Commentary
Wild Outcomes, More Confusing Trump Tweets
- by Philip Rosenstein , Staff Writer, February 6, 2017
Last night, we witnessed another unforeseen event — at a time when they have become commonplace. Ahead 28-3, well into the third quarter of the Super Bowl, the Atlanta Falcons had a 98.9% chance of winning the game, according to ESPN.
The Patriots had only a 1.1% of winning. Yet they did.
In both 2016 and 2017, odds defiers abound. Starting with Leicester City winning the English Premier League, which began the season with 5,000-1 odds to Brexit and the presidential election of Donald Trump. Statisticians, odds makers and pollsters just can't get a break.
What do all these unexpected, and in most cases wholly astonishing, outcomes have in common? They pose serious questions about the practice of predicting outcomes, particularly in politics.The harm to statistical polls and probabilities is such that President Trump feels comfortable making the astonishing claim that when polls are not in his favor, they are “fake news.”
advertisement
advertisement
At 7 a.m. this morning, the President tweeted: “Any negative polls are fake news, just like the CNN, ABC, NBC polls in the election. Sorry, people want border security and extreme vetting.” His message was in response to polls showing his travel moratorium is not popular with the public.
The President’s retort to unfavorable polls plays well with supporters, but it fails to grasp the complex nature of polling and statistics.
Polling is far from an absolute science — it is a mathematical way of comprehending sentiment and extrapolating that to understand a larger population. Developing predictions from those polls, as FiveThirtyEight, The Huffington Post and The New York Times did during the 2016 election, adds an extra layer of uncertainty.
A set of possible outcomes, no matter how unlikely, “can” happen. Just like the Patriots pulling out the win last night, and Donald J. Trump becoming president.
But the President’s tweet this morning doesn’t pass the logic test.
Republican pollster and co-host of ‘The Pollsters’ podcast Kristen Soltis Anderson replied to Trump's morning tweet: “You can say polling is broken or polling still works. ‘Negative polls are fake news’ but positive ones are not does not work.”
Trump’s approach to polling just doesn’t make sense. During his campaign, he consistently referred to favorable polls in campaign speeches and on Twitter — until they go against him. Then, he takes umbrage, crying mainstream media collusion, and now, “fake news.”
The conversation about polls should move on. How do we continue to make polls more accurate, while making polling methodologies and statistics more approachable and informative to the general public?
Pollsters need to refine their systems. In addition to mathematical parts of the equation, other subjective metrics, informing what the numbers say, should come into play. Both the public and the media should retain some skepticism — because sometimes, when you least expect it, a team or a president makes it just under the wire.
The HELL with polls...get RID of TRUMP and Pence as fast as possible.
Most of the predictions of the election outcome were wrong. But the polls, themselves, were generally right -- showing an electorate so equally divided that any "win" was generally within the statistical margin of error. Polls showed a race so close that a small push in one direction (Wikileaks, Comey) could have an effect -- and did. (For that matter, the prediction of the game outcome was wrong. But the data -- historical -- on which it was based, was correct.) There's not a crisis in polling, but journalists have hopefully learned that overconfident predictions of outcomes is a problem. As for Trump, he doesn't have a bias against polls or data, per se -- he's obsessed with tallying scores. But he's a giant crybaby bully who can't stand it when reality doesn't go his way, and is not only uninterested in learning anything that doesn't jibe with his "they love Trump" narrative, but appears to be unable to absorb knowledge.