• Undecideds set to determine close election

• How to explain 'Shy Trumpers' theory

• Candidates very rarely underperform averages

• Follow our 2024 US Election live blog here

• Get the latest data including current prices and volume of bets in our daily update

• View Betfair Exchange politics markets as you have never seen them before

As they stood this morning, these are the polling averages in the swing states according to Fivethirtyeight.

Arizona: Trump 48.0, Harris 46.8
Georgia: Trump 48.3, Harris 47.4
Michigan: Harris 48.6, Trump 45.9
Nevada: Harris 47.8, Trump 46.8
North Carolina: Trump 47.8, Harris 47.4
Pennsylvania: Harris 48.3, Trump 46.9
Wisconsin: Harris 48.8, Trump 46.7

What about the undecideds?

Note that the combined shares amount to between 94-96%. Come election day, that will be around 97.5-98%. In 2020, it was 98.1%. Thus, while we can see who is ahead as a snapshot of today's opinion, even f that snapshot doesn't move, we can't effectively forecast the race without predicting where that extra group of undecideds, or third party voters will end up.

One man discussing this around the clock is data analyst Carl Allen (@realcarlallen), whose book "Why the polls weren't wrong" takes aim at the industry of polling models and forecasters such as Nate Silver, and the obsession with 'margin' and what he sees as a false relationship with 'probability'

47-45 might be better than 45-35.
Nice area of research for aspiring analysts.

No debate 49-46 is better than 46-40 though

But again..spread is the best anyone can do right now. They are definitionally incapable of answering

What win % does a 2.5 point "lead" have?

😁 https://t.co/QHQHTS46Np pic.twitter.com/MEXPJSpl3S

-- Carl Allen (@RealCarlAllen) September 27, 2024

Allen notes that, first, candidates underperforming their polling average is extremely rare. Thus, the closer they are to 50%, the stronger their chance. Second, that attempts to allocate the don't knows are fraught with risk.

He notes that in recent UK elections, such as for London Mayor and the General Election, pollsters were observing double-digits undecideds. In some constituencies, 'Don't Know' had more support than any candidate. But because the UK tradition is to exclude or allocate them, overly confident leads were reported. Allen says his forecasts were substantially less bullish on Labour for those reasons.

Unlike the UK, these US polling averages count, rather than allocate, don't knows. Hence how we end up with the aforementioned low combined totals.

Does this explain 2016 and 2020 'misses'?

Now, rethink the last two US elections, where the polls were widely deemed to have been 'wrong', because Trump finished closer to Clinton and Biden than the polling average margins. In fact, the polling averages for Biden and Clinton weren't far wrong at all.

If we instead assume that Trump merely dominated the undecideds and squeezed third parties (in 2016, third party support collapsed very late), the numbers match up.

This is also quite a logical scenario. A variant on 'Shy Trumpers'. As I found time and again talking to real American voters throughout 2016, many found Trump's character repellent, but baulked at the idea of voting for Clinton. Most often, they would say they were undecided or thinking about Gary Johnson (a right wing libertarian). With hindsight, it is no surprise they ultimately fell in line with the Republican, even if reluctantly.

Harris is faring better than Clinton

So will this happen again? It is certainly possible that Trump wins the undecideds but his problem is that Harris is already close to 50% in the states she needs. As Allen points out, it is extremely rare for candidates to underperform their polling average. Clinton and Biden didn't underperform their polling average, but the 'margin', which is artificial without the complete numbers.

Here's another batch of polling averages. This time from Decision Desk, to compare with the averages listed at the top.

Decision Desk HQ/@thehill Battleground Poll Averages:

Arizona - 32 polls
🟥 Donald Trump (R): 48.4% (0.9)
🟦 Kamala Harris (D): 47.5%
[R+0.5 since last month]

North Carolina - 32 polls
🟥 Donald Trump (R): 48.4% (+0.5)
🟦 Kamala Harris (D): 47.9%
[R+0.3 since last month]...

-- Decision Desk HQ (@DecisionDeskHQ) September 26, 2024

As you can see, Harris is much closer to Biden than Clinton. She is very close to the magical 50% target in the critical states of Pennsylvania and Wisconsin, and/or clearly ahead. Those two, along with Michigan, where she is trading shorter odds, would almost certainly deliver the electoral college. Note, the numbers also suggest Trump is favourite to win Georgia, North Carolina and Arizona.

Also factor in that pollsters have probably adjusted their sampling in Trump's favour. That was certainly the case in the 2022 midterms and led to Republicans badly underperforming expectations. Remember this when hearing the constant refrain that Trump is fundamentally understated in the polls.

Does this mean Harris is sure to win? Of course not. There are still five weeks until election day. Events could move the needle and, in any case, the numbers remain very close. But moving forward, follow these polling averages, remember the rule about getting close to 50% and that candidates very rarely underperform the average. And follow those three key 'Blue Wall' states. If Harris wins Pennsylvania, Wisconsin and Michigan, she has a chance of winning the presidency well in excess of 90%.

Now read the latest US predictions from Betfair predicts