Are partisan pollsters calling the election wrong?

I have updated my model to exclude polls connected with a party.

The eye-catching Iowa poll – which gave Kamala Harris a lead over Donald Trump in the traditionally Republican state – is an outlier. Harris winning Iowa is unlikely (though not impossible). Most importantly, the poll points to a heartening scenario for the Harris campaign: Democrat gains among white voters in Iowa could mean easier than expected wins in the Midwestern Rust Belt states, even if Iowa itself remains a long shot.

The poll was conducted by Selzer, a firm with a reputation for astonishing accuracy. But in its wake, three other pollsters – all with links to either the Republicans or the Trump campaign – published figures very much at odds with Selzer’s findings.

There are several partisan pollsters in the mix right now. Campaigns commission polls and publish their headline numbers. And then aggregators such as myself, FiveThirtyEight and RealClearPolitics incorporate them into our models. But affiliations to a certain campaign implies an occasional willingness to direct the numbers in a certain direction (through carefully worded questions, for example). My model and FiveThirtyEight account for this.

There are still plenty of so-called partisan pollsters who call elections very well, however. So I still use them in my model, albeit with considerable “weightings”.

In 2022 I decided to exclude these types of polls altogether. It seemed poorer quality polls were flooding some races and rendering a lot of coverage irrelevant to the actual reality on the ground. Instead they were playing up the potential of each campaign to squeeze out more votes. By excluding them all together I ended up with a better model than most (it’s something I trumpet at every social occasion with breathless zeal).

Now, what if we were to do that here, excluding all polls – Dem or Rep affiliated – from my New Statesman model?

The New Statesman model says Harris has a 51 per cent chance of winning the White House. That accounts for all the polls out there. Exclude partisan pollsters and Harris moves to a 52 per cent chance. The median electoral college win for Harris goes up from 273 to 275.

It's in the individual states that these figures are telling (nationally these figures are marginal). In Wisconsin, Harris’s chances go up from 55 per cent to 59 per cent. In North Carolina they’re up from 40 per cent to 43 per cent. In Iowa they’re up from a measly 9 per cent to (oh, an also measly) 12 per cent. But in enough battleground states, the differences are not nothing.

The race, forecast by all and sundry as a likely toss-up, will remain a toss-up. Throwing out partisan pollsters doesn’t change that. But it does make it marginal. And if you were a betting man, forced into a decision, you would err in favour of Kamala Harris here.