Even on Election Day, analytics site FiveThirtyEight and its founder Nate Silver, which accurately forecasted Obama’s victories in 2008 and 2012 – prompting questions over whether Silver was the most accurate contemporary pundit – pegged Hillary Clinton’s chances of winning the presidency at 71.4%. And it certainly wasn’t the only outlet predicting a Clinton victory.
In a blog post, Silver called President-Elect Donald Trump’s victory “the most shocking political development of [his] lifetime.” He added: “In an extremely narrow sense, I’m not that surprised by the outcome, since polling — to a greater extent than the conventional wisdom acknowledged — had shown a fairly competitive race with critical weaknesses for Clinton in the Electoral College. It’s possible, perhaps even likely, that Clinton will eventually win the popular vote as more votes come in from California.”
A prominent SEO practitioner, who asked not to be named, said the election results also demonstrate how current polling methods are clearly flawed.
“They basically measure people with landlines who answer unknown numbers in the middle of the day,” he said. “I'm nothing like those people.”
Jay Friedman, COO of programmatic media planning and buying firm Goodway Group, agreed.
“Approximately 50% of the polls are phone-driven. Home phones are disappearing, cell phone users can easily block unwanted numbers and online polling hasn’t become sophisticated enough or fully normalized yet,” Friedman said. “In short, we need to figure out how to poll in today’s media usage landscape.”
That being said, according to Pew Research Center, 13% of Americans do not use the internet. With a US population of nearly 319 million, that’s more than 41 million people.
“How do most of the statisticians collect the data? Mostly via the internet,” wrote Dixon Jones, marketing director at marketing search engine, and SEO backlink checker Majestic, in an email.
“How disenfranchised are you if you … aren’t even on the Internet? My guess is …VERY. So the main lesson here is that you cannot ignore the people you cannot reach in polling! They probably are not an unbiased group! On top of this – the election was so vicious, people probably didn’t want to openly say their preference. Until America heals, that will continue I expect.”
Jonathan Chanti, senior vice president at influencer marketing platform Hypr, agreed.
“Polling is an outdated format for collecting data … people are … reluctant to share their feelings, which requires a deeper level of data and analytics on social patterns and interactions to understand what they are really thinking,” Chanti said. “This is why Trump [used software company] Cambridge Analytica to guide his decisions on marketing, and approached the campaign [in a] more targeted [fashion], and ultimately won.”
In addition, if the people surveyed aren’t honest about their opinions, Chanti noted the data is meaningless.
Further, Momentum CTO Jason Snyder noted an AI system called MogIA predicted a Trump victory last month.
In a story in Forbes, an incredulous Kevin Murnane even wrote: “This outcome appears exceedingly unlikely and it makes you wonder if we aren’t seeing the digital age version of the Chicago Daily Tribune’s notorious ‘Dewey Defeats Truman’ headline ‘reporting’ the outcome of the 1948 presidential election.”
And, funny enough, Forbes reported the 1948 prediction was wrong because “the people sampled in the polls were not representative of the American voting public” because data was collected via phones, which were more likely to owned by wealthy Republicans, and the pollsters gathering data had a lot of freedom to choose who they wanted to interview.
Further, per Forbes, the recent MogIA prediction was based on 20 million data points taken from US social media engagement. That’s because the candidate with the higher engagement rating has won the past several elections.
And this, in turn, begs the question whether we should be using AI and social for more accurate polling in future elections.
Snyder said algorithms might have had a hand in this, too, “serving up content that didn’t offer enough understanding of what’s really at stake.”
Bots also helped Trump win the election “in that what people were seeing wasn’t being generated by people at all,” Snyder said. “There was data indicating that one-third of all pro-Trump mentions were generated by bots. That’s important because mention volume on Twitter has been a pretty reasonable indicator for the outcome of elections and there’s been a lot written about it. Additionally, many automated sentiment analysis systems cannot identify sarcasm, which, in the case of Trump, is a meaningful and significant piece of data. So perhaps the combinations of AI used this year yielded some results that were misinterpreted.”
Further, Russell Jones, principal search scientist at Moz, noted our polling models don't take into account how easy/hard it is to vote.
“People's intentions … are thrown off if it becomes harder to vote. As reported here, there are [868 fewer polling places] because the Voting Rights Act was gutted. [North Carolina], for example, cut 27 locations,” he said. “Arizona and North Carolina, two of which had the most poll closures (and North Carolina which had last minute repudiations of even more egregious voting restrictions) alone were enough to make up the difference in the Electoral College to switch the election.”
At the same time, Moz's Jones said the outcome is not that surprising.
“The polls were right in that [Clinton] will have won the popular vote,” he said. “Where they were wrong were in swing states where things like voter suppression could make the difference, since the candidates were already within the margin of error.”
Friedman, too, said he doesn’t believe the data necessarily fell short.
“FiveThirtyEight had Clinton at a ~71% chance of winning as polls opened. That means that if we ran 100 simulations of the election, Trump would win 29 times,” he said. “When we play roulette in a casino and it lands on 0, we don’t say the data failed. Red/black in roulette has better odds than Clinton did, but a chance of greater than 0 means exactly that.”