Political pundit Silver explains precarious art of prediction
By Lou Fancher for San Jose Mercury News
If you're wondering whether Hillary Clinton will run for office or Republicans will recapture the White House in the 2016 presidential election, don't ask Nate Silver.
The data-crunching statistician and author of New York Times bestseller "The Signal and the Noise: Why Most Predictions Fail -- But Some Don't," is burned out on politics.
Or so he said, in a Cal Performances presentation on May 4 at Zellerbach Hall that was nonetheless mostly about politics.
Silver garnered national attention and a devoted following when his forecasts in the 2008 and 2012 presidential races tipped probability into near-perfect predictions. A generational shift from trusting hard-won human intuition to relying on push-button, computer-generated solutions that seem to fall like rain from a cloud, converged with Silver's founding of FiveThirtyEight, the digital forecasting publication he named after the total electoral votes available during a presidential election.
Even before he correctly called the outcome of 99 of 100 states (a total of his presidential predictions) and 35 of 35 Senate races, he'd conquered baseball.
His PECOTA (Player Empirical Comparison and Optimization Test Algorithm) debuted in 2003 as an early sabermetrics model for successfully predicting Major League Baseball players' performance.
In sports and gaming, football fanatics, poker players and Moneyball enthusiasts (stemming from Berkeley author Michael Lewis's book about the Oakland A's numbers-reliant, build-a-winning-team method), added feverish momentum to Silver's rising popularity. Housed at The New York Times from 2010 to 2013, the 36-year-old's FiveThirtyEight jumped to ESPN this year.
Silver began his 60-minute PowerPoint lecture in Berkeley with warnings.
Big Data has both speed and volume problems, he said, citing a single tweet that suggested the White House had been bombed and President Obama injured that caused the S&P 500 to immediately drop by $1.6 billion. "Making $1.6 billion decisions based on 140 characters is the problem," he said.
The impact of an imbalanced signal-to-noise ratio -- the central point he emphasizes in his book -- is extreme, especially in today's aggregating environment.
Amid too much information, made worse by bias, it's nearly impossible to detect the "signal" -- the relevant factors predicting the course of complex problems like economic trends, earthquakes, mortgage bubbles and global warming.
He showed graphics illustrating how an algorithm converted two factors from a Federal Reserve Economic Data report into 10.9 billion combinations an average person would have to consider.
Biased headline trolling adds to the problem, he said, creating false positives or patterns. Silver's appearance on a network television morning show on election day 2012 caused his Google search numbers to spike above Joe Biden's (but not above Justin Bieber's, which remained 20 percent higher than both Silver and Biden).
Silver said that instead of "leading with their gut" to make predictions, people should think Bayesian, as in Thomas Bayes, an 18th century English minister and statistician.
The Bayes theorem addressed chance based on probability, while admitting bias (Silver called this "Know where you come from"). The Bayesian path, paved with old-fashioned trial-and-error, Silver claims in his book, is a true "path to less wrongness."
"What's the big idea behind Google?" he asked the audience. "It isn't one concept. They do 10,000 searches on their system every year. They refine; go for continuous improvement."
Silver said the hybrid approach -- combining rational, objective data, common sense, human intuition based on lots of experience and an awareness that vast, fluid problems are harder to predict than games with rules -- would improve the art of prediction.
But even a lot of volume (Big Data) and high standards (science) won't fix the battle fatigue he finds in predicting political races. "You piss half the people off right away," he said, "the other half the next time."
Which is why he's turned to burritos. FiveThirtyEight is immersed in a nationwide search for the country's top burrito. Silver's prediction? He won't say, but he did offer what incautious folks will take as a hint: "We're identifying the top 64 now, and a good number are in California."