Nate Silver may have started the stat-heavy method of predicting elections. But a team at Microsoft led by applied economist David Rothschild thinks it can do more, and better. Microsoft Research Lab has big ambitions to wrangle market forecasts, prediction games, and web-centric polling into the world of data-based predictions.
Its inaugural product launched this week, a just-in-time prediction site for this year’s Senate, gubernatorial, and House elections in the US, embodies the goals of the group: to lower the costs and deepen the insights gleaned from traditional polling.
“Traditional polling is a very expensive way of engaging people,” Rothschild says. He argues that the field should be faster, cheaper and more adaptive in how it asks questions. That’s partly because not every predictable event offers the data of big political races. Polling data is sparse and infrequent for many House races, and even more so for one-off events. Many political races lack good fundamental data on incumbents or previous election results, and years of past scores are often missing in sports.
In the best traditional polls, polling companies attempt to represent a broad population by randomly calling thousands of people. Though Rothschild still uses these in his predictions (they’re still the best thing out there), he argues their value is declining. With the death of landline and rise of caller-ID and cell phones, people are picking up their phones less, which has driven down response rates from 35% from the 1990s to around around 10%.
“The Nate Silvers of the world and the Upshots and all these other places, they do a perfectly good job in aggregating the data that’s out there,” he says. “I worry that that data won’t exist in 4 or 8 years and we’re preparing for that time.”
Rothschild’s answer is to create more online surveys and innovative ways to cull data without using the phone. His goal is to quicken the pace of culling answers and broaden the range of questions. For example, in a series of daily polls of users on Microsoft’s XBox video game console, Rothschild was able to generate forecasts and models using the data and demographic information that closely mirrored the accuracy of poll aggregation.
There’s more where that came from on the internet, smartphones and video game consoles. The future is “moving away from the idea that everyone needs to be asked the same questions in the same setting to asking the right questions to the right person the right way,” he says.
Longer term, Rothschild’s hope is to take innovative predictive measures beyond elections to predict policy outcomes. “The real reason I care and you care about who’s going to win an election because that person or group of people are going to influence policy,” he says. ”I hope to be talking to you in two years about what the likely tax rate is going to be if candidate A or candidate B wins.”