David Plouffe called the Gallup Daily tracking poll the worst thing to happen to journalism in 25 years. The reason he said this is that it encourages journalists to misinterpret statistical noise as actual movement in the race.
Over this weekend and in the beginning of next week, this is going to come into play in a major way which will probably not be good for Obama.
Follow me below the fold to see why...
The Gallup Daily tracking poll, as you might know, is reported as a 3-day rolling average. In other words, they poll every day, but the number they report isn't the result they polled the day before -- it's the average of that result and the two days before it. This is to reduce the "margin of error", to prevent 'blips' in polling caused by a bad sample -- something that's unpreventable in polling -- from appearing as a trend. This does work, but only to a point.
Although Gallup does not release the raw results of its daily surveys, it is possible to estimate what their results were from day to day. And this is how I discovered the problem that's going to rear its ugly head on Sunday, and is probably going to color the coverage of the Republican Convention next week.
Today's tracking poll release shows Obama expanding his lead, but it's not because his poll numbers improved from Wednesday to Thursday. It's actually because the result on Thursday was better than the result on Monday, which was part of the result reported yesterday, but not in the one reported today.
Here's a chart of what the daily results probably look like (courtesy The Left Coaster).
August 20: O 45 M 43
August 21: O 45 M 44
August 22: O 45 M 45
August 23: O 45 M 46
August 24: O 45 M 44
August 25: O 42 M 48
August 26: O 48 M 40 (post-Michelle)
August 27: O 54 M 38 (post-Hillary)
August 28: O 45 M 45
[EDIT: Today's Gallup Daily result for August 28 is the average of the three bolded entries above -- when you take the average of those three days' polling, it gives you Obama 49 McCain 41.]
Now, what's going to happen in the next couple of days? I estimate that there's significant error in the last 4 samples and the reality is probably that Obama is ahead by 4-5 points, which would represent a small convention bounce.
In tomorrow's release, Gallup will replace the August 26 polling with the polling being done right now. If Obama doesn't show at least an 8-point margin in today's poll, tomorrow's tracking poll will show McCain gaining again.
But the big problem comes on Sunday, when the August 27 polling falls off the average. Obama showed a 16-point lead on that day's polling -- almost certainly a major error in his favor. When this major error suddenly leaves the average on Sunday, Gallup will report a major drop in Obama's support. The temptation for the talking heads to ascribe this to Sarah Palin will, of course, be utterly unavoidable, even though as I've clearly shown, this "drop-off" was etched in stone because of something that happened on Wednesday.
This has the potential to form a very unfavorable narrative for Obama going into the Republican Convention. I don't know what what we can really do about it, but one thing we could do is to stop paying attention to the daily Gallup results. They are inaccurate at best and misleading at worst.
<hr>
Footnote: Let's estimate what the next 3 days of Gallup tracking poll releases are going to look like. Assume that Obama actually holds a 5-point lead, 48-43, which would represent a respectable convention bounce. Let's be very favorable to Obama and assume that Gallup gets lucky and gets this result 3 days in a row. What is the graph going to look like on Monday?
Sat: O 49 M 42
Sun: O 47 M 43
Mon: O 48 M 43
It has the potential to be worse than this, too, but as you can see, having his lead drop from 9 points today to 4 on Sunday would actually be a good result for Obama.