Now that the election is over and the votes have been counted, I would like to compare Battleground270’s projections to the actual results. As it turns out, Battleground270 called 49 out of 50 states correctly for the Presidential election and all but one state in the Senate elections.
As you can see in the table below, Florida is the only state I projected incorrectly. My predicted margin of victory for Mr. Romney was 0.2%, making it the closest of any of the states. While it seems Florida will indeed be the closest state, it was actually Barack Obama who emerged victorious, by 0.9%.
However, just calling the state correctly is not a sufficient way to analyze an election projection. Instead, I have compared my projected margin of victory for each candidate to the actual margin of victory.
Looking at the results individually, it seems the biggest outliers from my projections were Arizona, Colorado, Michigan, and New Hampshire. Two of these states, Arizona and Michigan, were not truly taken seriously by either of the campaigns, and, at least in Arizona, there were very few public polls for the Presidential race in the weeks leading up to the election.
Colorado and New Hampshire, on the other hand, were hotly contested states that were both expected to be extremely close. Mitt Romney even made his last stop of the campaign in New Hampshire, usually a sign of strength for a state. However, both states went strongly for the President, by a margin of 4.7% and 5.8%, respectively.
In the Presidential election, Battleground270’s margins were not drastically off. On average, I missed the actual margin by 1.785 percent. Additionally, my projections had a slight Republican bias of 1.092 percent. Because the projections are made with respect to public polling (along with weighting and personal judgment), this bias can be explained mostly by looking at the polling, as we will do in a minute.
The story was a little different, however, in the Senate. While 49 out of the 50 states I called were correct, Democrats beat my projected margins by an average of 2.607%. And remember, that’s just an average. Some candidates like Senators Claire McCaskill and Bill Nelson ended up clobbering their opponents. Other races were closer to the predicted results, but in all races but three, the Democrat beat the projections.
The race I called incorrectly, North Dakota, was one of those states. Democratic candidate Heidi Heitkamp won in a state that voted for Mitt Romney by nearly 20%. Other Democratic candidates also did much better than expected. So what caused such a large difference in expectations and reality?
Karl Rove and the right-wing talking heads got at least one thing right this election cycle: the polls were wrong. However, where Rove and others were wrong is in the direction that the polls were skewed. During the waning weeks of the campaign, it essentially became GOP orthodoxy to denounce nearly all public polls. The idea was that pollsters were oversampling Democrats, and instead, they believed, Romney was actually winning or tied in most states and was likely to go on to win a landslide victory.
Obviously, that didn’t happen. Instead, as it turns out, most of the polls were actually skewed towards the Republicans (by ~1% in the Presidential election and ~2.5% in the Senate elections). As noted above, Battleground270 uses a weighted average of polls to help calculate the expected final vote. Therefore, Battleground270’s error for the Senate elections is probably at least somewhere close to the overall bias for polling this cycle.
There will be those who try to blame Obama’s larger-than-expected victories on Hurricane Sandy or perhaps other factors. But in reality, it is likely pollsters were too strict when screening for likely voters. They undersampled young voters, latinos, African Americans, and, overall, those who no longer use landline phones. What’s the common link between these groups? They all tend to vote much more Democratic. Let’s hope they don’t make the same mistakes in the future.