"Battleground Drops It weighting Or Shifts The Poll Numbers To The Liberal Side !"
( Or A How to Manipulate a Tracking Poll in one easy step)
We missed doing a "poll update" on the GWU Battleground Poll today, partly because we did not receive the usual PDF via email with the daily results. The results flipped on that survey flipped today from the 1-2% lead that they had been showing for John McCain for the last seven releases to a two-point Obama advantage (48% to 46%).
The difference is explained at the bottom of the slide in the PDF. "
On 9/29 - Weighting changed from Party ID, Race, and Age to Race and Age." So according to the graphic, they stopped weighting by party identification on today's release. Does that explain the reversal in vote preferences? By all appearances, it does.
I talked separately today to both Republican Ed Goeas earlier today, and his Democratic counterpart Celinda Lake, the two pollsters who oversee the Battleground survey. Goeas explained that their party weights had held Democrats to a three-point advantage on the initial party ID question ("Generally speaking, do you think of yourself as a Republican, a Democrat, or an Independent?"). With the party weight removed, he said, Democrats now hold a seven point advantage on party identification. With the party leaners included, Goeas said, the Democratic advantage grows to 9 points.
Goeas also said that removing the party weight had the effect of changing the age distribution. It increased the weighted value of respondents between the ages of 18 and 34 from 17% to 22% of their sample.
Last night, Nate Silver flagged the age of the Battlegound survey as a likely explanation for the fact that, until today, only the Battleground survey showed John McCain with a greater percentage of the national vote than Barack Obama. Nate was right to question the relatively small number of younger voters in the sample.
He pointed out that the U.S. Census Current Population Survey (CPS) showed 18-to-34-year-olds as a larger percentage of the voters in 2004 (23.8% -- while Silver imputed a slightly higher value, this spreadsheet on the Census page he linked to allows for a calculation of the more precise statistic).
In my National Journal column, set to appear tomorrow, I'll have more thoughts on the issue of weighting by age.
The quick version is that while there is no ideal "right" answer for the age distribution of the electorate, the CPS estimates from 2004 are not a bad place to start.
"The real issue," Celinda Lake explained by email, "is the turnout model." She continued (link added):
I believe there will be heightened youth turnout and our work together for Rock the Vote and our internal polling has shown heightened youth interest and turnout. I believe we have been underestimating the youth turnout with a traditional, conservative turnout model. We have moved to a model with higher youth turnout ans well as looking at age in the weights.
Bottom line: The change in the Battleground weighting today most likely accounts for their net four-point shift on their survey in Obama's favor. I'll have more details tomorrow when my column is posted