Relying on numbers doesn't always add up

The site FiveThirtyEight.com is in the numbers game, building a business out of the types of predictive analytics that have become so seductive in recent years as humans attempt to gain edges and assurances in chaotic times.

If there is a frustration with statistics, it is this: When the numbers get something right, those behind the data don't seem to mind the praise that comes with it. When the numbers get something wrong, there often tends to be what looks like an excuse — terms like error margins and outliers. Hey, it wasn't us. The data was good.

But the reality is closer to this: The numbers don't lie, unless we ask them to lie.

If we can branch out of sports momentarily to help prove this point, let's talk about the results of Tuesday's election. Heading into the polls, FiveThirtyEight gave Hillary Clinton a 71.4 percent chance of winning, while Donald Trump was at 28.6.

A lot of us — particularly in the media (and my hand is raised here, albeit from a distance as someone who doesn't write about politics) — either dismissed or ignored that data, treating a probability as a virtual certainty. Even though FiveThirtyEight offered caveats that Trump had a path to a narrow victory (one he secured, of course) and reminded everyone that he was within one of those "polling errors" of being ahead, Clinton supporters essentially said, "Quiet down."

They asked the numbers to stop telling the whole truth and only to confirm the part of the truth they wished to believe.

Conveniently, FiveThirtyEight also makes football predictions every week based on the same accumulation of data. Obviously football winners and losers aren't gauged by live call polling, but they are tabulated based on a number of strength vs. weakness factors.

I'd venture to say Vikings fans were less confident than Clinton supporters going into the week. But the numbers told us they were both in the same position to win — favorable, but with plenty of chances to lose. (That the game and election result were eerily similar is interesting, though not the point).

The New York Times was even more bullish on Clinton, giving her an 85 percent chance to win going into the election. Conveniently, the Times' Upshot forecast also used a football analogy to put things in perspective, reminding everyone Clinton's chances were the same as the "probability that an NFL kicker misses a 37-yard field goal."

Even in a calendar year where a number of kickers (including Blair Walsh) have missed notable kicks from even shorter distances, the other 15 percent was dismissed.

Ignoring the 15 percent is not the fault of the data. That's the fault of human nature and the compulsion toward confirmation bias.

Let this be a reminder that it's not just enough to amass data or even to understand data. You must be able to interpret data and keep it in its proper perspective — all while assuring that it is good data in the first place, which is a whole different discussion.