The Polls Aren’t Wrong. Nor Were They Wrong in 2016

By John Banks

In recent days, we’ve written about the “Biden Sweep” scenario, and how Wall Street has come to embrace it as a bullish expectation.

The thing that Wall Street fears is a government spending cut-off that leaves the U.S. economy in a deflationary hole. If policy gridlock cuts off the anticipated flow of upcoming spending, permanent job loss and recessionary scars could take down the U.S. economy and the stock market, too.

If Democrats take the levers of power in the White House and Senate, on the other hand, trillions of dollars will flow in 2021. The polls now strongly favor a Biden sweep. In consistency, breadth, and durability, the available poll numbers say Biden has the most favorable outlook of any presidential candidate in modern history, dating back to when scientific polling efforts began in 1936.

The race was close at one point, but that dynamic changed after the first presidential debate. Not only does Biden lead by historic margins, as far as presidential races go, he is forecast as ahead in key swing states like Florida and Pennsylvania, and competitive in historic Republican strongholds like Arizona and Georgia.

The shape of U.S. Senate races is comparable: Democrats are on offense, and Republicans are on defense, in a number of tight races where, just a few months ago, the GOP incumbent was expected to win handily. At the same time, the Biden campaign has a significant cash advantage, and is reported to be outspending the Trump campaign 50-to-1 in some media markets.

Wall Street is not a sentimental place, and investors are not thrilled with the possibility of capital gains tax increases, a greater regulatory burden, and higher labor costs under a Democratic administration.

But Wall Street is looking past those factors in anticipation of huge spending programs — including more trillions in direct-to-consumer relief via stimulus checks and unemployment benefits, a fair portion of which will flow into the stock market.

A “Biden Sweep” would also go far in reducing the risk of post-election chaos. If Biden wins a crucial swing state like Florida on election night, for example, the race could actually be decided fairly quickly, with the post-election lawyering kept to a minimum.

There will almost certainly be a flurry of shouting and protests in the immediate aftermath of Nov. 3, but a drawn-out legal contest, paired with armed stand-offs in the streets, is the nightmare scenario. The stock market would prefer to avoid that scenario, and a Biden Sweep would help do that. 

Some of you have written in to question the Biden Sweep scenario, and to argue with conviction that President Trump will win the 2020 election, no matter what the polls say.

For those of you who wrote in to express the view that President Trump will win, a strong source of conviction was the fact that Trump, as the challenger, defied the polling odds in 2016. If the polls were wildly wrong in 2016, this argument goes, then the polls could be wildly wrong this time, too.

But here is the thing. The polls were not wildly wrong in 2016. They were more or less accurate — just as they were in 2018.

The science of presidential polling, which began in the mid-1930s, is far from perfect. But polling methodologies, as expressed by dozens of different outlets across the political spectrum, have developed a better track record than many realize. The 2016 result did not mar that record, though many think it did.

This may seem confusing. How could the polling have been accurate, or even close to accurate, in 2016 if Hillary Rodham Clinton was the dominant favorite, and yet Donald J. Trump turned out to win?

To put it another way, doesn’t the fact of a 2016 Trump win — when so many expected he would lose — demonstrate the polls and the oddsmakers were way off?

No, it does not. The fact of Trump’s 2016 win, and the manner in which he won — which is just as important as the win itself, as we shall see — were consistent with the data. That has meaningful implications for the polling data as it stands right now.

To be specific, it means that Wall Street is not being irrational in betting on a “Biden Sweep” as driven by an increasingly favorable outlook for Democrats taking control of the White House and U.S. Senate.

The breadth and consistency of the data here, across a majority of outlets — including multiple polling outlets that have historically favored President Trump, like Rasmussen Reports and Fox News — is reliable in its aggregate message. When you look at the total picture, it isn’t fake news. The Biden Sweep really is the dominant scenario at this point. 

Does President Trump still have a chance to win? And does the GOP still have a chance to retain the U.S. Senate? Yes, of course. A chance is always meaningful. That is part of the 2016 lesson, too.

To understand what happened in 2016, and why “the polls were wrong” is not the lesson to draw, it helps to think about the difference between probability and outcome. The probability of an event, versus the one-off outcome of an event, are wholly separate things. Let’s look at some thought examples.

Imagine I have a cloth bag with five marbles in it. Four of the marbles are blue, and one is red.

I offer you a prize: If you can draw the red marble on your first try, without looking, I will give you $10.

You reach into the bag. Success! You draw the red marble, and you win the $10.

Before making the draw, your odds of success were 20%. There were five marbles, but only one of them (the red one) could produce a win.

After the draw — even though you picked correctly — the 20% assessment remains true. You picked the winning marble, but you didn’t change the odds. For the instance in which you won, it was simply the case that the low-probability occurrence worked out.

Let’s look at a more detailed example. It is the final hand at the final table at the World Series of Poker Main Event, a No-Limit Texas Hold ‘Em event. Two players remain, with precisely equal chip stacks.

In this final hand, both players go “all in” — putting all of their chips in play — immediately after the three cards of the “flop” are dealt.

After going all in, both players’ hands are turned face up. The dealer prepares to reveal two additional cards, the turn card and the river. Whoever wins the hand also wins the whole contest.

On turning their hands up, the first player reveals pocket aces. The second player has an inside straight draw, meaning, they need to draw one additional card, on the turn or the river, to complete a “straight,” a hand composed of five cards in sequence.

At the point of going all in, what are the odds of either player winning the hand, and the tournament?

The first player — the one with pocket aces — has an 83% chance of winning. She does not need to improve her hand to win. She just needs to avoid being “drawn out on,” meaning, the scenario where the other player draws the needed card to complete an inside straight.

