Still Unbiased:

My American Prospect study of press labeling continues to evoke responses from conservatives vexed by the finding that the press actually labels liberals more often than those on the right. In a lot of cases the objections are pure bluster. For example, Andrew Sullivan writes:

I ignored Geoffrey Nunberg's piece in the American Prospect in April, debunking the notion of liberal media bias by numbers, because it so flew in the face of what I knew that I figured something had to be wrong. (And I was too lazy to do all the enormously laborious number-crunching to refute it. So sue me.)

But then you're always hearing people plead physical laziness by way of justifying intellectual laziness.

Another critic, Edward Boyd, can't be accused of laziness of either sort. He has produced a new study that overcomes the obvious objections to an earlier one, which relied on a search of a database that contained 40 percent foreign newspapers. This study looks at a year's worth of articles in five major papers, and purports to show that conservatives are actually labeled more often than liberals are.

In the process of obtaining this result, though, Boyd decided to alter my methods in two ways, which introduced new problems. First, he changed the method of computing the averages in a way that penalizes the most liberal figures for the fact that the press mentions them less often than the most conservative figures. If he had used my original method on his data, his study would show that politicians from both sides have the same likelihood of being labeled. And second, Boyd changed the search method to one that turns out to produce a high proportion of false positives for several of the conservative figures on the list (unfortunately, Boyd didn't actually examine his results to see how well the method was working).

To correct this, I individually examined all the hits that turned up in three of the papers in Boyd's database to rule out false positives. When I then computed the averages for Boyd's database using the method I had used in my original study, I found that the average liberal in fact has a 51 percent greater likelihood of being labeled than the average conservative -- this against a 33 percent greater likelihood in my original study.

Those with a taste for this sort of thing will find a detailed discussion of Boyd's study here. For myself, I'm satisfied that my original claim was correct, and hereby retire from the label-counting game.

Still, there are some general points worth making. First, all these studies clearly refute the original claim made by Bernard Goldberg and other conservative media critics that conservatives are labeled much more often than liberals are. If you accept Boyd's figures uncritically, they show a slight preference for labeling conservatives; if you accept mine, the discrepancy goes the other way. But none of these studies have shown what anyone would call an overwhelming disproportion one way or the other.

Nor for that matter does the variation in labeling from one paper to the next seem to have anything to do with a paper's alleged bias. In fact, if you were going to rank the papers using Boyd's own figures according to the ratio with which they label liberals and conservatives, you'd wind up saying that The Washington Post (1.24) is a lot more conservative than The Chicago Tribune (0.76). Was Goldberg on to something, or is this just a lousy way to measure point of view?

What's more important, the studies all show that the labeling of politicians on both sides is pretty infrequent. On the average, a well-known politician is labeled in only about 5 percent to 10 percent of the articles in which his or her name appears, and a less well-known politician far less often than that -- this against the much higher labeling percentages for groups like Americans for Democratic Action or the Heritage Foundation.

In fact, what these studies really suggest is that the press doesn't label people nearly enough. It's true you occasionally run into a label that seems purely gratuitous, as when the Chicago Tribune writes:

Messiest Desk: Minnesota's liberal Democrat Paul Wellstone, who seems to have re-created his freshman dorm room in a 3-foot-square...

(Those slovenly liberals!) But even if labels are reductive, they generally do useful work, as in these examples:

President-elect George W. Bush announced Friday that he will nominate John Ashcroft, a stalwart of the Senate's conservative wing and a passionate opponent of abortion rights, to be attorney general.

Sen. Barbara Boxer, a staunch liberal, opposes a judicial appointment for Cox, who has a strongly conservative voting record.

Forty-five senators, ranging from conservative Republican Jesse Helms to liberal Democrat Paul Wellstone, backed a stronger measure calling for sanctions now.

Helms became a national advocate of conservative causes that he began to champion as a television commentator here in the 1960s.

Jeffords is a social friend to several conservatives. He joined a Senate quartet, the Singing Senators, with conservative Republicans Lott , Larry E. Craig of Idaho and John Ashcroft.

