VoteCastr’s Hyped Real-Time Polling Data Badly Botched Trump Calls Early On, But Startup Claims It Mostly Got Them Right Later (Updated)

Ballot Vote Stickers
Susan Montoya Bryan/AP/REX/Shutterstock

UPDATED: Nearly every pre-election poll showed Hillary Clinton leading Donald Trump heading into Tuesday — and even the Republic National Committee predicted Trump would fall 30 electoral votes short of winning the presidency.

Also getting the call drastically wrong: Controversial Silicon Valley startup VoteCastr, which claimed it was going to shake up the media biz by delivering real-time polling results in seven battleground states, defying the decades-long practice of news organizations holding off on those calls until polls close. The upstart partnered with Slate and Vice News to publish data collected from dozens of polling locations throughout the day on Nov. 8 with predictive models to provide minute-by-minute projections.

“For the first time in the modern political era, Americans will see what the network executives and campaign insiders have seen all along: the game as it unfolds,” the startup boasts on its website. The Palo Alto, Calif.-based company was founded by political operative Ken Smukler, and its chief strategist is Sasha Issenberg, a contributor to Bloomberg Politics, former Slate columnist and author of “The Victory Lab: The Secret Science of Winning Campaigns.”

The problem: VoteCastr’s data ended up being wildly inaccurate.

The startup’s numbers showed, wrongly, that Clinton would prevail in five of the seven states it was tracking. It’s unclear why the VoteCastr model failed, but ultimately its results proved to be less reliable than the traditional media’s polls the startup was angling to disrupt.

The final estimated tallies that Slate posted at 7:55 p.m. ET from VoteCastr incorrectly indicate that Clinton beat Trump in five states: Florida, Iowa, Ohio, Pennsylvania and Wisconsin. The vote counts were accurate in only two states: New Hampshire and Nevada.

[UPDATE, Nov. 11, 1:30 p.m. PT: According to VoteCastr’s Smukler, Slate’s final update used its numbers as of 5 p.m. ET and never posted the company’s projected end-of-day numbers, which “created the false impression that VoteCastr’s numbers were wrong and is a great disservice to VoteCastr.” He says the startup correctly called six of eight battleground states that it was tracking 3-6 hours before polls closed (with the exception of Florida and Wisconsin).]

That came after VoteCastr data earlier in the day suggested Clinton held a lead in several of the states, including Florida, Pennsylvania and Ohio. The closely watched experiment appears to have led some investors to have an overly optimistic expectation that Hillary Clinton would win the White House, with traders pushing up stocks based on the VoteCastr info, the Wall Street Journal reported.

Leading up to the election, media pundits had been critical of VoteCastr, arguing that the data could deter people from voting if they believed their candidate was leading — which is precisely the reason news orgs don’t report on polling data until voting closes.

VoteCastr also encountered a few snafus on Tuesday. At 2 p.m. ET, Slate said its data-visualization maps showed only early vote totals, which remained largely static throughout the morning, because of an unexpected technical delay in receiving real-time data from VoteCastr. Earlier, at around noon ET, VoteCastr said on Twitter that its Nevada projections had errantly included Green Party presidential candidate Jill Stein, who was not on the ballot. “We messed up and we are correcting the Nevada results accordingly,” the company said.