When it comes to hour-by-hour and, in some cases, minute-by-minute Election Day horse-race results, many political scientists say you’re better off avoiding it all. Close Twitter. Minimize your browser. Throw your phone in a lake.
That’s partly because while day-of projections, which are based on voter-turnout numbers and historical voting patterns, are good at providing a glimpse into the right now, they often don’t reflect bigger-picture trends. For example, younger voters, who tend to procrastinate on getting to the polls, and blue collar voters, who are less likely to have the kind of job that you can leave for an hour in the middle of the day, won’t always show up in voter-turnout tallies until after 5:30 p.m.
Day-of polls also aren’t good at detecting aberrations in voting patterns. If significant numbers of people who have never voted before show up to the polls, as Donald Trump has promised, or certain demographic groups, like Latinos, end up turning out in record-high numbers, turnout-based models won’t show that.
Read More: Here’s What You Should Watch on Election Day
Another reason to avoid day-of results, some political scientists argue, is that they may have the effect of depressing voter turnout in the evening hours. “It’s not just the presidential race you have to think about,” Elaine Kamarck, a senior fellow in governance studies at the Brookings Institution, told TIME. “It’s the down-ballot races in all the Western states.”
If Democrats in California can see that, say, Hillary Clinton is ahead in early results by a significant margin, Kamarck said, voters in that state may decide to avoid the hassle of going to the polls after work, which could help Republicans in both gubernatorial and statewide races, as well as local contests. Largely in deference to such fears, the biggest news outlets — ABC, CBS, NBC, CNN and Fox, as well as radio stations, newspapers, wire services and websites — redoubled their pledge before Congress to avoid calling a contest until the polls close in that state.
New, public-facing websites like VoteCastr.us promise to disrupt that tradition, which they see as unnecessarily precious in a 24-hour news cycle.
Starting Election Day morning, VoteCastr, in partnership with Slate, will publish up-to-date poll results every three hours throughout the day. Sasha Issenberg, a journalist and partner with VoteCastr, described the traditional news outlets’ “self-imposed gag order” as a disservice to the public interest and argued that that there’s “scant evidence” that calling elections before polls close has affected turnout.
VoteCastr, which is backed by a bipartisan team of election veterans, including Blaise Hazelwood, former political director of the Republican National Committee, and Ken Strasma, who headed up micro-targeting for Barack Obama in 2008, will collect data from thousands of precincts across 17 battleground states, borrowing a strategy long used by candidates, parties and super PACs.
In an interview with TIME, Strasma was cautiously optimistic about the accuracy of the effort. “Any poll depends on the pollsters’ assumptions of the makeup of the electorate,” he said. “So with VoteCastr, they’re looking at past polling and surveys and then, on Election Day, saying, ‘O.K., based on who we’re actually seeing turn out, we can reasonably expect these people in this precinct to vote in this way.” In 2012, the Obama campaign used virtually the same method to predict that in Hamilton County, Ohio, their candidate would win 56.4% of the early vote. In an article for Slate, Issenberg pointed out that their guess ended up being just two-tenths of a percent away from the final count.
Tim Saler, vice president of the research firm Grassroots Targeting and former deputy campaign manager to Florida Governor Rick Scott, was also encouraging, but reiterated that the real magic was in the modeling. In Colorado in 2012, the Mitt Romney campaign cited public polling showing their candidate doing well among independents, Saler said. But those polls missed the fact that people who were identifying as independent, because that’s the way they thought of themselves, were actually registered Republicans. So on Election Day, Romney did much worse in Colorado than his campaign’s projections showed, Saler told TIME. “You have to avoid falling into the trap of misrepresenting the baseline,” he said.
The take-home message, said Roger Tourangeau, president of the Association for Public Opinion Research, is that while many of these models have become increasingly sophisticated over the past decade, with vastly more public polling and deep historical records on voter behavior, none of it is perfect. The only surefire way to know who won an election is to wait for AP vote tallies, which will come in after polls close or early the next morning, or to count every last vote, including absentee, overseas and preliminary ballots — an impressive logistical endeavor that will be complete by the end of the week.
If you want to avoid that roller-coaster feeling Election Day, wait until the polls close to get excited.