You couldn't remove the labels from those examples without depriving them of a lot of their news value. And inasmuch as politicians are labeled as conservative or liberal in only a tiny percentage of the articles that mention them, it's a safe bet that underlabeling is the real problem. Granted, these aren't the only labels that the media use -- there are all those phrases like "far right," "left-wing," "progressive," "left-leaning" and occasionally "reactionary," "lefty," and "near-socialist." But taken together these are used far less often than the labels "conservative" and "liberal" are. And while party labeling is often informative enough, there are plenty of times when a judicious label would help the reader figure out what's going on.

Underlabeling is a particular problem when it comes to figures less well known on the national stage -- the people who ought to be labeled the most. In my own study, for example, I found that Helms was labeled about ten times as often as Mitch McConnell, and Wellstone more than six times as often as Nancy Pelosi. This means either that Helms and Wellstone are labeled too often or, what's far more likely, that McConnell and Pelosi aren't being labeled enough.

One other point about those alternative labels: I keep seeing conservative bloggers trumpeting studies that show that the phrase "right wing" and "far right" are used more frequently than "left wing" and "far left" are. Andrew Sullivan is the most recent blogger to pull out this chestnut, citing a study that shows that the former are used more than twice as frequently as the latter in The New York Times. But that observation is utterly meaningless. It's a good example of what happens when you turn people with only a dim understanding of statistics and method loose on these press databases.

Some simple points about method, then. Rule one: Check the hits you get. When you look at the occurrences of "right wing," "far right" and the like in the press, it turns out that the majority of their occurrences have nothing to do with American politics. Of the 9,700 instances of "right wing" or "left wing" in the Los Angeles Times, for example, 2,900 occur in articles that mention hockey, and another 4,400 occur in articles that mention nations like Israel, Japan, Germany, Italy, France, and so on, the vast majority of them dealing with foreign politics. Similarly, the 3,300 occurrences of "far right" and "far left" include 1,400 that occur within seven words of "photo" or "photograph," and 1,600 that contain the names of foreign countries. And with no way of knowing whether the left- and right-hand versions of these are balanced, what you're getting when you search on these terms is mostly noise.

What's more, even if there was a way of winnowing these instances down to articles that mention domestic politics, the frequency of labeling of any kind has to be relativized to the things that are being labeled. In Boyd's study of five newspapers, for example, the five conservatives in the group were mentioned in the press more than two and a half times as often as the five liberals. So if the "far right" and "far left" labels were distributed proportionately, you would expect a decided preference for the former over the latter.

Of course you can reasonably ask how often these labels occur in the vicinity of the names of particular groups or figures, which is what Boyd did in an earlier study. But in that case you have to include the other labels that the media use as well. A search on Wellstone's name in The Washington Times turns up only one occurrence within seven words of "left-wing" or "far left" in 533 occurrences -- a bit less frequently than "right wing" and "far right" are used of Helms. But before you start accusing The Washington Times of having a liberal bias, dig a little deeper and you'll also find examples like:

Paul Wellstone, the new radical Democratic senator from Minnesota, has trouble disguising his contempt for colleague Jesse Helms.

The Marshall meeting was stacked with outsiders imported by Mr. Wellstone's radical political machine.

Mr. Georgacas said his state's Democratic Party "has moved to the extreme left under [Democratic Sen. Paul] Wellstone …"

Far to the left of the Democratic Party remains Sen. Paul Wellstone of Minnesota.

Democratic Whip David Bonior, Minnesota Sen. Paul Wellstone and all sorts of lefty peaceniks have been drinking at the VFW Halls. . .

Not hard-core leftist Paul Wellstone, Minnesota Democrat, and not rock-ribbed conservative Jesse Helms, North Carolina Republican.

A near socialist like Paul Wellstone can work with the left's version of Darth Vader, Jesse Helms, on human rights in China.

The fact is that there's no way to enumerate all the possible labels, much less evaluate their distinctive connotations, and most of them are difficult to search on mechanically -- there's just too much noise in the data, particularly given the homonymy of the words "left" and "right." Anyone who wants to stay in this game would be well advised to stick to "conservative" and "liberal." But I'm out of here.