Public opinion polls leading up to the November presidential election have been telling a very clear and consistent story for several months now. But that doesn't mean everyone believes it. What should we make of accounts from people "on the ground," like this recent New York Times piece, suggesting that the polls are missing key pockets of Trump support in competitive states like Pennsylvania?
Concerns along these lines came up a lot in my interviews with Democratic activists for my book. Some said that a big problem with the polls in 2016 was that they missed important signs of support "on the ground." One related a story to me in which he was canvassing in a white working class Philadelphia suburb in 2016 that had traditionally been a deep Democratic stronghold, but people kept telling him they were undecided. Another told me that union workers he knew at a factory were showing up in MAGA hats despite a long history of voting Democratic. Another told me that the Clinton campaign kept having to downsize its speaking venues because crowds kept failing to show up in large numbers.
We're very good at looking back and seeing patterns that led to where we are now. It's a lot harder to interpret all the anecdotes you're hearing now and figure out what they point to.
Lots of people involved with that election have stories of evidence that ran against the polls but presaged a Trump win. The polls, of course, were not particularly wrong, but they notably understated Trump support in Michigan, Wisconsin, and Pennsylvania. Maybe they are this year, as well?
Probably not. The big miss from four years ago was pollsters not realizing how different the preferences were between college educated whites and non-college educated whites in those states. They've largely adjusted for that now. Is it possible that they're missing some other important bias in the polls? Sure. But the odds that we can figure this out ahead of time with evidence "on the ground" are pretty remote.
The problem is one of pattern recognition. We're very good at looking back and seeing patterns that led to where we are now. It's a lot harder to interpret all the anecdotes you're hearing now and figure out what they point to.
The 9/11 attacks make for a good example. In hindsight, there were plenty of warning signs, including accounts of Middle Eastern men suddenly enrolling in flight schools in the U.S., increased Al Qaeda chatter about possible attacks, the warnings in the Presidential Daily Briefing, etc. How did the Bush administration not see it coming? The pathways are obvious in hindsight, but before 9/11, they pointed in dozens of directions, with hundreds of potential targets, perpetrators, and dates.
What are the anecdotes telling us about next week's election results? Pretty much anything we want them to! Yes, there are lines at Trump stores, enthusiastic turnout at his rallies, huge boat parades -- he's got some energetic supporters, to be sure. There are also long lines of voters showing up for Democrats, a large number of former Republican officials endorsing Biden, people voting who haven't voted before to cast Trump out -- that sounds like some good anecdotal information supporting a big Biden win.
The fact is that no matter what happens next week, there will be ample anecdotes and "on the ground" evidence that suggested it was going to be this way all along, and we'll conveniently ignore the clues that pointed in the other direction.
So no, I don't think these stories from the ground are all that useful right now. It's not that they're wrong, it's that they point in every direction at once, and they're a great source of confirmation bias.
Fortunately, we have a pretty good tool that tells us what's likely to happen in an election that takes into account the fact that lots of different people in different parts of a state and the country can be experiencing different aspects of a campaign at the same time. It's called polling. It isn't perfect, but it's pretty honest and straightforward, and it's not trying to tell you what you want to hear.