Why polls got it wrong: shy Trumps, flaky Hillarys and Facebook events.

Electoral polls totally failed to predict the winner of the American presidential election. The intellectual elite was telling us that Hillary was going to win. But they got it wrong, from political experts to Ivy League statisticians. All their sophisticated models and analysis were useless. American voters decided that their destiny will take a different path.

Which reality did they miss? The answer is : shy Trumps and flaky Hillarys.


1. Shy Trumps

During the electoral campaign, Trump supporters were repressed by political correctness. They were constantly degraded as white supremacists, low-income, uneducated, angry white males.

On the land where the First Amendment was proclaimed, they were the victims of an intolerance campaign, originating from so-called ‘tolerance supporters’, people who pretended to defend pluralism and the right to be different. This intolerance even became violent:

A Maryland high school student wearing a red “Make America Great Again” cap was injured and taken to the hospital Wednesday morning after he scuffled with a group of anti-Donald Trump protesters who punched and kicked him, according to police. (…)
A parent who looked on as the students marched said she saw the Trump supporter belatedly join the marchers and begin to argue with them. She said she tried to defuse the situation and that the student argued with her, too. She said he told her, “I have my right to free speech too,” and that he looked at the crowd and said, “Go home.” She did not see the later physical confrontation.
Shy Trump after his coming out

Which consequence for the polls? Due to the social pressure, some Trump voters were too shy to admit their intention to the pollsters. Polling companies were perceived as belonging to the ‘other side’, as being representatives of the liberal establishment they were seeking to overthrow.

This phenomenon is not new, it is called ‘social desirability bias’. It is known as the ‘Bradley effect’ in the US, and as the ‘shy Tory factor’ in the UK.

Some data analysts still dispute the effect of Shy Trumps. Anyway, a second effect remains: flaky Hillarys.


2. Flaky Hillarys

What about Hillary voters? Apparently, some of them were like Hillary herself: unreliable, impossible to trust, and lacking emotions. They did not attend to her rallies, nor to the polling station.

Hillary meeting flaked by her supporters

Despite their so-called ‘stupidity’, Trump supporters were still smart enough to make their way to the polling station, and to vote for their candidate.

Shy Trump voting at the polling station

At the same time, many potential Hillary voters preferred to stay at home and watch TV:

Flaky Hillarys staying at home on election day

Which consequence for the polls? Some people were telling to pollsters that they will vote for Hillary, but on the election day, they lacked the enthusiasm to make a move to the polling station.


3. Attendance Prediction of Facebook Events

When we think about it, these shy and flaky behaviors happen all the time: if we look at any Facebook event, the number of people registered as ‘going’ is always over-estimated. Like shy Trumps, some people have no intention to show up at the event, but they still click ‘going’, because they are shy. They do not dare to confront or disappoint their Facebook friends.

On the other hand, like flaky Hillarys, other people kind-of-wish to come to the event, but ultimately, they flake. They don’t have enough motivation to show up.

The result? When we see a popular event on Facebook:

we no longer expect to find a large crowd:

because the reality can be different:

Such situations are frustrating for the people who really show up. They also give headaches to event organizers, who need to guess the real audience size. I faced this problem myself, when organizing the Kyiv Deep Learning meetup in Ukraine, a country with a very flaky culture.

The solution? An app that predicts event attendance: https://www.showup.tech

And with a better prediction of event attendance, we might better predict voter attendance. So that the disaster of the 2016 election never happens again.