It was four years ago next week that I launched the MysteryPollster blog with a post about party identification. Left leaning bloggers were up in arms at the time about surveys showing a post-convention bump for George W. Bush and attacked pollsters for what they believed was an over-representation of Republicans in the survey samples.
If not, you missed my June column on the subject and last weeks' Huffington Post article by Seth Colter Walls. The same argument has come up about three recent polls -- by CBS News, USA Today/Gallup and the Gallup Daily tracking -- that all showed John McCain receiving a convention bounce and taking the lead over Barack Obama.
"Interestingly," Walls wrote, "all three polls were... conducted using a higher sampling of Republican voters than in July, raising a question of methodology."
More disturbing to Walls was the apparent conflict with recent trends in voter registration statistics. "In a year in which Democrats have a lead of 11 million registered voters over Republicans," he wrote, "and have been adding to that advantage through a robust field operation, are pollsters over-sampling Republicans?"
The short answer is that the changes in party registration are not necessarily inconsistent with the apparent short-term boost in Republican identification. To understand why, and to help navigate the debate over this topic now and in the future, let's review some basic concepts.
Party ID is an attitude. As asked by most pollsters and studied by political scientists for more than 60 years, party identification is determined by asking what voters "consider themselves" when it comes to politics. Unlike demographic characteristics such as race and gender, attitudes can change.
While most American voters acquire an identification with either the Republican or Democratic party early in their adult lives, and tend to keep that sense of attachment throughout their lives, a small number do not. Pollsters and academics who have done panel studies that track the same individuals over time have found that a small number will sometimes shift their answer to the party ID question, particularly around the fall election campaign.
The changes are usually from the independent category to one of the parties and back again. Changes from one party to another over a short term are very rare. But it can happen, particularly when perceptions of the president or the presidential candidates help shift impressions of the two parties.
Over the last four years, for example, the Pew Research Center has noted a 5 percentage point drop (from 33 percent to 28 percent) in Republican identification, a 3 point increase in independents (from 32 percent to 35 percent) and a 2 point gain in Democratic identification. (The 2008 numbers were measured before the party conventions.)
Party registration is different. Yes, those who check the box for a party when they register to vote will usually tend to "identify" with the same party when asked by pollsters, but not always. Many states lack formal party registration. In some states, most voters are aware of their registration status and can report it accurately. In others, many have no clue. As a campaign pollster, I sometimes asked respondents to report both party identification and registration, and their answers were often inconsistent.
Walls cites an AP report on gains in Democratic registration in the 28 states that register voters according to party affiliation. Those trends mirror the long-term increase in Democratic identification cited above. But we should also remember that if the Republican convention served to re-energize Republicans and redefine the GOP brand around McCain and Sarah Palin, many voters may have shifted overnight from a sense of independence back to Republican identification. Their registration status is irrelevant.
The wording and placement of the question matters. Not all pollsters ask about party ID the same way. Some follow the lead of the University of Michigan's National Election Survey, asking respondents, "Generally speaking, do you consider yourself a Republican, a Democrat, an independent or what?" Some follow the slightly different version first created by Gallup that asks, "In politics, as of today, do you consider yourself a Republican, a Democrat, or an independent?" Many political scientists believe that phrase "as of today" makes the Gallup version more responsive to short-term changes.
Others have found that that they get different results when they ask about party ID at the beginning of the questionnaire rather than the end. Since most public pollsters ask the question at the end, the topics covered in the survey may sometimes cause a small number of respondents to report different leanings than they would have otherwise.
Keep in mind that some pollsters ask a follow-up question that pushes the initially independent about which way they "lean." And minor differences in the procedures interviewers use to ask the questions and record the answers can also create slightly different "house effects" from pollster to pollster.
All of these differences make it tricky to compare party ID results from poll to poll, and especially from pollster to pollster.
Scrutinize party composition, but with care. The one change since 2004 is that more and more pollsters are routinely including their party identification results in public releases. That is a positive development. The debate over whether pollsters should weight by party rages on -- and tends to be particularly acute at moments like this one, when party weighting is most hazardous -- but at least we are in a better position to evaluate each new poll.
The more we know about each new poll, the better.