A little bit of big data goes a long way.
“It is impressive, but also frightening, how quickly people are putting so much on the line on the basis of just the smallest snippet of data,” statistician and best-selling author Nate Silver (@fivethirtyeight) said to a group of SAP partners, users and employees Monday night.
|“Apply your human skepticism toward the data you produce,” Silver told SAP TechEd on Monday.|
The Dow Jones Industrial Average lost $136 billion within seconds of a bogus tweet from a hacked AP Twitter account in April. The index rebounded as quickly as it fell, but Silver’s example illustrated how people can behave irrationally when they encounter even one unstructured datum -- one that, in this case, was a hoax.
- Big Data ... Big Bias? Ideology, old wives tales and other predispositions can blind people to what data tells them. Silver cited 2012 U.S. Presidential election coverage by conservative Web site Drudge Report, which handpicked stories to forecast a Mitt Romney victory.
- Desperately Seeking Signal. The title nods to Silver’s 2012 book -- and a 1985 Madonna movie -- but the substance referred to some people’s desire to find meaning in anything, including random coin toss data from a Princeton University professor’s classes.
- Feature or Bug? Sometimes analytically derived outliers are an opportunity, but they’re usually just a case of adverse selection, such as mapping software that suggests you drive a route with little traffic -- not realizing that the road is uncongested because it is closed.
“The world is a competitive place,” Silver said. “Apply your human skepticism toward the data you produce on a day-to-day basis.”
Toward that end, Silver offered three general suggestions for big data better practices:
No. 1: Think Probabilistically
“Certainty sells,” Silver said. “But in the real world, knowing what you don’t know is part of the battle.”
Weather forecasters build uncertainty into their models, he noted. The farther into the future they look when predicting the path of a hurricane, the wider that potential track becomes.
“Weather forecasters learned early on to think that way, [which] conveyed to the public to think in terms of probabilities,” Silver said. “People who were always looking for the perfect answer may only be running in place.”
No. 2: Know Where You’re Coming From
“You want to be alert to what your blind spots and biases might be,” Silver said. “Of course, that’s a lot easier said than done.”
|Silver suggested that big data users, like Bayes’ Theorem, weigh new information against what they already know.|
Organizations that purport to be bias-free tend to be the most biased, according to Silver. These organizations frequently chose fictitious male candidates for job interviews over female candidates with identical credentials, in an experiment cited in former Facebook COO Sheryl Sandberg’s book.
“Often the worst things that we do cost us a lot more than the best things that we do,” Silver said. “If you tell me that you have no bias whatsoever, then I think you have a lot of bias that you’re not really comfortable with yet.”
No. 3: Try and Err
“More often than not, the wins you make from using data more effectively are little wins,” Silver said. “Big data, but little wins around the margins.”
So it’s important to make sure that your products work in the best laboratory of all, Silver stated. For example, Google constantly tests its products on its users because experiments and focus groups are no substitute for the real world.
“When you start out with your ideas, test them and adjust your release date based on how well your experiments go,” Silver said, “you converge toward better and better answers.”
Silver closed with Danish scientist and author Piet Hein’s poem “Road to Wisdom”:
The road to wisdom? Well, it’s plain
and simple to express.
Err and err and err again,
but less and less and less.