“Collaborative election monitoring via social media” – notes from a panel about Electionland at #ijf17

I’ve been at the International Journalism Festival in Perugia, and here are my notes from a session about “Electionland”, billed as the largest collaborative journalistic exercise to monitor an election ever. The panel featured Scott Klein of ProPublica, Fergus Bell the founder Dig Deeper Media, Ed Bice of Meedan, and was hosted by An Xiao Mina.


The idea of Electionland came out of conversations about how data could better be used to cover the US election. Against a background of the unprecedented unpopularity of both candidates, and one of them openly telling his supporters that the vote was going to be rigged, ProPublica spotted an opportunity to do a different type of election coverage. Scott Klein said that normally, ProPublica doesn’t really do election coverage in the traditional “Here are the numbers, and here is who won” sense – and he usually gets to leave the office early while other news organisations are hunkering down for an all-nighter.

Their position was that they could use social media signals to try and alert local reporters to problems on the ground with voting. There are estimates that in 2012 as many as 700,000 people in the US didn’t get to cast their votes because of long queues, and it is a problem that disproportionately affects districts with a higher percentage of ethnic minority voters. The typical way of reporting this is to have a reporter pick a few polling stations, and then drive around and see what’s up. They figured using data could help drive more informed reporting.

So they set up a massive newsroom operation on the night. In the end there were 11,000 journalists working on it. Not covering the politics or candidates of the election, but covering the process of the election. That broke down into 400 local journalists, 600 j-school students from 14 different schools, and 100 professionals in the live newsroom.

One thing I liked about this session, incidentally, was that actually I learnt a lot about the mechanics of the US election system. I hadn’t previously really grasped how decentralised and unregulated the states are, and how hands off the federal government has to be, leaving states free to pursue very different electoral styles. “Some states are very bad at running elections” said Klein, while others, like Minnesota, consistently have high turn-out and very low incidences of any problems. It still astonishes me really though that a land which in public declares itself to be the home of democracy and freedom can have people working so energetically to actively disenfranchise parts of the population.

Back to Electionland. So the newsroom had a “feeder” team, who were monitoring social media activity for reports of problems voting. They’d pass their info on to a “catcher” team, who would do further verification. Promising leads would then be passed on to local journalists. It was important to have that local dimension they explained, because a centralised office in New York wasn’t going to spot all the angles, whereas someone getting tips about their patch could quickly spot “Oh, that’s where the Mayor votes” or “That’s right where the African-American voting population is.”

Fergus Bell was on the panel, and he spoke a little bit about the training operation they had to set-up. The biggest challenge was that you can’t do a dry-run for election night. The process they settled on, then, was one where they held a central training session for senior people at the j-schools that were taking part, and they in turn trained their students.

They had to have an agreed centralised process, so what they did was crowd-source all the different terms that might be relevant if you were looking for reports of long lines or voter intimidation or registration issues. These were turned into complex Tweetdeck search queries, and other organisations like Facebook helped by providing access to tools like Signal and CrowdTangle. Each j-school was given a set of states to monitor, and told to set up exactly ten workstations in an identical way. That way the process was consistent across the country. The central newsroom kept an open hangout with all the distributed centres, so they could convey any new instructions across the country simply by walking up to a computer screen. The core team had a huddle every two hours during the course of the day and night to discuss whether the process was working, and make any changes necessary to the workflow.

Once the students had spotted potential leads, they put the social post into a centralised system, where further work could be done on it. Metadata from the tweet or post, for example location data and timestamp, was automatically extracted. The posts could then be tagged – Ed Bice described it as “Collaborative verification with structured tagging”

Across the night they generated something like 400 news stories – including video footage of electronic voting machines not working properly – that they could bring to a wider audience. And despite the tweetstorm protestations of the winning candidate, Bice said they saw no evidence of widespread voter fraud.

You can watch this panel in full online: “Electionland

Find all my blog posts from the 2017 International Journalism Festival.