Post-election update: I called every state except Florida and Ohio. In both states, the polls were simple wrong. In Florida, Biden was polling ahead of Trump, as per FiveThirtyEight, 50.9% to 48.4%, or +2.5, but Trump won by +3.4, which puts the polls out by 5.9. In Ohio, Trump had a narrow polling lead of +0.6, but ended up winning by +8.1. The increases in Biden’s numbers over the last couple weeks, which I based my predictions on, I think were real, but since the polls were out by a fairly large margin, above the margin of error, they weren’t enough to push Biden over the top. In Georgia, the polls were more actuate, with Biden polling +0.9 and winning, it seems today, by +0.2, and my method was able to predict the results.
The 2020 presidential election polls have been remarkably consistent over the past year. Since October 2019, Joe Biden has shown a consistent lead over Donald Trump in national polling averages.
The Real Clear Politics poll average shows Biden’s lead ranged from +4 to +6 early in the year, as he faced a tough primary campaign and Trump received a bump during the early days of the pandemic. Over the summer, Biden’s lead increased to as high as +10 on June 22 as the United States entered its second wave of the pandemic, before narrowing to the +6 and +7 range in the lead-up to the debates.
Following the first presidential debate on September 29, where a bombastic and combative Trump was widely criticized for his interruption-filled performance, and Trump’s coronavirus diagnoses late the next day, Biden’s polling numbers steadily increased to +10 by October 11, only to narrow back down to +7 following the president’s recovery.
The final presidential debate on October 22 was one of Trump’s few remaining chances to turn his polling numbers around. In a much more tame performance, Trump put forward his best possible case, focusing on the need to protect the economy during the pandemic and Biden’s 47-year record. But it wasn’t enough to swing the polls.
On the eve of the election, the Real Clear Politics average gives Biden a +6.5 lead over Trump. FiveThirtyEight’s poll average puts Biden’s lead at +8.5. The FiveThirtyEight election forecast gives Biden an 89% chance of victory, and Trump just a 10% chance.
Despite Biden’s consistent polling lead, however, many voters, especially Democrats, still traumatized by Trump’s unexpected victory in 2016, have trouble believing the polls. “They were wrong in 2016,” they say.
Trump’s 2016 victory was certainly a shock to many, with publication like HuffPost giving Hillary Clinton a 98% chance of winning, but the vast majority of 2016 election polls were actually within the margin of error. Polling firms have made adjustments to fix the few things that their models missed in 2016.
A look at the Real Clear Politics poll average in 2016 shows a much closer race over the 15 months before the election, with Clinton and Trump tying or close to a tie six times. National polls in 2016 also suggested that Clinton held a lead in the national popular vote, with the final polls putting Clinton at +3.2 on the eve of the election, and Clinton did indeed win the popular vote. It was only in the electoral college that Trump was able to find a winning margin.
I think that we can trust the polling numbers this time around, then, within reason.
The thing to keep in mind with election polling is that polls are nothing more than snapshots of the electorate at the time they are taken. A poll taken on Monday but released on Friday may already be wrong. A poll taken in June that has Biden up by +10 tells us nothing more than that Biden was up +10 in June. It’s only in the lead up to the election that we can really start making predictions.
Polls also have their own power to shape voting outcomes, particularly with the way they are reported in the media. A poll saying the Biden is leading by +10, for example, might make Trump supporters more likely to vote to prevent a Biden win, or Biden supporters less likely to vote because they don’t need to.
Because polls are snapshots of the electorate, and polls have the power to build on each other and turn small changes into consistent trends, we can use polling trends in the last week or two before the election to predict the results. If a poll on the Wednesday before the election shows a candidate leading by +2, then a poll on Friday leading by +3, then we can predict that trend could be leading by +4 by election day.
In the 2016 election, for example, the polling averages in the final two weeks saw Trump narrow the gap to Clinton to just +1.3, before going back out to +3.2 in the final few days before the election, telling me that the race was incredibly close, and the result was up in the air.
This year, I’m putting this method to the test on the entire electoral college map (using the YAPMS election map simulator).
Let’s start with the safe states, where Biden or Trump’s lead is well above the margin of error, which I’ve set at +/-4%.
That gives us 183 electoral college votes for Biden and 76 for Trump.
Then we have the likely states. These are states where Trump or Biden has a smaller lead, but still above the margin of error, and I don’t think the polling trend is bending enough to swing the vote.
That gets us to 258 electoral college votes for Biden and 124 for Trump.
Then we have two states that are right around the margin of error of +/-4%, Pennsylvania (20) and Nebraska D2 (1), which both favour Biden. With that, Biden crosses over the 270 electoral college threshold to 279, meaning that if the polls are within the margin of error, Biden already wins before even getting to the battleground states.
And finally, we have the swing states, where the polling data is within 2%. This is where my polling trends method comes into play. Looking at the way the polls are treading in each state, I have Biden winning Arizona (11), Florida (29), Georgia (16), Maine D2 (1), and Ohio (18) and Trump winning Iowa (6), North Carolina (15), and Texas (38).
That leaves us with a final electoral college total of Biden 354 and Trump 184.
The polling data, then, suggests a likely Biden landslide, but there are a few things to keep in mind this year.
As of today, over 95 million people have already voted early or by mail, up from 70 million a week before the election, meaning that late swings in the polls could mean less this year than they normally would, because so many people have already voted.
There is also an open question of the effects of voter suppression on the results. Trump has already said that he plans to declare victory either way tomorrow and challenge the counting of mail-in ballots in the courts, where Republican judges could rule in favour of Trump.
Even though the polling data is pretty clearly in favour of Biden, Trump still has a chance to win the election. The FiveThirtyEight election forecast runs 40,000 election scenarios, and shows Biden winning 89% of the time, but Trump winning 10%. In 100 elections, Biden wins 89 times, Trump wins 10 times.
A close result in any of the swing states could throw the election to the courts. But Biden’s lead is so big, and he has so many chances to get over 270 electoral votes, that the polls would have to be dramatically wrong for even the courts to swing enough states to give Trump a victory.
If Trump wins, in short, we should just stop polling, because it would be useless.
Thanks for reading, if you want to explore more content like this, please subscribe.