In 2016, the polls got it wrong. They failed to predict that Donald Trump was winning key battleground states. But a startup in San Francisco says it spotted it well in advance, not because of the "enthusiasm gap" — Republicans turning out and Democrats staying at home. Instead, the startup Brigade's data pointed to a big crossover effect: Democrats voting for Trump in droves.
The company built an app that asks a simple question: Which candidate are you going to vote for?
It's like what boots-on-the-ground organizers do. Though there is one big difference. In the physical world, most people aren't wearing their candidate button for the 18 months leading up to the election.
On the app, Brigade CEO Matt Mahan explains, people share their pledge and invite their friends.
"It's a little bit of a change from what most people are used to. People don't go around in the offline world broadcasting [whom] they're voting for," he says. "They may share it with friends or family. But I think many people feel that it's a fairly private decision."
The app seemed to work. It has nearly 200,000 verified users — not just signups or Twitter bots or trolls, but citizens the startup has crosschecked with voter registration records to confirm identity.
When Mahan looked at the pledges, he didn't see data among Republican voters to back up the trending hashtag #NeverTrump. Among registered voters on Brigade, 94.5 percent of Republicans pledged to vote for Trump and only 2.2 percent pledged to vote for Hillary Clinton.
That's roughly what you'd expect.
Here's where it gets fascinating: On the Democratic side, Mahan explains, "we saw something entirely different." Only 55 percent of registered Democrats pledged to vote for the Democratic nominee.
It's not the Bernie Sanders effect. This result is in the general election, after the primaries. It looks more like the Trump effect. Of Brigade's verified voters, 40 percent of registered Democrats pledged to vote for Trump.
Brigade saw this pattern back in September. But the company didn't trust it because its user base skews conservative. Even though Brigade was started by liberals in San Francisco, it went viral in Republican circles.
Once the election happened, however, it realized it was on to something after all.
Company analysts noticed that their data foreshadowed the outcome in states where Trump beat the predictions, as stated on Nate Silver's popular site FiveThirtyEight.
Let's take the swing state North Carolina. Trump beat the final prediction on FiveThirtyEight by 4.5 percentage points. According to the app, it was also a state where about 25 percent more than Brigade's baseline of Democrats pledged to vote for Trump. That is, when Brigade analysts compared North Carolina to the overall baseline of Democrats crossing over on the app — the analysts saw it happening even more, at this far higher rate there.
The same thing happened in another swing state, Pennsylvania, where Democrats on the app appeared 15 percent more likely to cross over.
(Small note: Michigan voter records don't provide party affiliation, so Brigade couldn't analyze that state).
In contrast, registered Democrats on Brigade were 30 percent less likely to cross over and pledge for Trump in Nevada. Trump underperformed FiveThirtyEight's forecast in that state by 1.2 percentage points.
These crossover effects break political science as we know it. For months, the media have been telling the story of how Trump has upended the Republican Party. Turns out there were early signs he's upending the Democratic Party, too — and possibly with the voters you'd least expect, like women.
Mahan's team is still poring over the data, and he shares a fresh finding. "Interestingly one of the data points we just pulled in the last hour indicated that a higher percentage of these registered Democrats crossing over to vote for Trump were women," he says.
In states with outcomes that didn't match the polling results (i.e., Trump did better than expected), Brigade saw white women registered as Democrats pledge their vote to Trump at a much greater rate (170 percent) than the country as a whole.
Copyright 2016 NPR. To see more, visit http://www.npr.org/.