In the wake of the election, Politico Magazine ran a story calling Donald Trump the “Black Swan President,” suggesting Trump’s election was the first we’ve seen Nassim Taleb’s concept of low probability, high-impact events in U.S. politics.
His colorful response to their request for his contribution to their story should have been the first sign of trouble for this piece, which seems limited to a very “Wikipedia” understanding of Taleb’s work. Trump’s election was no Black Swan event. The error here lies in the misunderstanding of Black Swan probability.
The Nature of Black Swans
The essence of Taleb’s theory is that the single observation of a black swan would undo a long-standing belief that all swans are white. In hindsight, it made sense that swans could be black, but biases prevented conventional wisdom from this possibility.
Black Swans satisfy three conditions, according to Taleb’s preface:
- The event is an outlier—it is outside the realm of regular expectations, and nothing of the past can point to its possibility.
- It carries an extreme impact.
- Human nature causes us to rationalize it in hindsight, making it seem as if it was predictable.
At this point, Trump may seem to have been an outlier, whose election is an extreme event to some, and there is no shortage of articles explaining in hindsight why he won.
While the “extreme impact” is effectively unknown (he hasn’t even taken office) and debatable, the root error is in the nature of outliers and probability. Had Politico’s authors read all the way to at least Table 1 in Taleb’s book, they would have read the differentiation between Mediocristan and Extremistan.
Mediocristan is where things tend to be more predictable, bland, and normal. Probabilities fit Gaussian bell curves (think political polls).
Extremistan, on the other hand, is Black Swan territory, where events with incalculable probabilities render Gaussian bell curves inapplicable.
By the simple understanding that Black Swans belong to Extremistan, where bell curves completely fail, we can empirically show that Trump was no Black Swan.
The Gaussian 2016 Outcome
Political polling is inseparable from Gaussian bell curves. Polls are always reported on with a margin of error (MOE), which is the same thing as the standard deviations (or sigmas) that you learned in statistics.
[Chart source: Wikipedia]
Bell curves are centered around a mean value, around which 68 percent of the data is within one sigma (or margin of error); 95 percent is within two sigma; 99.7 percent is within three sigma. While Black Swans aren’t Gaussian by nature, it isn’t uncommon to see them expressed as five or six sigma events.
The RealClearPolitics average of national polls (four-way ballot) lists ten polls in its final 2016 summary in the race for the White House. The final results were within the margin of error of nine out of ten polls. The only poll (Gravis) that wasn’t within one MOE still allowed for undecided voters to decide the race.
The outcome was a one sigma event—a predictable, Gaussian outcome, not a Black Swan event.
‘The Scandal of Prediction’
A true Black Swan outcome in the election might look like an Electoral College blowout by Bill Murray. It would have stunned pundits, since he wasn’t even on the ballot. In hindsight, we would rationalize it. Who wouldn’t want Bill Murray? Surely someone somewhere will have claimed to have a model that predicted it.
The 2016 election outcome was a Black Swan only in the minds of the ethically challenged media. Stop blaming pollsters and hysterically acting as if we’re now living in Extremistan.
Fire “data-driven journalists,” big data modelers, and anyone who ever said that Trump only had 28.6 percent chance of winning. Their “epistemic arrogance,” as Taleb would say, has polluted political journalism.