election analytics roundup

Here are a few election related links:

• Eva Regnier at the Naval Postgraduate School has a bootstrapping forecast model for the US Senate election. Her model uses forecasts from Simon Jackson and Drew Linzer. Information about the upcoming election (usually polling results) becomes available over time, and this information produces a sequence of probability forecasts for each race. Eva writes, “I suspect that these probability sequences are not optimal, i.e. they could perform better with respect to single-period probabilistic scoring rules, using the information they have available. I also suspect — this is a stronger claim — that they can be bootstrapped, i.e. that the right function could take the forecast sequence itself and produce a forecast that outscores the original at each update.”
• Sheldon Jacobson’s team at the University of Illinois has its Election Analytics site up and running again that is predicting the probability that the Republicans will take the Senate. His method uses Bayesian estimators that use available state poll results.
• Of course, Nate Silver has a Senate forecasting model on FiveThirtyEight and provides a nice discussion of how the polling data is used in the forecasting model.
• Matt Yglesias: The real problem with Nate Silver’s model is the hazy metaphysics of probability.
• Andrew Gelman has a post on poll sample sizes called “Was it really necessary to do a voting experiment on 300,000 people? Maybe 299,999 would’ve been enough? Or 299,998? Or maybe 2000?“
• In 1936, the Literary Digest ran a huge and very expensive poll to forecast the Presidential election. They collected about 2.4 million responses in a totally biased sample and predicted that FDR would lose. At the same time, George Gallup accurately forecasted the election with a sample size of only 50,000.

I’ve blogged about elections a lot before. Here are some of my favorites:

• why is it so easy to forecast the Presidential election and so hard to forecast the NCAA basketball tournament?
• drive carefully: you are 18% more likely to die in a fatal car crash on Presidential election days (Good news: today is not a Presidential election, so you should be OK).
• what data do election forecasting models use to make predictions? Hint: they don’t all use just polling data.
• forecasting the Presidential election using regression, simulation, or dynamic programming
• Moving from polls to forecasts: the polls say the election is a dead heat but analytics models may indicate that it’s a landside
• It’s the economy, stupid: the pieces of information you need to know to forecast the Presidential election