Enough With the Polls, Let the Voters Decide

There are a lot of predictions and models trying to tell us how the presidential race will end up on the morning of Nov. 4, the day after voters head to the polls in-person to cast their ballots. There is a deluge of models and polls saying that former Vice President Biden is headed for an easy victory. There are also a number of smaller but sizable dissenting voices saying that a last-minute come from behind victory for President Trump is well within the realm of possibility like it was in 2016.

Many pundits and journalists feel the need to offer predictions on the outcome. Based on this number or this swing district or this voting bloc, candidate “A” is going to win. I used to traffic in such notions that we are able to predict the outcome of presidential elections in a country of 300+ million people holding fifty separate state elections. Four years ago, I would have said Hillary Clinton was cruising to an easy victory, just look at the polls! It doesn’t take a data scientist to see another President Clinton in the White House, she has almost a 100% chance of winning according to the HuffPo model! I would have been wrong, along with thousands of other writers, if I had put that prediction into words. Luckily, I didn’t, so I don’t have to look at my old writing and kick myself. I’m admitting it here, to be candid.

In 2020, it seems that we know less about what will happen on Election Day than we did in 2016, and that says a lot since we should be learning from the mistakes and bad guesses of four years ago.

With Covid-19 changing just about every norm in life, from where and how we work, to kids learning at home, to voting patterns and habits, it seems nearly impossible to predict how some hundred million people will vote after our lives have been turned upside down for some eight months.

Both paths, a Biden victory, or a Trump victory, seem entirely plausible. In 2016, a Trump victory seemed out of the question. This time, it would be malpractice to assume that the polls giving Biden a lead have much meaning after they terribly missed the mark four years ago. Yes, some polls were right, many national polls correctly called the popular vote for Hillary Clinton, but as we learned, that is irrelevant. It’s useful for judging broadly the public’s sentiment about the race, but it has no bearing on the outcome of the Electoral College.

The issue in 2016 seemed to be late deciders breaking for Donald Trump. In 2020, there are fewer undecideds, so they say, which means polls might be more accurate. Maybe it is so, maybe it’s not.

The two pollsters who weren’t shocked in 2016 are once again breaking with the grain of their industry and predicting a Trump victory:

Two pollsters who weren’t blindsided by this are Arie Kapteyn and Robert Cahaly. Kapteyn, a Dutch economist who leads the USC’s Dornsife Center for Economic and Social Research, oversaw the USC/Los Angeles Times poll that gave Trump a 3-point lead heading into election day—which, Kapteyn notes, was wrong: Clinton won the popular vote by 2 points. Cahaly, a Republican pollster with the Trafalgar Group, had preelection surveys that showed Trump nudging out Clinton in Pennsylvania, Michigan, Wisconsin, Florida and North Carolina—all of which he won.

This year, both men believe that polls could again be undercounting Trump’s support. The reason is “shy” Trump voters—people reluctant to share their opinions for fear of being judged. Though the “shy voter” idea is thrown around a lot by both Trump supporters and Democratic skeptics, Kapteyn and Cahaly have specific insights into why, and how, Trump support might be going undetected.

Ever since cell phones became the norm and landlines phones started to become a thing of the past, polling has become less reliable and less predictable. Pollsters wrote large articles on the subject, noting the vast differences in responses for similar questions between cell phone uses and landline users. Add “Do Not Call” lists to complexities of polling, especially with regard to cell phone, and it was clear that the days of live call polling was coming to an end.

Pollsters started using other means, like the internet, to reach voters, but that also has flaws. There is no single “correct” way to poll the general public, there is a mix of ways to poll and gauge interest in a candidate. The rest is some secret sauce each pollster comes up with to massage their numbers.

The “shy Trump voter” loomed large in 2016, and could return in 2020. That doesn’t mean there are enough shy Trump voters to push him over the top.

With turnout hitting record numbers during early voting, interest in this election is off the charts. At the moment, it’s leading pollsters to predict one of the largest turnout elections in over 100 years. That might be true, but it might not. With so many voters voting early and by mail, we don’t know how many will end up voting in person on Election Day. The number could be down from previous years because many voters decided to vote early or by mail. It could also be a blowout with large numbers pushing the participation rate well toward or surpassing 60% of eligible voters.

