas had taken the copy; TBI was supposed to hand all of its election records over to the city, they said.
TBI was under no obligation to keep them from the public.
About a month after he returned to Maryland, Koumoutseas said, he destroyed the images, although the city charter requires all ballots to be kept intact until at least six months after the election. He said the city issued an order to TBI to destroy the images. Worcester and True said they were unaware of any communication by the city with TBI asking them to do so.
Elizabeth Milias and Chris Bryan, along with Clerk Kathryn Koch, became Aspen’s Election Commission, the body normally charged with certifying election counts and finding solutions to any potential problems, on March 23 – several weeks before the election and outside the normal timeline for the group to be appointed. Normally, the commission serves a two-year term, from July to July, and oversees one municipal election. True and Worcester, asked why the commission was not in place at the time, said they didn’t know, that it may have been an oversight. The city might have simply forgotten to appoint one after the 2007 city election.
Mayor Ireland said in a recent interview: “It just never came up.”
So Milias and Bryan came into their positions with no predecessors, not sure about their exact responsibilities. Worcester told the commission just before its dismissal that the group is traditionally “ceremonial.”
But this time around the commission was special because of the unique circumstances. Milias and Bryan expected to serve the regular two-year term, with four extra months at the beginning, Milias said. (Bryan, an attorney, did not respond to requests for an interview.)
And although Bryan said he would certify the election, it’s clear that both he and Milias took issue with the way the election was conducted. When they began asking questions about IRV and requested taxpayer cash for independent legal counsel because they felt Worcester and True were inherently conflicted as city employees, Milias and Bryan were quickly hushed, investigated and dismissed from the commission at the beginning of November.
The city told them, “Not only are we not going to give you the money, but you are out of line,” Branscomb said.
Former councilman Johnson found evidence through records requests that the Election Commission had been receiving e-mails from Marks, lobbying them to gain independent counsel. He also found that Milias and Bryan had met in private – which violates Colorado open meeting rules – over a hamburger the day before the election to discuss commission business.
Johnson said Marks had deliberately infiltrated the Election Commission and convinced the members to take action to cause a recount of the election, hoping that the recount would find that she had won the mayor’s seat.
Marks never requested a recount. But she did file a lawsuit against the city, saying it had violated the Colorado Open Records Act when city officials denied her access to the images of the ballots that were shown on the screen the night of the election – the same ballot images TBI took to Maryland the day after the election and the city had shown publicly in a May 7, 2009 test of the tabulation software. The commission agreed with Marks on most of her allegations.
City officials have claimed that Marks brought the allegations to the commission and the members followed her lead. And although Marks has partnered with Milias in a project called the “RedANT,” an online publication that is very critical of Aspen City Hall, Milias denied those claims. Marks is no longer a part of the “RedANT.” What happened since the initial heated debate indicates that at least Milias was on board with the allegations before Marks became the face of the argument. Milias was the one who pointed out the mistake in the way TBI’s system was counting the test ballots on election eve.
“I have a mind of my own,” she said in a recent interview.
True and Worcester maintain that the reason the City Council dismissed Milias and Bryan from the commission was that their term was over. It had nothing to do, they said, with the investigation. When asked why the city waited until four months after the regular June end of the term, they said it was another oversight, that the terms should have ended much earlier.
But Branscomb thinks the commission’s dismissal hinged on Johnson’s investigation.
“That is a blowback intended to chill the proper functioning of democracy,” he said.
It’s clear, the critics say, that the election system for May 5, 2009 was flawed. Even True and Worcester admit that things went wrong, although they maintain the city handled them in the best way possible.
So what can be done to fix the system?
The current Election Commission is taking a much more active role than any has in the past.
It is fully reviewing the large number of complaints from Marks, Branscomb and Milias. The complaints are, in effect, a compilation of the Marks lawsuit against the city, a complaint she filed with the district attorney’s office (which eventually decided not to pursue charges) and the result of thousands of hours the three have spent investigating what happened. They include alleged breaches of normal election protocol described in this story, violation of Secretary of State rules the city adopted before the election and, perhaps the biggest concern for Marks, the city’s refusal to release the ballot images.
The City Council has placed a measure on the Nov. 2 ballot, asking voters if they want to repeal IRV and return to the traditional runoff election Aspen used before it adopted IRV. If IRV is retained, the City Council may keep the True Method or switch to a different system.
So now, it’s the voters who have a choice to make.
Does Aspen retain IRV and potentially repeat what has happened over the last 15 months? Or does the electorate want to bring in the traditional runoff process that, while costly and cumbersome, never presented the counting-related problems of IRV?
The current commission is trying to figure that out. Members Bob Leatherman and Ward Hauenstein are trying to find a balance between answering the ire of the critics – Marks, Milias, Branscomb and Zimet – and educating the voters about the pros and cons of IRV and traditional runoff. It’s just a matter of finding that balance, Leatherman said.