You are not logged in. LOG IN NOW >

Taking a Deep Breath Over Google Supposedly Sending People to the Wrong Polling Places

BY Nancy Scola | Thursday, November 4 2010

Google's polling-place look-up tool used Voting Information Project data to point people to where they needed to go to cast a ballot. A Google spokesperson says that a few million people used the tool on and before this Tuesday.

Research conducted by the DC-based political software company Aristotle on Google's polling place look-up tool grabbed some big, gut-grabbing election day headlines on places like Fast Company ("Google May Have Directed Hundreds of Thousands of Voters to Wrong Polling Locations"), Politico ("Google Polling Place App Misleads Voters") and, eventually, inevitably, the Drudge Report. The allegation, in short, was that nearly a million civic-minded Americans might have been sent the wrong way after using Google's nearly ubiquitous tool to dutifully figure out where they were supposed to cast a ballot.

But dig a little deeper into this situation, and a couple things eventually become clear. First, while there do indeed exist some Americans that could have gotten bad voter information from the Google polling place look-up tool (Fast Company's reporting was sparked by an editor there who really did get the wrong address for his New York City polling place), the fears conveyed by those stories were wildly overblown. And second, bigger picture, our tolerance for a new model of collaborative, open-source democracy might not be quite on par with our ambitions for it.

To be sure, any bit of evidence that Americans are being thwarted while taking part an election is, of course, troubling. But at least some of the claims made by Aristotle and relayed by those publications just don't stand up. And none of them merit the hyperventilation.

A quick dip into the weeds is necessary here. While it was Google's app that grabbed the headlines, what the claims really called into question was the data powering their widget, which comes out of an effort called the Voting Information Project. That project, known as VIP, has worked for years to bring together state election officials, tech companies, and election advocates to standardize and publish data held by localities on where America's 200 million eligible voters are supposed to cast their ballot. At the moment, VIP partners include Google, the Pew Center on the States, the left's organizing outfit New Organizing Institute, and the conservative-leaning firm Engage DC. By streamlining and nationalizing that data, VIP made possible a bonanza of new democracy tools this election, from Google's own widget (used by election officials in, for example, Virginia and Hawaii, and campaigns including Lisa Murkowski's, and political party efforts like the National Republican Senatorial Campaign Committee's site) to other third-party tools like Foursquare's "I Voted" mapping project, Mobile Commons text-based poll locator, and even the sassy Your F*&king Polling Place site.

A chart supplied by Aristotle and posted by Fast Company showing an alleged error rate for Google's polling pace look-up tool for a dozen states.

Back to the study. Aristotle's research is said to have involved testing against Google's app, and thus VIP's data, a thousand or so households in twelve different states. Randomly selected by the company, those addresses were run against both the free Google widget and the polling place look-up tool called WhereIVote.com that is sold by Aristotle. (Fast Company reported that the test involved using states' own on-site look-up tools, but Aristotle confirmed in a call that the control used was its own site, which it says retrieves information directly from the official sites.) Differences between the two results counted as an error against Google. Aristotle's researcher then extrapolated from that sample an overall error rate for each state. That error rate ranged from .1% in Iowa to an eye-popping 18% in Washington State, the latter of which was said to account for more than 420,000 of the potentially misdirected voters in the Aristotle study.

But the Washington State Secretary of State Office's lead on the Voting Information Project finds Aristotle's findings absurd. "It's just impossible," said Paul Miller in a call yesterday. Miller serves as a senior technology policy advisor in that shop. Washington is, you see, a vote-by-mail state with the exception of one county, Pierce County. "It's difficult to understand what errors could be found with polling places," said Miller, "because we didn't put any polling place data into the project -- not even for Pierce County, because 70% of those people still vote by mail."

And Aristotle's Senior Vice President of Data Operations Igor Kyrylenko concedes that Washington State's upsetting error numbers aren't what they appear. "It's a little bit of a problem, I would admit," Kyrylenko said yesterday in a call. The trouble, he explained, came when Aristotle used just Pierce County addresses to test, but extrapolated a total error number using the number of households in all of Washington State. "The numbers in Washington are potentially a little bit wrong," said Kyrlenko. Pressed on why, if there was no polling place information from Washington State in the tool at all, the error rate found by Aristotle wouldn't be either zero percent or a hundred percent, nothing in between, Kyrylenko couldn't explain.

(It's difficult to test the findings at this point; both the Google tool and Aristotle's Where I Vote app came down after election day.)

A study of Google's polling place look-up tool conducted by the DC firm Aristotle grabbed some provocative headlines on election day.

