Making Predictions In A New Political Climate

Incorporating emotion into analytics models


Polling is a notoriously inexact science, and previous elections have seen pollsters get it tremendously wrong. Neil Kinnock’s loss to John Major in the 1992 UK elections and Alf Landon’s to FDR in the 1936 US presidential elections, in particular, stick in the memory. The 2008 US presidential election, however, was thought to herald a new dawn in political predictions, with renowned statistician Nate Silver offering another way that has, until now, proven extremely accurate.

Silver was one of the pioneers of sabermetrics in baseball, developing his PECOTA (Player Empirical Comparison and Optimization Test Algorithm) system to predict the future performance and valuation of major league players by comparing their data with 20,000 post-WW2 players. In 2008, he began applying his quantitative methods to politics, correctly calling 49 out of 50 states in the 2008 general election. In 2012, he called all 50 correctly. It seems, however, that Silver’s methods may finally have proven fallible.

The US elections have long delighted, confused, and entertained the world in equal measure, thanks to their extraordinary length and the often bizarre array of characters on show. This year’s Republican and Democratic primaries are, however, making previous years look like a quiet, orderly discussion between old friends. The fervor surrounding both Donald Trump and Bernie Sanders has shaken the foundations of the political establishment, rendering any attempt by pundits to second guess the results a near impossibility.

Donald Trump, widely considered a no-hoper when he started, is now a front runner in the race for the Republican nomination. Even with his repeated gaffes, often ludicrous remarks, and global mockery, his poll numbers have increased. This has caused panic in both liberal and conservative circles, with liberals seeing him as a dangerous lunatic, and conservatives seeing him as being unelectable when it actually comes down to the real presidential contest. Democratic hopeful Bernie Sanders was similarly considered a no-hoper, a sideshow in a primary that was supposed to be a mere formality for Hillary Clinton’s coronation. But Bernie fever has gripped the nation, particularly amongst the young, in a similar way that Corbynmania helped sweep socialist MP Jeremy Corbyn to the leadership of the UK Labour Party earlier last year. Sanders’s rallies have drawn huge crowds, and he has raised record amounts of money from individuals, dwarfing others in the contest.

Trump and Sanders in the US, as well as Corbyn and UKIP’s Nigel Farage in the UK, all represent a new kind of politics. This politics is characterized by mistrust in the establishment, and its engine is social media and the internet. Pundits have yet to really acclimatize to this new environment, and none more so than Nate Silver. While panic has set in over Trump’s ever-growing poll numbers, Silver has help steadfast to his conviction that Trump will not win, providing a supposed voice of reason in the insanity and fear that such a buffoon could potentially become the leader of the free world. In September, he told CNN’s Anderson Cooper that Trump had a roughly 5% chance of beating his GOP rivals. Obviously, Trump’s poll numbers have since soared, and his second place finish in Iowa, while not exactly what Trump would have wanted, still makes Silver’s 5% look vaguely ridiculous.

Silver himself recently admitted as much, acknowledging in a blog post that he’d been too skeptical about Trump’s chances: ‘Things are lining up better for Trump than I would have imagined. If, like me, you expected the show to have been over by now, you have to revisit your assumptions.’

Blake Zeff, the editor of the political news site Cafe and a former campaign aide to Obama and Hillary Clinton, has warned of the dangers of trying to make predictions based on models created in old political environments. Jeff said: ’This is an extraordinary, unusual, utterly bizarre election year, in which events that have never happened before are happening. That’s a nightmare scenario for a projection model that is predicated on historical trends.’ What was true yesterday is not necessarily true today, and that’s a problem for Silver and his team. Accounting for emotion is a difficult task in data analytics. Cold, hard numbers cannot take into account the sort of fervor that we are seeing, and a way must be found of doing exactly that.

How exactly they go about this it is difficult to know. Greater analysis of social media is an obvious place to start, but this focuses primarily on the young, an age group traditionally far more left wing than older generations who do not use social media to anything like the same degree yet vote in far greater numbers. The problems of relying too much on social media to try and predict elections was seen last year in the UK, with many arguing that one of the reasons the Conservative victory caught so many by surprise was that people were looking at social media too much.

One idea for how it could be taken into account, ironically, comes from the very man who managed to beat Donald Trump in the Iowa Caucus - Ted Cruz. Much of his success has been put down to an excellent ‘ground game’, an old school method of campaigning that was foresaken by many others in the race. Cruz’s team, however, gave it a modern twist, using a team of statisticians and behavioral psychologists to employ something called ‘psycho­graphic targeting’, in which campaigners alter the way they deal with potential voters based on a psychological and political profile created using information collected about that individual. His campaigns use of data is also ironic because Cruz has been a heavy critic of excessive government data collection, but maybe consistency didn’t show up as a big voter issue.

According to The Washington Post, the Cruz campaign has employed Massachusetts-based Cambridge Analytica to run the data-side of its operations. To develop its psychographic models, Cambridge surveyed more than 150,000 households across the country and scored individuals using five basic traits: openness, conscientiousness, extraversion, agreeableness and neuroticism. According to Cruz campaign officials, the company also used social media to do this, developing its correlations in part with Facebook data such as subscribers’ likes. The Cruz campaign then modified the Cambridge template, renaming some psychological categories and adding subcategories to the list, such as ‘stoic traditionalist’ and ‘true believer.’ The campaign also did its own field surveys in battleground states to build a more precise predictive model based on issues preferences. The Cruz algorithm was then applied to what the campaign calls an ‘enhanced voter file,’ which can contain as many as 50,000 data points gathered from voting records, popular websites and consumer information such as magazine subscriptions, car ownership and preferences for food and clothing.

The Cruz campaign has utilized all this information to make a concerted effort to tightly tailor outreach to individuals. For example, personalities labelled ‘stoic traditionalist’ are believed to be highly conservative, and would be spoken to in a way that was ‘confident and warm and straight to the point’, because that was the one deemed would have the greatest impact. Even campaign e-mails are tweaked according to this research. In emails, ’stoic traditionalist’ would receive very direct and to the point messages, whereas someone labeled ‘temperamental’ would receive a message that was inspiring, and became more and more positive as the conversation progresses.

Could analysts look to similar methods to more accurately predict results? Incorporating emotions into prediction models looks like a difficult task, but such psychological testing could go some way towards better gauging the tide of public sentiment. Trying to incorporate an entity as wildly erratic as Donald Trump into prediction models, however, may simply prove impossible, even for Nate Silver. 

University lecture small

Read next:

How Are Higher Education Institutions Using Analytics?