A political poll is only as good as our 2+2 stuff is
One of my favorite all-time brain teasers is this: How long is a million seconds, and how long is a billion seconds?
I love it because it demonstrates how profoundly ill-equipped most of our brains are to handle numbers. Consider: Picture an apple in your head. Easy, right?
Now picture 25 apples. Gets a little harder, but there it is, right? I’m sure you’ve got a 5x5 grid of apples in your brain. OK. Now picture 100 apples. Getting very difficult, bordering on impossible.
OK. Now picture 3,276 apples. The end.
Anyway, the million seconds vs. billion seconds thing is fascinating because it shows the gap in what a million and a billion is. To our ears, it’s nothing - this guy’s a millionaire, this guy’s a billionaire, whatever. But the answer? I’ll give you until the next paragraph …
A million seconds is a little more than
11-and-a-half days.
A billion seconds is a little more than
11,500 days (or nearly 32 years).
I bring this up A) to dazzle you with my math skills and B) as an introduction to the problem with presidential polling - namely, that we are really bad in separating poll numbers from potential realities.
“Most people, when they hear 80%, they think ‘ 99.9%,’ and that’s the problem,” said Patrick Murray, the director of the Monmouth University Polling Institute. “The problem isn’t the math we do, but the inability of the media to portray the level of uncertainty involved and the high probability of things happening that you wouldn’t expect, but happen with regularity.”
See: Clinton v. Trump, 2016.
I was talking to Murray about the 2020 election - and full disclosure, we discussed this about 12 hours before Trump’s coronavirus diagnosis - and the issues pollsters face when their numbers concerning future events are bandied about.
“The problem is polling always has a margin of error,” Murray said. “And it’s designed to be a snapshot of a population that we know who they are today. But election polling violates that, because election polling is about what’s going to happen in a population that won’t exist until elec
tion day.”
In short: Just because someone says they’re voting for Biden today doesn’t mean they’re going to vote for Biden in a month. Heck, it doesn’t even mean they’re going to vote, period.
But polls like Monmouth’s - routinely ranked as one of the best outfits in the nation - are then used by others (notably, Nate Silver at FiveThirtyEight) and put into magic math machines and that’s where you get things like “Joe Biden has a 80% chance of winning” (currently the odds at FiveThirtyEight) and Murray is … well, less than thrilled about it.
“It’s mainly because of things like that that make it difficult to convey polling will always have a layer of uncertainty,” Murray said. “It was never designed to be a precise measurement. Never designed to be able to predict things that will happen in the future. But these forecast models out there give you these very precise estimates, that someone has an 81.6% chance of winning the election, and there’s no such thing. The problem with the models and their proba
bilities is they’re limited to whatever they think is important in the election. It’s not like a deck of cards, where you know there is a finite number of hands you can deal out of any deck. Each hand has its own probability and you can calculate that. In an election there’s so many other things that can happen. There are infinite probabilities. So some of these modelers might be coming up with a million possible outcomes and telling you which outcomes have which chance of happening but there are more than a million possible outcomes. There are too many unknown variables.”
Like … well, like one of the candidates contracting a life-threatening disease.
In the end, know this: An election poll is just that - a poll. It’s a view of how people are thinking at that moment in time, not what they will actually do in the future.
So yes, Biden has - by Monmouth’s telling - a 5% to 7% lead in the national polls (that’s as of last Thursday.) And while it’s instructive, it’s not Nostradamus.
But still … come on, Murray, what it all means!
tell me
“If on election day Joe Biden is ahead by 5 or 6 points nationally, it would be an extremely low probability that Donald Trump should win the national popular vote,” Murray told me.
But remember: Margin of error. In fact, tattoo that on your eyeballs.
“But that means Biden could win the national vote by 2 or 3 percent, within the margin for error, and then it would be less surprising if Trump won the national vote, but it wouldn’t be surprising at all if he won the electoral college,” he said. “So this 5 to 7 point lead means Biden certainly is in the lead, but he’s just at the cusp of the electoral college being able to go in the opposite direction.”
In short: We’ve still got a few million seconds before we know anything for sure. Buckle up.