The Signal

A Signal mission: Putting a number on uncertainty

David Pennock
The Signal

Uncertainty means being unsure, at least a little, about the outcome of some event in the future. Who will win the Super Bowl in 2012? Who will be our President in 2013? I don't know. No one knows. Anyone who says they know with conviction is either lying, delusional, or has inside knowledge that I believe — well, with 99.9% surety — that no one in the world has.

Most people are happy to acknowledge that uncertainty exists: it's a part of life. We buy insurance precisely because we realize that the unexpected can happen. Many things are unpredictable, from Arab uprisings to Apple stock. But few go beyond admitting they are unsure to say exactly how uncertain they are, quantifying their doubt with a number, say 50.5% sure that Obama will prevail. What does it means to be "50.5% sure" of something? Part of our mission at The Signal is to report on predictions like this from all around the web, explain what they mean in straightforward terms, and encourage our readers, and leaders, to think more, and more clearly, about uncertainty.

[ Destination 2012: Check out Yahoo!'s election headquarters ]

If I flip a normal coin, the two possible outcomes — heads or tails — are equally likely. The chance of flipping heads is 50%. If I flipped the coin 1000 times, it would land on heads about (though not necessarily exactly) 500 of those times, in no particular order. Similarly, the chance that I roll a six on a normal die is one in six, or about 17 percent. Coins, dice, cards, and roulette wheels are engineered to have a very predictable form of unpredictability. They're built precisely to have a known probability for every possible outcome. As long as they're engineered well and no trickery is involved, it's easy to quantify precisely the uncertainty in coin flips and dice rolls.

In social systems like elections or sports with millions of moving parts and billions of possible factors, reasonable people can and will disagree about how likely any given outcome is. To get an accurate estimate of, say, the chance that Obama wins reelection, the best we can do is seek some kind of consensus forecast that incorporates as much information as possible. In some ways the election is like a coin flip where no one knows exactly how likely heads is versus tails, in no small part because it depends on so many variables: unemployment rates, real estate prices, wars, scandals, approval ratings, and disasters, to name a few, and every combination of the above.

A proven method of forecasting such events is to cast a wide net: to gather many predictions from a diversity of sources and merge them together, a process that some people call crowdsourcing that taps into what James Surowiecki termed the wisdom of crowds, a phrase that's been used and abused widely but that does have a growing scientific basis (e.g., see Surowiecki's book by the same name and Scott Page's book The Difference).

That's the approach we take at The Signal. We gather information from a variety of sources (some of which are themselves information aggregators), including prediction markets, polls, web search, social activity, games, and historical precedent, and combine it together to produce a Consensus Prediction that is as unbiased, accurate, and timely as we can possibly make it. We will strive for maximum transparency, reporting where our data comes from allowing curious users to browse the raw source data whenever possible.

From now until election day on November 6, 2012, we will create and publish a growing number of Consensus Predictions on nearly every aspect of the US elections, including the Party, Presidential, Senate, House, and Gubernatorial races, and related economic and financial issues, from the unemployment rate to the price of gas, all updated in real time to catch the impact of political tremors and earthquakes alike, as they happen. We will add interactive polling and games ("fantasy politics", anyone?) in 2012 to keep the site fun and engaging — and to provide us with even more information to pump back into our predictions!

The Signal blog, which you are reading right now, is your written companion to the numbers. Here we will explain what the numbers mean, highlight changes as they happen alongside breaking news, and tie together all the content and activities — predictions, polls, and games — under The Signal's umbrella. To get a feel for The Signal blog, you can read our past articles originally published on the Yahoo! News politics blog The Ticket. We're exciting to join an incredible team of Yahoo! News writers and editors, Yahoo! Research scientists, and academic experts to provide an objective, quantitative lens on the election and beyond.

David Pennock has a Ph.D. in computer science from the University of Michigan and currently heads the algorithmic economics group at Yahoo! Research. Follow him on twitter @pennockd.

Other popular Yahoo! News stories:
Fourth accuser urges Herman Cain to 'come clean' about harassment
Accuser's press conference cuts Herman Cain's odds in half in prediction markets
More than 49 million Americans — even more than thought — live in poverty, new measure finds

Want more? Visit The Signal blog or connect with us on Facebook and follow us on Twitter.

View Comments (2)