No Hope, No Cash, No Jobs, Now Casting Trump

“Kevin Bacon, please don’t die” The punchline of the old meme leaps to mind when I try to make sense of the US presidential election. “We used to have Bob Hope, Johnny Cash and Steve Jobs, now we have no hope, no cash and no jobs” goes the setup. Now, it appears the American voters hope a Trump card will fix all of the above (perhaps minus the bacon).

I used to like Donald Trump’s TV-show The Apprentice. In fact, I liked it so much, I read two of his books: How to Get Rich and Think Like a Billionaire (how can you not love those titles?). They didn’t work on me, though. I’m still not rich.

There are several candidates for digital topics in this election. The Clinton e-mail server. The Russian hacker attacks. The late night/early morning Trump twitter rage. Fake news stories. Facebook’s alleged liberal bias. Suspicion of hacked voting machines. The role of social media and the lack of publishing ethics in new media. But what strikes me most is the shortcomings of the polls and predictions. Most had Clinton as a clear winner. The Trump victory surprised many. There are theories about why you can’t trust polls, for example that the selection is biased or that answers tend to favor the more “politically correct” candidate (whereas actual votes are anonymous and carry no such stigma). Another reason could be that those who answer polls do not constitute a representative selection, for example as higher-income voters may be more active in polls, or pollsters have difficulties finding respondents in some groups which may not have landline phones or for some other reason.

Google used to claim it could predict things like how flu spreads and, that’s right, election results through analysis of search results. There’s even a word for it: “nowcasting”. Did Google pick the winner this time around? Turns out the answer is “yes”. But then again, no. Trump led search for most of the election campaign (Fox News being the congenial source). This is simply looking at how many searched for Trump, the assumption that those who do would also vote for him. That’s a bit of a stretch, there can be many other reasons why one would search for a candidate’s name. This is the weak link in big data analytics. It can find relationships, but not necessary causality. In the 2012 primary elections, Google Trends was right about 50% of the time. Just as accurate as a coin toss, in other words. Google Flu – the service that used nowcasting to predict flu spread, was taken offline in 2014 after having made false predictions 100 out of 108 weeks. So yes, Trump was more popular in search which was in line with the outcome of the election. Except, the opposing candidate won the popular vote, so Google Trends was wrong, while it was right at the same time. The conclusion is that nowcasting is a shaky tool for predictions. That no system is perfect. That numbers can be deceiving.

You never know. But in any case: Kevin Bacon, please don’t die.