Has the race really changed?
As the US rolls into the final days of the election campaign, expect the polls to tighten further. This contraction may or may not reflect the state of the race; the polls were always absurd.
Election geeks on both sides of the aisle pore over RealClearPolitics.com polling data. This is the Superbowl of politics, and we check the scores constantly to see how our team is doing. Nine days ago, on October 12th, Biden led Trump by 10.2 points in the two-way national average. Today, on October 21st, Biden is still up by a difference of 7.5 percent.
RCP averages for state polls tightened over the same period. Biden led by 7.2 percent in Pennsylvania on October 12th, today the difference is 4.9 percent. Most battleground states are much the same over the same period: Wisconsin, 6.3 and 4.6 percent, Minnesota, 9.0 and 6.3 percent, Florida, 3.5 and 2.1 percent, and so on.
To understand what may happen in the poll over the next few days, we need to understand the sorta-failure of 2016. Then we can make some projections about 2020.
The ghosts of 2016
The public widely regards 2016 polling as a failure. Hillary Clinton led the RCP average by 3.3 percent in the national polls, so she was going to win. Instead, she only beat Trump by 2.1 percent, and he won the electoral college. The difference between the average of the polls and the final result was actually tiny. Much maligned Rasmussen Reports hit the bullseye at Clinton plus two percent.
But there’s more to the story. Nate Silver points out that the average error in presidential polls for the 21 days before the election is around 4 percent. In 2016 the error was 3.1 percent at the national level. The weighting of Silver’s methodology is different than RCP’s straight average, but intuitively if the election day error was 1.2 percent, then the error on the other end, around three weeks out, has to be much higher. History seems to bear this out; on October 18th, 2016, Clinton led by 7.1 percent in the RCP average.
Looking at individual pollsters, a Monmouth poll ending October 16th, 2016, had Clinton up by 12 points. By election day, Monmouth fell to a more reasonable but still far off six percent lead. Other polls are not as easy to compare, but around the same time an Associated Press-GfK poll had Clinton up by 14 and an ABC News poll had Clinton up by 12.
Let’s think about this. In 1996, Bill Clinton beat Bob Dole 49.2 percent to 40.7 percent (with 8.4 percent to Ross Perot). Was it reasonable at any point to think Hillary Clinton would beat Bill Clinton’s margins? Or any of Barack Obama’s margins? Or Reagan’s margins in 1980? No. So why would anyone ever release a poll that says so?
Why do polls tighten? Do events drive poll changes? It’s hard to draw conclusions from the 2016 election cycle. Nate Silver argues that Comey’s letter to Congress immediately sank the polls, and probably cost Clinton the election. He also notes that she was already in the danger zone. She also lost because of perceived corruption, falling minority turnout, choosing to do government business on a personal email server in her basement, and a ton of other personal Samsonite.
Certainly, most news media outlets believe news drives polling and voting, thus the Herculean effort to shield Biden from criticism. Twitter censored the New York Post’s story on Hunter Biden’s emails because they were obtained illicitly, but illegally obtained tax information published by the New York Times is just fine.
With the media security blanket, why have polls still moved in Trump’s direction? Such stories might move the actual vote a percent or two one way or the other, which might be critical in the electoral college in a knife-edge election, but the tightening of the race may be more about polling returning to reality.
So what lies ahead in the next weeks? An alternative explanation to events driving polls is that as the race winds down, the polls tighten to what pollsters believe is reasonable, irrespective of the underlying data. While numbers are straightforward, weightings and estimates of turnout require subtle judgment calls.
Biden is likely to win, as most polls point that way. On the other hand, there are subtle signs that Trump could win. RCP’s Sean Trende points out the mismatch between Trump’s steady 44 to 46 percent approval numbers and forties election polling. That level of approval could get Trump 46–48 percent of the vote, and he’s shown he can win at that level. Trafalgar Group had uncanny success predicting state-level results in 2016 and 2018, and predicts a Trump win. An analysis by JPMorgan highlights Republican registration advantages and early voting patterns in key states.
The mixed indicators point to a close but sharply divided race, not a blowout in either direction. A Trump win could be anywhere from two percent negative as in 2016 up to a percent or two in the black. A Biden winning percentage starts at beating Trump by one or two percent. Will he come in at 3.9 percent ahead (Obama’s margin in 2016) or 7.2 percent ahead (Obama’s margin in 2012)? Maybe Joe Biden will overperform Barrack Obama, but would you take that bet?
Pollsters, universities, and news organizations that conduct polls are in business to make money. Non-credible polling hurts reputations and costs real dollars in the out-years. While we can’t predict the winner, we can predict the polling. A smart businessman would constantly examine underlying assumptions used for weighting, especially for the last poll before the election. Consciously or not, this tweaking of the models will lead to intuitively acceptable results.
A nice, safe, estimate would be Biden at plus two to four percent…
A nice, safe, estimate would be Biden at plus two to four percent, depending on who the pollster believes might ultimately win. A prediction of Biden at +3 percent with a three percent margin of error would cover anything from Trump breaking close to even in the popular vote to a fairly large victory of Biden at six percent.
A few pollsters will be outliers in each direction, but not by huge margins. As the polls converge, the RCP average will likely be in the Biden +2 to +4 range. CNN published a poll ending on the 4th of October with Biden in the lead by a whopping 16 percent. Ronald Reagan beat Walter Mondale in 1984 by 18.2 percent. Would CNN publish a poll that Biden is the second coming of Ronald Reagan on this November 2nd? They might, but most people wouldn’t take that bet.
It’s just possible that polling in the final days of the election becomes more about protecting reputations and future business than about accurate predictions.
Brian E. Wish works as a quality engineer in the aerospace industry. He has spent 29 years active and reserve in the US Air Force, where he holds the rank of Colonel. He has a bachelor’s from the US Air Force Academy, a master’s from Bowie State, and a Ph.D. in Public and Urban Administration from UT Arlington. The opinions expressed here are his own. Learn more at brianewish.com.