Everyday funny clips about this election circulate in the trendspace. Today’s features a fellow ostensibly representing the Trump campaign attempting to cast doubt onto stories that cite polls. He asks the reporter, “Which Polls?” She replies: “All of them.” The narrative of the Trump campaign at this point is one of an underdog promising to surprise everyone once again. The problem is that the polls never indicated that Trump was losing, at least not in the last fifteen months. If you have been reading my website or medium postings, you know that I was early on board to predict that Trump would win the GOP nomination. This was obvious as he had the most Twitter activity and was leading in the polls.
You read that right. Trump was leading in the polls in July 2015, he was not trailing. There was no shift in the mathematical universe that made 2+2=5, Trump was leading. It was a safe bet.
Take a look at this chart including all national Republican polls beginning June 2015, n=567.
Note the mess at the start. Who knew in May 2015 that Trump was seriously running? I didn’t. Once he gets serious, t-minus 500 polls he surges ahead of Bush and never looks back. The others included on this chart are Rubio and Cruz, as Kasich was never really a threat to Trump. In fact, Walker had faded so badly by this particular start date that he isn’t even included in the dataset, despite leading some polls a few weeks before the arbitrary start date of June 1, 2015.
The Election Algorithm
Nate Silver got the Republican nomination wrong. Why? Because he lost track of the algorithm of the election. This is a process that uses a particular set of procedures to count the preferences of people. Our best way to do that is by asking them their preferences. Silver during the early phase of the nomination lost sight of this, substituting the judgement of people telling other people who to prefer, thus his emphasis on endorsement count. Instead of his well established rigorous approach the public received something other than an aggregated model of opinion models. He switched out the good stuff for crap.
In his account, Silver “acted like a pundit.” He goes through a number of sets of arguments for why he ignored the polling evidence, but they really get down to the idea that he really hoped that Trump wouldn’t win. Beyond that, you might say, if the media knew that Trump was winning, why didn’t they “stop him.” To paraphrase Bart Simpson: what do you think they were trying to do? Silver, like many others, were trying the Paul Anka strategy: just don’t look. In this metaphor, the public is Homer, and Trump is Lard Lad.
More theoretically, people weren’t turning to Silver for his sense of the absolute probability of Trump’s nomination. They wanted the results from his aggregation appliance, not his view of history. Silver’s Bayesian theory of history, while interesting becomes an example of the problem of depth hermeneutics. His reading strategy was to apply a high inertial Bayesian model to quantitative data. Applying the same approach to discontinuous qualitative data doesn’t really work. This has been a problem for data journalists in general — there are very rarely times when interesting stories can be transformed with simple data scraping and quantitative analysis. This happens to academics frequently as they move between fields, it isn’t that no one in the other field thought of your ‘simple’ insight, they did, there was a problem with it, and they tried something else to solve it.
The Trump Paradox
No one saw Trump coming because they chose not to see, not because there was a lack of numerical evidence. The gravity well of Trump’s candidacy warped the light of publicity and made it difficult for even seasoned data journalism folks to report the story accurately early. Trump was winning.
Some final graphics. First, all 670 polls. This reading of the election winds all the way back to January 2015.
Here is the very early going. I have changed the colors to be helpful — Bush is blue, Trump is Orange. Bush is in slow, steady decline, Trump is rocketing up. I guess the Trump underdog story was true in March 2015. Although this was back when Uptown Funk by Bruno Mars was passing Taylor Swift’s Blank Page for number one song, and Same Sex Marriage was illegal in dozens of states. In other words, we are talking about tempestuous polls, two years ago.
There was no error in polling, just wishful thinking by the establishment.
So yes, all of the polls. The same polls that induced candidate Trump to become nominee Trump. Would he have continued into January if he had been trailing in the polls? No.