Tracking Polls and the Dead Cat Bounce
I’ve been a little discouraged for the last few days by the polling news. The conventional wisdom is that Obama got a bounce and Romney did not. But is that really true?
The best way to gauge the impact of short term events are daily tracking polls. The tracking pollster uses the same demographic assumptions and polling methodology for each daily poll. This insulates the results from changes in methods, questions, or models, which should highlight the impact of daily events. The two best tracking polls are Rasmussen, which tracks Likely Voters (LV), and Gallup, which tracks Registered Voters (RV):
An initial look at the tracking polls seems to offer conflicting answers. Rasmussen indicates that the Obama bounce has dissipated, while Gallup shows Obama coming out with a big lead. Which one is correct?
The answer is not readily apparent, because of a fundamental problem with comparing those two polls. Rasmussen uses a 3 day average, while Gallup uses a 7 day average. If you compare the polls by their end date, you are comparing two completely different periods of time.
To work around this, I decided to compare the two polls not by their end date, but by their start date. Instead of comparing the Rasmussen results from 24-27 Aug with the Gallup results from 21-27 Aug, let’s compare the Rasmussen results from 27-29 Aug with the Gallup results from 27 Aug – 2 Sept. This may give greater insight into the effect of daily events, as the points of inflection for the two tracking polls should be in sync.
The results are interesting. See below for Romney’s and Obama’s support levels over the convention period:
The results shows a couple of things. First, Gallup and Rasmussen do track more closely when start dates are compared. Second, Rasmussen is more volatile than Gallup, with wider swings in magnitude. This can be expected since Gallup averages across 7 days instead of 3, which would attenuate daily spikes. Finally, Gallup is a lagging indicator, with the impact of daily events showing up 4 days after Rasmussen.
Rasmussen shows that each candidate received a bounce during and immediately after their convention, and that each bounce began to dissipate soon after the convention. Romney’s bounce was less noticeable because it was offset by Obama’s bounce. Obama’s bounce has since dissipated, and it appears that Romney will come out of the convention period ahead of where he was before it.
Why does Gallup not show this? Because they have not reported that data yet. Until yesterday, Gallup was still reporting a mixture of pre- and post-DNC data. Today was the first release composed entirely of post-DNC polling data. Based on correlation between the two polls, over the next few days we can expect the Gallup to mimic what Rasmussen has done over the past few days: shift from an Obama lead to a Romney lead or at worst a tie. Looks like that Obama bounce was a dead cat bounce after all.
The correlation between Gallup and Rasmussen is even stronger when comparing the midpoint of their date ranges. It looks like Rasmussen leads Gallup by 3 days on the nose:
Gallup consistently polls 1 point higher for Obama and 1 point lower for Romney than Rasmussen, a well-documented result of the difference between registered voter and likely voter screens. The spike in support for Obama on Rasmussen last weekend was clearly a reflection of registered voters getting temporarily motivated by the emotional appeal of the convention. But like new years resolutions, it does not appear that their enthusiasm lasted very long.
The Gallup release this afternoon should show a 1-2 point drop for Obama and a 1-2 point rise for Romney. The Obama lead should drop to 3 today and return to 1 tomorrow, where it has been since before the conventions. Yeah, I’m calling my shot.