Three tips for poll obsessors (like me), in a presidential election year, followed by a few insights from Nate Silver’s FiveThirtyEight.com. The first two below are old and obvious, so feel free to skip to No. 3, which is not.
- Polls are not predictions. Everyone knows this, but most of us ignore it most of the time. They are fuzzy-focused snapshots of a moment that has already passed by the time you read them. Fuzzy refers to the fact that, given the margins for sampling error (which we tend to ignore when we look at polls) means that a poll showing candidate X ahead by plus or minus three points means he might be ahead by six or the race might be tied. And by the time you read the poll, the situation may have changed. This is a pretty good argument for paying less attention to them, but I can’t help myself.
- Obviously, in a presidential election, national poll numbers (in addition to be being fuzzy and out of date) are almost irrelevant; it’s the swing states that matter.
- We don’t know which states those are. This is less obvious point, and the analysis is pretty surprising. FiveThirtyEight.com recently looked back at the last four presidential elections, and focused on the states that the estimable Cook Political Report had identified as the key swing states five months ahead of the election (which is where are now). Then they looked at the final election results to see if those “swing” states turned out to be very close in the actual election. Cook’s list turned out to be “wrong,” much of the time, not in predicting which way the swing states would swing, but in predicting which states would be close enough on Election Day to merit the status of “swing” states. It’s a fairly impressive bit of research. In each of the last three presidential elections, Cook’s list of swing states in June turned out to contain more states that didn’t end up being close in the end than states that were.
Cook’s June swing-state list also, in three of the last four presidential elections, left off several states that turned out to be much closer. The really big example of how wrong these June lists turned out to be was 2016, when, of four states identified as the key swing states, two of them turned out to be close, and two didn’t.
But – and here’s the big deal – NINE other states that weren’t on the June swing-state list turned out to be quite close (including our own beloved Minnesota). This is not about the folks at Cook being dumb. They’re not. They’re smart. It’s about the unknowability of November from the standpoint of June.
I absolutely agree with the point of the 538 piece. It’s not that the Cook team isn’t smart; it’s that the future is unknowable, which is really annoying. (And I know, too, that I will continue to notice when the list of predicted swing states is adjusted. I only hope this analysis will help me keep my shirt on when that occurs.)