How your eyes can deceive you in a political campaign
When I was working in City Hall in a large-ish midwestern city, the Mayor used to say to me “that is never going to be heard on the sidewalk.”
He was referring to a political truism that has a number of sayings attached to it.
“Signs don’t vote.”
“The chattering classes…”
The point of all these sayings is that your eyes have a way of deceiving you. Economists Daniel Kahneman and Richard Thaler described these inclinations as “cognitive biases.” Contrary to how classical economics described people as “rational,” prospect theory of behavioral economics describes people as “irrational in certain predictable ways.”
In fact, the original cognitive bias identified by this theory is the representativeness heuristic bias. In effect, people take their limited experience and generalize it to the population as a whole, even though the people they interact with are not a representative sample.
Consider the lengths pollsters go through to ensure that their results are representative of the population as a whole. The start of a poll is simply calling people who are likely to vote and asking them their opinion.
Next is where the art comes in. The pollster needs to develop a model of who he or she thinks is going to vote. If you get that model wrong, your results will be wrong. That’s what happened in 2016, when many pollsters did not balance their results based upon the voter’s level of education. Since Clinton received much higher support among college-educated voters than Trump, if the polling results had a higher percentage of college-educated respondents than the percentage that actually turned out to vote on election day, it would seem, as it did, that Clinton had stronger support than she did.
Similarly, in 2012 Romney was convinced on election night that he had won. The problem was that his model predicted that a similar percentage of the Black population would vote as in a normal election. However, with Obama on the top of the Democratic ticket, a higher percentage of the Black population turned out to vote than ever had before. The result is that Romney’s polls underestimated the number of Black votes that would be in the final tally, and so also underestimated Obama’s support.
The differing models pollsters use is where you can get variation among the predictions from the various pollsters. But if pollsters trained in politics and statistics have a hard time determining how a carefully constructed sample of respondents can be generalized as representative of the entire population, imagine how hard it is for just a regular person to do.
When I teach economics, I describe the following example to my students to illustrate this phenomenon. My daughter smokes, as do most of the people in her social circle. As a result, she is convinced that most people smoke. In fact, only about 14 percent of people smoke, according to the CDC, or about one in seven, a very small percentage. But because everyone Kelsey knows smokes, she thinks her group of friends is representative of the entire population.
Political operatives need to train themselves to ignore their senses, because they know their impressions will lead them astray. Instead, they have to use data and other indicators to objectively analyze the situation. Even so, given the fact that we want our candidate to win, it is devilishly difficult to silence our inner voices. Consider how my own lack of enthusiasm for Biden led me to predict that Michael Bloomberg would win the nomination once my preferred candidate Elizabeth Warren dropped out.
Full disclosure: I am now an enthusiastic Biden supporter who has already voted for him. But in the primaries, I was a Warren fan, and those feelings obviously impacted my analysis.
People who have not worked in politics, however, are unaware of how their senses can betray them. Consider the concern expressed by some that the polls are underestimated Trump’s support. After all, he is holding huge rallies.
People will attend rallies for a lot of reasons. They may attend rallies out of curiosity, even if they don’t intend to vote for Trump. They may be Trump groupies, following his rallies from state to state (such people do exist). People like that essentially get double-counted since they are part of the crowd at multiple rallies.
And then there are the genuine local Trump supporters. But consider the following. Trump recently held an impressive rally near the border of Nevada and Arizona with as many as 10,000 in attendance. Very impressive, right? But in Nevada and Arizona, both swing states that are leaning Democratic according to the polls, more than two and a third MILLION people have already voted early, more than a week before the election. If everybody at Trump’s rally voted for him, that would represent just 0.0043 percent of the total vote so far, not even counting the people who have yet to vote. In other words, these rallies look big, and they are, but they represent only a tiny slice of the actual electorate.
Consider another example. Yesterday, I read perhaps the stupidest oped I have ever read in The New York Times. That’s really saying something, by the way, but in general, Times editorials tend to be well-written, well-researched, and compelling. In fact, I have submitted to them repeatedly, only to be turned down repeatedly.
But this editorial was idiotic. Written by Shawn McCreesh and titled “Can we trust Pennsylvania’s polls,” the oped argued that since the writer lived near a Trump store that many people went to, and that many of his family’s friends are Trump supporters, then the polls must be wrong about Biden’s Pennsylvania support.
I can relate, by the way. I live in Macomb County, Michigan, one of the three counties credited with putting Trump over the top in 2016. He still has a lot of support here, with many people proudly displaying Trump signs. That does not mean that the state overall will vote for Trump, however. Macomb County is big, but if the rest of the state votes for Biden, the Macomb County vote won’t matter much.
The point is that the idle impressions of Shawn McCreesh and myself are virtually useless in assessing how millions of people are voting across our states. The fact that the Times published an oped with such flimsy evidence, filled with baseless hand-wringing, is another black mark against its editorial board, who have made some bad calls of late.
By the way, if you detect a note of bitterness, you’re probably right.
Anyway, does this mean that the polls are right and Biden is destined for the White House? I can’t give you a better answer than anyone else. In fact, even the people who spend all their time studying the polls such as Nate Silver of FiveThirtyEight, G. Elliott Morris of The Economist, and Nate Cohn at the Times all admit that they could be wrong. The only thing I’m saying is that the polls have a much higher likelihood of being right than the unscientific impressions you and I get from our friends and neighbors. We should let go of our biases and just go out and vote.