Note: This story has been updated.
Ever since the arguably skewed CNN poll of a few weeks back, conservative voters have been looking at the methodology of polling companies with an increasing amount of skepticism. The fact that most polls have used a model that tries to mimic the voter turnout in 2008, when Democrats beat Republican turnout by 7 points (as opposed to presidential elections like 2004, where turnout between the two parties was relatively even), has not improved this state of affairs.
And now, the suspicion of poll bias appears to have yet one more piece of evidence to support it. Via a tweet by anonymous poll analyst NumbersCruncher comes the following graph showing the degree of oversampling of Democrats employed by the most recent polls (all of which show Obama leading Romney):
Blogger keithbacker at Battleground Watch sums up the graph this way:
The real take-away which I have mentioned the times I blog national polls is that many of those national polls are HORRIBLE for Obama, namely the ABC/Washington Post and CBS/New York Times polls where you have large Democrat over-samplings but rather small leads for Obama. This means if Obama doesn’t meet or beat his stellar 2008 turnout advantage he’s in for a drubbing on election day.
These over-samplings serve a few purposes but mainly drive down enthusiasm for Republicans while assisting the Obama campaign with “bandwagon” supporters who simply like being on the winning team (they’re real and they count).
This is a telling bit of information, and shows the power of polls to become a self-fulfilling prophecy. Granted, most of the polls using a high Democrat oversampling may be doing so simply as a means to estimate conservatively what could happen, if things stay the same from 2008. This is not unreasonable, given the enthusiasm gap that has recently been measured, but that enthusiasm gap is itself arguably reinforced by polling. The Romney ground game will need to push back aggressively if it wants to make up for that negative feedback loop.
UPDATE: Libertarian pollster Emily Ekins writes in to dispute the graph, while pointing to other problems with polls. From her email to the Blaze:
First, the graph only shows the party-id breakdown among all survey respondents, (which is all these pollsters tend to report). However, what we need to see is the party-id breakdown among likely voters, since presidential election match-up numbers are based on likely voters, not all survey respondents. Second, these surveys tend to report party-id without sorting Independents; however, most Independents lean Democratic or Republican. Sorting Independents tends to diminish the party id gap. However, I’d be interested if whether these polls get a more Democratic or Republican “cut” of Independents, which could skew the results. Third, the graph’s horizontal axis presumes each sample should have an equal number of Democrats and Republicans, since it’s titled “% of Dem Sample Advantage.” However, survey data typically indicate that more people do self identify as Democrat than Republican and more people self identify as conservative than liberal. All this graph really tells me is that Democrats like Obama, and the more who self identify as Democratic, the higher Obama’s share will be in the survey.However, what we ultimately want to know is how many likely voters self identify as Democratic and Republican. We can turn to Election Exit Poll data for a reference: we find that since 1992, self identified Democrats lead self-identified Republicans by an average of 3 percentage points among actual voters. (2008 is a clear outlier) This of course doesn’t even account for how Independents lean.
Party ID 1992 1994 1996 1998 2000 2004 2006 2008 2010 Avg Democrat 38 36 39 39 39 37 38 39 35 38 Republican 35 36 35 37 35 37 36 32 35 35 Independent 27 27 26 24 27 26 26 29 29 27Source: Roper Center
The leads us to the real issue, which is whether these pollsters’ likely voter models are correct. I found it extraordinarily difficult to track down how these pollsters define “likely voters”, and they only report party ids for all survey respondents, not likely voters. We want to know if these surveys (below) which show Obama leading, have accurate likely voter models, and if the share of likely Democratic voters is reasonable. From these data alone, we can’t tell.