Prior to the start of the Democratic National Convention last week, we introduced a convention bounce adjustment that adjust polls taken during the convention period based on historical averages of convention bounces from elections past.
Presently, this adjustment has us subtracting approximately 6 points from Barack Obama's numbers in each poll conducted over the weekend. So a poll that shows it Obama +4 will be adjusted to a McCain +2.
Over the course of the next week, however, the adjustment will begin to reverse itself, eventually deducting from McCain's rather than Obama's numbers. Specifically, over a period of approximately 2-3 weeks from the Republican convention, it will subtract about 2 points from McCain's numbers:
Under ordinary circumstances, convention bounces are fairly predictable, and I am convinced that the adjustment would be a Good Thing. However, this year's conventions are unusual for at least four reasons, a couple of which we could anticipate in advance and a couple of which we perhaps could not:
1. The conventions are conducted nearly back-to-back, which has never happened before.
2. Both conventions are being held very late. Ordinarily, conventions are held earlier in the summer, where bounces might linger for a longer period in the absence of other campaign news.
3. McCain picked his VP immediately following the Democratic convention; VP selections also typically produce a bounce, although we had not built in an adjustment for this.
4. Hurricane Gustav could materially impact the Republican Convention in unpredictable ways; the first day of the convention has already effectively been canceled.
Given these conditions, I think the convention bounce adjustment may be introducing more noise than it is removing, and Occam's Razor would dictate that we should remove it unless we clearly feel that it makes the model better. On the other hand, I don't want it seem like I'm changing the process in midstream simply because I don't like the results I'm getting (although keep in mind that as of a few days from now, the convention bounce adjustment will begin to benefit Obama rather than McCain).
So, I'm going to throw the question out to you guys. If a filibuster-proof majority of 60 percent of you think we should dispense with the convention bounce adjustment, we will do so. Otherwise, we will keep it. Vote early and often.