In case the new issue of PS: Political Science and Politics is still on your junk mail table, here's a primer on the journal's recent publication of 13 distinct predictions of the 2012 election: Five academics predict an Obama victory, five predict a Romney victory, and three say it's too close to call.
And here's a prediction I feel good about: Five of them will be correct.
All 13 of the predictions in this peer-reviewed journal are the product of fundamental models, which examine broad historical trends that influence elections rather than simply aggregating polls and prediction markets. Some of the models use polls as a guidance, but the focus is on information like economic indicators, incumbency, past election results, the state of war, and other lofty data points divorced from public opinion surveys.
I wholly endorse the idea of academics working alongside journalists in the popular election prediction industry—obviously—but PS looks silly publishing these forecasts at the end of September. Models are useful in painting a broad electoral picture six months ahead of time, before public opinion has coalesced. They typically cannot account for the narrow margins of victory that shake out weeks or days before polls open. Relying on fundamental models in October is like relying on pre-season baseball predictions in October. I would look stupid—or at least delusional in my fandom—if I forecasted the Philadelphia Phillies winning the National League East today, when they are eliminated from the running, even though they were pre-season favorites.
(The Signal's model, not included among these 13 examples, uses a combination of fundamental data, polls and markets, with the emphasis shifting to the latter two factors as Election Day approaches.)
Models are still relevant in October—just not for their predictive power. There are two similar but distinct reasons why we forecast elections. The first is to have some clue as to the results ahead of time, a job we must now cede to the pollsters and prediction markets. The second is to glean insight into the forces that govern voter opinion. The popular and academic press regularly conflate these two goals.
The model I developed with Patrick Hummel, for example, has found that the recent movement in an economic indicator—whether it's up or down from six months ago—is much more important than the overall levels. We also found that there are diminishing returns to economic indicators after the second quarter.
Here are some similar insights from a few of the newly published models:
- Douglas Hibbs updates his well-known model of "bread and peace," which concludes that elections correlate with just two data points: real disposable income and US military fatalities due to offensive wars. The model borders on subjective this cycle; Hibbs designates Afghanistan as an offensive war for the incumbent Democratic Party, whereas it was as defense war for the Republicans for its first seven years.
- Helmut Norpoth and Michael Bednarczuk rely entirely on the results of the New Hampshire primary results.
But, the journal does not focus on these compelling insights. Instead, it focuses on the results, which are not only too late to be relevant, but center on expected vote share, the wrong question. Stakeholders, from individual voters to campaign managers, care about the probability of victory more than the overall estimated vote share.
When it comes to research, while fundamental models can illustrate sweeping historic correlations, polls and prediction markets are necessary to provide the granularity that allows us to calibrate the value of the events that occur within a campaign. Publishing outdated forecasts is a parlor game that makes academia appear out of touch with reality and behind the popular press.
David Rothschild is an economist. He has a Ph.D. in applied economics from the Wharton School of Business at the University of Pennsylvania. Follow him on Twitter @DavMicRot.