National political polls showed Hillary Clinton was going to win in 2016. And then she didn’t… And polls leading into 2020 mostly had Joe Biden winning. And then he did….So as a pollster, I often get asked: “why were the polls so bad in 2016, and what happened to correct them for 2020?”
The problem with this question is that the general election #polls of 2016 were in fact more accurate than the general election polls of 2020. What was misleading is the interpretation of the polls...
Let’s start with the polling aggregate leading into the 2020 election put together by RealClearPolitics:
In looking at the highlighted aggregate ‘spread’ we see that the polling average had Biden ‘winning’ by 7.2 points. Looking at actual votes nationwide, we see Biden only won the popular vote by about 4.5 points, meaning the polls in 2020 were off by roughly 2.7 points.
There are a couple things to keep in mind here before moving on:
First, this is not an aggregate of ‘voters’; this is an aggregate of ‘likely voters’. This is pollsters’ attempt to be predictive, knowing full well that not everyone who answered these questions is going to vote, or vote with who they said they would come election day.
Second, and most important, this is an aggregate of what would be the popular vote, not the electoral college. Every poll in this aggregate is ‘nationwide’, meaning that pollsters are looking at what would be the popular vote. Problem is, we don’t vote for a president by popular vote; we vote by state in the electoral college. So this, to me, is like looking at statistics about two baseball teams to see which would be the likely winner in a game of baseball, and then having them play a game of hockey...it's just a different set of rules.
With all of that said, we can easily see how pollsters were in fact more accurate in 2016 than they were in 2020.
Looking at the 2016 spread, we see that the polling average had Clinton winning (the popular vote) by 3.2 points. In 2016, Clinton actually won the popular vote by about 2.1 points, meaning that in 2016 the polls were off by about 1.1 point. OR, about a point more accurate than the 2020 polls.
What I’d like to say about both of these instances, is that the polls were only off by a point or two. That’s not too shabby.
So before asking about a poll’s accuracy, we need to take a step back and ensure we understand what the poll is actually looking at and how it relates to the point being made. For example, the 2016 narrative was that Clinton would win the presidency (which is won through the Electoral College), but the polls were actually looking at the popular vote (which is not how the presidency is won...yet).
In a post coming soon, I’ll share more about key poll characteristics to look at to help us better interpret them.