Romney supporters were up in arms regarding a Pew poll released two days ago of 1,956 registered voters that showed Obama leading Romney by 10%, 51%-41%. Much of the controversy centered on Pew’s sample of registered voters, which included 459 self-identified Republicans compared to 813 Democrats, with the remaining 684 declaring as independents. Critics argued this partisan distribution was heavily and unfairly skewed toward Democratic respondents. They pointed out that the partisan difference in 2008 – a strong Democratic year – was only 7- in favor of Democrats according to the CNN exit polls, and that two years ago the parties were evenly divided in the midterm elections. But the Pew poll suggests that partisan advantage has almost doubled.
How can Obama be up by 10%, particularly when the forecast models I’ve discussed in previous posts based on the fundamentals suggest that in contrast to 2008, this race is practically a dead heat? To understand the seeming discrepancy, you should keep four key points in mind. First, remember that Pew doesn’t simply report the raw sample results; they do adjust the sample to bring it in line with broader demographic variables such as race, ethnicity, and even cell phone use. So, Obama’s final margin is based on a weighted sample of 38% Democrats, 25% Republicans, and 33% independents – not the 41% Democrat subsample that some sources reported yesterday.
However, Pew does not adjust based on partisan self-identification – that is, they don’t try to “match” the sample to some national distribution of Democrats and Republicans. Critics who think they should do so are mistaking the nature of partisan self-identification in these polls. Often, they confuse self-professed party identification with party registration data. Party identification is an attitude – not a demographic figure akin to one’s religion, or race, or ethnicity. As such, although partisan identification is generally stable on the whole, it can change, and often does in response to how survey respondents are reacting to the current race and candidates. As evidence, consider this chart from Pew that shows the changes in party self-identification dating back one year.
You can see that the number of self-professed Democrats and Republicans has varied by as much as 8% during this period. Yes, yesterday’s sample contained a relatively high number of Democrats, but the average partisan Democratic lean across this period is about 7%. Keeping in mind that there is always some random variation in probability sampling, a 10% Democrat skew is not unexpected in a Pew poll, given changes in past partisan distributions among registered voters.
Third, as I noted many times before, polling organizations like Pew all have their own house formulas for choosing a survey sample and deciding how to weight survey results. Pew’s methodology has consistently produced results that show Obama leading Romney in head-to-head matchups. Other polling firms, such as Gallup, show this to be a much tighter race. It is for this reason that I have long cautioned against relying on any single poll to gauge the state of the race. But neither should you discount any poll, such as Pew, just because you don’t like their results. Instead, I prefer to rely on the aggregate tracking polls, such as those at RealClearPolitics or Pollster.com. Although they also have their own methodological bias based on how quickly they adjust trend lines to the latest polls, and the degree of “smoothing” they use to handle polling outliers, I think they give a better indication of the actual state of the race than does any single polling outfit.
What do they show? As of yesterday, the RealClearPolitics “poll of the polls” has Obama up over Romney by 2.7%, 47.3%-44.6%. The Pollster.com aggregate poll, meanwhile, shows the race dead even with both candidates pulling in 46.1% of the vote. For what it’s worth, the TPM poll tracker has Romney up by .6%, 46.3%-45.7%, but since this is a left-leaning partisan site I don’t normally look at the polling data there.
The aggregate polls, then, indicate that this is a very close race – one that is too close to call. That is consistent with at least some of the political science forecast models I’ve discussed in previous posts. Note also that at this point most polling firms are still sampling registered and not likely voters for the simple reason that their likely voter screens don’t achieve a high level of accuracy until we get closer to Election Day. Historically, for reasons dealing with education, income and other factors related to voting, the pool of likely voters skews a bit more Republican than does the one based on registered voters.
A final point. Several of you have emailed arguing that based on Electoral College forecast models, Obama has a much bigger lead than is indicated by the popular vote. At this point, however, I’m not ready to pay much attention to state-level polling and Electoral College projections, in part because there’s not always enough state level polls on which to gauge the likely outcome. But even at the national level, polls are still not nearly as predictive as they will be in another month, as this graph based on research by Chris Wezlien and Robert Erickson indicates:
Moreover, history says that it is highly unlikely that the winner of the popular vote is not going to take the Electoral College as well. Yes, there are exceptions – and this year might be one of those. Of course, at some point shortly before Election Day, these state-level polls will become numerous and accurate enough to project state winners and to calculate an Electoral College result. Until then, however, I don’t see much point paying any attention to state-level projections. The aggregate polls tell us pretty much what we can hope to know at this stage of the race – and they suggest a very close contest.
Actually, aside from the Wonks, nobody really pays much attention to the polls until after Labor Day.
But, if you must be looking at early polls, I suggest those where they only poll “Likely Voters”. Year in and year out, they prove to be the most accurate. In a year such as this, many of the disillusioned who voted for “Hope and Change” received way too much change and have lost hope, and thus will likely not be voting.
Since the turnout will likely decide this election, likely voters will play a much larger role in the accuracy sweepstakes.
There are lots from which to choose; my favorite is Rasmussen, but reading more than one is always a good idea.
Sheldon – Most of the polling outfits will switch over to likely voter samples after Labor Day, when their likely voter screen get a bit more accurate. You are right, however, that polls of likely voters are, generally, more accurate.