Is Donald Trump going to lose Iowa? That’s the conclusion many US pundits came to after a bombshell poll over the weekend. That poll, conducted by the psephologist Ann Selzer, put Kamala Harris three points ahead of Trump in Iowa, despite Trump having comfortably won the state by almost ten points in the past two presidential elections.
So Iowa could tonight return to swing state status. In past elections voters in the state have backed Reagan, Clinton, Obama, and Trump: now they might turn to Harris. However, at the same time as the Selzer poll was published, a contradictory but less-covered poll indicated another strong Trump victory. This poll from Emerson College concluded that Trump is still ten points ahead of Harris. Once again, US pollsters have been left scratching their heads. How could two polls just two days apart with the same sample size in the same state be so different?
One problem for pollsters is the increase in early voting in recent US elections. This year, at least 76 million voters cast their ballots before polling day. The pandemic changed early voting, making it harder for pollsters to tell whether shifts in polling reflect changes in voters’ preferences or if they had returned to pre-election patterns. Complicating matters, Axios analysis says more voters have misled pollsters on purpose and, as the NYT’s Nate Cohn warns, numerous ‘late deciders’ will have a similar effect: telling pollsters their voting intention before changing their minds on the day.
Every election, the make-up of these early voters changes. In 2020, a third of Republicans voted early compared with almost half of Democrats. This year, 36 per cent Republicans and 39 per cent of Democrats voted early. In this election, the difference may be down to women who make up 55 per cent of early voters. Selzer says women voters were the main factor in her poll since as a whole they prefer Harris to Trump. But there’s still some issue with the polling as a whole. Selzer says: ‘As neither of the major candidates gets to 50 per cent, there’s still a little squishiness in what could actually happen.’
The effect of ‘squishiness’ is less of a problem for California. Despite early voting starting 29 days before election day, polling consistently places Harris 25 points ahead of Trump (and the state has consistently voted Democrat since 1992). But using polling to judge the outcome in swing states is tougher, and in possible swing states tougher still. Some 37 per cent of Iowans have already cast their ballots and the polling cannot determine whether the result is on a knife edge or not.
Such differences in polling compared with election outcomes can be staggering. In the 2020 presidential election, a record 101 million Americans – some 63 per cent of the electorate – voted early during Covid. For pollsters, that election was the most inaccurate in 40 years. Pew Research found that 93 per cent of polls overstated Biden’s support, leading to a much closer race than expected.
Pollsters say their surveys show that the current election is the closest since 2004. But even still, the polls still miss the true result by a small margin. Nate Silver’s former polling site FiveThirtyEight says:
Even if all polls were conducted on election day itself (no temporal error) and took an infinite sample size (no sampling error), the average poll would still miss the final margin in the race by about two points.
It’s a pretty damning statement from one of America’s top polling sites, but it points to the reality of how the number of possible election outcomes increases as the polling narrows but the margin for error remains the same or widens. This finding is also shared by the American Association for Public Opinion Research which says that polls conducted within the last two weeks of a US campaign are wrong in either direction by 5.1 pts on the state level.
The margin for error for the seven swing states (Arizona, Georgia, Michigan, Nevada, North Carolina, Pennsylvania, Wisconsin and perhaps including Iowa as number eight this time around) is widening. In these states, some 93 electoral college votes (99 if you include Iowa) are up for grabs. That’s around a fifth of the total number of electoral college votes.
The seven swing states have polling aggregates of two points or less yet maintain wide error margins. This means that if the polls are off by two or three points, as previous elections suggest that most are, then landslides are still possible for both candidates.
The message from pollsters is that the closeness of this election cannot be overstated. Nate Silver ran 80,000 simulations of his election model of which Harris came first in 40,012 of them. The outcome was closer than flipping a coin: ‘Empirically, heads wins 50.5 per cent of the time, more than Harris’s 50.015 per cent.'
Listen to election coverage from Freddy Gray and Kate Andrews on the Americano podcast:
Comments