We Are One Month From The Election. How Accurate Are National Polls?

Election day is now just 30 days away. Democratic nominee Joe Biden currently has an 8 point lead in national polls over Republican nominee Donald Trump, according to the FiveThirtyEight polling average. But isn’t one month still plenty of time for a comeback? How safe is Mr. Biden’s lead?

To answer these questions, I looked at national polls from all presidential elections since 1980 (a total of 10 elections). These national polls were helpfully compiled by G. Eilliot Morris (who is currently employed by The Economist) in an R package called ‘politicaldata’.

I wanted to compare the accuracy of polls 1 month out to those taken right before the election. To do this, I found the average absolute error of all national polls taken about 1 month before the election (25-35 days before to be precise) and compared this error to the same calculation for polls taken within a week of the election (1-7 days). The table below shows these calculations:

We can quickly see that the polls were better some years than others. In 1992 and 1996 polls were not very accurate with 1 month to go; 2000, 2004, and 2008 conversely were within 1 point of the final result. We also see that polls taken within a week of the election were more accurate in some years, such as 2012 and 2016. For other cycles, like 1980, the picture a month before the election was actually less blurry.

The most important row in the table is the last row which shows the average absolute error over all 10 of the elections. We can see that the average absolute error 1 month before the election is 3 points, which is actually not too far off the 2.5 point average absolute error for polls one week before the election. So even though 1 month may feel like plenty of time for the race to change, polling history suggests that our picture of the race now is close to as good as it will get.

This is not to say that the polls on election day are still going to show an 8 point Biden lead! And that average absolute error is just that, an average. If we instead have a large polling miss like in 1992 or 1996, (and that miss underestimates Mr. Tump) the final popular vote margin could be Biden +2.4. Trump would then probably be a narrow favorite in the Electoral College due to his relative strength in the battleground states. That could happen of course, but is not likely. It is important to remember too that a polling miss could also benefit Mr. Biden.

What I am saying is that we should pay attention to the horse race polls now and not simply through our hands up and say it’s still too far away from the election to look at polls. We will have a better picture of the race on election day than we do now, but not a much better picture.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: