An ABC News/Washington Post poll has Barack Obama leading John McCain by a a 52-43 margin. Noting that this is Obama’s “first clear lead” during the campaign, the Post declares that the “shape of the race” is “altered.”
But the Post may be overstating things. First, the tracking polls don’t show a substantial alteration. Rasmussen’s has Obama up by 2 points; Gallup’s has him up by 3. This is familiar territory. In addition, an Ipsos-McClatchy poll has Obama leading by a narrow 44-43 margin, while ARG has Obama up by 2 points. In my opinion, the Ipsos-McClatchy poll is more accurate than that of ABC News/Washington Post when it comes to measuring the number of true “undecideds.” As to the Obama-McCain split, who knows?
The second possible problem with the ABC News/Washington Post poll relates to the tricky issue of party identification. The McCain camp has noted that Democratic party identification in the Post poll was 16 points higher than Republican identification. The Post’s polling director has responded that actual party ID numbers among likely voters in the sample had the Democrats at plus six points. It was only when people who offered no original party ID were asked whether they leaned one direction or the other that the number jumped to plus 16 for the Dems.
Party identification has been pretty volatile since 2004 — it strongly tiltled the Democrats’ way for several years and then seemed this year to tilt back towards the Republicans a bit. As a result, I don’t think anyone knows what the proper sample mix for a national poll should be. That said, it seems indisputable that a poll with a sample in which Democrats have a 16 point edge is unreliable. Nor would I place great trust in a poll where the Democratic ID edge is six points with “leaners” tipping overwhelmingly Democratic.
A state poll in a state where voter registration is known raises somewhat different issues. Consider, for example, the ARG poll of Pennsylvania that I discussed yesterday. Its sample included 14 percentage points more Democrats than Republicans, a gap that matched state party registration figures. But exit polls in the 2006 election suggested that the Democratic edge was only five percent. Which number is more realistic for purposes of polling this election?
The disparity in party affiliation between 2006 exit polls and current registration numbers presumably reflects, in significant part, a wave of Democratic registration related to the Pennsylvania primary in April. Many of those who registered at that time may actually be independents; some may even be Republicans who wanted to promote “chaos.”
But this doesn’t mean that a poll that whose sample matches current registration figures is skewed. Indeed, it seems to me that as long as the sample of Democrats polled includes in proper proportions those who registered as Democrats without being true Dems, the poll should not be skewed. For the “false Dems” will respond as such when the pollster asks them whether they favor Obama or McCain, just as they will on election day. (The more problematic issue is what they will tell the pollster about their registratration and/or identification).
There’s no way to tell whether the ARG poll includes “false Dems” among its Democratic respondents in the proper numbers. However, we do know that, despite the 14 point Democratic edge in the sample, Obama was only four points ahead in the poll. And this result was in line with other recent Pennsylvania polls.
To comment on this post, go here.