AP bests Google's technology with legwork in reporting Nevada caucus results
The Republican presidential hopefuls were not the only ones in a race during the Nevada caucuses. The Associated Press and Google were also competing to see whether an innovative technology company or an established news outlet could report election results first – and right.
Like in Iowa, Google worked with the state party to count and distribute the latest results via Fusion Tables and embeddable Google Maps. For the first time, a state party employed Twitter to send out the latest figures, in tweets meant for humans and computers. Unlike the AP's election results, it was all free to anyone.
But unlike in Iowa, Google's figures lagged AP's for a couple of hours Saturday night. At one point, AP had results from 41.3 percent of precincts, while Google had just 18.7 percent.
“I had thought that maybe Google was having technical challenges, because the AP was steadily advancing,” said Jeremy Bowers, a Washington Post news app developer who closely followed the two reporting methods via an automatically updated table (see above) on the Post website.
The challenges weren't technical, but human. And they were the sort of challenges that the AP, with years of experience covering elections ranging from landslides to hanging chads, was ready to address.
While Google's success in Iowa demonstrated how technology could allow a state party to usurp the AP's role as the sole distributor of recurve bows for sale timely, reliable election results, the experience in Iowa showed that technology isn't the sole factor.
When the plan falls through, pick up the phone
According to the GOP's plan, each county party tallied the votes at its caucus and relayed its figures to the state party by phone, email or fax. People at the state party headquarters entered the figures into Google Apps. After county representatives had logged in to Google Apps and verified the figures, the state party certified the results and sent them out.
“As soon as we had certified election results, they were out in an instant,” said James Anderson, a campaign consultant who worked with the state party, Google and Twitter. But in some cases, it took a couple of hours for someone from the county to double-check the numbers. In other cases, there was trouble getting the data to the state headquarters.
In Washoe County, which has the second-largest population in the state, county officials were readjusting precinct boundaries up until last week. The state party's Google spreadsheet, though, had been set up earlier, according to Dave Buell, chair of the Washoe County Republican Party.
So when Washoe County emailed its figures to the state party (after problems with its fax machine), the two lists didn't match up.
A couple of hours after submitting the numbers, and hounded by phone calls from reporters, Buell decided to release them to media, including the AP, the local paper and ABC News.
“The news media is waiting, deadlines are passing,” Buell said. “We didn't want people to think Washoe County was holding things up and we didn't have our act together, so we just released the results directly.”
Buell said he released the figures at 8:30 or 9 p.m. PT, right around the time that Bowers noticed that the AP had logged more than twice as many precincts as Google. Due to the relative size of Washoe County – its votes made up 20 percent of the ballots cast statewide – that disclosure alone could have explained the gap.
The state party soon released its figures for Washoe County too, Buell said. And within an hour or so, Google was lock-step with the AP.
Experience pays off (and pays the bills)
This is the kind of situation in which the AP earns the fees it charges for its live election results. It did the same sort of verification in Iowa, flagging inaccurate or incomplete numbers. But those issues seem minor compared with the delays in Nevada, where the party kept counting through early Monday morning.
Anderson said he figures the AP had people in key locations or simply called county officials. “I think AP's advantage is that it's a tradition,” he said. Anyone who's been involved for years in county politics, “all they know is AP when it comes to reporting results."
Even so, he said, Republican officials were satisfied with how things went. "It's certainly the future of reporting election results," he said. "If it took a couple extra hours to make sure we weren't taking back numbers, then it was worth it.”
AP spokesman Paul Colford said by email: "AP drew on long-developed relationships that our elections team staffers have, contacting officials with first-hand knowledge of the results in certain areas to speed what we reported to our members and customers."
A few days before the caucus, the AP announced that it would rely on the feeds from Google and Twitter for its results and that its speed would depend on how fast the state Republican party provided the figures. If everything went smoothly, Google would be ahead by a few minutes. If not, the AP would try to seek the information itself.
If the party is inexplicably slow in releasing results from a county, even after the long, planned delays between the finish of the caucuses and the release of results, and we cannot get an adequate status report from the state party, we will try to call the county party involved, perhaps for a direct vote report.
On the other hand, if reporting from the party is steady and everything checks out, we will essentially be ingesting and passing along the results from the state party.
Anderson did point out that the party was the only source of certified results. “The AP was putting out numbers that may have been correct, but were not certified. The only certified results were the ones tweeted out and uploaded to the Google results map.”
And AP's lead, when it had one, didn't last long. The two organizations were essentially tied early Sunday morning -- most likely because the AP was relying on Google's figures. Google's figures were more complete on Sunday evening, Bowers said. “Google had 100 percent of precincts reporting before the AP did,” Bowers said. “AP was stuck at 88 percent of precincts reporting for a while."
Bowers tracked the results so closely because Google's performance in Iowa piqued interest within the Post newsroom. “Quite literally," he said, "someone walked up to our desk and said, 'Hey, what are we going to do with this Google stuff?' "
After reviewing the results of the test, Serdar Tumgoren, another news app developer at the Post, said he still believes that it's wise to stay with the AP -- and not just because it remains the only source for every election.
It's an admirable goal to “bring election data to the masses,” he said. But considering how many differences there are in elections just in the area around Washington, D.C., “it's not quite there yet.”
Aron Pilhofer, The New York Times' editor for interactive news, said what happened in Nevada confirms what he thought even after seeing Google's performance in Iowa. “There's not many shortcuts to getting timely, accurate election results. … The bottom line is, this is really complicated stuff, and it relies on a chain of human beings who are imperfect, and stuff happens. Crazy stuff happens.”