US Presidential election - Party like it's 2016
What would happen if we just repeat the polling errors of history?
One of the difficulties in learning a little about statistics and polling is that you start to learn about the important concepts that people throw around. And you find that they are very different to what you hoped they were.
“Margins of error” in polls, for example. At one point, I thought they were some expression of confidence, or how close each poll was likely to be to the true result - perhaps derived from some analysis of how accurate they had been in previous years.
Oh, what a naive, hopeless fool I was. The “margin of error”, “MOE”, or “confidence interval”, reported faithfully alongside a poll’s top-line results, has absolutely nothing to do with how accurate the pollster has been in the past, nor does it reflect how accurate they expect to be this time around. What it does give you is a single measure of a single source of error; one amongst many, and almost irrelevant for most purposes.1
But even though this reported margin of error isn’t doing the job we thought it was, the track record of polling - and what kind of uncertainty this introduces - is still of huge interest. We can use them to see how our current picture would turn out if the polls have been “just as right, just as wrong” as they have been in previous years. After all, we have no reason to believe that the size of this year’s errors will be much different.
Fine, I’ll do it myself
At a national popular vote level, the calculation is simple. Borrowing a chart from a FiveThirtyEight article, we can see that the average of the polls varied quite a lot over the Presidential and other kinds of election, with a bit more error in favour of Democrats in recent cycles.2
Nationally, Harris is currently ahead by 2.8ppts (FiveThirtyEight), 3.5ppts (Silver), 2.2ppts (RealClearPolitics). Let’s call it three percentage points.
Referring to the table, we can “run back” the polling biases, and apply them to our current situation. For every election in the last quarter-century (Presidential plus mid-term polling records) Harris’s win-loss record at the popular vote would have looked like the below. Blue is a Harris popular vote win, red is a Trump popular vote win.
So, if the polls stay as they are right now, but polling errors occur as they have historically, then we still have two cycles where Donald Trump would win outright.
Rather notably, these are the only two years where Trump was in fact involved as a Presidential candidate. His presence in a contest does seem to throw off pollsters.
And - note again - this is the popular vote.
Newsflash - the Electoral College continues to exist
Of course the popular vote is not how the US decides its elections.3 Instead, it uses votes via state delegates to the Electoral College. So we need to look at the state polling, the errors that occur there, and and how they might have the potential to skew the current picture.
Here’s the current polling averages for some of the most important states - as expressed by one of the main polling aggregators: FiveThirtyEight.
So, if we now assume that the polling at this point is absolutely correct - that is, every vote actually ends up precisely at the polling average, then Harris wins a) whole bunch of states not on the chart (219 Electoral Votes) + NH + VA + MN + MI + WI + NC and - just + PA) which results in narrow Harris win. She totals 286 Electoral Votes4 to Trump’s 252.
But we can already see how fragile this is. If the national popular vote could swing away from the polling averages by 4ppts or so in some years, how much more could the more volatile and unpredictable states go? Fortunately, we can work this out too.