Leading up the 2020 election, the polls indicated Joe Biden would win handily. For the experts, pundits and media, the question really became whether it would come in a landslide – assisted by a blue wave and control of the Senate – or a smaller victory in the Electoral College.[1] Few believed it would be close or gave Donald Trump much of a chance. That was not the case at The Offence,[2] which evaluated the 2020 swing state (and district) polls and compared them to the difference by which the polls were off in predicting their respective 2016 contests, referred to as Trump’s polls-to-performance (PTP). The electoral analysis indicated that, were the pollsters to have failed to correct the errors of 2016, the election would be exceptionally close due to the tightening of critical races in the closing weeks:
| Electoral Votes | Trend to 2020 Election | 2016 PTP* | Trend + PTP | |
|---|---|---|---|---|
| Arizona | 11 | -1.79 | 1.80 | 0.01 |
| Florida | 29 | -1.46 | 2.50 | 1.04 |
| Georgia | 16 | -1.20 | 1.39 | 0.19 |
| Iowa | 6 | 1.33 | 6.31 | 7.64 |
| Maine-2 | 1 | -2.61 | 9.89 | 7.28 |
| Michigan | 16 | -6.20 | 4.33 | -1.87 |
| Minnesota | 10 | -9.06 | 5.78 | -3.28 |
| Nebraska-2 | 1 | -2.79 | -2.80 | -5.59 |
| Nevada | 6 | -4.63 | -1.32 | -5.95 |
| North Carolina | 15 | -0.66 | 4.96 | 4.30 |
| Ohio | 18 | 0.83 | 6.53 | 7.36 |
| Pennsylvania | 20 | -3.61 | 4.62 | 1.01 |
| Texas | 38 | 0.97 | 1.49 | 2.46 |
| Wisconsin | 10 | -7.47 | 6.67 | -0.80 |
Consequently, as early returns came in showing Trump with leads, while the media looked dazed and confused, there was little surprise here, nor was their confidence those leads would remain steady.
But rather than attempt to predict the election, which is reserved for those who provide odds, the analysis was presented to assess what could be developing as the election approached. It was conducted on a state-by-state (and district-by-district) basis, believing some potential errors may be due to internal state demographics and the methodologies of polls, some of which only survey a single state.
What were the election results? As it stands, Biden will win the Presidency, the specific electoral margin yet to be determined. This was not, however, the purpose of the analysis. Rather, it was intended to determine the accuracy of the polls, for which the total vote – or who wins – is less relevant than the difference in share, as measured by points. Among the twelve states (and two districts) included – those determined therein to be the critical battlegrounds – how did the analysis compare to both the actual results and the expert evaluation at the much-lauded FiveThirtyEight, as measured by Trump’s over/underperformance?
| Election Results* | FiveThirtyEight Projection | FiveThirtyEight Difference | The Offence (Trend + PTP) | The Offence Difference | |
|---|---|---|---|---|---|
| Arizona | -0.39 | -2.60 | 2.21 | 0.01 | -0.40 |
| Florida | 3.37 | -2.50 | 5.87 | 1.04 | 2.33 |
| Georgia | -0.29 | -1.00 | 0.71 | 0.19 | -0.48 |
| Iowa | 8.20 | 1.50 | 6.70 | 7.64 | 0.56 |
| Maine-2 | 7.69 | -1.60 | 9.29 | 7.28 | 0.41 |
| Michigan | -2.70 | -8.00 | 5.30 | -1.87 | -0.83 |
| Minnesota | -7.20 | -9.10 | 1.90 | -3.28 | -3.92 |
| Nebraska-2 | -11.10 | -2.20 | -8.90 | -5.59 | -5.51 |
| Nevada | -2.70 | -6.10 | 3.40 | -5.95 | 3.25 |
| North Carolina | 1.40 | -1.70 | 3.10 | 4.30 | -2.90 |
| Ohio | 8.20 | 0.60 | 7.60 | 7.36 | 0.84 |
| Pennsylvania | -0.70 | -4.70 | 4.00 | 1.01 | -1.71 |
| Texas | 5.80 | 1.50 | 4.30 | 2.46 | 3.34 |
| Wisconsin | -0.70 | -8.30 | 7.60 | -0.80 | 0.10 |
Polls are unlikely to ever perfectly match election results. The most accurate collection of polls will find half of them slightly overestimating a candidate’s support while half slightly underestimate. The closer the PTP differential is to zero, the more accurate any single poll is proven to be.
On the aggregate, FiveThirtyEight applied its advanced modeling to the published polls and underestimated the Trump vote in 13 of the 14 elections – all but Nebraska-2, which had relatively few polls upon which to project a result – doing so by an average of 3.79 points, erroring by an absolute value of 5.06.
