Headlines across the media post-election agreed: President Obama may have been granted another four years in the White House, but Nate Silver won the election. His analysis, which takes into account the size, quality, and recency of polls, accurately predicted 50 out of 50 states. Silver’s feat is impressive, considering the outcome of the 2012 election was lost on pundits and political scientists across the country. Most political scientists and economists base their forecasting models solely on economic fundamentals, and have had a subpar record in predicting election winners since the 1980s when it became popular. As a result, I don’t find these models so useful, especially this year.
In his New York Times blog Fivethirtyeight, Silver shows that for the last five elections, not including this year’s, models averaged astandard error of about eight points. Economist’s and political scientist’s had their worst year in 2000, predicting that Gore would win by a margin of about twenty points.
There are several problems with election forecasting models based on the economy. For starters, Silver says, they make their predictions too far ahead of time — easy for Silver to say, who continually updated his prediction up until election day. Also, I question the relevance of these models, which for the most part only predict the popular vote, not venturing as far as the electoral college.
University of Colorado professors Ken Bickers and Michael Berry gained notoriety for their prediction that Democrat Al Gore would win the popular vote and Republican George Bush would win the electoral college in the 2000 election. Their model factors in unemployment rates and changes in real per capita income on the state level. This year they predicted Republican Mitt Romney would win by a landslide, taking nearly every battleground state. (I have yet to see a mea-culpa article from them.)
There are other interesting factors that some political scientists take into account, such as Douglas Hibbs’ Bread and Peace model, which looks at per capita real disposable personal income over the incumbent president’s term, and cumulative U.S. military fatalities in overseas conflicts. Hibbs’ results revealed that President Obama would receive 48 percent of the vote share. This is an unusual one. I don’t know how military fatalities could be relevant in predicting an election, especially since foreign policy has taken a backseat to the economy this year.
It seems as though many models factor in irrelevant metrics. Then there’s political scientist Alan Abramowitz, who has held among the lower standards of error, has accurately predicted the popular vote (but not the electoral college) winner in every election since 1988, including this year’s election. His forecast is based on the candidate’s approval rating at the end of June, the growth of the economy, and the value of the “incumbency factor.” I think Abramowitz has been successful, comparatively speaking, because he employs very relevant metrics — approval rating and economic growth.
We have to remember that according to Vavreck, voter’s attitudes about the state of the economy are intimately connected to their approval of the incumbent party. Many voters tend to vote retrospectively, and if they are informed about anything, it’s their own economic condition. And Silver admits there is a connection between incumbents and the economy, saying, “ … ruling parties seemed to be having an easier time keeping hold of power when the economy was good.”
This year’s economic situation was complex and mixed, like we’ve been talking about. Perhaps this is the reason prediction models have been all over the place. Maybe an unstable economy produces unreliable presidential predictions. But generally, as depicted in this graph, fundamental models have done a poor job of making accurate predictions. “The ‘fundamentals’ models, in fact, have had almost no predictive power at all,” Silver says in his blog post. “Over this 16-year period, there has been no relationship between the vote they forecast for the incumbent candidate and how well he actually did — even though some of them claimed to explain as much as 90 percent of voting results.”
Lynn Vavreck’s theory supplies the fundamental metric that these election models are missing: How candidates set the agenda and tell voters what they should be thinking about as they make their decisions. Yes, current economic conditions are vital to voters’ judgements, but whether the president in a good economy can tout that good record, or if the insurgent can point out failures in his opponent will also shape voters’ decisions.