Do National Economic and Political Conditions Affect Ideological Media Slant?

Abstract

Political scientists are interested in the influence of the news media on politics. However, relatively few studies investigate whether or not ideological slant in news coverage changes systematically over time. If it changes systematically, what factors explain the changes? This study argues that external conditions, such as national political and economic situations, influence ideological media slant at the aggregate level. To examine this argument, “macro media bias” is measured quarterly by gauging the relative size of liberal and conservative news stories regarding domestic issues from 1958 through 2004. Utilizing ARIMA models, this study reveals that the news media tend to negatively react to government spending. Also, economic conditions, such as unemployment and inflation, significantly explain changes in the relative number of liberal and conservative news stories.

[Supplementary material is available for this article. Go to the publisher's online edition of Political Communication for the following free supplemental resource(s): the coding keywords, detailed coding rules, and alternative regression results.]

Keywords:

• ideological media slant
• macro media bias
• economic theory of news making
• organizational theory of news making

Notes

1. In contrast, ideological slant at the individual level means whether or not certain media outlets are liberally or conservatively biased at a given time.

2. Groseclose and Milyo (2005) measure media bias by examining who is more frequently cited in news. Gentzkow and Shapiro (2010)measure media slant according to the relative size of liberal and conservative language or expressions in news coverage. Ho and Quinn (2008 p. 356) measure media bias by examining “whether the editorial board supported the Supreme Court majority or not” in editorial pages. These measures seem highly correlated to each other. Kahn and Kenney (2002) also show that bias in editorial pages significantly explains bias in news coverage.

3. See http://www.stateofthemedia.org/2008

4. Prior research (e.g., Gans, 2004; Mencken, 1927; Sparrow, 1999) indicates that reporters are organization men and women. That is, the hierarchical structure of media firms affects news reporting.

5. On the other hand, Sparrow (2006) also speculates that journalists are less likely to take specific political/partisan positions and more likely to report balanced news because of their professionalism and the interaction between their sources and themselves. According to him, “indexing” and “horse-race” campaign coverage can be understood as journalists' efforts to mitigate professional uncertainty.

6. The news media's reaction to politics can be the mere reflection of the public's thermostatic reaction. In order to consider this possibility, this study controls public policy mood in the following statistical analyses.

7. These journalists' role perceptions have not changed notably since the 1960s.

8. This study assumes that news media outlets respond to changes in national economic and political conditions in a similar fashion. This is based on the so-called homogeneity hypothesis of news making (e.g., Sparrow, 1999). However, Cook (2006) argues that pushing this hypothesis too far can be problematic. For instance, Arnold (2004) shows that local newspapers differently cover congressmen compared to prestigious newspapers, such as the Times. Of course, Arnold (2004) also finds some similarities in news making between local newspapers and prestigious newspapers. This study discusses more about this topic in the final section.

9. However, Woolley (2000) criticizes this approach. Especially, he questions the appropriateness of using this approach to count the number of news stories related to specific issues without verifying the contents in abstracts.

10. 10% sampling is determined based on sampling theory (Bartlett, Kotrlik, & Higgins, 1994; Cochran, 1977). According to Cochran's formula for calculating sample size, 10% sampling produces enough samples for this study.

11. The summary of coding is presented in Table 1.

12. Introducing liberal proposals can be considered as liberal if they are positively described in news by reporters or their sources.

13. Standardization means that

14. Certainly, it is possible to construct the series without the standardization process. The unstandardized series tends to heavily reflect the movements in more covered issues, such as the issues of race, the environment, and welfare. The less covered issues are less likely to be reflected in the unstandardized series. The major findings in this study are not altered by using the unstandardized series. Of course, it does not mean that the results are identical. For instance, according to the alternative regression results, changes in the national economy (GDP) do not significantly affect the unstandardized macro media bias series. It may mean that national economic growth generally affects news rather than news of more covered issues. The results in Model 2 are not altered in terms of statistical significance. The significance of the presidential liberalism variable is also changed, but note that presidential liberalism is also produced through the standardized process. Hence, it is not quite meaningful to directly compare them.

15. Intercoder reliability refers to the extent to which independent coders evaluate a characteristic of a message (a news story in this study) and reach the same conclusion.

16. Due to its limited economic resources, this study could not train multiple coders and check all or many of the news stories. The sample may not be sufficiently large. However, since the stories are randomly sampled, the results from this sample should be representative and informative.

17. This does not include the agreement percentage regarding the relevance of news stories. When the relevance of the selected stories is considered simultaneously, the agreement percentage increases over 90%. Of course, the Krippendorff's α statistic also increases.