The most recent high watermark for voter turnout was Barack Obama’s historic win in 2008, with 57% of eligible voters casting a ballot. In 2012 and 2016, the number dipped to around 55% participation.

Could 2020 push that number higher as states provide more avenues to vote than standing in line on Election Day?

All of this is to say that predicting the outcome of the election, even for pollsters who seemed to have gotten it right four years ago is extremely hard. For that matter, pollsters like Trafalgar and USC Dornsife don’t have very long track records with their experimental polling methods. The Dornsife model has been tested to be accurate against several elections since then, but that doesn’t mean it’ll be accurate in the U.S. presidential election this year like it was in 2016.

Trafalgar seemed to poke and prod states like Wisconsin, Michigan, and Pennsvulannia and find that Trump’s support was understated in 2016 polling, and they came up with numbers that showed him winning. No one paid attention at the time, they seemed crazy.

They might still be crazy in 2020, we don’t know yet. Catching lightning in a bottle is hard once, let alone twice. That doesn’t mean it can’t be done, but it would sure be embarrassing for the entire polling industry if once again they were off so badly.

That brings us to the question of whether polling firms can afford to be off badly again. Some of them weren’t too wrong in 2016, only off by a few points, but the race was so tight that Trump ended up winning by less than a point in states where they said Hillary would win by 4 or 5 or 6 points.

If you look at it from that perspective, polling isn’t an exact science and was never claimed to be. Every poll has a margin of error, but few pay attention to it. Releasing poll numbers is the equivalent of saying “here are some numbers, but they might be off by a few points.” Most of the public and media tend to hear the first part, “here are some numbers,” but examining the cross tabs and margin of error is boring television and doesn’t make headlines.

I would like to see fewer poll numbers reported in headlines, and more analysis produced on the actual issues and where the issues resonate across the country. Polls start to become a tool to influence rather than reflect public opinion.

Take the most recent ABC News/Washington Post poll of Wisconsin. The results are staggering for Joe Biden, with a purported 17-point lead in the Badger State:

Wisconsin is leaning blue, with Democrat Joe Biden building a 17-point lead over incumbent Republican President Donald Trump in a new ABC News/Washington Post poll released Wednesday.

The news organizations say 57 percent of people who identified themselves as likely voters favored Biden, 40 percent favored Trump, and 2 percent favored Libertarian candidate Jo Jorgensen.

Even in the highest of high watermark years for Democrats, Barack Obama only managed a 14-point victory in 2008. Donald Trump outright won Wisconsin four years ago. For this poll, giving Joe Biden a 17-point lead to be credible, all context and history must be thrown out the window. Without any historical context, you could conclude that this poll seems legitimate. However, history screams that it seems unlikely, if not impossible, for Biden to win Wisconsin by 17 points. ABC News and the Washington Post have the same historical context, it’s somewhat surprising this poll would even be released given how much of an outlier it seems.

When does the constant flood of poll numbers become a thing of the past? I would say the answer is never because it would remove one of the easiest and quickest ways to gin up a headline that will garner clicks and attention.

Candidate “A” has a 17-point lead over candidate “B” will always draw eyes, even if only to scoff and shrug it off.

Let the people decide the election. It would seem that even in 2016, many Americans, including most Trump voters, were aware of the polls showing Hillary Clinton with high single-digit leads across the board. That didn’t seem to stop them from voting, but it did create a lot of anxiety leading up to the election. It also created a deep sense of mourning and betrayal on the part of Hillary Clinton voters who were being spoonfed data like this:

Looking at that chart, it’s obvious why so many Hillary supporters were devastated and caught completely off guard on election night. The models and predictions were so strong, how did this happen?

It happened because polling became gospel, and models like the one above are built on polls, so the models became even stronger gospel.

Let’s put the polls down for a while, let the voters breathe, and decide who will run the country for the next four years without being told who’s going to win.