By contrast, The Offence’s analysis overestimated his support by 0.35 – erroring by an absolute value of 1.90 – underestimating and overestimating the potential Trump share in an equal number of elections – seven (7). Between the two, The Offense came closer to the vote differential in 13 of the 14 battlegrounds, all but Minnesota. This is not intended to discount the work at FiveThirtyEight, which provides well-developed and valuable analyses on elections, and assisted in their forecasting a Biden win, but rather to highlight the inevitable consequences of poor polls, from which their diligent and detailed assessments were conducted.
What does this mean for polls? That is largely for the pollsters to decide. But irrespective of their claims heading into the election,[3] they did not correct for the systemic failings built into the 2016 polling. Thus, The Offence’s analysis, which simply adjusted the 30-day trend in differential to election day by the margins of 2016, rather accurately projected how close the vote would be in the key races, and in combination, the overall election for President. Was the electoral result correct? No, but that was never the intent, and electoral votes can easily tip with fractions of a point in only a few states, i.e. Arizona, Georgia, Michigan and Pennsylvania, which is why the Trump campaign continues to fight the results in these very states.[4]
Will the Mail-in Votes Deliver Their Mail-order President?
Curiously enough, when the final counts are certified and the lawsuits are completed, the margins of Biden’s 2020 win may prove quite similar to Trump’s 2016 win, either by the total vote differential among the decisive states or the Electoral College vote.
That said, with recounts expected, one would do well to consider that there was an unprecedented number of absentee/mail-in ballots this year, estimated to be roughly 80 million, twice the previous record.[5] It will be interesting to see how many of these ballots, which heavily favor Biden, fail to survive additional scrutiny, and if rejected ballots are sufficient to alter the results of any state’s election. To this effect, the Trump campaign filed suit in Pennsylvania claiming a violation of the Equal Protection clause of the Constitution due to the alleged uneven requirements of absentee/mail-in ballots, as compared to traditional in-person voting.[6]
How reliable are absentee/mail-in ballots? In 2018, Pennsylvania rejected 4.4% of their respective ballots. They received roughly 2.6 million this year,[7] a rate which could subject 114,400 ballots to disqualification, if has not already done so. This is greater than Biden’s current lead of roughly 54,000 with 99% reporting. Unfortunately for Trump, it is possible these ballots have undergone a form of ‘ballot laundering,’ in which the verification envelopes have been discarded and their contents, the ballots, have been stored away in combination with others that makes such post-count evaluations difficult, if not impossible.
And in Wisconsin, which is reported to have received nearly 2 million absentee/mail-in ballots, and currently shows Biden with a lead of approximately 20,500 with 99% reporting, there is no provision for voters to cure – or correct – disqualified ballots. In 2018, 1.7% of Wisconsin’s 150,114 such ballots were rejected. Although that total was only 2,517, a similar rejection rate of 2 million ballots would result in 34,000 disqualified/rejected ballots. It is probable the majority of improper ballots has already been rejected as part of the vetting system in place; however, the potential disqualifiable ballot total is greater than the current vote margin. Is it possible that additional scrutiny could turn Wisconsin?
Given the uphill climb the Trump campaign is facing, it is exceedingly unlikely any combination of recounts and lawsuits will be consequential enough to change the ultimate outcome of the Presidential election, but even if any of them alter the counts they will not significantly affect the poll analysis.
The Offence’s pre-election assessment of the polls was designed to forecast how ‘surprisingly’ tight the election may be and what the nation would be facing on election night if the polls were, once again, wrong. As it turned out, because of how close the races proved to be, election day became election week, America was left waiting for a declared winner, the win was disputed and lawsuits became the new subject of the day.
A more honest evaluation by an impartial media could have presented this possibility as more likely than unlikely.
Instead, the raw polls led many to confidently predict decisive Biden wins in the critical states, and thus the election. They were largely wrong. Why was The Offence’s review so much more accurate? The 2016 polls were included. Ironically, one could say the election was as close as should have been expected because the polls were right…
…by how wrong they remained.
Note: Figures are as of 11/11/20. This article will be updated when the election results are finalized.
[1] Nate Silver, Biden’s Favored In Our Final Presidential Forecast, But It’s A Fine Line Between A Landslide And A Nail-Biter, (FiveThirtyEight, 2020)
[2] The Editor, Donald Trump Reelected 279 – 259, if the Polls Are as (In)Accurate as They Were in 2016, (The Offence, 2020)
[3] Alexandra Hutzler, How the Pollsters Changed Their Game After Getting the 2016 Election Wrong, (Newsweek, 2020)
[4] Alana Abramson and Abigail Abrams, Here Are All the Lawsuits the Trump Campaign Has Filed Since Election Day—And Why Most Are Unlikely to Go Anyway, (Time, 2020)
[5] Processing, counting and challenging ballots in Wisconsin, 2020 (BallotPedia, 2020)
[6] Eric Boehm, Trump’s New Lawsuit Looks Like an Attempt To Cancel All Mail-in Votes in Pennsylvania, (Reason, 2020)
[7] Fredreka Schouten, Why mail-in ballots in Michigan, Wisconsin and Pennsylvania were counted so late, (CNN 2020)
The Offence 