18. Certainly, the state of the national economy may be measured in various ways. For instance, the Conference Board's Composite Index of Coincident Indicators can be used to measure the state of the national economy This index is constructed from four time series chosen by the Conference Board: payroll employment, personal income, industrial production, and manufacturing and trade sales (Conference Board, 2001). However, the main findings are not altered by replacing the GDP variable with this index. This study uses the GDP variable instead of the composite index because the index includes the information on unemployment.

19. Of course, there are other measures regarding presidential issue stances. For instance, Zupan (1992) applies the Americans for Democratic Action (ADA) scores to presidential position taking on ADA-identified legislation. Some (McCarty & Poole, 1995; Poole & Rosenthal, 1991) utilize presidential position taking on congressional votes to measure presidents' relative policy stances. Others (Bailey, 2007; Erikson et al., 2002) use mixed measures. Wood and Lee's measure and other presidential issue stance measures are highly correlated with each other. While these measures are biannual or annual data, however, the presidential liberalism series is quarterly.

20. This presidential liberalism series is on http://people.tamu.edu/~b-wood/index_files/replication. The series is also produced through the standardization process.

21. The DW-NOMINATE scores are presented at http://voteview.com/default.htm

22. Wlezien (1995) shows that the public thermostatically reacts to government spending.

23. Stimson (1991) includes “race” as a social issue. Wlezien (1995) does not separately include government spending regarding the “race” issue in his analysis.

24. Generally, this study follows the criteria Wlezien (1995) uses. All education spending items under the education categories are considered as education spending. As welfare spending, the items in the “children and family,” “welfare,” and “welfare others” categories are included. “Housing” items are considered as social spending. As health spending, the items under the “health service (senior),” “public health service,” and “health” categories are included. The “pollution abatement” category is also included as social spending.

25. Certainly, there are other ways of measuring government spending, such as using the current dollar.

26. For more information about the data, see http://www.unc.edu/~jstimson/Welcome.html.

27. To examine this argument, ARIMA models (Box & Jenkins, 1976) are utilized. The macro media bias series shows the first-order autoregressive and moving average process: ARIMA (1,0,1). No statistically significant seasonality is found in this series. In all models, the AR(1) and MA(1) variables show statistical significance, and the coefficients of these components are within the stationary boundary. Also, the q statistics from all of the models illustrate that no significant autocorrelation exists between residuals after regression. According to their statistical significance, the AIC/BIC statistics, and the likelihood ratio test statistics, the constant terms are unnecessary in all models in the tables.

28. The economic variables are lagged (one quarter) in the models.

29. Note that the numbers in the tables are coefficients, and t statistics are in parentheses.

30. The public opinion measure (policy mood) is a quarterly series and included as a lagged variable. The event variable is included as a current term.

31. Unemployment is measured as the unemployment rate growth .

32. The responsiveness of the news media to economic conditions is comparable with presidential responsiveness to economic conditions. Presidents, according to Wood (2009), respond to specific economic conditions in different manners. Erikson, MacKuen, and Stimson (2002) also find that the public tends to express more liberal issue sentiments when unemployment increases and more conservative issue sentiments when inflation is high.

33. The political variables share similar characteristics. For instance, when Congress is liberal, more liberal policies tend to be produced. Hence, this study examines them separately. Statistically, including all of them in one model causes typical multicollinearity problems, such as highly inefficient results.

34. Also, note that the macro media bias series and the economic variables are quarterly, but the spending variables are annual, which may explain the sensitivity of the news media on the economic conditions.

35. In order to more thoroughly examine the theoretical reciprocity between political actors including the news media and the public, we need to utilize vector autoregression models and moving average simulations. This is beyond the scope of this study and will be discussed in another study in detail.

36. Of courese, some news outlets are more liberal than the Washington Post (Gentzkow & Shapiro, 2010 Groseclose & Milyo, 2005; Ho & Quinn, 2008)

37. According to Groseclose and Milyo (2005) and Ho and Quinn (2008) the ideological issue positions of news outlets are spread rather than converged on the median position.

38. Of course, this study recognizes that some media outlets are more liberal than others. As we can see in Table 1, the New York Times is a liberal news media outlet, which is quite comparable with other empirical studies (Gentzkow & Shapiro, 2010 Groseclose & Milyo, 2005; Ho & Quinn, 2008) However, the topic this study investigates is how the news media respond to changes in external conditions, which is different from the previous topic: whether or not media outlets are biased in a specific way.