Yes, Virginia, there is a media bias

Media bias can often be hard to spot. After all, a whole lot of it takes the form of editorial choices: what goes on A22 instead of A1; what gets 2 minutes instead of 10 seconds; what goes on the news at all. We news consumers don’t get to see the process that results in those choices, and we—obviously— don’t get to see the stories that the media chooses to leave on the copy room floor. And as Al Franken and others have written, the main press biases are not political, but self-interested: they like the news hot, fast, and cheap.

But sometimes there is political bias, and there’s just no missing it. Take for example this Associated Press writeup of the Gallup poll I mentioned in my last post:

Democrat Wesley Clark, in the presidential race for less than a week, is tied with President Bush in a head-to-head matchup [.]

Clark, a retired Army general, garnered 49 percent support to Bush’s 46 percent, which is essentially a tie given the poll’s margin of error. [.]

In the head-to-head confrontations, it was Kerry at 48 percent to Bush’s 47 percent; and Bush’s 48 percent to Lieberman’s 47 percent. Bush held a slight lead over Dean, 49–45 percent, and had a similar advantage over Gephardt.

(My emphasis) The margin of error of this poll was 3 points, plus or minus. Which means that all of these results are inside the margin of error; an honest, unbiased reading would say that the head-to-head matchups are universally too close to call. If Clark, ahead 49–46, is “essentially tied,” why then is Dean, behind 45–49, not also tied? Within the framework of the poll, these are indistinguishable results.

Also, the numbers themselves compare apples to oranges: the Clark 49–46 is among registered voters; the Dean 45–49 is among all adults. The Dean number for registered voters? 46–49, mirror image of the Clark result. Dick Gephardt polls 46–48 among registered voters—a closer result than Clark’s. Yet Clark is tied and Gephardt is behind.

Is Will Lester of the Associated Press unfamiliar with the meaning and application of statistical sampling error? Is he unable to read poll results and match up like with like? I doubt it. Much more likely that Mr. Lester and his editors could not (consciously or otherwise) accept that a popular wartime President could be losing to a guy who has been in the race for 5 days—let alone a French-looking sandwich nibbler like Kerry—and likewise could not accept that he could be tied with an unelectable Northeast ultraliberal or a washed up pantywaist with no eyebrows. Lester and the AP instead spun the numbers in a clearly biased way.

Hint for those not paying attention: the bias was not liberal.

UPDATE: Of course, it just might be possible that Will Lester is more familiar with the application of sampling error than I am. Kevin Drum of Calpundit writes in an email:

Unfortunately, I think Will Lester has the statistics right. Since
the 95% confidence interval is 3%, that means that 3% = two standard
deviations. Clark and Bush are separated by 3%, which is two standard deviations, and therefore you cannot (quite) say with 95% confidence that Clark is actually ahead of Bush.

However, Bush is ahead of Dean by 4%, which is more than two standard deviations. You can therefore be 95% confident that Bush is probably genuinely ahead of Dean.

And also see Ted Barlow of Crooked Timber in the comments below.

I still think it’s odd that Lester quotes different sets of numbers in different paragraphs, without making it clear that he’s doing so; and that it just so happens that he picks the sets of numbers that are generally more favorable to Bush and less favorable to Dean and Gephardt.

But my accusation of bias is clearly without merit, so please consider it retracted with apologies.

Filed under: politics/2004