Tracking the pollsters
This will be our last check-in with these targeted pollsters until Election Day. We’ll revisit the final polling and forecasts on November 3rd and then compare them to the actual results when we have them. Ideally, we will know enough on November 4th to make a rough comparison. It is very likely that the final count will not be known for a week or so afterwards – but that does not mean that we will not know enough within 24 hours of the polls closing to know who won.
Biden’s lead in the FiveThirtyEight polling average remains relatively steady at plus 9.1 points. This is minus 1.6 point decline from last week’s Biden +10.7 points, but that was the absolute high point of Biden’s lead in the polling average this year. With one week to go, we’ll keep an eye on the polling to see how close pollsters get to the final national result and see how close the state polling averages are to the final result in several battleground states. First up is a sample of pollsters organized by FiveThirtyEight ratings. Here are the most recent polls by pollster. The change from one week ago is in the parentheses beside the margin. Only the tracked pollsters who have released polls since last week’s post are listed below.
The Best: First Tier (A+) pollsters:
Siena College/New York Times Biden +9 (+1)
Above Average: Second Tier (A to B+) pollsters:
IBD/TIPP (tracking) Biden +7 (+2)
SurveyUSA Biden +10 (no change)
Average: Third Tier (B to B/C) pollsters:
YouGov: Biden +12 (+1)
USC Dornsife (tracking) Biden +11 (-1)
Ipsos Biden +8 (+2)
Morning Consult Biden +9 (no change)
Mediocre: Fourth Tier (C+ and worse) pollsters:
HarrisX Biden +4 (-3)
Survey Monkey Biden +6 (-2)
Rasmussen Biden +2 (-3)
As you can see, only one of the four Best and two of the four Above Average tracked pollsters released a national poll this past week. All of the Average and three of the four Mediocre tracked pollsters did so. You can expect to see results from the better categories in the next week, perhaps even more than one poll from some of them. The better pollsters are being more cautious, but they also tend to be higher cost polls as most of them use live calling, while others in the lower categories use IVR or online methods which allow them to poll more quickly and cheaply - but perhaps at a cost in quality (we’ll see how they do next week).
Next we will look at the polling averages by state (and Congressional district where applicable), according to The Upshot. For the states, we’ll focus on whether pollsters are collectively getting right the race in the battleground states instead of how individual pollsters are faring. Since some pollsters are only in a few of the states and others only poll once or twice in the entire election cycle, it’s better to stick with the polling averages here. Again, the change from one week ago is in the parentheses beside the margin.
Arizona: Biden +4 (-1)
Florida: Biden +3 (-1)
Georgia: Biden <1 (-1+)
Iowa: Biden +3 (no change)
Maine-02: Biden +4 (-1)
Michigan: Biden +9 (no change)
Nebraska-02: Biden +7 (-1)
New Hampshire: Biden +11 (no change)
North Carolina: Biden +3 (no change)
Ohio: Trump +2 (+2)
Pennsylvania: Biden +6 (-2)
Wisconsin: Biden +7 (-1)
Texas: Trump +3 (+1)
As you can see, Biden did not increase his margin in any of these states/districts. However, where there is change it is small and a narrowing is to be expected (but not guaranteed) in the final weeks.
Consilience among forecasters
In 2016, the forecasters had very different projections. Folks don’t remember it that way because all the models gave Clinton a better shot of winning than Trump. However, election forecasting is based in probabilities not just a simple predicted winner. Huffington Post gave Clinton a 92% chance of winning. That’s even better than any model except The Economist gives Biden this year. The Upshot gave the Democratic nominee an 85% chance of winning. And FiveThirtyEight gave her just a 70% chance of winning.
So, how are these forecasts different? Consider the odds that something awful might happen to you today. Would you feel safe with a 30% chance that it will happen? How about an 8% chance? Those are very difference chances: 3-in-10 versus <1-in-10. They are so different, in fact, that the forecasts told us one important thing: the race was likely volatile.
There are two different types of forecast models. The first is a “polling model” which incorporates, you guessed it, polling data. The second is a “political science model,” which according to Larry Sabato “uses election fundamentals such as the state of the economy and the incumbent president’s approval rating to predict either the popular or the electoral vote.” The former is the kind of forecasting I’ve referred to frequently in this newsletter, but the latter is employed by serious forecasters like Sabato. It is also fair to note that there is a true polling model which uses only polls and a hybrid model in which election fundamentals are used along with polling to project a result. FiveThirtyEight has used both the true and hybrid models in the past, but the model commonly referred to is their hybrid model.
Since the political science model does not incorporate polling it can make final projections of election results in advance of the election – often weeks in advance. And those projections can be accurate. Typically, they include factors such as the power of incumbency, approval ratings, and the state of the economy. However, Sabato believes that this year a “simplified model” that only uses Gallup approval ratings to project the outcome is likely a better predictor. This is because of how partisan politics in the US is today and because the pandemic has skewed perceptions/importance of the economy among voters. Sabato explains it here.
Here’s a table from Sabato’s Crystal Ball containing potential projections for this year’s election using the simplified model:
As you can see from this table, the better Trump’s net approval rating the more likely the probability he will win. What the source note describes is that the probability tracks how well incumbents who were running for reelection did over a 60 year period based on net approval ratings from Gallup in late October of the election year. According to Sabato: “Based on a net approval rating of -12 in the most recent Gallup Poll, the model now predicts that Joe Biden will receive 344 electoral votes to 194 for Donald Trump. The model now gives Biden an 86% chance of winning the electoral vote.”
So, the simplified political science model forecasts Biden with an 86% chance of winning the election next week. That is strikingly similar to the FiveThirtyEight polling model (hybrid) forecast of Biden with an 87% chance of victory. The Upshot’s polling-only model projects a Biden winning 357 EVs to Trump’s 181. Sabato’s model projects Biden with 344 EVs to Trump’s 194. Again, strikingly similar forecasts. The Economist, which is new to the forecasting game this year, has Biden a near lock on the probabilities with a 95% chance of winning. However, their projection on the electoral vote is Biden with 352 EVs to Trump’s 186. Once again, that is strikingly similar to the other polling models and the simplified political science model.
In science, this is called consilience. It is when evidence from different academic disciplines or methodologies converge on a single conclusion. We often hear it called “consensus” when discussing climate change, but consensus is a political term not a scientific one (which contributes to the misunderstanding of the science actually says in this case). Concerning climate change, there is a consilience that human activity is causing global warming. And this year, we see a consilience of forecast models converging on a Biden victory of about 350 electoral votes.
Anxieties
I know a lot of you are anxious, scared, even panicky about the election. My advice to you if the campaign is affecting your emotional and mental health is to stop paying attention. Get off social media and ignore the political news. Seriously. If my blog doesn’t help calm you, then stop reading it. Elections always get crazy at the end in one way or another. If you cannot deal with it, then don’t. Tomorrow, I am going to do something a little different and explain what worries me about the election. If you think it will stress you out, then skip it. I don’t think it will because regardless of any rational and irrational worries I have, I am optimistic about the chances of a Democratic victory in general and a Biden victory in particular. As a student, one of the things I liked better about exams than papers was that whether you liked it or not the exam was going to happen and it would be over at a time certain. An election is similar to an exam. It’s going to happen next Tuesday. No amount of anxiety is going to change that. So, relax. Please.
If it helps, keep organizing. (Even if it doesn’t help you should probably keep organizing.)