In 2005, University of California-Los Angeles political scientist Tim Groseclose and University of Missouri economist Jeff Milyo published a study in the Quarterly Journal of Economics (QJE) claiming to provide quantitative evidence of what they call “strong liberal bias” in the media.Footnote 1 Their estimates place 18 of the 20 national news outlets to the left of the centrist US voter.Footnote 2 Not surprisingly, this claim has received a tremendous amount of media attention, particularly after Groseclose published a book based on the QJE results titled Left Turn: How Liberal Media Bias Distorts the American Mind and made appearances on “The O'Reilly Factor” and other news programs.
Few scholars of the political media would deny that media organizations tend to have different slants on the news. These differences are often significant and appear to be driven in large part by economic factors such as consumer demand and media competition.Footnote 3 It is also true, as Groseclose and Milyo correctly note, that most journalists in the United States tend to be liberal and to vote Democratic.Footnote 4
However, these two facts do not necessarily imply that American media outlets have an overwhelming liberal bias. The policy preferences of reporters are only one of many possible influences on the content of the news. Numerous other competing journalistic norms and practices exist that limit the extent to which reporters' personal views influence their reporting.Footnote 5 As a result, previous studies of partisan bias in reporting on presidential elections have generally not found consistent results.Footnote 6 So why did Groseclose and Milyo (hereafter GM) reach such different conclusions? A closer examination of their method reveals that their estimates of media bias—and Groseclose's extensive extrapolations from those findings in Left Turn—rely on questionable assumptions about the processes generating citations of think tanks and interest groups by reporters and members of Congress, respectively.Footnote 7
GM's model is built on the assumption that the advocacy process in which members of Congress cite think tanks and interest groups in floor speeches somehow parallels the journalistic process by which reporters cite those groups in their reporting. This assumption is the basis for their mapping of media outlets onto a comparable ideological scale as members of Congress and the public (refer to their QJE article for technical details). If the press is unbiased, GM suggest, media outlets will cite think tanks in news reporting in a fashion that is “balanced” with respect to the scores assigned to the groups based on Congressional citations, which were measured during the 1993–2002 period. Any deviation from their definition of the political center (a composite based on a weighted average of House and Senate adjusted ADA scores) is thus framed by GM as bias.
Many objections can be raised to GM's methodology, the significant extrapolations that Groseclose makes from those findings in Left Turn, and the ungenerous tone of his responses to his critics (whom he repeatedly dismisses as “left-wing bloggers”). In this contribution, however, I will focus on GM's identifying assumption that the processes generating journalistic and Congressional citations to the think tanks and interest groups in their sample are identical. Specifically, I show how three plausible deviations from this assumption provide alternative explanations for GM's finding that the media are overwhelmingly liberal.
Asymmetries in Technocratic or Subject-Matter Expertise
First, liberal organizations in GM's sample—like the Council on Foreign Relations and the Center on Budget and Policy Priorities—may have more technocratic or subject matter experts than their conservative counterparts. Such a discrepancy may result in a greater number of citations by the press relative to Congress if reporters are more likely than politicians to seek out recognized experts in their field (rather than just like-minded views).
To illustrate, assume that there are two kinds of political stories. In the first, the press interviews ideological experts about policy debates. These stories are frequently written in a “he said, she said” framework but some may be written in a slanted format that corresponds to the reporter's point of view. For the sake of argument, we'll stipulate that citations in these stories have an identical data-generating process to that of Congress. But let's also assume journalists are expected to consult technocratic or subject matter experts about important trends or recent developments in the news. Under journalistic norms, citations to these experts are not always “balanced” by an opposing expert or a quote or argument from the other side. For instance, reporters might consult the respected tax and budget analysts at the liberal Center on Budget and Policy Priorities about a technical policy issue without also quoting the conservative Heritage Foundation. Likewise, as Paul Waldman pointed out in a Media Matters critique of GM's study (Reference Waldman2005), the National Association for the Advancement of Colored People, which receives the third-most media citations of the groups considered by GM, may be quoted in stories on racism without a quote from an opposing group—a plausible practice for stories in which “balance” would not be expected or even appropriate.
It follows from these premises that if there are more generally recognized technocratic or subject matter experts on the left-to-center side of the spectrum, then a study using the GM methodology would place the media on the liberal side of the ideological spectrum.
