It’s the Economy, Stupid, says the University of Colorado
People have become increasingly skeptical and dissatisfied with major election polling — and with good reason. All too often polls no longer seem to reflect demographic reality or eventual election results. Recently several excellent articles have been written evaluating the pro’s and con’s of election polling and their questionable predictive ability. These articles examine various relevant issues; accuracy, demographic sampling versus voter demographics or political party affiliation, national versus state versus local polls and how they relate, or the relative merits of one polling company’s proclamations versus another… The media just adores this very contentious issue — reporting and debating the ‘horse race’ provides endless fodder for the front page. However, while mucking around in the weeds, the forest is all too often ignored; for the most part, polls which have become extremely numerous over the years, have simply become far less accurate — or perhaps this is just surrendered as “something that goes without saying.” Polls that do manage to accurately predict one election frequently fail to do so in other elections.
One might protest, “So what, polls are the best thing we have!” Or “what else can we use?” Well, it turns out that there actually may be a better way to predict presidential election results. A rather obscure university model has proven to be far more reliable than public polling — or pretty much any other model. Developed by two University of Colorado (CU) political science professors, Kenneth Bickers and Michael Berry, this analysis method has proven amazingly accurate at correctly projecting Electoral College results on a state by state basis covering an amazing 30 years and eight presidential elections.
Their most recent analysis, released October 4th, projects that Romney will receive 330 of the total 538 Electoral College votes (five votes more than their August prediction). President Barack Obama is expected to receive 208 votes; well short of the 270 needed to retain the presidency. A University of Colorado press release explains:
“The key is the economy, say political science professors Kenneth Bickers of CU-Boulder and Michael Berry of CU Denver. Their prediction model stresses economic data from the 50 states and the District of Columbia, including both state and national unemployment figures as well as changes in real per capita income, among other factors.
“Based on our forecasting model, it becomes clear that the president is in electoral trouble,” said Bickers, also director of the CU in DC Internship Program.”
Bickers’ and Barry’s previous analysis was published in the peer-reviewed journal PS: Political Science & Politics along with a dozen other models. The journal has published similar models every presidential election year since 1996. This year’s collection displayed the widest split in outcomes, Berry said, with five predicting a Romney win, five Obama, and three defaulting to a toss-up. Out of all of these models, only two focus on economic conditions rather than the popular vote. The Bickers and Barry model is the only one that uses more than one state level measure of economic conditions in addition to other national and state level factors. The University of Colorado press release expounds on these factors:
“In addition to state and national unemployment rates, the authors looked at per capita income, which indicates the extent to which people have more or less disposable income. Research shows that these two factors affect the major parties differently: Voters hold Democrats more responsible for unemployment rates while Republicans are held more responsible for per capita income.
Accordingly — and depending largely on which party is in the White House at the time — each factor can either help or hurt the major parties disproportionately.
Their results show that “the apparent advantage of being a Democratic candidate and holding the White House disappears when the national unemployment rate hits 5.6 percent,” Berry said. The results indicate, according to Bickers, “that the incumbency advantage enjoyed by President Obama, though statistically significant, is not great enough to offset high rates of unemployment currently experienced in many of the states.””
The model was developed recently, and uses high quality data from the Bureau of Economic Analysis (BEA) in the US Department of Commerce and the Bureau of Labor Statistics in the US Department of Labor. Relevant data from those sources for each of the previous eight presidential elections was then analyzed. The model had an average error rate of 28 Electoral College votes and five states. Looking forward, the researchers mention several caveats, including that states which are very close to a 50-50 split may fall in an unexpected direction due to factors not included in the model. Several of the 2012 election swing states may fall into this category. This type of ‘backcasting’ is a well established method to validate model projections. Of course, the only way to know how well such methods will actually function is to see how well their projections match real world results over time. Interestingly, Bickers and Barry also indicate that Governor Romney has a 77% likelihood of winning the popular vote.
Just how historically accurate is this model’s backcasting for very tight election years? Consider the author’s statement:
“…the 2000 election is of particular interest because no forecasting model published in advance of that election correctly predicted George W. Bush as the winner. Because our model is predicated on the notion that during close elections, the Electoral College winner may not win the popular vote, it is critical that our forecast classify this election accurately. In 2000, the model correctly classifies 47 states, most notably Florida, which Bush was estimated to win with 51.2% of the two-party vote. The state’s final certified result awarded Bush a razor thin majority of the two-party vote at 50.004%. The only states incorrectly classified in 2000 are Pennsylvania, West Virginia, Arkansas, and Louisiana. Despite these inaccuracies, the model expected Bush to win 274 electoral votes to Gore’s 264, an error of a mere two votes.” [emphasis added]
The most recent national polls vary from Romney leading by 5 to Obama ahead by 5 — a 10 point spread. Swing state polls are similarly schizophrenic, although Obama holds the lead in the key swing states of Ohio and Pennsylvania. He also continues to hold a majority in polling results of minorities, although less so than during the 2008 election. These factors imply an Obama win. Further confounding the issue, however, is the fact that independents are polling strongly in favor of Romney, which suggests a Romney win. Many of the polls are so close that some experts are even discussing the unlikely situation of an Electoral College tie (an interesting subject itself, but unlikely to occur). All of these key variables and muddled polling internals make it very difficult to draw any meaningful conclusions about the 2012 presidential election from polling data.
The University of Colorado model promises an efficacious prediction method based on key economic and unemployment indicators that are entirely divorced from polling vagaries. If successful, this suggests that presidential elections are about the stewardship of the national economy, rather than the minutia of gaffes, political commercials, campaign strategy, or a candidate’s rock star status — which would be a very positive indicator for our Republic. It’s always possible this will be the one year the model fails, but if the University of Colorado prediction matches its promise, in a few short days the nation will see a substantial win for Governor Romney.
Note: All thirteen election models can be viewed on the PS: Political Science & Politics website at http://journals.cambridge.org/action/displayJournal?jid=PSC.
Initial version originally published at AmericanThinker