Gallup Throws Doubt On Political Polling

Editor & Publisher noted yesterday that Gallup reported a shift in party identification in the US; Democrats outpoll Republicans in party identification — by a single point:

In a (perhaps) historic shift, more Americans now consider themselves Democrats than Republicans, the Gallup organization revealed today.
Republicans had gained the upper hand in recent years, but 33% of Americans, in the latest Gallup poll, now call themselves Democrats, with those favoring the GOP one point behind. But Gallup says this widens a bit more “once the leanings of Independents are taken into account.”

However, that’s not exactly how Gallup itself headlines its results:

Americans are about as likely to identify as Republicans as they are Democrats according to a review of recent Gallup polls. However, once the leanings of independents are taken into account, the Democrats gain an advantage. Democrats have been on par with, or ahead of, Republicans in party identification since the second quarter of 2005.

A one-point difference, especially in a poll of 1,000 adults, falls within the margin of error in any case. It’s interesting and revealing to see E&P cast this as a “historic shift” when Gallup reports that this has been unchanged for almost a year. Either E&P has a strong case of analytical illiteracy or they want to put a spin worthy of John McEnroe on what amounts to a lob.
But one issue does arise from this fresh polling, one that undermines practically every political survey published in the past couple of years. If party identification is so close as to be a dead heat, why do polling services routinely underrepresent Republicans? CBS polls routinely overpoll Democrats so badly that their results are hardly worth the effort of analyzing. Their last major poll had a disparity between Democrats and Republicans of thirteen points — which they corrected to a nine-point difference. Surprise! It found that Bush’s approval numbers had dropped!
Here’s another that CQ noted earlier this year. Despite undersampling Republicans by 5 points, the NSA terrorist surveillance program managed to garner majority support, with or without warrants. CBS actually had their raw numbers correct in this poll, but “corrected” them through weighting to give Democrats a five-point edge. Again, it should surprise no one that this allowed CBS to report that Bush had suffered another drop in support in May of last year.
Now we have a benchmark against which to measure these polls. Any poll purporting to take the political temperature of the American electorate that doesn’t reflect these rather steady numbers should receive the scorn it deserves. The polls that we have seen thus far in 2006 all fail the sampling test, and most of what we saw last year fares no better. Too bad Editor & Publisher didn’t think to actually review media reporting on polls based on this information.