David M Rothschild on Posted on

In the first academic paper I ever wrote, I compared the performance of poll and prediction market-based forecasts in the 2008 election. Naturally, I described the accuracy of the forecasts, but I also talked about the timeliness of the forecasts. Prediction market data has a lot of added value in that it is always fresh, while polling, especially back in 2008, was not updated very regularly. How could it be? Traditional polling with telephones and probability-based samples is expensive. At the time, I was told that I should compare the prediction market-based forecasts at the times that the poll-based forecasts were fresh. That standard academic mindset of thinking about accuracy at any cost, restriction, or timeliness is wrong. Resources in politics are allocated regularly and it is critical that the market intelligence is updated when decision makers need it, and time-granularity allows researchers to better understand the true impact of events.

Non-traditional polling, using online/mobile-based opt-in samples, allows some polling to provide the necessary time granularity to answer many new and meaningful questions about sentiment. Of course, it takes a careful understanding of both the data collection and analytics to make non-traditional polling accurate. We believe we have a cracked that with the modeling and post-stratification we do with data collected through Pollfish’s mobile polling system. This type of polling is cheaper, flexible, and more timely than traditional polling. And, even if you are concerned about the accuracy, consistent methodology makes the changes in levels verifiably meaningful.

This year we have been polling 12 issue questions weekly to gauge how the voting public are reacting in their public policy preferences to the election cycle; changes in support for gun control shows the value of high time granularity. In the immediate aftermath of the massacre in an Orlando nightclub there was a lot of concern over the type of semi-automatic weapon used by the shooter.

Right after the massacre strong support for gun control surged. While two months later the support is still slightly higher than before the massacre, it is notable how quickly support came back down in just 2-3 weeks. Weekly polling allows us to better gauge the ephemeral effects of massacres on support for gun control.


The topline views obscure some interesting demographic shifts; Republicans and Democrats may have quickly fallen back to pre-massacre positions, but independent voters remain elevated. This type of reaction to an event makes sense; only a small portion of the electorate is really open to change over short periods of time. Thus, it is critical to understanding public policy to use the more advanced analytics of non-traditional polling to examine demographic sub-groups, along with the topline results.


This work is joint with Tobias Konitzer and Sam Corbett-Davies: PhD candidates at Stanford University in communications and computer science, respectively.