Both GMFootnote 8 and Groseclose (157–160) acknowledge that journalistic and Congressional citation processes differ and concede that such differences could bias their estimates. They argue that the bias could run in either direction,Footnote 9 but concede that the concern described above is “the criticism we have heard the most frequently.” To address this point, they provide an appendix showing that their results do not differ substantially when they control for what they describe as “possible measures of the ‘quality’ of a think tank,” including whether it has a closed membership and whether its staff titles include “fellow,” “researcher,” “economist,” or “analyst.”Footnote 10 However, several prominent conservative think tanks have consciously aped the tropes of the center-left establishment (such as fellows and closed memberships) while discarding their commitment to technocratic scholarship.Footnote 11 It is therefore not clear whether these indicators, which put the Family Research Council in the same category as RAND and the Council on Foreign Relations, capture the technocratic or subject-matter expertise of think tanks.Footnote 12
Asymmetries in Public Relations and Marketing Skill
GM's methodology also doesn't allow for differences in the relative success of liberal and conservative groups in marketing their work to the media versus Congress. One possibility is that conservative groups devote more effort to promoting their work to members of the Congress than the press or are more effective at doing so, which could skew the results. For instance, the Heritage Foundation focuses extensively on influencing legislators.Footnote 13 If they succeed in doing so and members of Congress end up citing Heritage more than the press does (as GM find), GM's methodology would interpret this discrepancy as proof of media bias rather than the success of Heritage's legislative outreach. Likewise, if liberal think tanks and interest groups devote more effort to or are more effective at promoting their work to the media rather than Congress, GM's methodology would again interpret such a difference as evidence of media bias.
In Left Turn, Groseclose notes this concern as well as the concern about technical expertise described above, but points to the correlation between GM's estimates and those of Gentzkow and Shapiro, who identify the phrases that best distinguished Democrats from Republicans in Congressional speech during 2005 (e.g., “death tax repeal”) and measure their prevalence in newspaper reporting during that year. Based on this relationship, Groseclose concludes that while “such arguments are important in theory … they are not in practice” (176).
However, the relationship between the estimates of the two sets of authors (r = .36 for the six outlets that both coded) is highly sensitive to outliers.Footnote 14 It is true, as Groseclose notes, that if the Wall Street Journal is excluded due to possible problems with Gentzkow and Shapiro's data collection process, the two sets of estimates are strongly correlated (r = .94).Footnote 15 However, as Figure 1 shows, the relationship is driven almost entirely by the difference between a clump of mainstream news outlets (the Los Angeles Times, New York Times, USA Today, and Washington Post) and the conservative Washington Times, which is an outlier in GM's data. If the Washington Times is instead excluded from the data, the relationship between GM's estimates and those of Gentzkow and Shapiro reverses completely (r = −.90).
Comparing estimates of media slant
More generally, Gentzkow and Shapiro explicitly state that their data “do not show evidence of an economically significant bias relative to the benchmark of profit maximization”Footnote 16 and disavow any interpretation of their measures as measuring bias relative to “a benchmark of ‘true’ or ‘unbiased’ reporting.”Footnote 17
Asymmetries in Citations Over Time
Finally, the patterns of citations by media organizations may be sensitive to the periods for which the organization's citations were coded in GM's data. A spokesperson for the parent company of the Wall Street Journal, which was coded as liberal by GM, pointed out that “the researchers' ‘study’ of the content of The Wall Street Journal covers exactly FOUR MONTHS in 2002” while other outlets were coded for much longer periods. As the spokesperson added, “the relative newsworthiness of various institutions could vary widely” over time. In particular, the period for which GM code Congressional citations (1993–2002) does not match the intervals for which they code citations by news organizations, which vary widely in both duration and the specific time period that was coded.
This concern is buttressed by John Gasper's finding that GM's estimates of the ideological locations of media outlets are not stable over time. One explanation for Gasper's results is that changes in the political landscape over time may have differential effects on the patterns of journalistic and Congressional citations of think tanks and interest groups. In particular, changes in the issue agenda and configuration of power in Washington may make certain groups more likely to be cited in the press than Congress (and vice versa).
GM's results may be sensitive to these differences since the periods for which citation data were gathered varies dramatically between news organizations. Several were only coded for very brief periods. For instance, the New York Times, USA Today, the Journal, and the Washington Post are all classified as liberal on the basis of less than one year's worth of data from the 2001–2002 period.Footnote 18 Most of this coverage took place after the September 11, 2001 terrorist attacks—a period in which the news was dominated by the US government's response to 9/11. Bush administration officials were therefore frequently quoted during this period, which could paradoxically cause GM's approach to interpret the media as having a liberal media bias if articles quoting an administration official on one side were balanced by a quote from a left-of-center group (e.g., stories on the debate over the treatment of prisoners by the United States that included a quote from Amnesty International as well as the Bush administration).
Unfortunately, the issues described here undermine the key assumption of GM's model, which is not otherwise identified.Footnote 19 Absent further research to validate that assumption, it is not clear how much weight we should put on their estimates. How, then, should scholarship in this area proceed? I would recommend Gentzkow and Shapiro's methodology as a better starting point. Their assumption that the frequency of language characteristic of partisan speech in news reporting can be used to estimate media slant seems like a reasonable approach, though more research is clearly necessary.
Beyond these measurement concerns, GM face a deep epistemological problem that plagues all research on “media bias”—the lack of a well-defined and widely-accepted definition of unbiased reporting. Even if we accept their assumptions, it is not clear that we should endorse their argument that journalistic deviations from the center are evidence of bias.Footnote 20 In some cases, as the comedian Stephen Colbert has famously argued, reality may have a liberal bias.Footnote 21 In other cases, it may have a conservative bias. The center has no monopoly on truth and we should not ask the media to follow it slavishly. Democracy works best when the press is independent, not narrowly centrist.
