The States of the Nation project: Reuters/Ipsos launches a deep-dive, 50-state presidential poll

Our new tracking poll of 15,000-plus Americans lets users fine-tune turnout projections and create their own scenarios for the outcome of the 2016 race. If the election were held today, Clinton would win by 108 Electoral College votes.

Predicting the winner of the U.S. presidential election may hinge less on understanding which candidate voters prefer than on forecasting who will actually vote.

It’s a relatively straightforward exercise to get an accurate estimate of which candidate is most popular. It’s much harder for a pollster to figure out which voters will cast a ballot. At best, a poll result is an informed estimate based on historical data and assumptions about how various factors might affect turnout. How will turnout be affected by having the first female candidate from a major party? Can Donald Trump bring out more white male voters? Will Hispanics be motivated to turn out in larger numbers this year?

So, Reuters and our polling partner, Ipsos, created a polling election tool that allows users to experiment with the critical variable in this election – turnout – to see how the results may be affected.

The Reuters/Ipsos States of the Nation project combines a massive online survey that gauges support in every state for the two main presidential candidates with the ability to set different turnout rates for a variety of demographic groups. The tool then runs millions of simulations to project how those factors translate into votes in the Electoral College.

The survey – the largest presidential tracking poll ever attempted – asks up to 16,000 Americans every week which candidate they prefer, ensuring the site has the most up-to-date data available. The survey is so large that we will be able to report polling data each week for almost all of the nation’s 50 states and Washington, D.C.

And that granularity matters, because campaigns often turn on mobilizing voters, and sometimes on very small and specific groups of voters.

Some polls focus on registered voters, a well-defined set of people. But using that cohort leaves open the question of who among them will show up to vote. Hence polls like ours focus on a subset of registered voters - likely voters.

The Reuters/Ipsos team currently estimates overall turnout in the election will be around 60 percent. But our turnout model is nuanced: It estimates 70 percent of white men will cast ballots and 60 percent of African-American women will go to the polls. Candidates who believe they appeal to young voters will almost certainly try to do better than our estimated 13 percent turnout rate. President Obama built his electoral success on turning out more young and black voters in 2008.

How would Hillary Clinton do, for example, if Hispanic voter turnout was closer to 38 percent than our estimate of 30 percent? How might Donald Trump do if African-American turnout slumps to 50 percent? Our tool lets you answer those questions.

We’ve prepared a number of specific scenarios on the tool to show what might happen, for example, if large numbers of young voters decide to stay home on polling day, or if there’s a surge of white men at the polling booths.

But the real power of the tool lies in the ability of users to build their own scenarios. You can adjust turnout by sex, age cohort, race and ethnicity, income range, and party affiliation to see how those factors affect the overall result.

So how does it work?

When you first come to the tool, you will see what we project to be the most likely outcome if the election were held on the day of the most recent polling.

As of Aug. 19, for example, our projection showed Hillary Clinton’s chances of winning at more than 95 percent, and by a margin of 108 Electoral College votes.

But that’s based on our default model of a 60 percent overall turnout rate. There are ways to look at our polling data.

How might Trump win, for example?

There is no doubt, right now, that Trump has strong support among whites. He leads Clinton by 44 percent to 39 percent among all white voters. But he is supported by just 20 percent of minority voters. So to win, he would have to see his core supporters come out in large numbers while others stay home.

In the Trump Wins scenario visualized above, we adjusted turnout to be higher than expected for Republicans and older, white Democrats, and lower than expected for Democratic minorities and younger, white Democrats. In that scenario, Trump’s chance of winning is 75 percent, and by about 26 Electoral College votes.

Perhaps you want to tweak those parameters, or build your own scenario from scratch. Go ahead. You can change the turnout for any of the groups by clicking on them and adjusting the turnout rate to see what happens. Or you can remove groups or add new ones.

If you click on “add a group,” you can construct a demographic group by selecting from the boxes below. To choose white male Republicans aged 18-45, earning less than $75,000, for example, click on “White”, “Male”, “18-30”, “31-45”, “<$25K”, “$25-50K”, “$50-75K” and “Republican.” The site will tell you what percentage of the population that group comprises, and what our projected turnout rate for the group is. Drag the nearby turnout slider left or right to change the turnout rate.

You can add multiple groups to your scenario, but you can’t pick overlapping groups. So if you picked all men as your first group, you can’t also pick only Asian men as a second group, because Asian men are a subset of all men.

In general, changing turnout rates for large groups of people – all women, for example – will have a much bigger impact on the results than picking smaller groups.

You can share your scenario on Facebook or Twitter with the share buttons.

No one can know for sure, until election day, how accurate anyone’s 2016 projections will be. Our survey provides only a snapshot of current voter preferences; candidates do rise and fall in the polls from week to week. But most polling failures have come about less because the pollsters misunderstood voter preferences and more because the pollsters failed to accurately estimate turnout. In the end, the only voters who matter are those who cast ballots.

The British plebiscite on leaving the European Union was the most recent example. In the days before the June vote, most pollsters projected the “remain” vote ahead. But almost all of the polls underestimated turnout among older voters and overestimated it among younger voters. As a result, predictions were significantly off.

Similarly, polls of adult Americans, or even of the smaller population of registered voters, may well be out of step with voters who are truly motivated to cast a ballot. Adding to the complexity is the Electoral College. Voters don’t directly choose the president; the Electoral College does. Each state gets one Electoral College vote for every seat it holds in the U.S. Congress. The formula gives states with smaller populations disproportionate influence on the outcome.

It also means that a handful of voters in a single state can carry huge weight. Think back to the 2000 presidential election, which was ultimately decided by a few hundred votes in Florida. George W. Bush won many of the smaller states and Florida, which gave him the majority in the Electoral College, while Vice-President Al Gore won the national popular vote but lost the presidency.

States of the Nation is distinct from our other presidential tracking poll, which can be found at the Reuters Polling Explorer site. The two polls have different sampling sizes, reporting periods and update schedules, so their results will differ.

Typically, the Polling Explorer tracking poll is updated at least twice a week, on Tuesdays and Fridays. States of the Nation, a much bigger survey, is updated once a week, on Fridays. Polling Explorer is intended to reflect the changing pulse of the voter, putting out updates based on a five-day rolling average. States of the Nation employs a weekly average. The Polling Explorer poll sometimes detects short-lived shifts in voter preference that may not last long enough to show up in the larger States of the Nation survey a few days later.

Editor’s note: Maurice Tamman is an investigative reporter at Reuters. Ken Ellis is  chief technology officer of Reuters News Agency. Clifford Young is president of Ipsos Public Affairs (U.S.)