You are using an outdated browser.
Please upgrade your browser
and improve your visit to our site.
Skip Navigation

New Prediction Model Doesn’t Prove Romney Will Win

A widely publicized political science forecasting model by Ken Berry and Michael Bickers in Colorado projects that Romney should win big.  

My take? Add it to the pile.

There are dozens of political science models. Some are good, some are bad. Some say Obama will win. Others say he will lose. Is there any reason to believe that this model is any better or worse than the others? At the very least, there’s not much reason to assume it’s any better than the competition. The press release reveals it’s predicted every election since 1980. If it hadn’t, that would be pretty embarrassing. In the eight presidential elections since 1980, six have been 7+ point blowouts in the two party vote. So all you need is a model that gets 2000 and 2004 right without missing the other six elections. Call it 2-0, in my book.

I can even prove that it’s not difficult: there are a million other political science models running around with a similar claim to accuracy over the last eight elections. And do those models point toward a clear Romney or Obama victory? In the aggregate, the political science models point toward a competitive race, but there are models showing a clear victory for either side. The Colorado model has a mirror-image counter by Drew Lizner of Emory University, who gives Obama a 99 percent chance of victory. The FiveThirtyEight model integrates economic variables and it finds Obama with a modest advantage, and so does a model built by John Sides, Lynn Vavreck, Seth Hill for the Washington Post. The famous Abramowitz “Time For a Change” model points toward an extraordinarily tight race. So, contrary to popular belief, the fundamentals do not clearly point to a victory for either side. And not only do the economic-based models show a tight race, Obama’s approval rating is at about 48 percent, which most agree is a product of fundamentals like economic performance.  

Just for good measure, I have one specific issue with this particular model: it relies on different economic metrics for Democrats and Republicans. I won't dismiss the possibility that voters judge Democrats and Republicans by different economic standards, but it is highly suspect, at least in my judgment, to assume that the public responds differently to the parties based on so few elections. We only have a handful of post-war presidential elections and it's already tough to make highly confident models based on all the available data in the first place, let alone when you start treating half of the elections differently. 

Do economic factors point toward a decisive Romney victory? Such a win is possible, but the fundamentals-based models point toward a close race. For every model showing Romney in the lead, there's a model going the other direction and there's no reason to presume that yesterday's model will prove any more accurate than its pro-Obama counterparts.