Now that we are turning our attention to the general election, an important question comes up often: When it comes to tracking the presidential "horse race," does it make any sense to follow the national polls?
After all, as we were reminded eight years ago, we do not choose our presidents with a national popular vote. Instead, we vote for presidential electors awarded on a winner-take-all basis within each state (and, in Maine and Nebraska, by congressional district). As a result, the most appropriate way to track the presidential contest would seem to be with state-level polls. The presidential campaigns will be devoting most, if not all, of their internal survey research budgets to state-level polling.
So why look at national media polls? Let me suggest four reasons.
By October, we will likely see tracking surveys released weekly (or perhaps more often) in the dozen or so most competitive "battleground" states. For now, however, state-level polling is far more sporadic.
As of this writing, the most recent available surveys in Ohio, Iowa, Colorado, Nevada, New Mexico and New Hampshire are more than 20 days old. We have seen new releases in only two of the battleground states so far this week -- in Wisconsin and Michigan -- and both are automated surveys from Rasmussen Reports, a source still viewed with skepticism by many of my pollster colleagues.
At the national level, however, we have seen four new surveys in the last week, including the daily tracking polls produced by Gallup and Rasmussen. Those two both show significant gains by Barack Obama since Hillary Rodham Clinton endorsed him and departed the race.
Hopefully, we will see more statewide tracking in the fall, but for now, national polls provide the most current data.
At the risk of overgeneralizing, the big national media polls tend to be more rigorous in the way they call their sampled numbers. In practice, especially at this early stage of the campaign, the national polls typically use longer field periods (five or six days) and call back numbers more often when no one answers the first time. Such polls are also more likely to try to select a random respondent within each sampled household rather than interviewing whomever answers the phone.
While these practices should, in theory, produce more representative samples, pollsters disagree on how much practical difference they make. We often relax our procedures closer to Election Day in order to allow for faster turnaround when voter preferences are shifting.
Again, not all surveys at either the state or national level are created equal. I can think of statewide polls that are as "rigorous" as any national survey and national polls that use less exacting methods. However, my point is that the national media surveys typically take a different approach to methodology that we ought not ignore.
At least three national survey organizations -- Gallup, the Pew Research Center and CBS/New York Times -- are now routinely interviewing respondents on their cell phones in order to reach those Americans in "cell phone only" households.
As of late 2007, one in six American households had wireless service only (no landline) and would thus be missing from traditional telephone samples. Gallup has been doing supplemental cell phone interviewing since January and has found that the additional cell phone interviews may produce samples slightly more supportive of Barack Obama. As Gallup and others continue to interview by cell phone, they will be able to determine whether those preliminary findings are real.
If so, the survey organizations ready to interview over cell phones will have an important advantage. Unfortunately, federal law [PDF] bars pollsters from conducting automated surveys over wireless phones, and many statewide surveys are now conducted using the automated "interactive voice response" (IVR) methodology.
Beyond The Horse Race
Gary Langer, director of polling at ABC News, is fond of advising us to "throttle back" on our obsession with horse-race polling results. While it may be harder for some of us to follow that advice than to stop blinking or breathing, his argument is grounded in an important characteristic of most national media polls: They tend to go deeper into "internal" measures that tell us far more about perceptions of the candidates, the campaign, the issues and the direction of the country than about which candidate is ahead or behind at any moment.
The budgets of most national polls typically allow for longer questionnaires that probe these topics in depth. Again, there are more than a few exceptions, but many statewide surveys we follow ask only a horse-race question and little else.
My point here is not to argue that national surveys are "better," only to urge against putting on blinders and focusing only on polls that yield state-level numbers.