Four Days Out: What If The Polls Are Just As Wrong As They Were In 2016?
As I post this we are only about 90 hours from the polls opening on election day. No doubt you have heard Trump supporters deride the polls indicating a large Biden lead, to include in key states, as “fake news” with the attending triumphant cry that “the polls were wrong 2016 and they are wrong now.” I decided to assume that the polls today are every bit as wrong today as they were four days before the election in 2016 and evaluate how that would impact Tuesday’s election.
I took nine key states that Trump won in 2016 and which have sometimes shown Trump behind this year. I compared that to polling on November 4, 2016 (that was four days before the November 8 election). In both cases I used Nate Silver’s 538 polling data. This data reflects a compilation or aggregation of numerous polls, a poll of polls.
For each of these nine key states I compared Trump’s predicted result by the polls on November 4, 2016 vs. his actual result in the election. With the exception of one state (Arizona) Trump outperformed the November, 2016 polling on election day. That difference in Trump’s favor became “2016 Error” in the chart below (for Arizona the forecast on November 4, 2016 equaled the result so the error was zero).
That error was then used to handicap the current polls for each of the nine states. Trump was credited with full amount of the 2016 error in the current polls.
The results are summarized in the chart below. By way of explanation, for each of the columns of polls the number reflects Trump’s relative position to his democratic opponent. For example, Trump was ahead in the polls by 0.3 in Florida four days from the election in 2016 and is down in the polls by 2.0 to Biden in Florida today. The 2016 polling error is then calculated as the difference between the 2016 election and actual election result. In Florida that error was 0.9. The amount of that error is then subtracted from Biden’s current lead to show that even if the Florida polling is off by as much as it was in 2016 then Biden would still win by 1.1%.
Two things should be emphasized. First, Trump won all of these states in 2016. Even assuming the polling error is just as bad today as it was in 2016 then Florida, Pennsylvania, Michigan, Wisconsin, North Carolina, Georgia and Arizona all flip to Biden. Under this model Trump hangs on to hold Iowa (even though currently behind in polling there) and wins Ohio (where he was once behind but has moved slightly ahead in recent polling).
Second, the assumption that the polls are just as biased against Trump today as they were in 2016 is probably wrong. Pollsters are in no mood to repeat their embarrassment from 2016. Accordingly, pollsters have worked hard to tweak the demographics of their polling to catch more voters who are likely to vote for Trump. Many pollsters are more concerned that they may have overcompensated and are now inflating Trump’s standing in the polls.
As stated this model predicts that seven states won by Trump in 2016 will flip to Biden in 2020.
Assuming Biden wins all the states Clinton won, and the seven states this model predicts, he will achieve a net gain of 117 electoral college votes over what Clinton received. Biden would receive 350 electoral college votes, a landslide greatly exceeding Trump’s claimed record “landslide” of 306 electoral college votes.
Assuming Trump wins all the other states he won in 2016, he finishes with 188 electoral college votes.
The predicted 350 electoral college votes Biden would receive is 80 more than the 270 required to win. Biden could lose any of the six states he is predicted to flip and still win big. In fact, he could lose any two of those states and still win. If Biden lost both the two biggest states of Florida and Pennsylvania he would still win easily, even if the polls today are just as wrong as they were in 2016.
Let’s assume Biden loses all the states where the adjusted forecast has him winning by less than 2%. That credits Trump with Florida, Pennsylvania, North Carolina and Georgia. Biden loses 80 electoral college votes but still rests exactly on the 270 he needs to win. However, the closeness of this call shows that while the path to victory for Trump is difficult, it is possible. [Faithless electors have been excluded from these calculations].
It should be pointed out that this study also assumes both candidates win all the other states their party carried in 2016. This seems a safer assumption for Biden than for Trump. Texas is surprisingly in play with Trump holding a current lead of only 1.3% and a few polls actually calling it for Biden. The Trump Campaign has made noise about flipping Nevada (and its six electoral college votes) which went to Clinton by less than 2 points in 2016. However, the polls show Biden ahead in Nevada by over 6%.
COMPARISON TO PRIOR STUDY
I did a similar analysis on October 13th. It is summarized in the chart below.
There were two general trends between the studies. First, Biden’s lead eroded at bit, but only slightly. By four days out in 2016 Clinton was crashing the polls. Second, as a result of that 2016 Clinton poll crash the adjusted 2016 error was less in this four day out analysis than it was in the analysis for October 13th. The average polling error from the sampled states on October 13th was 3.38% whereas four days out from the election it was only 2.8%.
Simply put, compared to 2016 Biden enjoys a much bigger lead than Clinton did with only four days remaining before the election. That lead has remained largely stable, declining only slightly over the last several weeks. The pollsters still called it wrong four days out in 2016, but not by as much as they had in mid-October of that year.
As a consequence Biden actually picked up a state, Georgia, in this four days out analysis that was credited to Trump in the October 13th analysis.