Reached by phone, Aristotle CEO John Phillips emphasized that his statement on the errors his researchers found in the Google tool -- “As millions of Americans went to the polls on Tuesday, Google offered to provide voters with accurate information about their polling place. Unfortunately, according to investigations by various reporters and observers, and an Aristotle review of the tool, the technology used likely produced erroneous information for some voters." -- spoke about only "some" affected voters. Moreover, Phillips said that he thought that the blame here lay not with the Voting Information Project, necessarily, but with Google. (To get back in the weeds for a moment, 11 of the 12 states in Aristotle's research are "feed states" in VIP parlance, meaning that the source data comes directly from state election officials in XML form.)

Phillips floated the idea that perhaps Google was drawing "crop circles" around entered addresses and then simply directing people to the nearest polling place, which isn't necessarily the one assigned to their actual precinct.

On Phillips' distinction of "some," with Washington State out of the picture, the number of potential errors in Aristotle's 12-state study drops down to some 300,000. North Carolina contributes 43,000 to that total, and state election officials there don't see what whatever errors might exist in the data have to do with Google. "I have a really hard time believing that they [Aristotle] matched the two and came up wrong," said Marc Burris, Chief Information Officer for the North Carolina Board of Elections, in a call yesterday, "because Google has data as current as we provide it to them." The data feed provided to the Voting Information Project is, said Burris, the very same one that powers the state's Board of Election site that Aristotle has implied it used for its own testing. Burris takes the whole matter personally. His office, he reported, hadn't gotten a single call from a North Carolina voter who had used the tool and gotten bad information. "And I would know," he said, "because this thing is kinda my baby."

In other words, voters are on occasion going to get bad election day data, whether they get it from Google or from their local election officials.

Conceding that errors do happen when it comes working with huge caches of election data from multiple sources, the New Organizing Institute, a lead VIP partner, cites Aristotle's combative reputation in the industry and its competitive interest as a vendor offering a for-fee product similar to the one Google made available for free. "Aristotle has a history of attacking organizations it feels competes with its business and has done so again in attacking the data Google is using in its web application for polling places," said the New Organizing Institute's Education Fund in a statement.

Part of this story is, to be sure, that a vendor like Aristotle might understandably feel big-footed out of things by a company like Google -- which, despite its good-guy reputation, is indeed a vendor in its own right.

For Google's part, a Google spokesperson laughed when told of Phillips' suggestion that the company built a polling place look-up tool that used crude "crop circles" to figure out the right place to send voters. That approach would, indeed, likely produce error rates in the high double-digits, not one or five or seven percent. While declining to specify hard numbers in public, a Google spokesperson said that the tool responded to a few million different requests for polling place data this election day, while only a relative handful of errors came in through the reporting mechanism baked into the site. (As a back-up, the Google tool also returned to voters a link to their state polling place look-up tool.)

People who have spent time in the thick of American elections will tell you that, in practice, our democracy is hardly perfect. That's part of the problem that the Voting Information Project says its attempting to solve, by applying some of the principles that are driving the overall push towards more accessible, participatory, and collaborative politics. VIP is a challenge to the top-down model driven by a single government entity or a single vendor.

Open source's potential, say advocates, is matched by a possibility that with increased freedom comes increased potential for trouble. But the collaborative approach, says VIP participants, is also has its own self-checks. Mindy Finn is a partner at Engage DC, which worked with the VIP effort to manage the data relationships with states. "The whole idea is that we're not trying to supercede election officials," said Finn. "We're a partner here."

"We're all about working with election official to produce really excellent data," said Anthea Watson, the Voting Information Project's project manager. "With any data, there are errors. But we publish everything in an open-source way, and we invite people to help us to help voters."

North Carolina's Burris, echoing a sentiment generally being expressed by cash-strapped states, says that working with the Voting Information Project has been a positive experience. "Anything that can help get people to vote is a great thing," he said, "but especially if we don't have to pay for it." The worry of many involved in the Voting Information Project, and among supportive observers, is that those blaring headlines of election day -- suggesting scores of Americans being denied the chance to participate in American democracy -- only serve to scare off those open-minded government officials like Burris who have reached out of their bureaucratic silos and taken a chance at working with outsiders to participate in an evolving, experimental, collaborative effort to make elections work better. No one, says participants, wants a single voter sent off track. And Fast Company has, indeed, found three voters given bad information by the Google tool, and even the White House seems to be the victim of some bad data.

"Anything above [a] zero percent [error rate] causes people concern," agreed Burris. But the bigger question becomes whether such an intense focus on the regrettable errors found in innovative approaches to doing democracy means that the perfect gets made the enemy of the good-and-getting-better.

Figuring out new ways of doing American democracy, say advocates, isn't about perfection. It's about the alternative.

Note: A sentence on VIP partners was tweaked to more accurately reflect the groups participating in the project.