After two years of political posturing and endless campaigning, the 2020 election season is coming to a close. We will soon know whether we will have another four years of the 45th president (Donald Trump) or, alternatively, a 46th president (Joe Biden).
According to conventional wisdom, President Trump is the underdog. The national polls have former Vice President Biden up by roughly eight points, and most election forecasts predict that he will win handily. Biden is also leading in public-opinion polling of the battleground states, though his advantages there are marginal relative to his purported national lead.
Between the Economist model that gives Trump a 4% chance of winning and any given POLITICO headline, the commentariat’s confidence has reached such a fever pitch that it is evocative of one of the most egregious examples of hubris in American history: the November 3, 1948 headline of the Chicago Tribune, “Dewey Defeats Truman.” In actuality, President Harry Truman dispatched his Republican challenger and made a mockery of polls that had him behind by as much as 15 points.
The open question, then, is whether Trump will replicate Truman’s upset or, rather, become the first president to lose re-election since George H. W. Bush in 1992. My data and intuition point to a Trump re-election being more likely than not, with the president leading 305 to 233 in my estimation of the Electoral College vote. This is not to say that Trump is heavily favored, but he has a slight advantage.
Have the Pollsters Cleaned Up Their Act?
There are two primary groups of people: those who look at the polling errors of 2016 and say, “I will never trust the polls again,” and those who pretend that 2016 never happened. I am in the middle of these camps.
The 2016 polls accurately called the winner of the national popular vote, Clinton, while overestimating her margin by about a percentage point. Pollsters were far less accurate in the Rust Belt, where Clinton’s support was overstated. In fact, only one pollster—the Trafalgar Group—managed to call Trump’s surprise victories in Michigan and Pennsylvania.
2018 was not much better for the mainstream pollsters, who miscalled important races in Florida, Iowa, Indiana, and Ohio, and generally underestimated Republican support in peripheral contests (e.g., the Ohio and Michigan Senate races). All things considered, it is not clear that the mainstream pollsters have rectified their polling issues in the Trump era. And when polls are within the margin of error, I consider them to be a nonfactor in predicting election results.
Trump’s Minority Support:
Over the past four years, Trump has waged a targeted campaign for the support of non-white voters, all the while keeping his eye on the white, working-class voters who propelled him to victory in 2016. It is my prediction that Trump will maintain his 2016 support among white voters, while expanding his support among black and Hispanic Americans.
Individuals predicting a Trump loss often theorize that he has lost support among white, working-class voters, but I am reluctant to draw this conclusion. While a recent POLITICO article cited several polls suggesting that Trump has lost such support, the theory is reliant on anecdotes and false equivalencies. 2016 polls understated Trump’s support among white voters, relative to his performance in exit polls. Thus, comparing 2020 polls to 2016 exit polls—instead of 2016 pre-election polls—is intellectually dishonest. Additionally, given that nearly two-thirds of those who did not vote in key Rust Belt-states are white members of the working class, I am convinced that for every suburbanite and voter over 65 who is flipping to Biden, there is a Trump voter in waiting.
Furthermore, Trump is likely to command more support among voters of color than any Republican candidate in the post-Cold War era—with the exception of George W. Bush in 2004. Consider that Rasmussen Reports, the most accurate pollster in 2016, pegs Trump’s support level among black Americans at 30%. Emerson College, which found that Trump is at 19.4% support among black Americans, is a bit more modest. However, even 19.4% exceeds Trump’s 2016 support among black Americans: eight percent.
State polls affirm these numbers. One example is the University of New Orleans’ poll of Louisiana, which found that Trump enjoys 28% support among black Americans overall and a one-point lead against Biden (43% to 42%) among black males. This is a major swing from the university’s 2016 poll, which found that Trump had just 4% support among black Americans in Louisiana.
Trump is also performing very well among Hispanic Americans. Emerson pegs his support level at 40.7% among Latinos, which is a major improvement from 29% in 2016. State polls tell a similar story, with Marist College finding Trump ahead by six points—52% to 46%—among Florida Hispanics.
Trump Leads in the Battlegrounds:
In the United States, presidents are elected by the Electoral College, a group of 538 men and women who are selected via democratic elections in each of the 50 states and Washington, D.C. And this system necessitates that candidates win states that are consistently competitive—i.e., “battlegrounds.”
In 2020, there are essentially six battlegrounds—Arizona, Florida, Michigan, North Carolina, Pennsylvania, and Wisconsin—and one peripheral contest, Minnesota. Out of those seven, I estimate that Trump will win six (AZ, FL, MI, N.C., PA, and WI) and lose one (MN), giving him a 72-vote win in the Electoral College. Here are my underlying assumptions about these states:
Arizona—Trump: The 2018 election of U.S. Senator Krysten Sinema, a Democrat, is a commonly-cited rationale for Biden winning this state on November 3rd. However, midterm elections are historically brutal for the incumbent party. Consider that, in 2010, Democrats lost several Senate seats in states that Obama went on to win in 2012. Additionally, Republicans have a party-affiliation advantage; Democrats have not been successful in turning out the mail-in vote, a metric for enthusiasm. It also does not hurt that Trump increased his support among Hispanic Americans, a key constituency in Arizona. Polling is a wash, with the advantage seesawing between Trump and Biden in recent days.
