Should You Trust Slate’s Early Numbers?

What’s actually going on out there has been pretty much a mystery, but Slate is aiming to change that. Photo: Sara D. Davis/Getty Images

Once upon a time, Election Day was largely a mystery, with the image of likely outcomes created by preelection polling being reasonably solid unless turnout patterns, as occasionally reported by mostly local media outlets, produced something big and shocking.

Then came the era of widely leaked exit polls, during which anyone wanting to get an early take on an election spent a lot of time obtaining, publicizing, and sometimes overinterpreting the absolute most recent set of “exits.” After the exit-poll “crash” of 2002 and then the egregious exit-poll errors of 2004, the media sponsors of the exit-poll system began cracking down on distribution of the data and the potential for leaks. Why even do the exits? Because they are invaluable for interpreting election results, even if they don’t necessarily predict who is going to win where by how much.

In any event, the reduced availability and credibility of exit polls left a deep, gnawing gap in the lives of a lot of people on Election Day, whether they were journalists paid to report the results or just citizens anxious to get a glimpse of their country’s future. The wait was made all the more painful by the knowledge that deep inside the major campaigns there is all sorts of real-time data and analysis going on.

This year, efforts are under way to feed that need, most ambitiously by Slate, which in partnership with a new data-analysis firm called VoteCastr and an army of volunteers at actual polling places, is offering what it describes as the kind of rich information campaigns enjoy.

There are three kinds of information Slate is using: actual early-voting data (mostly completed over the weekend), its own “proprietary” preelection “large-sample” polling, and “real-time” reporting of turnout patterns as discerned by its volunteers and perhaps some other sources. All this goes into a model that also includes 2012 data, and comes out as a rolling estimate of where the actual vote stands for each major candidate in eight battleground states: Colorado, Florida, Iowa, Nevada, New Hampshire, Ohio, Pennsylvania, and Wisconsin. As of this writing, Slate has estimates for all of the states other than Colorado, and Clinton is in the lead in all of them. But the system isn’t just one of statewide projections based on a mysterious mix of inputs: It offers county-by-county real-time estimates, too. For example, right now VoteCastr shows Clinton winning the key central-Florida Hispanic center of Osceola County by a 55-39 margin.

As Slate warns readers, you cannot take any of these estimates to the bank, because they can and will change during the day as different “batches” of voters show up at the polls. As the estimates become less and less based on very solid early-voting tabulations (at least in the states with significant early voting), they could get more and more volatile. By the same token, estimates for states that have very high early voting — such as Florida and Nevada — may already be pretty close to the final projection.

It is impossible to judge at this point how accurate this whole system really is. We don’t know what the final estimates will look like, and anything before that is fascinating but not necessarily all that informative. If Slate’s estimates wind up being pretty accurate, and particularly if they differ from what the public polls predicted, it will add to the growing critique of old-school random-sample live-interview polls as being less reliable than those the campaigns conduct privately — which Slate is going to so much trouble to emulate. Four years from now we may have forgotten about Slate’s experiment, or acknowledged it as a breakthrough. In the meantime, enjoy!