With Hillary Clinton’s imminent departure from the campaign, we have a little bit more time to incorporate some new features. One thing I’m thinking about doing is finding a way to include national polling in the averages. I have a relatively intuitive solution to the problem. But I’m not sure if I like it so I’m hoping to rely on the wisdom of crowds (e.g. you guys) to sort everything out.
Right now, our state-by-state polls show a Barack Obama ahead by a tenth of a percent (0.1 points) against John McCain, if you extrapolate them outward to a popular vote based on the relative turnout each state had in 2004.
However, if you apply our methodology to national polls, they show Obama leading McCain by a somewhat larger margin — 3.1 points.
There’s nothing sacred about national polls. In fact, we tend to pick up more information from the state-by-state polls — we simply get more volume out of them. The total weight of all current national polls is 26.11; for state polling, it’s 74.16.
However — we are arguably throwing some information away when we ignore national polls, and particularly the perspective of pollsters who might not weigh in at the state level very often.
One workaround would be to combine the state-by-state and national numbers into one global average, and then adjust the state-by-state polls to match the combined average. Presently, for example, the joint average would look as follows:
State Polls 74.26 Obama +0.1
National Polls 26.11 Obama +3.1
TOTAL 100.37 Obama +0.9
If you combine the state and the national polls according to our weighting method, Obama “should” be ahead by 0.9 points in the popular vote. Instead, we actually have him up 0.1 points based on the state polls alone. So the solution would be to add 0.8 points to his polling in each state; this is what we’d call the “national polling adjustment”. Of course, at other points his numbers might be adjusted downward. It all depends on what’s happening with the national polls. We wouldn’t give deference to the national polls, but we would give them some say on our model.
Do we like this approach? Or do we think it ruins the purity of what we’ve got going on now?