Want these election updates emailed to you right when they’re published? Sign up here.
Hillary Clinton’s chances of winning the presidency have held fairly steady in the FiveThirtyEight models over the past 10 days. The polls-only forecast currently gives her an 88 percent chance of winning; since Aug. 7, her chances according to that model have been between 83 percent and 89 percent. The polls-plus forecast puts Clinton’s chance of winning at 78 percent; over the past 10 days, her chances according to that model have been between 76 percent and 80 percent. Indeed, Clinton’s post-convention bounce has stuck around so long that Donald Trump has been reduced to tweeting out a poll that showed him close but still losing.
A lot of Democrats I know have started to talk as though this election is over. They point to the fact that no candidate since 1952 who was leading at this point in the election cycle, a few weeks after the conventions, has lost the popular vote. So if Clinton’s lead in the polls is clear and the polling leader at this point in the campaign has never lost, why aren’t Clinton’s chances of winning according to our models even higher?
There are a couple of answers.
Our polls-plus forecast looks at economic indicators, in addition to polls. The state of the economy has historically influenced the election, with a better economy helping the party holding the White House. The indicators we use — jobs (nonfarm payrolls), manufacturing (industrial production), income (real personal income) and others — currently show a decent but not great economy. The polls-plus model sees Clinton’s healthy lead in the polls and an economy that historically would presage a close election and so expects the race to tighten.
Our polls-only model looks only at the polls. And although no candidate since World War II has come back to win the popular vote after trailing at this point in the campaign, there have been races in which the polls bounced around a lot between now and Election Day. Candidates have made up 6 to 8 percentage points — roughly the size of Clinton’s lead right now in the national polls — before. So there is precedent for a big enough share of the electorate to change its mind that Trump could come back. It certainly wouldn’t be easy for Trump — he’s the overwhelming underdog, but it’s not impossible for him to win.
Simply put, the polls aren’t perfect at this point in the cycle — there’s still a good deal of uncertainty inherent in trying to predict who will win the election and by how much based on the polls. We have more than two months until the election, and polls have coverage error, measurement error and non-response error. You can see this margin of error in the table below, which shows how election results in past years compared with the polling averages two to three weeks after both major parties held their convention.
|DEMOCRATIC CANDIDATE LEAD|
|CAMPAIGN||POST-CONVENTION POLLING AVG.||POPULAR VOTE||ABSOLUTE DIFFERENCE|
So every polling leader at this point in the election cycle went on to win the popular vote.1 But look at the “absolute difference” column. Although the race didn’t change much down the stretch in the past few elections, that isn’t always the case. In 1996 and 2000, for example, the difference between the candidates’ level of support in the polls at this point in the race and the final vote margin was 5 percentage points or more. Al Gore’s lead on George W. Bush in 2000 narrowed enough in the last couple of months for it to be one of the closest elections ever. And 1996 and 2000 are not aberrations.
In 1980, Jimmy Carter, after a successful convention, was tied with Ronald Reagan according to the national polling average.2 But Reagan ended up winning by nearly 10 percentage points. Reagan’s gains were greater than the deficit Trump faces now.
Just four years before Reagan’s victory, Gerald Ford nearly pulled off the greatest comeback in the modern polling era. Ford was down by double digits in 1976. But aided by a decent job approval rating and perhaps by Carter’s missteps, Ford narrowed Carter’s lead and even inched ahead in the final Gallup poll of the campaign. Ford didn’t win another term, but he proved that a race can change even after both conventions are in the rearview mirror.
Perhaps the most interesting potential precedent for the 2016 campaign is 1968.3 Richard Nixon was up by 8 percentage points and opened an even larger lead in the fall. But Hubert Humphrey was able to consolidate a previously divided Democratic base (as Trump needs to do with Republicans) and cut into Nixon’s lead. Humphrey was also aided by President Lyndon Johnson’s rising approval ratings and the original October surprise (an announcement by Johnson that the U.S. was halting the bombing in Vietnam). Humphrey ended up losing in the second-closest presidential election of the 1900s. It’s not too difficult to imagine Republicans rallying behind Trump — perhaps Clinton is hit with a scandal or WikiLeaks drops an October surprise.
Indeed, half the presidential elections included in the table saw the difference between the candidates’ support change by more than 5 percentage points. Three races experienced a shift that was greater than Clinton’s current lead in the national polls. There are a lot of undecided voters this year, so we could still see a large shift.
Of course, that could benefit Clinton if she picked up most of those voters. Also, some supporters of Libertarian Gary Johnson and Green Party candidate Jill Stein may eventually back Clinton. And her superior ground game could increase her support. We don’t know which way the polls may change. This election season has been so odd that I’m not willing to dismiss any scenario.
So how safe is Clinton’s lead at this point? Our polls-only and pulls-plus models try to quantify just that. But as a sanity check, we can build a simple model based solely on these polling averages and the final result. That is, a simple linear regression model based only on how well the national polls from this point in the campaign have predicted the final margin.
The model’s result looks a lot like our forecasts. A candidate ahead in the national polls by 7.5 percentage points (as Clinton is in the polls-only forecast) wins the popular vote 93 percent of the time, according to my simple model. Our more rigorous FiveThirtyEight polls-only model gives Clinton a 90 percent chance of winning the popular vote.4
What all of this tells us is that Clinton is probably going to win the presidency but that no Democrat should take it for granted.