Florida—Trump: There are few factors cutting in the Democrats’ favor in the Sunshine State. Every incumbent president has won Florida since Ronald Reagan in 1984, including George H. W. Bush in 1992, and Republicans won the governorship and a U.S. Senate seat in 2018. While Democrats have a party-registration advantage, the polls are within the margin of error and the Republican registrants trail their Democratic counterparts by just two points in early voting. One factor could admittedly threaten Trump’s reelection—Americans over 65 turning against him because of his handling of the COVID-19 pandemic—but this shift is likely to be counterbalanced by Trump winning over Hispanics who did not vote in 2016 or who previously voted for Clinton.
Michigan—Trump: Republicans are the historical underdog in Michigan and, despite the gubernatorial victories of Republican Rick Snyder and Trump’s surprise win in 2016, Biden initially appeared to be favored here in 2020. Also to Trump’s detriment, Democrats handily outnumber Republicans in terms of party affiliation. And Republicans performed poorly in Michigan in 2018. That said, the Trafalgar Group has Trump up by two points, Republican-affiliated voters constitute a plurality of the early votes cast, and Biden’s lead in the polling average is within the margin of error. This was the closest call of any state I assessed.
Minnesota—Biden: The Land of 10,000 Lakes could be the most closely-fought state in the 2020 election. It went for Clinton by 1.5 points in 2016, and current trends indicate that Biden will just barely continue the Democrats’ 48-year winning streak. Biden’s lead in the polls is within the margin of error, but Democrats had a good 2018 in Minnesota and they are holding their own in early voting.
North Carolina—Trump: Public-opinion polling shows a slight Biden lead, but it is well within the margin of error—and polls in the Old North State consistently underestimate Republicans. Democrats lead marginally in the number of early votes cast, though this state has ancestral Democrats who may be voting for Trump despite their party registration. The president’s relatively large 2016 winning margin, 3.7 points, makes me think he will overcome countervailing winds and once again win the state’s 15 electoral votes.
Pennsylvania—Trump: Polls here are within the margin of error, and rudimentary analysis makes it clear that Trump’s support is being understated. Trump will also benefit from increased support among black Americans. Democrats have a massive early-vote lead, but there is an enthusiasm advantage for Trump and his voters are likely to pull through on Election Day—just as they did in 2016, when they made him the first Republican to win Pennsylvania since 1988.
Wisconsin—Trump: Biden leads Trump by roughly six points, but this advantage is within the margin of error and Trump outperformed his Wisconsin polls by 7.2 points in 2016. Trump’s chances are boosted by the Democrat-to-Republican parity in partisan affiliation, Republican Governor Scott Walker’s near-win in 2018 (despite a toxic national environment), and an eight-point Republican advantage in early votes cast. I will be holding my breath for the final poll from the Trafalgar Group, but it is clear that Wisconsin is an increasingly Republican state where Trump has a small advantage.
Final Assessment:
Thanks to elevated support among black and Hispanic Americans and stable support among white Americans, Trump maintains a slight, 72-vote advantage in the Electoral College. Trump’s losses relative to 2016 will be marginal, and he is likely to counteract them by increasing his support among demographics that have not traditionally voted Republican and by turning out some white, working-class voters who stayed home in 2016.
It is unlikely that Trump will win a plurality or a majority of the popular vote. However, recent polls have gone a bit haywire in projecting double-digit leads for Biden. Even the “mainstream” election forecaster Nate Silver warned against this phenomenon, noting, “[T]here’s usually quite a bit of herding in the final national polls, as people are often reluctant to stray by more than 2-3 points in either direction from whatever consensus develops.” I am willing to take Silver at face value and declare that Trump will trail in the popular vote by a maximum of three or four points, not seven or eight as recent polls have suggested.
Finally, if my predictions materialize and Trump wins tomorrow, the 2020 election will be an extraordinarily embarrassing “Dewey Defeats Truman” moment for the mainstream consensus. It could even mark the end of the mainstream polling industry—and the beginning of an era dominated by free-thinking, independent pollsters like the Trafalgar Group.
Hi Declan, I’d love to understand more why/what you think the Trafalgar group is doing differently? I find the methods used by most polls as reasonable, but maybe with larger error bounds than they’d admit.
The Trafalgar Group attempts to identify silent Trump voters by asking a series of questions. Their methodology made them the only pollster to accurately call the 2016 results in MI, PA, and WI.
