Breitbart Site Distorts HuffPo Piece on Gallup Poll to Claim Bias

One of the reasons I don’t bother doing this on a more regular basis is that I have only so many hours in the day, and other priorities. That alone wouldn’t keep me from writing about something I felt was important, and certainly outright misrepresentation by media (even right-wing media known for its bias and distortions like Andrew Breitbart’s ventures) could be considered important.

The problem is, at much of the insular, self-referential alternate-reality universe that is today’s right-wing media, these distortions happen with such regularity that it would take many more hours than I have in my day even if I did nothing else. That’s why we have Media Matters, I guess. ;o) However, in this case, the distortion was brought to my attention by a conservative-leaning friend on Twitter who obviously believed what it had to say, and it was also so blatant that I just had to stop and write a quick debunking.

Everyone knows that as we head into the last month before a quadrennial Presidential election, the stakes are the¬†highest they get in American politics. As such, it’s not surprising that this is the month in which some of the nastiest of the dirty tricks get pulled, and also when all attention is focused on any little twists and turns that might affect the race, if it is close. One of the closest-watched indicators are the polls. This is how everyone from the conservatives to the liberals discovers how the candidates are doing, day to day, and so it is probably inevitable that questions about the pollsters’ impartiality might arise. Truth be told, if there were genuine evidence that any one pollster was putting their firm’s thumb on the scale in favor of one candidate or the other, I’d be interested in such a story, too. In 2010, for instance, Nate Silver of the excellent FiveThirtyEight blog at the New York Times, pointed out that Rasmussen Reports was statistically biased and inaccurate in favor of the GOP. But Nate Silver is a statistical analyst and poll-watcher par excellence. The hacks at Breitbart are nothing more than conservative mouthpieces, always on the lookout for whatever makes Democrats or liberals look bad.

So it is perhaps not surprising to read this at Breitbart’s site:

WE TOLD YOU SO: HUFFPO BUSTS GALLUP POLL BOOSTING OBAMA BY TINKERING WITH SAMPLE ON EVE OF ELECTION

Conservative critics of mainstream media polling are feeling vindicated today after the left-wing Huffington Post revealed the fact that the Gallup poll likely altered the percentage of non-whites in its sample in the weeks after the Democratic National Convention, likely boosting President Barack Obama’s approval rating–and, possibly, his poll margin.

Other than the annoyingly shout-y headline about the Huffington Post “busting” Gallup – one of the nation’s oldest and best-respected polling firms – the claim is indeed the sort of signature-Breitbart style of eye-catching. It certainly would be significant news if Breitbart (or even HuffPo…heck, anyone) had caught Gallup fudging its statistics to benefit the Obama campaign or Democrats in general.

Of course, as is also usual for Breitbart, they’re betting readers won’t click through to read the HuffPo article they link to which supposedly “proves” the charge. Sadly, they are probably correct in this estimate of their readership. Readers do not come to Breitbart’s site for rigorous analysis or fact-based news, they come to have their already-existing beliefs about the way certain thins are validated and reinforced. So it shouldn’t come as a surprise that when one does click through to that article, one finds a very different story:

The estimated nonwhite percentage of the sample for the past five weeks was as follows:

Huffington Post's small spreadsheet of Gallup's non-white sampling numbers for September through early October

Given the size of Gallup’s weekly adult sample, typically over 3500 respondents, the increase from the previous four weeks to the first week of October is very unlikely to be explained by random variation alone. This change, however, does bring the nonwhite share of Gallup’s sample much more closely into line with the 2010 Census estimate of the nonwhite share of the U.S. adult population which was 31.8 percent.

So, contrary to what Breitbart’s reporter claims, the switch in sampling of non-whites isn’t a nefarious plot to go in the tank for Obama in the final weeks of the campaign, it’s a correction of earlier numbers,¬†either due to sampling error, or to a higher estimate of non-white voting population. Even that is a guess, since Gallup (as the HuffPo article makes clear) doesn’t release the racial composition of its daily tracking sample. The numbers in the above graph are estimated based upon respondents answers. In other words, no one but Gallup truly knows.

The author of the HuffPo article, Alan Abramowitz, a political science professor at Emory University, goes on to opine that “the increase from the previous four weeks to the first week of October is very unlikely to be explained by random variation alone.” I’m not sure I agree, given that the number goes up eight-tenths of a point, holds steady for a week, then goes DOWN more than two points before going back UP to slightly less than three points above where it had been. I’m not the expert that Abramowitz is, (then again, neither are the partisans at Breitbart.com), but a fluctuation of two or three points first one way, then the other, seems well within the realm of random statistical variation. More than that, though, is Abramowitz’s statement that those numbers are his own ESTIMATES of what Gallup’s own, real numbers must look like. In other words, in the absence of the real numbers from Gallup, Abramowitz is extrapolating and guessing to get the above chart, and then making tentative assertions based on his extrapolations.

If you read Breitbart, however, all such nuance and qualification is lost. Instead, you’re told (several paragraphs down in the piece) that the reason might be because the Justice Department is strong-arming Gallup. Their evidence for this is a link to yet another right-wing site, Tucker Carlson’s Daily Caller, which says the DoJ filed an unrelated lawsuit against Gallup, though – hilariously – they also say in a separate article that this lawsuit, though announced, has never even been filed. It gets even murkier after that, with “evidence” in the form of quotes from their own pundits claiming without evidence that the polling numbers for this election are somehow employing “absurd assumptions about turnout.”

When you’ve got to quote yourself as evidence that your suspicions are correct, you might have a credibility problem. Beyond that, though, is the simple fact that Professor Abramowitz, in his HuffPo article, was making an educated GUESS at what Gallup’s racial sampling might be, and his professional opinion as a political science professor is that if he’s correct, “the increase in the percentage of nonwhites in Gallup’s latest weekly tracking poll sample is a welcome development that should produce more accurate estimates of not only presidential approval but also support for the presidential candidates among registered and likely voters.”

Exactly so. But you’d never know – or even hear – it, if all you read was the Breitbart piece.