Election Polling: Are Democrats Oversampled? (Pt. 2)
Some Republicans have complained that election polls are biased in favor of Democrats. The polls oversample, or include too many, Democrats, they argue, to fairly represent the true intent of the electorate. To understand this issue, The Christian Post spoke with Scott Keeter, director of survey research at Pew Research Center.
A recent post on a blog called Battleground Watch had a chart showing the percentage point advantage for Democrats over Republicans in the samples of a few recent polls. Generally speaking, with one exception, the greater proportion of Democrats in the sample the greater the lead for Obama in the poll. This graph was shared often across conservative blogs as evidence that pollsters were oversampling Democrats.
Another conservative blog has since been created, unskewedpolls.com, that weights recent polls according to the partisan makeup of the electorate using data from Rasmussen Reports. As of Sept. 29, the blog's weighted average of recent polls shows Mitt Romney with a 7.4 percentage point advantage.
Get Our Latest News for FREE
"It's an age old complaint and it doesn't only come from the right," Keeter explained, because in 2004 it was Democrats who complained that polls were showing more Republicans in their sample.
While these conservative bloggers suggest the pollsters are intentionally surveying more Democrats or weight their samples to favor Democrats, this is not the case. Most polling organizations do not weight their samples for party identification.
"We don't know what the party identification balance is in the country at any given time. That's an attitude. One of many we're trying to measure. So, we don't have any basis for pre-engineering that, either through weighting or through quotas or any method like that," Keeter said.
Unlike demographic characteristics, party identification (not the same as party registration) can change often during an election cycle. Any polling organization that weights its sample according to party identification, therefore, risks imposing its own bias about what it believes the electorate will look like on Election Day.
As pointed out in part one of this series, polling organizations using random digit dialing to select their sample. The partisan breakdown of the sample is a reflection of the randomness of the poll and the questions used to select the likely voters from the sample.
There are two possible reasons, Keeter said, that recent polls show an advantage for Democrats when respondents are asked which political party they identify with.
First, there are demographic changes taking place in the United States that favor Democrats. In particular, the proportion of non-whites, especially Latinos, in the electorate is increasing. Non-whites tend to identify with the Democratic Party more often than whites.
"The demographic future of the U.S., given the loyalties of Hispanics and other non-whites, really favors the Democrats," Keeter said.
Second, and more importantly, the Republican brand has suffered in recent years. It began, Keeter said, during the second term of President George W. Bush. And, while the Democratic brand has declined also, especially in 2009 and 2010, it has not declined to the same degree as Republicans.
As pointed out in part one, pollsters do make judgments about which respondents are likely voters that will impact the partisan makeup of the sample. There are many factors that can be taken into account when deciding who is likely to vote, such as enthusiasm, understanding how to vote, or experience voting. Different polling organizations use different methods to determine a likely voter and these methods are not always disclosed.
"There are things we could probably do in our samples to make our polls less Democratic and more Republican if we took individual questions and gave them more weight in the likely voter scale," Keeter said. "This process of building a likely voter scale in the polls in something that's very individual for different pollsters. It's also the case that most pollsters don't disclose in great detail how they do it. That could be having an impact on the differences among polls, but it could also be having some impact on what your bottom line is."