The second player — the one with an inside straight draw — has a 17% chance of winning. He is behind on the flop, but has two opportunities, via the turn card and the river card, to complete his straight draw.

With two cards to come, the player drawing to the straight has a 17% chance to win, because of the way the math works out. There are four cards out of 45 that will complete the straight — which wins if it completes — and two opportunities (the turn and river) to draw.

Under these circumstances, the player will get their straight roughly 17% of the time. The other 83% of the time, the draw cards don’t change the outcome of the hand, and the other player wins. 

Now, let us say that, in this particular instance, player two gets the straight. He draws one of the four cards needed, and wins both the hand and the event. He wins it all.

If the draw is completed, and player two wins, would that mean the poker announcer was wrong in stating that, at the time of going “all in,” player one had an 83% chance?

Would it have been inaccurate for the announcer to say — before the turn and river were dealt — that player two had only a 17% chance of winning?

Again, the answer is no. The assessment of 83% versus 17%, in this example, would have been correct. Drawing to the straight was the low-probability event. But sometimes the low-probability event works out. That is what it means to have a chance: If an event is mathematically likely to occur 17 times out of 100, then roughly 17 times out of 100 it will actually take place.

When Donald J. Trump won the presidency in 2016, he did so with the equivalent of drawing to an inside straight. His path to victory was highly unlikely, but the numbers worked out.

This is where path dependency matters. The manner in which Trump won shows that it made sense to see Clinton as the favorite, just as the player with pocket aces in our given example would be the favorite — even though, 17% of the time, they still lose.

In the 2016 presidential election, nearly 129 million votes were cast for the Democrat and Republican candidates, and Hillary Clinton won the popular vote tally by more than 2.8 million votes.

But Trump won the electoral college — the thing that matters — by a margin of fewer than 80,000 votes across three crucial swing states: Michigan, Pennsylvania, and Wisconsin. If not for that razor-thin path to victory, Clinton would have won the Electoral College. She didn’t win, of course, but the point is understanding the shape of probabilities versus the actual outcome.

In an election where nearly 129 million votes were cast for the top contenders, and the Democrat won the popular vote, Trump’s margin of victory was dispersed across exactly the right states, with a small enough vote count, in terms of the deciding votes, to fit inside Ohio Stadium with 25,000 seats to spare.

Polling prognosticators who gave Clinton a better than 80% chance of winning certainly felt foolish, and shocked, the day after the election. But based on how the contest played out, they were not wrong. If a number of factors had not lined up exactly right, things would have gone the other way.

The point here is two-fold. First, the 2016 polling data was not invalidated by Trump’s victory. That has implications for the present day, and for the accuracy of polling data in general. The data remains useful, and believable, when it comes from a wide array of sources over a meaningful period of time.

Second, it doesn’t make sense to say, “The polls are wrong now because they were wrong in 2016” — because the polls were not all that wrong. The media was shocked at the result, to be sure — while some who expected Trump to win were not — but that is wholly separate from the odds of the thing.

As for the 2020 election, there is reason to argue Trump is in the same situation he was in before.

Based on the numbers we are seeing, across a wide range of sources — again, including multiple polling sources historically favorable to Trump — the president is once again the underdog, drawing to an inside straight.

This seems largely due to the impact of the presidential debate, and the events that unfolded after it.

Now, can President Trump still win? Of course he can. A double-digit chance is still very much a chance. That is the other message of 2016: If someone has any kind of double-digit odds of winning, from, say, 10 to 30%, they are still in the game in a major way.

Professional poker players, who experience probability match-ups over and over again, understand this reality intuitively. Something with odds of happening 10% of the time will actually happen a fair amount of the time. Ten percent is not zero. If you play a lot of poker, you will see 10% outcomes routinely.

The public does not intuitively grasp this, however, because probability is a weird and bizarre thing, especially in relation to historic one-time events.

“Well if he drew an inside straight once,” some of you might say, “maybe he can do it again.”

To which we say, again: Yes, of course! There is no reason to assume the contest is over or the die is 100% cast. But at the same time, odds are odds, and probability has no memory, which means the outcome of a past event doesn’t impact the next one.

President Trump also has obstacles that would be tough for any presidential incumbent to overcome.

The single biggest one is probably the fact that incumbent presidents tend not to get re-elected if the election comes in the aftermath of a recession.

President George H.W. Bush arguably lost the 1992 presidential election because of recession conditions. Bush had a booming stock market, but unemployment was high. President Jimmy Carter, who lost to Ronald Reagan in 1980, was also dogged by recession woes.

This matters for 2020 because, not only is the United States working through a recession, a significant portion of the country is experiencing what feels like a Great Depression, complete with food bank lines and vanished job prospects for tens of millions of service-sector workers.

The pandemic is also impacting the election in ways that are harmful to the incumbent. For example, the senior demographic — voters aged 65 and older — are trending Democrat over Republican for the first time in decades, in large part because they are upset with the pandemic response.

Seniors are the group most likely to experience severe impacts from COVID-19, and also the group that most intensely misses the ability to hug their grandkids. That matters a lot for Florida, a must-win state for Trump where Biden is now polling ahead. 

Again, President Trump can still win. We aren’t saying otherwise, and neither is Wall Street — the market moves decisively in favor of dominant-odds scenarios on a regular basis (and changes tack on a dime if something different happens).

But the polling data has shifted substantially against Trump in consistent and credible ways, across too many data outlets to ignore, and the claim that “the polls were wrong in 2016” is not correct. It is a misread of probabilities versus outcomes — and the data right now really does favor a Biden Sweep.