Seeking the Truth in politics, economics, history, and religion...
Thursday, November 5, 2020
What happened to the polls?
It wasn’t very far into the counting on Tuesday night when many people started attacking the polls. There were definitely problems with some polls, but much of the anger was premature.
One major problem with many of the early allegations of bad polling was that the counting was ongoing. At the point where many people, notably Trump supporters, were attacking the polling, states had often reported only about 60 to 80 percent of their vote.
In this year’s election, many of the battleground states followed a pattern that The Dispatch has called the “red mirage.” Because early in-person votes were counted first, Joe Biden often started with a healthy lead. As Election Day in-person ballots were counted, Donald Trump closed the gap and often took a lead. It was at this point that many casual observers assumed that the polls were wildly inaccurate.
But the story doesn’t end there. There was an abnormally large number of absentee ballots this year. When these votes were counted, the gap closed once again. In some states, mail-in ballots allowed Biden to take the lead. In others, the outcome is still up in the air.
Votes are still being counted so it is still premature to say exactly how close the polling was this year, but it looks a lot closer on Thursday morning than it did on Tuesday night. Having said that, some polls were better than others and some races look more problematic than others.
I did my last analysis of the presidential election polling on Oct. 27, a week prior to the election, but the polls didn’t stop coming in at that point. Real Clear Politics has a handy page that sums up battleground state polling (excluding Georgia and Nevada, which you have to look up separately) as it stood on Election Day. Here is a breakdown that compares these numbers with the actual results currently reported by the AP. My Oct. 27 article covered several additional states, but for brevity, I’ll concentrate on the ones that have been decisive or are still in play.
Arizona – RCP spread – Bden + 0.9 The First Oct. 27 – Biden + 2.2 Current count- Biden + 2.4 (88 percent reporting)
Florida – RCP spread – Biden + 0.9 The First Oct. 27 – Biden + 1.2 Current count – Trump + 3.4 (99 percent reporting)
Georgia – RCP Spread – Trump + 1.0 The First Oct. 27 – Trump + 0.4 Current count – Trump + 0.4 (99 percent reporting)
Michigan – RCP spread – Biden + 4.2 The First Oct. 27 – Biden + 9 Current count – Biden + 2.7 (99 percent reporting)
Nevada – RCP spread – Biden + 2.4 The First Oct. 27 – Biden + 5.2 Current count – Biden + 0.5 (75 percent reporting)
North Carolina – RCP spread – Trump + 0.2 The First Oct. 27 – Biden + 1.2 Current count – Trump + 1.4 (94 percent reporting)
Pennsylvania – RCP spread – Biden + 1.2 The First Oct. 27 – Biden + 4.5 Current count – Trump + 2.6 (89 percent reporting)
Looking at the data, a clear pattern emerges. The polling in every state tightened noticeably in the last week before the election. As I frequently point out, polls are not predictive. They are snapshots of current opinion. They are also not exact. Every poll includes a margin of error that is typically between about two and four points.
So far, the largest error seems to be Florida where polling swung by about four points. Pennsylvania currently has the second-largest error, but votes are still being counted there and the president’s lead is shrinking as absentee ballots are counted. Nate Silver of FiveThirtyEight noted on Twitter that Biden’s margin on the absentee ballots so far has been enough to eventually put him in the lead, which will also close the polling gap.
If we exclude the two outlier states and average the current polling error for the remaining battleground states, we find that the polls were only off by an average of 1.3 points. That’s pretty dang close. That margin may shrink as well as more votes are counted, particularly in Nevada, which is running behind. Even including Florida and Pennsylvania, the polling error was only 2.4 points. That’s still not far off and well within the margin of error.
Admittedly, the shift in all cases was toward Donald Trump, but that does not necessarily mean that pollsters were biased against Trump. There is some evidence that late-deciding voters went for Donald Trump, which would have impacted the accuracy of the final polls. New York Times exit polls show that Trump won voters who decided in the last week by a two-to-one margin. There were also big demographic swings that confounded pollsters. This year, Trump gained with minorities but lost ground with white voters.
No one pollster seems to have gotten it all right. Trafalgar is becoming the darling of the right for showing a two-point Trump lead in Florida. However, Trafalgar also showed Trump up by three points in Arizona, five in Georgia, and two in Michigan, Those surveys were outliers that overestimated support for Trump.
The best way to watch the polls is to look at averages and trends. Don’t obsess over individual polls. Polling averages are generally pretty close since they smooth out the difference between the Trafalgars and other polls that swing in the opposite direction, such as the Nov. 1 Quinnipiac poll that showed Biden leading by five in Florida. The big takeaway from the polling average for Florida that showed Biden with a less-than-one-point lead should have been that the race was too close to call.
The polls were not exact in 2020, but they never are. They aren’t intended to be. The polling averages were mostly very close, however, and Florida may turn out to be the only state where the polling average did not correctly pick the ultimate winner. Reports of the polling industry’s imminent demise seem to be very premature.