The polls are close, but that doesn’t mean the results will be

A normal polling error in either direction could give either candidate a decisive victory.

It’s been quite the weekend for high-profile poll releases. Last night, Ann Selzer’s new poll showed Kamala Harris leading by 3 points in Iowa. And this morning we got polls of all seven major battleground states from NYT/Siena. 

But our forecast has been hovering right around 50/50 since mid-September. Donald Trump gained ground in mid-October, and Harris has regained just a little bit now, but it’s always remained comfortably within toss-up range. So if you believe the polls, we’re coming up on the end of the closest presidential race in 50 years. Harris leads by about 1 point in our national average — though our estimate of the national popular vote, which is mostly not based on national polls, shows a slightly wider margin than that¹ — and the battleground states are even closer. Donald Trump has a 0.3-point lead in Pennsylvania, while Harris has small leads in Michigan (D +1.1) and Wisconsin (D +0.9). 

However, that doesn’t mean the actual outcome will be all that close. If the polls are totally accurate we’re in for a nail-biter on Tuesday night. But a systematic polling error is always possible, perhaps especially if you think pollsters are herding — only publishing results that match the consensus. And because things are so close, even an average polling error would upend the state of the race.

Now it’s important to note that polling error runs in both directions, and it’s pretty much impossible to predict which way it will go ahead of time. Harris could beat her polls or we could be in for a third Trump miss. But both scenarios have one thing in common: they’d turn election night into a relative blowout. 

Our forecast of the popular vote shows Harris winning by 2.1 points — but it has a standard deviation of 3.1 points. That number is mostly based on how accurate — or inaccurate — national polls have been in elections since 1936. The actual result should fall within one standard deviation of our forecast about two-thirds of the time — which means anywhere from about Harris +5 to Trump +1. Anything within that range should be considered a “normal” polling error — still in the thick part of the probability distribution rather than being out in the tails.

But no one really cares about the popular vote. What about the battleground states?