Are the polls hiding a Harris landslide?
Pollsters may have over-corrected their failure to fully capture Trump support in '16 and '20
For weeks, a small but growing number of US commentators have smelled a rat. The polls are wrong - or so they claim. Specifically, they charge that polling companies - so traumatised by their failure to fully capture the level of Donald Trump’s support in 2016 and 2020 - have overcorrected, and are now artificially inflating the former president’s numbers and underestimating Kamala Harris’. Over the weekend, they seized on fresh supporting evidence when the respected Iowa pollster Ann Selzer, who correctly predicted the scale of Trump’s victories in the Hawkeye state in the last two elections, showed Harris winning it by three points. Aha! - the doubters cried - if Harris is winning in Iowa, she’s headed for a landslide nationally. But is she?
There is certainly some evidence to support their claim. Iowa aside, the weekend also saw the publication of eye-catching polls in Ohio and Kansas. In the former, once considered a swing-state but no more, Trump’s lead was a mere three points. In the latter, which hasn’t been won by a Democrat since 1964, Trump’s lead was an eyebrow-raising five points. If those numbers are even close to being accurate, it seems highly implausible that Harris won’t win the all-important trio of Michigan, Pennsylvania and Wisconsin - and comfortably. Furthermore, as NBC News’ Chief Political Analyst Chuck Todd recently told me - if you study national polls beyond their headline figure, you often find strikingly odd demographic breakdowns - with traditionally Democrat-voting groups surprisingly leaning towards Trump, and vice-versa. Are the pollsters fiddling with their numbers to produce a headline figure that shows a near-tie?
If that’s true, what could be motivating pollsters to risk yet more potential embarrassment? There are two likely explanations. The first is that they’ve calculated the reputational jeopardy of under-estimating Harris’ support is less than the risk of under-estimating Trump’s. If their numbers have failed to accurately predict the Republican vote for a third consecutive general election, it will inevitably lead to accusations of either incompetence or bias. As a result, it is entirely feasible the polling companies are over-weighting traditionally Trump-leaning groups, such as rural voters, despite having a relatively limited sample size, risking exponential error in their topline figures. Over-correction, they may have figured, is easier to excuse than failure to correct in the first place.
The second explanation is that the pollsters are ‘herding’ - in other words, only publishing results that are broadly in line with their competitors’ for fear of the reputational damage of being wrong. The rise of polling aggregators, such as 538 and RealClearPolitics, has also increased the pressure for polling companies to produce results broadly in-line with the national average. Whilst such sites have their uses, especially when it comes to identifying trends, they have allowed political journalists to become lazy. In essence, it has become much easier to check your favourite polling aggregator each day to see if the headline numbers have moved than to spend several hours combing over the demographic breakdown tabs of various swing-state polls.
The key flaw with the argument that the polls are under-estimating Harris can be found in the identity of those advancing it. Almost without exception its most vocal proponents are those who most wish it were true. Confirmation bias is everywhere these days, on both the right and the left, and my hunch is that many Democrats are clinging to the hope that the polls have over-corrected because the truth is harder to swallow - that this election remains too-close-to-call and that their nemesis, Donald Trump, could be about to return to the White House.
There’s one other thing to note: maybe the polls weren’t so wrong in ‘16 and ‘20 after all. Whilst it’s certainly true that polling averages didn’t fully capture the scale of Trump’s eventual vote-share, it’s also true that most voters who decided at the last-minute broke for Trump. In other words, those final polls may not have been entirely wrong, it’s just they were simply snapshots of the race a few days out before a small but significant number of voters shifted towards Trump. If last-minute deciders break decisively for Trump again tomorrow, the doubters will have won a hollow victory. The polls will, indeed, have been wrong, but they’ll have been underestimating Trump once again.