On November 2, Quinnipiac University Polling Institute released a new national presidential poll that places President Obama’s approval rating at 47 percent – up six points from last month’s Quinnipiac poll. The poll also had him at least five points ahead of all the Republican primary contenders, and indicated 49 percent of respondents think the president deserves re-election, up from 42 percent last month. The poll results appear to be positive news for President Obama, but demographic data shows that Democrats were sampled more than Republicans by a 13 point spread. This calls into question how random sampling could appear to favor one party over another, and whether the Obama administration should be confident in embracing the poll results.
Presidential Approval Poll Analysis
Quinnipiac University surveyed 2,294 registered voters via random phone calls to landlines and cell phones between October 25 and October 31. An additional sample of 869 likely Republican primary voters was surveyed regarding likely primary votes. Quinnipiac University identifies likely voters based on self-identified voter registration. Weighting was used to adjust raw data in accordance with data from the most recent Census. Only demographic criteria was weighted, including gender, age, education, race and region. The margin of error for the general voters was +/-2.1 points, and for the Republican primary sample was +/-3.3 percent.
The survey closely followed NCPP standards and the scientific method. Interviewers were bilingual and called during varying times of day and week. Software was used to ensure that both listed and unlisted phone numbers were being dialed.
Despite all reasonable efforts being used to ensure random sampling, the pool of registered voters who participated in the poll consisted of 36 percent Independents, 35 percent Democrats and 22 percent Republicans. All political party identification was self-identified and unweighted. The split among Independents was 48 percent leaning Democrat and 35 leaning Republican, with 9 percent not leaning and 6 percent leaning toward a different party.
Why is the Party Affiliation Spread so Large?
Quinnipiac University Polling Institute methodology states that respondents were strictly contacted at random. There is no explanation for the 13 point spread between Democrats and Republicans besides the consequences of sampling. When the same poll was conducted by Quinnipiac University last month, the spread was much smaller with 33 percent Independents, 31 percent Democrats and 28 percent Republicans. Independent voters leaned Democratic by 40 percent, Republican by 43 percent, 9 percent not leaning and 6 percent leaning toward a different party. For comparison, a separate Gallup poll indicates that as of 2010, 38 percent of Americans considered themselves Independent, 31 percent identified as Democrats and 29 percent identified as Republicans. Despite this, when questioned about general ideology by Gallup in the same year, 42 percent identified as Conservative, 35 percent identified as Moderate and 20 percent identified as Liberal. These findings suggest that many of the Independents who identified themselves as having no preference between Democrats and Republicans may follow a Conservative ideology.
The Case Against Party ID Weighting
If Quinnipiac had weighted the survey for party identification, they would have placed more weight on the responses provided by self-described Republicans, causing the spread between the parties to more closely mimic the national spread of party identification, which is typically estimated by using exit poll results from previous elections. While this would have made the poll appear more balanced, several experts, including the NCPP and renowned pollster Mark Blumenthal, discourage weighting for party identification. The argument against this type of weighting is that party identification is not constant in the way demographic factors such as age and gender are. A respondent’s party identification could fluctuate based on current events, or even as a response to the poll questions that preceded the demographic categorizations. Traditionally, polls that are unweighted for party identification tend to indicate that majorities of respondents identify most with the party of the candidate who was highly rated. Blumenthal says it is possible that feelings about particular candidates can dictate a respondent’s party identification at a given moment, especially when the respondent is a moderate Independent and is asked to which direction he “leans.”
Weighting for party identification was usually considered good practice prior to the 1980s, when respondents were more likely to permanently adopt party affiliation. In modern years, the group of Americans who consider themselves Independent has been on the rise. Recent polls overwhelmingly indicate that more people describe themselves as Independent than Democrat or Republican. Popularity trends among third parties and ideological movements, such as the Green Party and the Tea Party, further divide respondents.
Those who do not support weighting for party identification conclude that this type of weighting is more equivalent to weighting the response to an opinion question than it is to weighting a demographic, and would result in inaccuracy. Critics of party identification weighting believe that as long as a poll is conducted scientifically, the random sample will usually reflect the views of the general population. Although both parties tend to cry foul when polls show a large spread between parties, the perceived oversampling tends to shift based on general public opinion of candidates, and their respective parties, at a given time.
The Case For Party ID Weighting
Some polling experts, including Kristen Solits, Director of Policy Research for The Winston Group, a Republican affiliated public opinion research and strategic consulting firm, argue that party identification weighting can be a useful way to neutralize sampling errors and gain a clearer prediction of election results. Proponents of party identification weighting make the point that, although stated party identification in non-election day polls is flux, election exit polling suggests that Americans tend to stand with their party when voting in presidential elections. This is evidenced by the fact that party identification in exit polls has remained relatively static over the past twenty years. Soltis states that if party identification reported at exit polls is looked at as a relatively constant demographic, similar to geographic region or education, it could be used to eliminate some of the sporadic fluctuations characteristic of non-election day polls, via weighting.
Interpreting the Presidential Approval Poll Results
Democrats are likely to argue that the Quinnipiac poll indicates positive results for both President Obama, and the Democratic Party as a whole, while Republicans will argue that the poll was flawed due to oversampling of Democrats. Similar results from additional polls could signal positive response to recent events such as the Gaddafi execution or Obama’s nationwide tour promoting his jobs bill. If additional polls show lower approval ratings for Obama, and fewer respondents identifying as Democrats, the theory that this particular poll was inaccurate will become more credible. Any poll, when considered alone, should not be regarded as factual without several other scientific polls backing its findings.
Quinnipiac University. November 2, 2011 Poll, October 6, 2011 Poll. Accessed November 3, 2011.
Jones, Jeffrey. Democratic Party ID Drops in 2010, Tying 22-Year Low. Gallup. Accessed November 3, 2011
Saad, Lydia. In 2010, Conservatives Still Outnumber Moderates, Liberals. Gallup. Accessed November 3, 2011
NCPP. The Good & Bad of Weighting the Data. Accessed November 3, 2011
Blumenthal, Mark. Weighting By Party. Mystery Pollster. Accessed November 3, 2011.
Soltis, Kristen. Party ID: The Case for Weights and Historical Margins. Pollster.com. Accessed November 3, 2011.
Decoding Science. One article at a time.