Legal Commentary

Everyone loves to talk about how conventions affect presidential candidates' chances of winning. This talk typically focuses on the candidates' standing in national polls; more specifically, these days, the focus is on national tracking polls. As just one, but hardly an exclusive, example, here's Jim Lindgren over at the VC. Among other things, Lindgren writes that "The Obama bounce, which peaked at a 6% lead in the Rasmussen poll released last Tuesday, is gone."

But Obama fans should wait before quaking in their effete, liberal chamois boots, and McCainiac-Palinophiles might want to hold off planning to overturn Roe, slash taxes for high earners and bomb-bomb-bomb Iran. Turns out that a quickly disappearing, 6-point bounce is precisely what history and statistics predict.

At least, it's what Nate Silver over at fivethirtyeight.com did predict more than three weeks ago, based on data from past conventions. Here are the basic facts Silver emphasizes:

1. Bounces don't start to show until roughly the third day of a convention.

2. "From there, the bounce accelerates quickly, peaking approximately 6-7 days from the start of the convention -- that is, the weekend afterward, if the convention runs from Monday through Thursday."

3. Bounces "then dissipate[] in a roughly linear fashion over the next 3-4 weeks".

4. The "average peak bounce [is] about 6 points." In other words: "a 6-point convention bounce represents par. If a candidate gets a bounce larger than 6 points, that can be considered to be a good sign. If the bounce is smaller than 6 points, that can be considered to be a bad sign." [emphasis in original.]

Now, things are a bit screwy this year given the back-to-back nature of the conventions, which is atypical. So, as Silver noted, it's not clear how precise a guide data on previous cycles will be. Nevertheless, here's a chart he put together to project what would happen to relative poll standing over the period starting with the Democratic [note the "ic", there for accuracy] National Convention (click for a larger version).

The key fact here is that based only on historical patterns, Silver was predicting a net McCain gain of 2 points, relative to the day before the DNC started. Using the Gallup daily (which has its unit root-induced problems), the race was exactly tied in Gallup's August 24 release. This means that, using Silver's logic, McCain will be behind the curve if he isn't up 2 in today's Gallup daily.

Having said all that, the national trackers aren't worth much at all. As a professional statistician, I highly recommend Silver's site. It's one of the best places on the net if you want to understand the real state of the race in terms of electoral votes.

Updates below the fold....

Update:I just looked up the Rasmussen daily from August 24, and it had Obama up by 3; today's Rasmussen daily shows a tie. So by this metric, McCain has gained 1 more point than the historical norm. (I hope I don't have to explain that that is nowhere near a statistically significant difference.)

Update #2:Today's Gallup daily is out, showing McCain up by 3. So McCain has gained 1 point over the predicted gain (for today) in both polls. Putting the two dailies together provides stronger evidence against the one-sided null hypothesis of a McCain gain of 2 or fewer points. I doubt that one could reject that null, but, at least for those who want to focus on national trackers, this is good news for the McCain crowd.

Update #3:It occurs to me that some/most/all of the data that Silver used in his model is probably from non-moving average polls. Because today's trackers imply bigger effects for last night's interviews than for Wed night interviews, McCain's bounce could be seen as larger. On the other hand, drawing that inference requires one to know the August 23rd-only numbers, which isn't possible from the moving avg trackers (again, see Brad Delong's post linked above). Moreover, my guess is that if one took the same tack for Obama's bounce, one would find that it was considerably larger than usual as well. So it looks like both candidates may have had larger than usual bounces, with the net effect of the larger-than-usualness roughly washing out, at least up to sampling error.