US Election: Were the polls really that wrong?

Barely had the results started tumbling out on Election Day when fresh obituaries began to be written about the polling industry. How could polls be so wrong again, was the refrain.

But…I have a contrarian take. I do not think they are quite as wrong as being made out to be.

First off, I find poll aggregators like RealClearPolitics and FiveThirtyEight very useful, contrary to the ‘advice’ of ‘traditional’ pollsters. Of course professional pollsters would rather you look at only THEIR poll and no other because they want you to believe theirs is the best and most reliable poll around. But for a user like me, an aggregator averages and levels the extremes between the more Republican-skewing polls like Trafalgar and Rassmussen and the more Dem-skewing ones like Emerson/NYTimes.

I particularly like RCP because it includes Trafalgar that FiveThirtyEight had a lot to say about.

So let’s look at the RCP averages.

For Michigan, the final RCP average was Biden by 4.2. The result was Biden 2.1. That’s an error of 2.1. Not unreasonable.

For Georgia, the final RCP average was Trump by 1.0. Biden leads there by 0.1. A very small error then.

For Arizona, the average was Biden by 0.9. Biden leads by 1.0. If you complain about a 0.1 error, you’d remind me about the “Any excuse for a tyrant” saying.

Lastly, Pennsylvania. The average was Biden 1.2. Biden leads by 0.5.

I want to talk about Ohio and Florida, the two bellwethers that went Red bigly. The final RCP average for Florida was Biden 0.9 and for Ohio it was 1.0.

The polls did get both states wrong by a bigger margin than the ones above. BUT I want to know how pundits thought Biden was winning either state with those numbers.

Here’s an example of somebody who did. Krystal Ball.

She predicted Biden winning Ohio and Florida in addition to Arizona. I know these are just gut feels but I want to know what such gut feels are based on when the polling indicates that they are too close to call and that Trump could well win them. I mean, if you expect Biden to beat Trump more handily than Obama against Romney, you had better have a very good reason.

Here’s Saagar’s own prediction which is interesting:

305 votes. As they say, dude just won the internet! The Rust Belt trio plus Georgia and Arizona. Exactly the states Biden is set to flip.

That is, IF you interpret the polls well, you could in fact get a good feel for what would pan out in the election.

Let me give my own example. A week before the election, a friend and I made our own maps for fun. We both predicted Biden winning Wisconsin, Michigan and Arizona but not Pennsylvania. And given how close Pennsylvania is, that seems to have been a fair call. We also thought Arizona was going to be close and hence had doubts about whether Biden would win. Here’s the kicker: neither of us live in the US! It was just a composite of the data, reports from various outlets and our own, uh, gut feel.

I don’t know exactly what he did but it was probably similar to my thumbrules which were:

1. If you are betting on a state to flip, it should have Biden leading by at or close to the margin for error – 5%. The reason for that is even after a normal polling error, Biden would probably still win the state. Wisconsin and Michigan were offering leads of that range, Pennsylvania wasn’t. Ohio and Florida was simply out of the question.
2. You could slightly punt on a Southwest state like Arizona because ‘migration’ from California and general growth was turning it a slight shade of blue. There were indications of this too in the Democrats’ performance there in 2018. No such indications in Florida.
3. No state that had stayed with Clinton in 2016 was going to flip for Trump. Hence we ignored the close polling in Nevada. Again, Nevada is close to California so…

I submit then that polling can be useful IF you remember that they are estimates looking into the future and have an inbuilt potential to be…wait for it….WRONG. What you cannot do is expect mathematical precision from a poll. It cannot even be as accurate as a weather forecast. The weather operates on the forces of nature and cannot change its mind. Humans can. Humans can also lie or troll pollsters.

Remember all of that and you may still get a good read of where things are at from the polls. No need to throw the baby out with the bathwater.