November 7, 2012

Note: About half a week after this post was published, Florida declared that Obama had won the 29 electoral votes of that state, giving him a total of 332 electoral votes to Romney’s 206.


Wednesday Morning Quarterback

We used this same byline eight years ago, after we failed to correctly predict the outcome of the Bush-Kerry battle. Our postmortem analysis revealed the perils of using Internet polls, which had erroneously pointed to a Kerry victory in Florida and Ohio. We learned our lesson.

Now back to the present. In our post uploaded on the night of November 5, we put forth the five most likely outcomes of this election. They all pointed to an Obama victory, but with different electoral vote counts: 303, 332, 319, 299, and 290 in order of probability, with the odds ranging from 3.8% to 2.8%, respectively. It now seems that one of the top two will prevail, depending on Florida’s vote count. Our students are elated, having learned the arts and science of mathematical modeling and seen how well it works. It is an art because of the assumptions embedded in the model, which require judgment instead of following hard-and-fast rules. It is a science because of the logical nature of a mathematical model. After our 2004 fiasco, we improved in 2008, pointing to an Obama win with 364 electoral votes as the most likely outcome—missing by one because Nebraska split its vote in favor of Obama. We feel comfortable with the model, and we will continually seek to refine it.

Nearly three billion dollars later, the election is over. We’ve received emails containing both complaints and congratulations after the public saw the probabilities that we projected. Perhaps, to better explain our methodology, it is worthwhile to provide an analogy. Here is what we do: we feed state-by-state polling data into our model, and crank out the numbers in the form of a distribution of electoral votes won by a candidate, with their associated probabilities. Rather than directly predicting the outcome of the election, the histogram provides insight into how many votes a candidate might win and how likely he or she is to win them. Imagine that a gambling man (or woman) might take the information to place a bet. Unlike betting on a roulette wheel, not all electoral outcomes are created equal—rather, the electoral vote counts for a candidate can go from 0 (very unlikely) to 303 or 332 (the two most likely) to 538 (again very unlikely) instead of the mere 38 outcomes of a roulette wheel. The histogram can be used for the house to set betting odds to guide the ever hopeful gamblers to place a bet. Not intending to advocate gambling, this is probably the best analogy we can make. As it stands now, Florida was (and is) a 50-50 proposition for either candidate resulting in almost equally likely outcomes in our calculations (3.8% and 3.6% for the 302 and 332 outcomes, respectively).

Today, you will undoubtedly be bombarded with analysis by the news media and political strategists, now that they have the hindsight to see what went wrong or right. I heard plenty on the radio this morning.

The need for analysis has lessened; the election is over. Now, the country has to tackle the many ills present in our land. We should demand that the elected officials work together. Posturing and partisan politics have to stop. We commented earlier on the need to address our deteriorating public infrastructure, education, housing, environmental and entitlement programs, and the unsustainable budget deficits. Politicians know these issues, and now, they must put aside politics and act to fix them.

- Sam Chiu