“A linebacker’s job is to knock out running backs, to knock out receivers, to chase the football” – Ray Lewis
We’ve had a bit of discussion here (and off-the-books) about election predictions (with some mild hand-wringing that all those “Obama is still in the lead” folks, like Nate Silver, are late to the game). Since the election is coming soon, and the consequences for another four years after, I figure I can spare a few more posts on the topic.
Let’s consider the simplest of all possible prediction models – the Real Clear Politics average of polls. On the face of it, poll averaging may not be the most exact form of measurement. But, knowing a thing or two about system biases (I estimate them for a living … sometimes) I will say that it’s not unreasonable to average a set of precise but biased results to estimate out the remaining poll bias. So, how well has the RCP average of polls predicted election outcomes in the past? Let’s look at the last two elections.
At this point in 2008 the RCP average was Obama 50.4 – McCain 42.8, for a total of 93.2% “decided”. The final outcome was Obama 52.9 – McCain 45.7, for a total of 98.6% committed to one of the major party candidates. In terms of relative percentages, RCP had it 54.08-45.92 while the final was 53.65-46.35 … not bad at all. Further, the “undecideds” who chose a major candidate broke 46.3% for Obama and 53.7% for McCain – pretty even.
At this point in 2004 the RCP average was Bush 48.8 – Kerry 46.0, for a total of 94.8% “decided”. The final outcome was Bush 50.7 – Kerry 48.3, for a total of 99% committed to one of the major party candidates. In terms of relative percentages, RCP had it 51.48 – 48.52 while the final was 51.21 – 48.79 … again, pretty darn close. Further, the “undecideds” who chose a major party candidate broke 45.2% for Bush and 54.8% for Kerry.
The current RCP average in the 2012 election is Romney 47.9 – Obama 47.0 for a total of 94.9% “decided”. If the undecideds break 50-50 and 99% end up choosing one of the major party candidates the outcome looks to be around Romney 49.95 – Obama 49.05, razor thin but Romney ahead. For the popular vote to go to Obama, those undecideds would have to break in his favor 60-40. (For the popular vote to go 50.2 – 48.7 as Nate Silver predicts, those undecideds will have to break 76-24 in favor of Obama … and Silver will be a legend.)
The current slate of polls tell another disturbing thing for the president though. Romney’s 47.9% average comes with a standard deviation of 1.8. Obama’s 47.0% comes with a standard deviation of 1.4. Of the 10 polls in the current average, 6 have Obama at 47%, 2 have him at 45%, one at 48%, and only one at 50%. Romney, on the other hand, is at 50% in three polls, 49% in 1, 48% in 1, 47% in 3, 46% in 1, and 45% in 1. What does this mean? Well, probably not much. But if you want to weave a narrative into it the obvious one his this: Obama is locked in tight at 47%, while Romney is at 50% when the “undecideds decide” and 47% when they don’t. An ever-so-mild indication that the undecided break might actually go in Romney’s favor.
Tomorrow’s a new day, and the polls will be what they will be – as will the election. Much campaigning to do, and many more October surprises to come.