COLUMBUS, Ohio-Sept. 26, 2012-There’s a lot of noice out there right now regarding the polling of the race for President of the United States. Some polls show Democrat Barack Obama up by a few points, while others show it is a really close race. Almost no polls show Republican Mitt Romney with a meaningful lead. That does not necessarily reflect what is really going on.
The reasoon Obama seems to enjoy a lead over Romney is straight-forward, and by now, well-explained: it is because the survey sample includes more Democrats than Republicans.
The next question is the most important: “What is the correct mixture of partisans in a national poll?”
Different pollsters have different answers. Most, especially national media pollsters, have apparently decided that their baseline partisan mix should be based on the results of the 2008 presidential election exit polls. This is problematic for many reasons – chief among them being that election exit polls today are terribly flawed. Given the growth of vote-by-mail and early voting, it is no longer reasonable to assume that the people who show up to vote in person on Election Day are representative of the voting public as a whole. And then, depending on what time of day an exit poll sampler is gathering data, and at what polls they are gathering data, information can be badly misrepresentative of the electorate they are trying to measure.
Of course, most would agree that the 2008 electorate was not a normal presidential election turnout. The reasons for this are well-documented – the first time an African-American candidate topped a major party ticket and the collapsing economy being the two most obvious. Knowing this, it is also obvious that pollsters should not use 2008 turnout as their model for surveys gauging this year’s presidential race. That some still do is disturbing because it is tantamount to an admission they would rather be politically correct than statistically correct.
The other factor largely ignored by most pollsters is the rise of the TEA Party movement. Even Rasmussen does not include the TEA Party as a factor in his weighting of survey data for the presidential race. Instead, he has said recently that he uses some formula based on an aggregation of 2004 and 2008 election turnout data.
I appreciate that effort, and it brings Rasmussen closer to reality than most other pollsters, but in order to properly gauge election turnout, some calculation of the effect of the TEA Party movement has to be included, For us pollsters, this is where things get dicey, because we don’t have any presidential election experience in the TEA Party era. We have to make our own turnout models based on an aggregation of previous presidential election turnouts and the 2010 midterm elections, which admittedly is very difficult.
It has been said that public opinion research is part science and part art. This process of weighting our survey samples is the artistic part of the process. It’s just that some pollsters are more creative than others.
This morning’s release of new surveys of Ohio and Florida voters shows Obama with huge leads over Romney, but those New York Times/Quinnipiac polls also include a 9 percent Democratic Party advantage over Republicans. This is likely because Quinnipiac no longer weights its surveys for partisan affiliation. It apparently gave up the practice in order to work with the New York Times, whose polling gurus have long eschewed such a step in processing polls. They simply believe it is proper to take whatever sample they get, and they got far too many Democrats in these two key states.
One more point about university-based polls to measure presidential elections: I have no first-hand knowledge that the Quinnipiac callers skew their interviews to help Barack Obama, but you should know that all of those survey interviews are conducted by college kids, the vast majority of which are very likely to support Obama over Romney. It simply introduces another opportunity for error to be injected into the polling process.