A Note on Polling Averages

“Democracy isn’t the objective; liberty, peace, and prosperity are. We want the human condition to flourish. Rank democracy can thwart that.”

— Sen. Mike Lee (R-UT), admitting his preference for fascism over democracy yesterday on Twitter. When he says “liberty” he means property. When he says “peace” he means order. When he says “prosperity” he means for the few. Conservative activists and thinkers have been making this argument for years, but GOP office holders have been quiet on it. This is what they mean when they say to you, “we don’t live in a democracy, we live in a republic.” Actually, we live in a democratic republic. For now.

Mark your calendars. Apparently the 30-day mark is the point where Rasmussen flips a switch and turns from trying to shape public opinion to reporting it. Its latest national poll has Biden up by twelve points. Twelve! On September 30th they had Biden up by eight points. All of their polling before that had the race essentially even. Has public opinion moved that much that quickly? Not according to other pollsters. Biden has been leading in the FiveThirtyEight polling average by at least six points since June.* On September 23rd Rasmussen had the race at Biden +1 and on September 16th it was even. The polling average on those dates was Biden +7.3 and Biden +6.9 respectively. And those averages included the Rasmussen polls.

Before we continue, let’s take a look at this table from The Upshot. Based on current polling averages, here are the several scenarios that could likely occur were voting to be held today.

As we discussed earlier in the week, Biden is so far ahead right now that he would win even if the polls have the errors seen in 2016 and 2012. Additionally, Biden would win 341 electoral votes today if all the currently close states broke for Trump. The polling this week has been terrible for Trump and for the GOP more generally. Biden’s lead has significantly increased in the past two weeks.

Biden is now up +9.8 in the FiveThirtyEight polling average. The Upshot has Biden +10. Real Clear Politics has Biden +9.7. What are polling averages and what do they tells us? Well, first of all it is important to know that polling averages, like polls, are not forecasts. Polling averages try to predict future polls, not the outcome of the race. That is, the average tries to understand where the race is at tomorrow rather than on election day. So, it is still a snapshot in time.

Forecasts, on the other hand, attempt to predict (“project” is probably the better term) the result on election day. They incorporate polling, but also historical trends, partisan make-up, demographic trends, and other phenomena (often varying among forecasters). It’s not clear what each forecaster’s “special sauce” is - that seems to be considered proprietary information - but most will give you an idea of what kinds of things they include in the mix. Some forecasters - like FiveThirtyEight - will also have a “polls-only” forecast.

Even though forecasters try to project the final outcome, the forecasts are still rooted in probability. If you hear someone say FiveThirtyEight forecasted Clinton to win, that would be accurate, but not entirely correct. FiveThirtyEight’s forecast projected a 70% probability that Clinton would win. Since 70% is greater than 30%, it would be accurate to say 538 projected Clinton to win. But, it would be inaccurate to say that FiveThirtyEight said that Clinton would certainly win.

Are all polling averages the same?

No.

FiveThirtyEight and The Upshot do sophisticated calculations to arrive at their averages. Real Clear Politics does not. It uses a simple average – something I have used also in earlier posts when I was discussing recent polling – and incorporates fewer polls than FiveThirtyEight or The Upshot.

FiveThirtyEight collects as many polls as it can each day. Except for those pollsters that have been banned by them for fabricating data, FiveThirtyEight aims to include every poll. Each day as it updates the polls (and that happens throughout the day, not just at one time) it recalculates the polling average. It uses a methodology that weights recent polls more than later ones. The method allows for a tapering off of older polls rather than a hard cut-off (like, say, four weeks). Better pollsters are also weighted more to recognize that those pollsters have better track records of success than mediocre pollsters. FiveThirtyEight also weights likely voter polls more than registered voters or adults. This is more of a factor earlier in the campaign season than now as most polling in the final month is of likely voters.

The Upshot has its own polling average that relies on the polls collected by FiveThirtyEight. According to its methodology statement: “The estimates adjust for a variety of factors, including whether a poll represents likely voters, whether other polls have shifted since a poll was conducted, and whether a pollster has leaned toward one candidate in a state or nationwide. Polls are weighted by recency, sample size, and by whether they're conducted by a firm with a track record of success.” So, while The Upshot does its own calculations its method and sources are very similar to FiveThirtyEight’s.

Real Clear Politics, on the other hand, uses a simple average with a hard drop-off date for older polls. This makes the average easier to replicate, but it does not factor into account the relative quality of the pollsters or give more recent polls the relative weight they deserve. Nevertheless, as of today, RCP’s polling average is only 0.1 point off from FiveThirtyEight’s.

Forecasts are more difficult to understand because forecasters use a variety of methods to arrive at their forecasts and some of that methodology is proprietary (which is usually described as their “special sauce”). Forecasters like FiveThirtyEight will tell you the kinds of things they put in their special sauce, but they don’t tell you exactly how they mix it. There is a simple type of forecasting which uses only polling.

Trump wins on the economy question, but not on the economy.

Trump keeps winning on the economy question. And voters continue to list the economy as one of their top issues. In Florida, for instance, Ipsos (B-) finds that “Economy and job creation” and “Coronavirus/COVID-19 outbreak” are by far the most important issues. So why is Trump doing so poorly? Well, he does terribly on the question of handling the pandemic. And what he and others apparently misunderstand is the relationship between the economy and job creation and the handling of the pandemic. Isolating the issues in a survey interview focuses voters away from the overall environment, but asking voters who they plan to support is a question that incorporates over-arching concerns. And when that happens, it is difficult to disentangle “the economy” from the pandemic response.

In the vice presidential debate last night, Pence repeatedly brought the discussion back to the economy as if it were some wild success for them. And he lied repeatedly about the administration’s response to the pandemic, which polls show no one is buying (most likely because they have been living it). Pence acted like they were separate things. This is something he might have gotten from the polling questions, but it is not a sophisticated way of understanding what voters are saying. This is baffling because Trump and his supporters have been complaining for months about how the pandemic has been affecting the economy. In fact, it would be a stronger argument to blame the economy on the pandemic except for one thing. It’s pretty clear that most voters blame the administration for its handling of the pandemic. Maybe that’s why they would rather adopt Mike Lee’s preference of fascism over democracy.

* One day in early June the average was Biden +5.9. Prior to that Biden lead by at least three points since the beginning of March.