It's September! Time to Start Paying Attention
September 3, 2024
Philadelphia. Tuesday morning. September 3rd. I am traveling both for work and pleasure right now. It has been a little difficult to get posts out, but I should manage at leasts another one this week. After about mid-September you may start getting a lot more posts from me until we get through November. Fair warning!
Today marks the unofficial start of the fall election season. This is the time that most analysts think we can start taking the polling seriously. Of course, taking polling seriously means understanding what polling tells us. Any particular poll gives us the cliche “snapshot in time.” Polls do not predict election outcomes.
There is a lot of confusion out there among people about whether the polls are accurate or not. Often, people are thinking of election forecasts rather than polling. For instance, in 2016 the national polling was pretty damn accurate. Almost all late polls correctly had Clinton winning the popular vote and within the margin of error for her 2.1% win. Of course, she actually lost the election, but that was because of the electoral college. The only polling that can really help with that would be if there were state-level polling going on during the final weekend of the campaign. Polling is expensive, and state-level polling was not very good back then. Because of these two things, the outfits that sponsor public polling largely decided not to conduct more that final weekend. And that was because no one really believed that one candidate would win the electoral vote but lose the popular vote, even though it happened just 16 years earlier.
When people think of how the polls missed in 2016 they are often thinking of the election forecasts. In 2016, 538 (styled FiveThirtyEight back then), Princeton University, and The Upshot at the New York Times were the major forecasters.1 Forecasts utilize polls as inputs, but often include other things like how states normally perform, economic conditions, etc. Princeton and The Upshot gave Clinton a 97% and 99% chance, respectively, of winning the electoral vote and thus the election. 538 gave Clinton a 71% chance. Nate Silver, who founded 538 but no longer runs it, said at the time that it was a miss, considering how well 538 did in 2008 and 2012. However, a 29% probably that something will happen is a very real chance that it will. Would you cross a street if someone told you there was a 29% chance you would get hit by a car? Princeton and The Upshot clearly missed, but 538’s forecast was accurate enough. A shift of a few thousand votes across states that typically supported Democratic nominees would have won Clinton the presidency.
In 2020, there was a different problem with the polling and I am surprised that it is not discussed more openly. It was the preoccupation with margin of lead that was in error. In W. Joseph Campbell’s excellent history of presidential polling, Lost in a Gallup, he focuses his final chapter on how badly pollsters missed on the 2020 election but we talk about it less because Biden won. There is definitely truth in that; no one really cares how much the polling misses if it gets the winner right. That also happened in 2012, where most polling showed a much closer race but Obama, the eventual victor, nonetheless in the lead. People still fret over polling because of 2016, since everyone got the winner wrong. In 2020, Biden was up by as much as ten points in the polling near the end of the election. He won by 4.5 points. The thing people don’t talk about is that while the margin was wildly off, most polling did an accurate job of getting Biden’s vote share.
There is a tendency to think, in the absence of other information, that Biden +10 means a landslide victory. However, it does not if he is polling at 52%. Not in a two-person race. And 52%, which some pollsters had him at, is just about what he got (it was closer to 51%). Tony Fabrizio, who is Trump’s pollster, famously says “Democrats get what the poll.” I don’t think this really translates to non-presidential races, but it sure seems to in presidential races in recent years. The polling in 2020 largely got Biden’s vote share correct - at least nationally (there were still state-level misses) - but because so many eventual Trump voters sat in the undecided category until the end, the margin of lead made it appear as though a landslide was on the way. The embarrassing thing about this is that it is very easy to see that you cannot win a landslide victory in a two-person race with 52% of the vote. Fabrizio knew 2020 was closer than it appeared for this very reason. Yet, still, the media just would not report on this obvious and glaring complication with their analyses. I won’t get into why they didn’t, although I do think one reason has nothing to do with business - they, like most people, see the margin and think that is the end of the story.
The thing to do these next several weeks is to pay attention to Harris’s topline number. That is most important. If she can stay above 50% in most polls most of the time, she is a good bet to win. Remember also, that she has to win the electoral vote to be president so the polling in the battleground states is very important. Unfortunately, state-level polling has not been very reliable. Focus on the top pollsters (e.g., Siena, Marist, Marquette Law, ABC News, Fox News). Some states have very good state pollsters who know their states well (e.g., Ann Selzer in Iowa and Florida Atlantic College). Do not get excited if you see Harris up by 3 in North Carolina but polling at 46%, but feel free to feel good if she is at 52%. The polls will only give us hints about what people are thinking at any given point in time, but if the electorate is not erratic (as they were in 2016) or indecisive (as they were in 1980) they should give us a good idea of who is leading. And right now, that is Vice President Kamala Harris. Here are the latest polling averages:
538: Harris +3.2
New York Times Poll Tracker: Harris +3.0
Silver Bulletin (Nate Silver): Harris +3.5
The Economist: Harris +3.5
RealClearPolitics: Harris +1.8 2
Check out each of these poll aggregation sites for information on battleground states. One thing that is interesting is that some have recently started tracking Florida and Texas as potential battlegrounds. Trump’s lead in those states is only about five points. Can you imagine what people would be saying if Harris had a lead of only five points in California and New York. (She doesn’t, by the way. She is up over 30 in California and over 15 in New York.)
Today it is 538, The New York Times Poll Tracker, Silver Bulletin, and The Economist.
As I have noted previously, I include RCP here to show how well Harris appears to be doing since they have a methodology that allows suspect Republican-leaning polling firms to not only be considered in the average, but give them the same weight as top quality and proven pollsters. RCP also does not appear to weight polls by how recent they are, giving much older polls too much consideration in calculating the average.