Data, polls, aggregators, and confirmation bias

Who is “winning” the U.S. presidential race?

No one is winning the race, because it depends on what people do on Election Day. There is no “ahead” or “behind.” It’s not like a basketball game where the team with a 20-point lead is far more likely to win, and the team with a 20-point deficit must potentially change its strategy to get back into striking distance.

But we are so desperate for a “winning” and “losing” narrative that we obsessively consume poll data about who won the debate or who is winning Pennsylvania.

About those polls

We all have a favorite in this race. So just as when watching a sporting event, we want our favorite to win, and we need to know how they’re doing right now, at this moment. Of course, individual polls have substantial margins of error of a few percentage points, so they tend to bounce around a bit — which is where the poll aggregators come it. Poll aggregators put all the poll results into a model, apply complex weighting protocols based on how dependable the pollsters have been in the past, and then generate set of numbers indicating who’s really ahead right now, by how much, and how likely it is that they will win.

The polls obviously get different results based not only on survey methodology but random chance. If the people clicking on an web ad or answering the phone today for a pollster are a little more Trump- or Harris-friendly than the ones yesterday or the ones that answered a different poll, you’ll get a different answer. Even the same methodology will generate different results just based on random chance.

But putting that aside for a moment, the poll aggregators get very different conclusions as well. As I write this on September 19, the poll aggregators look like this in the national popular vote and in the critical Keystone State of Pennsylvania (the spreads don’t all equal the difference in the poll averages due to rounding)

Poll aggre-gatorHarris (national)Trump (national)Spread (national)Harris (PA)Trump (PA)Spread (PA)
New York Times49%47%Harris
+3
49%48%Harris
+2
FiveThir-tyEight48.3%45.5%Harris +2.848.2%46.8%Harris +1.4
RealClear-Politics49.3%47.4%Harris +1.948.5%47.5%Harris +1.0
Nate Silver48.9%46.2%Harris +2.748.7%47.5%Harris +1.2

There’s a lot of discussion about these numbers right now. In particular, I’ve seen criticism of Nate Silver as being biased towards Trump because people claim his work is funded by Trump backer Peter Thiel. Silver denies this.

What is the truth?

The truth is, none of this matters.

No one is winning right now. One person will win on Election Day. Before that, none of this matters.

Polls measure what people who answer polls tell pollsters, not how people vote. That is not a balanced sample of the electorate, since it misses people who don’t answer pollsters — and it’s very hard to tell what kind of bias that introduces.

Measures to “unskew” polls based on perceptions of their bias are inevitably based on the unskewing entity’s politics. No one unskews a poll that favors them.

Poll aggregators are making their own assumptions about which polls are worth more and which are worth less. And they disagree. Which is why Harris’s “lead” in national polls — of which there hundreds — might be 1.9% or 3% based on who you ask.

The whole rickety tower of aggregators aggregating polls of people who might or might not represent the electorate invites people to add their own interpretation. And that interpretation always looks the same.

If you favor Trump over Harris, you say that the polls and aggregators are missing his strength and her weakness.

And if you favor Harris over Trump, you say that the polls and aggregators are missing her strength and his weakness.

People used to criticize the bullpen of the Red Sox by saying “No lead is safe.” With polls and aggregators, it’s not a question of whether a lead is safe. There is no lead.

It’s a data-soaked exercise in confirmation bias that has everything to do with what makes us feel good and very little to do with what matters in this election.

Believe nothing about the race. If this election makes you feel insecure, it should, because nobody knows what’s going to happen. That’s hard to live with.

Do something.

Research the candidates’ actual positions, not made up crap about fake sales taxes and post-birth abortions.

Talk to the people who matter. I have relatives in Pennsylvania, you can bet I’m talking to them.

Volunteer or contribute to the candidates whose positions you favor.

Become a poll worker to assure the integrity of the election.

And learn to live with uncertainty. Because fighting over polls and poll aggregators isn’t going to get anybody elected.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.