The 2020 Presidential Race: It Ain’t Over Till it’s Over

 We might as well stop the presidential election now and declare Joe Biden the winner.  At least that is the consensus of the presidential prediction machines that political pundits and the media are

pouring out.    Much like in 2016 where nearly all the predictions had Hillary Clinton a certain winner over Donald Trump, the same mistakes are possibly being made again this year. But to invoke two Yogi Berra lines, “it ain’t over till it’s over,” and it appears to be “Deja vu all over again.”

            The Princeton Election Consortium gives Biden a 93% chance of winning. The Economist says it is a 87% probability, and Nate Silver’s FiveThirtyEight, the darling of presidential prediction pundits, as of August 17, 2020 gives Biden a 72% chance of victory.    Others such as 270toWin given Biden ample electoral votes to  become the next president.  Let’s declare the election over, save ourselves a lot of time and money, and make Joe Biden number 46.

            But’s let look at this again.  In 2016 the Princeton Election Consortium’s final prediction was a 93% probability of a Clinton victory. The 270toWin site reported that Larry Sabato’s Crystal Ball, AP (Associated Press), the Cook Political Report, NBC, NPR, and others all gave final predictions of a Clinton victory. And FiveThirtyEight on November 8, 2016 (election day), gave Clinton a 71.4% probability of winning.  Obviously reports of Trump’s demise in 2016 were greatly exaggerated. The question is why so many predictors  made so many mistakes and potentially why they have not learned from their 2016 mistakes when forecasting for 2020.  As I pointed out in this publication in 2016, there are basic mistakes of analysis that pundits make and now prediction models are amplifying.

            Begin first with polling,  National polls do not matter because we do not elect the president by national popular vote.  Instead,  it is the electoral college that matters. US presidential elections are really 51 (50 states plus District of Columbia ) elections governed by different rules when it comes to voter eligibility and rules.  All that really matters is the race to get to 270 electoral votes.  Large popular vote leads in national polls may make one feel good but they do not necessarily translate into  electoral college victories. 

While most of the prediction models understand the electoral college issue, they nonetheless still  fail to appreciate that polling needs to be done at the state if not even at the county level to understand the micro trends impacting presidential elections.  As I have repeatedly pointed out, and do so again this year, it is a few swing voters in a few swing counties in a few swing states that will decide the presidential election.  More specifically, ten percent of the voters in 11 counties found in seven states will decide who gets to 270 electoral votes—10/11/7/270 is what the election is about.  Simply put, most if not all of the presidential prediction models work from polling at the wrong level of analysis.

A variation of this problematic analysis in 2020 is an argument being floated that  at some point the national public opinion  polls are showing such a large led for Biden over Trump (compared to Clinton versus Trump in 2016) that it necessarily means or translates into an electoral college victory for the former.  Nice theory but  not necessarily the case.  This argument is a form of the ecological fallacy where one tries to infer  the behavior of individuals based on group behavior.  Assuming that a really large national lead for Biden  will translate a victory for him statistically is wrong.

Second, as any good pollster will tell you, polls are snapshots in time and not tools of prediction or iron laws of certainity.   Polls tell us what will happen if an election were held today, not what is going to happen in 30, 60, or more days.  Lots of things can happen and change the political landscape.  In our heavily polarized era, it is probably not too many voters changing their views—there are really very few swing voters in the old-fashioned sense of changing partisan affiliations back and forth—but instead whether specific voters show up to vote or not.    Turnout mobilization is more of an issue than most models can predict.  They fail to capture how enthusiastic or motivated voters are, or assess properly what the remaining undecided voters in a few swing states will do when they decide to vote.  It was those who  did not vote or the undecideds who broke overwhelmingly in Pennsylvania, Michigan, and Wisconsin that decided the election by less than 85,000 votes.  Going into 2020, the models are making these mistakes again.

Three, all of these prediction models similarly err in thinking that campaigns and candidate strategies do not matter.  Consider Clinton in  2016. Bernie Sander beats her in caucuses and primaries in Minnesota, Wisconsin, and Michigan.  It then gets to the general election and what does she do?  Largely ignore campaigning in these three states and she winds up losing the latter two and almost Minnesota too.  In the closing days of the general election she runs off to Texas to campaign.  During the general election, as I calculated, Trump and Pence made far more campaign stops and appearances than Clinton-Kane, and in the critical Midwest, including Minnesota, Wisconsin and Michigan, Trump showed up and campaigned.  If campaigns matter, Trump proved that in 2016.   Presidential prediction models and pundits simply failed to predict bad campaign choices.

The same could happen in 2020.  So far Biden has run a lackluster campaign, assuming much like 2016 that being against Trump is enough.  Already Trump has appeared in Minnesota  on the first day of the Democratic National Convention while Biden announced a virtual campaign for  that state.

I remember back to May 2016.  I was invited to a national conference on combating corruption, hosted by Center for American Progress in Washington, D.C.  In attendance were the luminaries of DC insider politics, and me, the odd person out.  At one point the talk turned to  the inevitability of Clinton’s victory  and how her transition team was already formed and legislation being prepared.  I pointed out that from the vantage point of the Midwest her victory was not inevitable or even probable.  The scoffs were intense and  I was lectured on the certainty of the prediction models which foretold her victory.

The flaw in these models is to think predictions are iron laws of destiny.  They are not. Some pundits in the past chattered out “demographics are destiny” in arguing that it was inevitable that Democrats would enjoy majorities for decades as the US population diversified.  The quality of campaigns, choices by candidates, and other political variables can intervene to impact elections.  Going into the 2020 presidential general election which has only started,  a lot can happen.    It ain’t over till it’s over as Yogi Berra once said, and right now all the predictions are setting up for  them to miss the mark and be Déjà vu all over again.