The doublethink among Republicans is that Trump’s massive rallies are a sign of his strength, and yet at the same time his base is “silent” in polls. Trafalgar’s method is to ask people who they think their neighbors will vote for, in an attempt to correct for the phenomenon of respondents answering the way they think the pollster wants to hear. There are many reasons to regard this with scrutiny. Some are intuitive – it is unlikely that Trump voters’ desire to give the response that pollsters want to hear, if it exists, is stronger than their desire to “own the libs”. Another reason is that Clinton’s defeats in the “blue wall” states (Michigan, Pennsylvania, Wisconsin) are attributable more to depressed turnout in cities than to higher Trump turnout. Trafalgar is correcting for the wrong phenomenon here. Polls have to be off by WAY more than they were in 2016 for Trump to win. I think that’s possible, but if it happens it will be because turnout for Biden was depressed.
1. Hillary Clinton got solid turnout/margins out of Philadelphia, Detroit, and Milwaukee for a Democrat, just not the unique turnout that Barack Obama was able to achieve amongst inner-city African-American voters. Her bigger issue was the trends away from Democrats in blue-collar historically Democratic leaning areas like Erie County, Pennsylvania, Luzerne County, Pennsylvania or Macomb County, Michigan. So far in 2020, all the evidence ranging from voter registration trends to donations points to a better performance for Trump in those types of areas. Higher turnout in counties like Bay County, Florida and Cambria County, Pennsylvania relative to their registered voter pool also helped Trump as much as normal but not Obama-level turnout in urban areas for Hillary Clinton.
“Hillary Clinton got solid turnout, just not the kind that Barack Obama relied on to win those states” is not the own you think it is. Suburbs were a much bigger problem for Clinton than blue collar areas that have been trending Republican for a while now. There were only two counties in Pennsylvania that voted for Obama in 2012 and Trump in 2016. Moreover, if we recognize that a large part of Biden’s popularity is because of his association with Obama, I imagine he’d have a much easier time winning Obama’s coalition than Clinton did. This, together with the lack of an October surprise in the vein of the Comey letter, leads me to believe that Biden does not have the same problems Clinton did. He might not recreate Obama’s coalition, but I think he will do a better job.
2. You do realize that just because you show up at a rally does not mean you are comfortable telling a pollster or co-workers that you are voting for Trump. That statement shouldn’t be rocket science. It is no secret that those who most strongly believe that there are secret Trump voters are suburban women which can otherwise be describes as “polite society”. This is the group that the Democrats are projecting ridiculous increases for Biden. While Trump is no doubt weakening amongst suburban white women, I highly doubt he is bludgeoning amongst those voters as much as polling suggests. More important than lying to pollsters about voting intentions is low response rate and specifically a low response rate amongst those most likely to support the president.
Your conflation of a pollster that someone does not know with a co-worker who someone spends a lot of time with is concerning. If the “shy Trump voter” effect were true, then we would expect to see this effect practically erased by online polls which are both anonymous and don’t require a person to speak to anyone. We do not. Again, the final results will certainly be less decisive for Biden than polling suggests, but your response does not deal with the fact that it has to have gotten *worse* since 2016, despite the fact that pollsters have now corrected – probably too much so – for failing to sample Trump voters. Moreover, the undecideds two weeks out, who were around 15%, broke heavily for Trump in 2016. This is a far more convincing explanation for why he won than the “silent Trump voter” effect. This number was three times as high in 2016 than 2012, and it looks like this year resembles 2012 more than 2016 in that respect.
3. Put it this way, a neverTrump Republican is more likely to respond to a poll that a blue collar Trump Democrat, which explains why polls show the Democratic Party more unified than the Republican Party in the Rust Belt when the opposite is the case. In short, it is very difficult to reach a voter in northwest Minnesota, the north-central region of the T in Pennsylvania, the Upper Peninsula of Michigan, and northwest Wisconsin. Thus, it is very difficult to get a sufficient representative sample of voters in these type of areas which is the goal of polling unless you stay in the field for a significant period of time. Lastly, an independent voter in Cambria County, Pennsylvania votes very different from an independent voter in Delaware County, Pennsylvania and the problem with many of these polls is that the independents they chose to sample were more likely to be in the latter category than the former.
This is the most coherent argument of the three you gave. Without access to the polling data, I cannot say one way or the other whether, for example, ABC/WaPo decided that they were going to oversample blue-collar towns to account for the effects you describe. But it doesn’t matter. Your explanation for why someone would be willing to own MAGA gear and attend Trump rallies but not to answer anonymous polls over the phone or online is lacking; especially considering that *Donald Trump is the president right now* and his supporters have only gotten more vocal since then. Put simply, your silent Trump voters were people who didn’t know they’d be Trump voters until two weeks before 11/8/16.
So this aged pretty well (to be fair, so did predictions that Biden would carry or come close to Trump in states like Ohio and Florida)
Fair, but a lot of my assumptions ended up being accurate, e.g., President Trump winning a bigger share of the minority vote than any modern Republican with the exception of George W. Bush and not having much slippage among older voters.
As far as my electoral map, I would roughly compare my accuracy to that of the Cook Political Report: https://cookpolitical.com/analysis/national/national-politics/bidens-path-270-widens-trumps-path-narrows-texas-moves-toss.