VoxEU Column Politics and economics

Fair and balanced after all? The bias of the US press

Is the US media biased? According to a controversial new book, it is – and, perhaps surprisingly, to the left. This column takes a different analytical approach and argues that the press is actually much closer to the average voter’s sentiments than we might think. Might all these claims that the media is biased in one direction or the other be adding a whole new set of distortions?

Citizens typically obtain information about relevant policy issues via the mass media. Moreover, they might make judgements about the relative importance of issues – whether crime or the economy is the most important problem facing society, for example – by comparing the relative amount of media coverage issues receive, otherwise known as “agenda-setting” (McCombs and Shaw 1972).

Of course, journalistic norms dictate objectivity in coverage. But objectivity might fail in practice. For example, the owners, editors, and journalists of media outlets might have partisan agendas, and might slant their news coverage in an attempt to influence the views of readers or viewers. Alternatively, since they are for-profit enterprises selling a product, media outlets might slant their news coverage to cater to the pre-existing ideological tastes of readers or viewers.

If (i) media bias is mainly driven by the preferences of owners, editors, or journalists, (ii) most media outlets are biased in the same direction, and (iii) the media have some persuasive power, then media bias may have pervasive and long-term political and policy effects. Suppose, for example, that most media outlets in a country are biased to the left. Then the country is more likely to have a left-leaning government, left-leaning policies, and policymakers focusing on issues relevant to left-leaning citizens.

In a recent and influential paper, Tim Groseclose and Jeff Milyo (2005) classify the ideological position of US news outlets based on how closely they resemble different US congressmen. The “bridge” by which Groseclose and Milyo connect media outlets with congressmen is through their relative propensity to cite think tanks (in a non-negative manner). A media outlet is classified as more conservative the more it cites think tanks that are cited frequently by conservative congressmen, but rarely by liberal congressmen.

Groseclose and Milyo conclude that the US media display an overall liberal bias, i.e. most media outlets are to the left of the average American voter. A less publicised and discussed finding is that US media outlets are also relatively moderate, in the sense that they are located between the median Democratic and median Republican congressmen. The results in Groseclose-Milyo are likely to become salient in the public debate over the next year, as Groseclose has recently published Left Turn: How Liberal Media Bias Distorts the American Mind, which draws heavily on their findings.

Because of its novelty, prominence, and stark conclusions, the original Groseclose-Milyo paper has been subject to several critiques.

  • First, as Brendan Nyhan (2005) argues, the “think-tank bridge” between media outlets and congressmen is unlikely to be straightforward, since the media and congressmen often cite think tanks for very different reasons.
  • Second, the Groseclose-Milyo findings appear to be quite sensitive to influential observations, as well as the exact years studied. For example, John Gasper (forthcoming) shows that the liberal bias essentially disappears if we exclude the citations of a single group – the National Taxpayers Union – from the analysis.
  • Third, although Groseclose and Milyo can directly compare media outlets with politicians, they can only compare media outlets to voters by assumption. They assume that once specific biases of the US electoral system are taken into account, the ideological position of the average American is the same as the ideological position of the median member of the US Congress as revealed by his or her roll call voting record.1

As Andrew Gelman (2011) notes, if interest groups are able to influence elections or congressional voting or both, and if they tend to be conservative, then the median member of the Congress is in fact more conservative than the average American.

We believe that in order to pin down the absolute partisan bias of media outlets relative to voters, it is necessary to compare media outlets directly with voters. This is what we do in our paper “The Balanced US Press” (2011a).

In the paper, we propose a simple new method for placing newspapers, interest groups, political parties, and voters on the same ideological scale. The method uses data on state-level ballot propositions – initiatives and referendums. We exploit the fact that newspapers, parties, and interest groups routinely make endorsements for or against these propositions, and citizens ultimately vote on them. When a newspaper disagrees with a majority of the voters on a proposition, the newspaper has clearly taken a position that is to the left or right of the median voter. We average these cases to create an index of conservatism for each newspaper in our sample. We do the same for interest groups and parties.

We find that, on average, newspapers are located almost exactly at the median voter in their home states. In California, where we have the most data, newspapers are probably slightly to the right of the median voter. We find no evidence that the US press exhibits a liberal bias.

Our second finding is that newspapers are moderate relative to interest groups and political parties. That is, although newspapers exhibit some variation in their ideological position, they tend to be much closer to the median voter than most interest groups. This is consistent with the second finding noted above from Groseclose and Milyo.

These patterns are clearly shown in Figures 1 and 2. The figures display the estimated ideological positions of newspapers and interest groups in California and all other states, respectively. In both figures, endorsers are divided into 7 groups based on their index of conservatism (the “right-of-median” score). In turn, each figure is divided into 2 panels, with interest groups, parties, and county median voters on top, and newspapers on the bottom. Inspection of Figure 1 reveals that newspapers in California are more moderate than interest groups – the groups tend to be located in the most extreme bins, while newspapers are concentrated in the intermediate bins. Even newspapers, however, exhibit a bi-modal distribution, with more moderately liberal and moderately conservative newspapert than centrist newspapers. The same pattern appears in Figure 2, with the exception that the pooled distribution of newspapers in other states is more centrist than in California, exhibiting a single-peaked rather than a twin-peaked distribution.

Figure 1

Figure 2

A third finding relates to differences across issue areas. Newspapers appear to be more liberal than voters on social and cultural issues such as gay marriage, but they tend to be more conservative on economic issues such as the minimum wage. It is tempting to speculate on these patterns. For example, they are consistent with a world in which newspaper owners allow their editors and journalists – who tend to be left-leaning – to take a liberal stance on social/cultural issues, but not on economic issues, on which the owners themselves, and advertisers, may have a larger stake.

Finally, we might be concerned that the editorial and news sections of newspapers exhibit noticeably different biases. If this is the case, then although the study of ballot propositions allows us to compare voters and the editorial pages of newspapers, it does not allow us to compare voters with the news sections of newspapers. Using two existing methods for measuring relative bias, we compare the editorial and news sections of a large sample of newspapers.2  We find that, on average, the news and editorial sections of newspapers have almost identical partisan positions – if anything, news sections are slightly to the right of editorial sections overall.

Economists often point out that rational citizens are less likely to be influenced by media bias if they learn that this bias exists. However, we might also worry about the political and policy consequences of wrongly persuading citizens that media outlets have a left-leaning position overall, when in fact they are balanced. That is: What happens if citizens are convinced to “undo” a bias that does not exist?


Gasper, John T (forthcoming), “Shifting Ideologies? Re-examining Media Bias”, Quarterly Journal of Political Science.
Gentzkow, M and JM Shapiro (2010), “What Drives Media Slant? Evidence from US Daily Newspapers”, Econometrica, 78:35-71
Gelman, Andrew (2011), “Thoughts on Groseclose book on media bias”, AndrewGelman.com, 28 July.
Groseclose, T (2011), Left Turn: How Liberal Media Bias Distorts the American Mind, St. Martin’s Press, 19 July.
Groseclose, T and J Milyo (2005), “A Measure of Media Bias”, Quarterly Journal of Economics, 120:1191-1237.
Nyhan, Brendan (2005), “The problems with the Groseclose/Milyo study of media bias”, Brendan-Nyhan.com, 22 December.
McCombs, ME and DL Shaw (1972), “The Agenda-Setting Function of Mass Media”, Public Opinion Quarterly, 36:176-187.
Puglisi, R and JM Snyder, Jr (2011a), “The Balanced US Press”, NBER Working Paper No. 17263.
Puglisi, R and JM Snyder, Jr (2011b), “Newspaper Coverage of Political Scandals”, Journal of Politics, 73:931-950.

1 These biases are the lack of representation for Washington D.C., the over-representation of small states in the Senate, and the presence of gerrymandered districts in the House.
2 One is based on the relative propensity to use phrases used more by Democratic or Republican congressmen (Gentzkow and Shapiro 2010). The other is based on the relative propensity to cover scandals involving Democratic or Republican politicians (Puglisi and Snyder 2011b).
3 These findings are not included in the NBER version of the paper, but they are included in a later version of the paper available here.

2,415 Reads