Framing the Right Suspects: Measuring Media Bias

Abstract

This article examines ideological bias in six large daily newspapers and The Associated Press. The media examined are three to six times more likely to associate ideological labels (or frames) with organizations (think tanks) with a conservative orientation than think tanks having a liberal orientation. This tends to frame the analyses done by conservative think tanks as less objective than the analysis done by liberal think tanks. Regression results suggest that approximately three-fourths of the explained differential in framing rates is due to media bias. The rest is primarily explained by the differential in the “quality” of think tanks.

Notes

¹See the literature survey in Fico and Cote (1999) for a fuller description of this literature and Carter, Frederick, and McCabe (2002) for an example of this literature.

²See Watts, Domke, Shah, and Fan (1999) for a summary of the various studies supporting and rejecting the hypothesis of media bias. Also see D'Alessio and Allen (2000).

³See Matthes (2009) for a survey of articles on media framing studies. This survey examines framing in a number of contexts beyond pure political issues but gives a fair assessment as to how framing analysis grew in importance in the Communications literature from 1990 through 2005.

⁴Groseclose and Milyo (2005, pp. 1204–1207) and others distinguish between slanting coverage and bias. They define slanting as deliberately choosing a viewpoint orientation for the story regardless of the underlying facts. Bias is considered to be more of a subconscious reflection of the media's general ideological orientation without a deliberate decision to present a story from anything but a neutral or objective viewpoint. This article cannot distinguish between these possibilities. Even if the decision to attach labels is made less purposefully and a subconscious reaction to the conclusions of differing think-tanks, they still represent the bias animating the perspective of the reporter and has the framing effect discussed above.

⁵Even though The Associated Press is a wire service, it will be referred to as a publication throughout the rest of the article to simplify the language.

⁶A detailed description of the search terms for each database is discussed in the appendix and all data used in this article is available from the author.

⁷FAIR is a non-profit press watchdog group with a liberal orientation.

⁸The source for this information is www.opensecrets.org. Open Secrets collects, among other things, public information on campaign donations and other activities from the filings these candidates make with the Federal Election Commission. It makes the information from the three most recent election campaigns publicly available. Because this information is based on public filings, the data is limited to contributions of $200.00 or greater.

⁹There was only one contributor at Brookings who made a contribution to the McCain campaign and upon further investigation it turns out he died before the election so that ostensibly conservative analyst is no longer at Brookings.

¹⁰All test searches were done for time periods not included in this study. See Table 2.

¹¹For instance, The Washington Post completely changed its classification system in the late 1990s (in the middle of one of the 2-year periods in which data was not collected).

¹²A duplicate citation might occur when an article was updated multiple times, included in multiple publications or included a correction. The largest impact was on the citation count for The Associated Press.

¹³Of the nine think tanks examined below, with one exception, this only happened with The Manhattan Institute and Hoover Institution.

¹⁴This automation rules out the possibility that a coder would incorrectly conclude that one ideological label was associated with a particular think tank, when in that story another label (or no label) was the appropriate ideological label associated with that think tank.

¹⁵This is procedure recommended in Neuendorf (2002, p. 142) when there is only a single coder. Neuendorf (2002, p. 159) also recommends that the reliability sample be no smaller than 50 observations and need to be no larger than 300.

¹⁶These tests of intercoder reliability are discussed in Neuendorf (2002, chapter 7). Pearson's rho is appropriate when data is recorded in a numerical ordering such as a counting of occurrences.

¹⁷The raw data used to construct Table 3 and in the analysis below is available from the author.

*Most common frame for each think tank is highlighted in bold.

**Manhattan Institute and Hoover Institution were omitted for some periods for some publications.

†Center for American Progress only has results for the last period (01/2005–12/2006).

¹⁸Including the two foreign policy think tanks, CSIS on the right and CFR on the left, does not significantly change this difference. With these included, one-fourth of all references to politically right think tanks include an ideological frame and the percentage of references to politically left think tanks remains effectively unchanged at 2%. This comparison encompasses 93% of the total citations in the study.

¹⁹The liberal media watchdog group, FAIR, identifies CAP as center/left, whereas it identifies RAND and Brookings as centrist (Dolney, 2008).

²⁰This number is not adjusted by the number of researchers or employees. In part, this is because not all of the analysts are economists and may make such comparisons misleading. It may also be the case that the public reputation of the think tank would presumably be increasing in the volume of high quality research and much less related to the number of articles per scholar.

²¹The only articles excluded from the count where the non-peer-reviewed articles found in the papers and proceedings (May) issues of the American Economic Review.

^aFrom Groseclose and Milyo (2005). Ideological index for median voter is 50.1. Higher values indicate more liberal.

²²The values for the other four newspapers in this study are USA Today (63.4), The Washington Post (66.6), The Los Angeles Times (70.0), and The New York Times (73.7).

²³The value of each of these measures is set to zero when either TTI or NNI is unknown (when the publication is The Associated Press or the think tank is The Center for American Progress).

²⁴The negative one scalar is used to make the value of the variable increasing in the difference between the think tank and newspaper ideologies and make it easier to compare to the other difference variables in Table 6.

²⁵Typically, using the logarithm of a variable as an independent variable is done because the predicted relationship suggests that the dependent variable is related linearly to percentage changes in the independent variable. In this case, I am merely interested in testing whether the impact is decreasing in the number of publications in a given period. Adding one to the variable is necessary because in a large number of observations think tanks have zero relevant publications at which logarithms are undefined.

²⁶I would like to thank an anonymous referee for suggesting this estimation technique. I used Stata to estimate this model. The procedure command was “glm variable list, family (binomial) link (logit) robust.

*Significant at the 1% Level.

†Significant at the 5% Level.

‡Significant at the 10% Level.

²⁷The only exception may be for the liberal Center for American Progress. The coefficient on its dummy variable indicates is associated with a liberal frame about 32% (at the mean value of the independent variables) more often than other think tanks. However, there are only seven observations so coming to any firm conclusions based on this estimate is difficult.

²⁸This impact is calculated using the impact from change in ideological orientation from the TTI, NNI, and INT variables.

²⁹This estimate is fairly robust across all models and publications except for The Washington Times. If the change in publications were done first, the change in quality would account for one percentage point of the difference and the change in think tank ideology would have accounted for the remaining 15 percentage point difference.

³⁰If that think-tank's name is included in the article multiple times and that name appears in the same paragraph with one of the ideological identifiers at least once, that article will be counted as one observation, no matter how many additional times that think-tank's name appears with (or without) that identifier in the article.

³¹The Proquest [Wall Street Journal (Eastern Edition)] database uses check boxes to limit the search to only news content. As a result the search phrases used with the Wall Street Journal appear in Table 2 to capture information broader than just news content but the searches were limited to only the news content.

³²The search engines used with these databases will automatically include plural forms. It will also include the non-hyphenated version of terms. For instance, searching for the term non-profit will also capture all the articles in which the word is written nonprofit or non profit.