Monday, December 01, 2008

My 2008 Election Predictor

I mentioned about a month ago that a long time ago, I had made a model that used state voting trends to predict the outcome of the 2008 election. The model only gave a result at a given swing, the amount of change between the two parties, nationally, so its most predictive feature was simply finding how much of swing between 2004 and 2008 would be required to for the Democrats to win. The answer, according to the model, surprisingly little. Due to trends that saw a lot of swing states leaning more Democratic, a swing as small as 0.7% produced a Democratic victory. The thing to keep in mind though, is that Bush won the last election by a margin of 2.5%, meaning that for a swing between 0.7% and 2.5%, the Democratic candidate would win the electoral college, but could lose the popular vote by as much as 1.8%. This meant that the Democrats had a huge advantage going into the 2008 election, regardless everything else that happened. In the end though, there was a 9.3% swing in Obama's favor.

So how accurate was the model? Looking at the actual results, it turns out that Obama could have won outright with just a 0.7% swing, so that, amazingly, was correct. More surprisingly, a swing of just 0.1% would have produced a tie, something my model did not find. Overall, the model produced an average error of 5%, so taking that into account, this is what my model predicted a 9.3% swing towards the Democrats would have produced.



I was quite surprised myself to see that my model predicted that North Carolina and Virginia would be swing states. My model also did well in predicting the inevitability of Nevada and New Mexico becoming blue states. In fact, to keep both of these states red would have required a 5.6% swing towards McCain.

Of course the most obvious mistake was Indiana. As Indiana trended more Republican than the rest of the country in the last two elections, the model predicted it would continue to do so, and Obama would lose Indiana by over 13%. Instead, Indiana showed the second largest shift towards the Democrats this election. The largest shift, and ultimately my models largest error, was Hawaii. Clinton and Gore had won Hawaii by quite wide margins, but Kerry did really poorly, so the model incorrectly predicted that Hawaii was becoming more Republican when in fact it appears that Hawaiians just didn't like Kerry. The model did have Obama winning Hawaii by a decent 15%, but he actually won by 45%.

Additionally, the model got Missouri and North Carolina wrong, but they were within the 5% margin of error that I used on the map. The model was fairly accurate for both, they were just so close that even small errors created wrong results. Most of the big errors involved states that were so heavily in favor of one particular candidate, that it still produced the correct end result, such as in the example of Hawaii.

No comments: