POLITICS, COMPETITION, AND THE MEDIA BY CHUN-FANG CHIANG B.S., NATIONAL TAIWAN UNIVERSITY, 1999 M.A., NATIONAL TAIWAN UNIVERSITY, 2001 A.M., BROWN UNIVERSITY, 2004 A DISSERTATION SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY IN THE DEPARTMENT OF ECONOMICS AT BROWN UNIVERSITY PROVIDENCE, RHODE ISLAND MAY 2008 c Copyright 2008 by Chun-Fang Chiang This dissertation by Chun-Fang Chiang is accepted in its present form by the Department of Economics as satisfying the dissertation requirement for the degree of Doctor of Philosophy. Date Brian Knight, Advisor Recommended to the Graduate Council Date Ross Levine, Reader Date Pedro Dal Bo, Reader Approved by the Graduate Council Date Sheila Bonde, Dean of the Graduate School iii VITA Chun-Fang Chiang was born in Taipei, Taiwan on October 14, 1977. She received her B.S. in Mathematics in 1999 and M.A. in Economics in 2001 from National Taiwan University. She entered the Graduate Program of Economics at Brown University in 2003 and received the her A.M. in Economics in 2004. Her dissertation is concentrated in the field of political economy. iv ACKNOWLEDGMENTS I would like to express my appreciation to my advisor, Brian Knight, for his guidance, insight and encouragement. He introduced me to interesting topics and helped me formulate research questions. His constant help and stimulating suggestions kept me on the track of doing research. Without him I would not have completed my dissertation. I am very grateful for having an exceptional doctoral committee and would like to thank Ross Levine and Pedro Dal Bo for their support and encouragement. They always gave me inspiring feedback and made important points that improved my dissertation. I would like to express my appreciation to Kaivan Munshi, Andre Foster, Kenneth Chay, Sheetal Sekhri, Ying Pan, Masaki Yamada, and Ruben Durante for their helpful comments and discussion. I would like to thank Angelica Spertini for tremendous administrative and emotional support. I would also like to thank Celeste, Sybil, and Wenchy, all of whom have been extremely supportive. I would also like to thank my husband, Jeng-Daw Yu, who is always traveling to visit me without complaint. Lastly, I would like to thank my mother for all the hardship she has been through to raise her children. v Contents 1 Media Bias and Electoral Competition 1 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Literature Review 4 1.3 1.4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Theories of Media Bias . . . . . . . . . . . . . . . . . . . . . . . . . 4 1.2.2 Empirical Evidences on Media Bias . . . . . . . . . . . . . . . . . . . 5 1.2.3 Measure of Media Bias . . . . . . . . . . . . . . . . . . . . . . . . . . 5 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.3.1 Preferences and Action . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.3.2 Electoral Competition and Media Bias . . . . . . . . . . . . . . . . . 7 1.3.3 Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Evidence from Local TV News . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.4.1 Measure of media bias . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.4.2 Measure of Electoral Competition . . . . . . . . . . . . . . . . . . . 10 1.4.3 Econometric Specification and Results . . . . . . . . . . . . . . . . . 12 1.4.4 Baseline Specification Results . . . . . . . . . . . . . . . . . . . . . . 13 1.4.5 Potential Problem of the Daily Measure . . . . . . . . . . . . . . . . 14 1.4.6 Alternative Media Bias Measure . . . . . . . . . . . . . . . . . . . . 14 1.5 Evidence from Media Believability . . . . . . . . . . . . . . . . . . . . . . . 15 1.6 Interpretations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 1.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2 Political Differentiation in Newspaper Markets 24 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 vi 2.3 2.4 2.5 2.6 A Model of Newspaper Competition . . . . . . . . . . . . . . . . . . . . . . 29 2.3.1 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.3.2 Case I: Monopoly Newspaper . . . . . . . . . . . . . . . . . . . . . . 30 2.3.3 Case II: Two Newspapers . . . . . . . . . . . . . . . . . . . . . . . . 30 2.3.4 Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Data and Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 2.4.1 Newspaper Competition . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.4.2 Newspaper Reading and Political Demographics . . . . . . . . . . . 32 2.4.3 High-Speed Internet Supply . . . . . . . . . . . . . . . . . . . . . . . 34 2.4.4 A Test Using Aggregate Readership Data . . . . . . . . . . . . . . . 34 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 2.5.1 The Case of a One-Newspaper Market . . . . . . . . . . . . . . . . . 35 2.5.2 The Effect of Newspaper Competition . . . . . . . . . . . . . . . . . 36 2.5.3 Alternative Explanation – Heterogenous News Demand . . . . . . . 37 2.5.4 Reverse Causality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.5.5 Effect of Newspaper Competition on Getting Information Online . . 40 2.5.6 Influence of the Internet . . . . . . . . . . . . . . . . . . . . . . . . . 41 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3 Media Bias and Influence: Evidence from Newspaper Endorsements 58 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.2 Related Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 3.2.1 Sources and Measure of Media Bias . . . . . . . . . . . . . . . . . . 60 3.2.2 Media Influence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.3.1 Voter behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.3.2 Newspaper Endorsements . . . . . . . . . . . . . . . . . . . . . . . . 65 3.3.3 The Influence of Endorsements . . . . . . . . . . . . . . . . . . . . . 65 3.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 3.5 Empirical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 3.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 3.3 vii 3.6.1 3.7 Persuasion Rate and Counterfactuals . . . . . . . . . . . . . . . . . . 71 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 viii List of Tables 1.1 Electoral Competition and Media Bias in 2000 (Using Voter Return in 1996) 19 1.2 Electoral Competition and Media Bias in 2000 (Using State Polls in 2000) . 20 1.3 The Variance of Media Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 1.4 Alternative Media Bias Measure . . . . . . . . . . . . . . . . . . . . . . . . 22 1.5 Media Believability in 2002 and 2004 . . . . . . . . . . . . . . . . . . . . . . 23 2.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 2.2 One-Newspaper Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 2.3 Markets in Competition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 2.4 Effect of Newspaper Competition on Reading a Local Newspaper . . . . . . 52 2.5 Change in Newspaper Competition . . . . . . . . . . . . . . . . . . . . . . . 53 2.6 First Stage Regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 2.7 Political News v.s. Non-political News . . . . . . . . . . . . . . . . . . . . . 55 2.8 Effect of Newspaper Competition on Getting Information Online . . . . . . 56 2.9 Readership of the New York Times and Wall Street Journal . . . . . . . . . 57 3.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 3.2 Effect of Newspaper Endorsements on Vote Intention . . . . . . . . . . . . . 78 3.3 Effect of Newspaper Endorsements on Candidates’ Favorability . . . . . . . 79 3.4 Effect of Newspaper Endorsements on Voters’ Preference . . . . . . . . . . . 80 3.5 Influence of Top 20 Newspapers in 2000 . . . . . . . . . . . . . . . . . . . . 81 3.6 Counterfactuals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 ix List of Figures 2.1 One-Newspaper Market - Hartford, CT . . . . . . . . . . . . . . . . . . . . 48 2.2 Two-Newspaper Market - Boston, MA . . . . . . . . . . . . . . . . . . . . . 48 3.1 Dates of Newspaper Endorsements in 2000 and 2004 . . . . . . . . . . . . . 76 x Chapter 1 Media Bias and Electoral Competition 1.1 Introduction Mass media provides people with information to make decisions. However, the news media is viewed widely as biased (Baron, 2004; Goldberg, 2002; Alterman, 2003). During presidential campaigns, the media may provide biased information about a candidate’s political ideologies and their background. The media may also provide unbalanced news reports in favor of a specific candidate. On the one hand, media bias in an election could be motivated by an incentive to impact the electoral outcome. On the other hand, media bias could come from the desire to cater to major preferences of consumers in the media market. By observing the effect of electoral competition on media bias, this paper aims to distinguish the sources of media bias. Intuitively, if media bias is generated to cater to consumers’ preferences, there will be more media bias in favor of the Republican candidate when there is more Republican support in an area. However, when the area is not a Democratic or Republican stronghold, there should be less media bias. If media bias mainly comes from the desire to affect the election outcome, we should observe increased media bias when the election is more competitive, since that is when the media is more influential. The latter mechanism is somewhat intuitive but not very clear. Therefore, I construct a simple model to explain how electoral competition will increase media bias under the assumption that the journalist cares about the election outcome. 1 2 In the theoretical model, media bias is generated by media outlets. When deciding the optimal bias, the media outlet faces a tradeoff. The journalist may increase the probability of having the preferred candidate elected by generating more media bias. However, the reputation of the media firm may suffer in the process. The model shows that the marginal benefit of media bias is larger in a close election. As a result, the model has the prediction that media bias is greater when the election is more competitive. In the empirical analysis, two different data sets are used to investigate the electoral competition effect on media bias: local news archives from the Norman Lear Center in 2000 (Kaplan and Hale, 2001) and the believability surveys conducted by the Pew Research Center in 2002 and 2004. From the data of local news archives, media bias is measured by the relative length of presidential candidates’ speaking time in local TV news stories prior to the 2000 presidential elections. The electoral competitiveness is measured by vote returns in the last presidential election and the daily price of vote share in the Iowa election market. The empirical results show that media bias is greater in media markets where the election is more competitive. The believability survey conducted by the Pew Research Center provides a subjective measure of media bias. It asks people to rate the believability of the local media and the national media. The data shows that the believability of daily newspapers is lower in media markets where the election is more competitive, while the believability of national papers is not affected by electoral competitiveness. Reported believability of the news outlets could fall as competition increases because people are more anxious or more skeptical when an election is close. However, these unobservable characteristics should also affect the believability of daily newspapers as well as national newspapers. Since the believability of national newspapers is not significantly affected by the electoral competition , we know that the effect of electoral competition on the believability of daily newspaper is not simply because people are more anxious or skeptical. These empirical results imply that media bias does not come from the majority’s preferences, but rather it comes from the desire to affect election outcome. The term “media bias” has been defined in several ways. In Barron’s paper, media bias 3 is defined relative to the truth (Barron, 2004). The ASNE1 found that some people believe that media bias is “favoritism of a particular social or political group”, which is closer to the meaning of media bias in this paper. Under this definition, journalists could report the truth but favor one particular political party in his/her selection of information to be reported. In the theoretical model, media outlets can increase the probability of having their preferred candidate elected by increasing the candidate’s positive coverage. In the empirical analysis of local news archives, media bias is measured by the relative length of candidates’ speaking time. One assumption behind this media bias measure is that both candidates try to maintain a positive image and deliver clear information about themselves to the public. Therefore, when a reporter gives one candidate more sound bites, the reporter helps the candidate deliver messages to voters and also increases public familiarity. While these assumptions are hard to test, this measure is easy to calculate and can be obtained without any subjective judgement. This paper relates to the emerging literature that investigates the sources of media bias. The literature can be divided into two approaches: the demand side approach and the supply side approach. From the demand side approach, media bias originates from consumer preference. Consumers may like to see news that confirms their beliefs (Mullainathan and Shleifer, 2005). Even if consumers would like to choose media that delivers accurate information, their choices may be still affected by their prior beliefs (Gentzkow and Shapiro, 2006); therefore media outlets may have media bias to cater to consumers’ beliefs. In terms of targeting groups, the media could deliver more news to large groups due to the increasing-return-to-scale technology and advertising financing of media firms (Stromberg, 2002). From the supply side approach, it could reflect the preferences of journalists (Baron, 2004), editors, or owners (Besley and Prat, 2004). The empirical results suggest that media bias is derived from the desire to affect election outcomes and therefore provide support for the supply side approach. However, when there is more than one TV news station in the media market, the motivation to affect election outcomes could also come from the demand side, affected by the audience that shares political views with the media firm. 1 The American Society of Newspaper Editors (ASNE), “Perceived Bias,” p. 4, www.asne.org 4 However, the finding from the believability survey can not be completely explained by the demand side factors. There was only one daily newspaper in most media markets, and if there were more than one daily newspaper in the media market, respondents were asked to rank the one that they were most familiar with. Therefore, the demand side factors are unlikely to completely explain the decreased believability of the daily newspaper in a competitive election. If we assume that the believability of the media is related to consumer welfare, the consumer welfare is weakened when the election is more competitive. In terms of election outcome and voters’ welfare, if the media bias is generated to cater to the majority’s preference during the election, the election outcome is less likely to be affected. However, if media bias is coming from the motivation to affect election outcome, undecided voters may get biased information, which may result in an undesirable outcome. The paper is organized as follows: section 2 is the literature review, section 3 presents the model, section 4 and section 5 illustrate the econometric specifications and empirical results. Section 6 discusses interpretations of the results. Section 7 concludes. 1.2 1.2.1 Literature Review Theories of Media Bias There is little consensus about why the media is biased or how the competition between media firms affect media bias. Most theories of media bias try to provide a convincing story about where the media bias originates and predictions of how competition between media firms affects media bias. Under the assumption that readers hold beliefs that they like to see confirmed, Mullainathan and Shleifer (2005) provide a theory about how consumer preference and the competition between newspapers affect media bias. They find that if consumers’ beliefs are homogeneous, competition among media firms will not reduce media bias. Moreover, if consumers’ beliefs are heterogenous, duopolist newspapers will differentiate themselves by reporting extreme news so that they can charge higher prices. Gentzkow and Shapiro (2006) make the assumption that consumers will update their belief based upon the information they receive and, in addition, will choose a media firm 5 according to reputation. In their model, consumers care for the reputation of media firms. However, consumers’ judgements are still affected by their beliefs, therefore media firms will slant the news to cater to consumers’ beliefs. 2 Media bias may also come from the supply side. Baron (2004) explains the persistent media bias based on incomplete information and the career interests of journalists. In his theory, media bias originates from journalists whose career interests lead them to sacrifice current wages for future opportunities. A profit-maximizing news organization tolerates bias only if it gains more on the supply side than it loses on the demand side. Competition from a news organization with less bias could force a higher bias news organization to reduce the discretion granted to its journalists. However, competition between two news organizations with opposing biases can result in more biased news than with a monopoly news organization. 1.2.2 Empirical Evidences on Media Bias While there is some evidence indicating that the media may cater to consumers’ beliefs, there is no clear evidence about the source of media bias from the supply side. Dalton, Beck, and Huckfeldt (1998) find that the editorial stance of local newspapers is correlated with local perceptions of candidates in the 1992 presidential election. Erikson (1976) and Gunther (1992) also provide similar findings. However, this evidence does not strongly support the demand side stories, since it is hard to distinguish between the demand side factors and the supply side factors. 1.2.3 Measure of Media Bias Media sends out information by news stories, editorials, advertisements, or other formats. It is often more difficult to detect media bias from news stories than from other information sources since news stories are usually not slanted in an obvious way. Groseclose and Milyo (2005) provide a way to measure media bias in the news stories of various media outlets. 2 Their model predicts that when consumers’ beliefs are homogeneous, competition between independently owned news outlets can reduce bias since more news outlets will increase the probability that consumers receive the true information/feedback. This prediction is different from the prediction of Mullainathan and Shleifer (2005). However, the effect of competition between media firms is not clear when consumers are heterogeneous. 6 They counted the citations of think tanks made by the media and computed a score by comparing those citations to citations of think tanks in speeches made by members of Congress. Gentzkow and Shapiro (2006) use candidates’ sound bites in TV news stories to generate a measure of media bias. This could be an objective measure since it is directly calculated by counting sound bites in news stories. In this paper, we also use this method to measure media bias. 1.3 The Model In this section, we present a model to explain how the electoral competition affects the media’s motivation to report unbalanced news. Under the assumption that media owners care about the election outcome, the model predicts that there will be more media bias when the election is more competitive. 1.3.1 Preferences and Action There are two candidates in the election. Candidate D is from the Democratic party and candidate R is from the Republican party. Citizens are divided into three groupsDemocrats, Republicans, and Independents. Democrats and Republicans will vote for their candidate. Independents will vote for Democrat or Republican candidates depending on the information they receive from the media. Let πD , πR , and πI denote, respectively, the fraction of voters who are Democrats, Republicans, and Independents. There is only one media outlet. Two candidates are sending out positive information about themselves in equal amounts to the media. However, the media outlet can choose to give one or the other candidate more positive news reports. Unbalanced news reports will hurt the media firm’s reputation and affect their long run profits. Therefore, the media firm will monitor the news reported by the journalist working for the firm. The journalist cares about who will be elected in the election and the wage affected by the quality of news. Suppose that the journalist is a Democrat, the journalist’s utility function can be expressed by the following: U = αPD uD + (1 − α)u(Q), (1.1) where PD is the probability of candidate D to be elected, uD is the utility he receive if 7 candidate D is elected. u(Q) is the utility affected by the news quality. Let 1/2 + b be the fraction of Democrat information covered in the news, where b is the media bias towards the Democratic candidate. Assume that an Independent’s vote decision depends on the single information he randomly receives from the media, the probability of voting candidate D, pID , is equal to 1/2 + b.3 I be the fraction of Independents voting for the candidate D. The probability of Let rD candidate D to be elected is I I PD = P rob(πD + rD πI > πR + (1 − rD )πI ) I = P rob(rD > 1.3.2 (1.2) 1 πR − πD ) + 2 2πI (1.3) Electoral Competition and Media Bias We are interested in how electoral competition affects media bias. Let |πR −πD | πI be a measure of the electoral competitiveness. We can classify these situations into the following three cases: D| > 1 . In this case, the election is not competitive. PD is either 1 or • Case 1: |πRπ−π I 0. The bias has no effect on the probability of winning for candidate D. The optimal bias b will be 0. D| • Case 2: |πRπ−π < 1 and πD − πR > 0 I In this case, there are more Democrats in the media market. Given the probability of an Independent voting for the candidate D, pID , the fraction of Independents voting I is approximately distributed as a normal distribution with for the candidate D, rD mean pID , and variance pID (1 − pID ). Let σ = (pID (1 − pID ))1/2 , we have 1 πD − πR PD ∼ + b)), = Φ( ( σ 2πI (1.4) ∂PD 1 1 πD − πR = φ( ( + b)). ∂b σ σ 2πI (1.5) and 3 The information from the media firm can affect an Independent’s decision by increasing the candidates’ charisma (Andina Diaz, 2004). The assumption that an Independent’s decision depends on one information from the media could be too simple. However, as long as pID is a function of b with the properties pID (1/2) = 1/2 and pID 0 (b) > 0. The results we have in the model will not change. 8 ∂PD ∂b is decreasing in |πR −πD | . πI Given the utility function in equation 1, the first order condition with respect to b is αuD ∂PD + (1 − α)u0 (b) = 0, ∂b (1.6) where u0 (b) is u0 (Q)Q0 (b). u0 (b) is negative because bias will hurt the performance of the journalist. The second order condition is αuD ∂PD + (1 − α)u00 (b) < 0. ∂b (1.7) From the first order condition and second order condition, we have the effect of |πR −πD | πI on the optima b∗ : ∂b∗ D| ∂ |πRπ−π I =− ∂P ∂ ∂bD |πR −πD | ∂ πI D 00 ∗ αuD ∂P ∂b + (1 − α)u (b ) = 1 R φ0 ( σ1 ( πD2π−π + b))) 2σ 2 I − ∂PD αuD ∂b + (1 − α)u00 (b∗ ) < 0. (1.8) Therefore, when the election is more competitive, the journalist will choose to generate more bias. • Case 3: |πR −πD | πI < 1 and πD − πR < 0. In this case, there are more Republicans in the media market. Similar to case 2, the Democratic journalist could increase PD by choosing positive b and equation (8) still holds. Therefore, bias b will still increase when the election is more competitive. However, in case 3 we can see in b as long as πD −πR 2πI ∂PD ∂b is increasing in b, while in case 2 ∂PD ∂b is decreasing + b < 0. Therefore, the marginal benefit of generating positive bias is greater in case 3. Intuitively, compared with case 2, the Democratic journalist will generate more bias. Moreover, from the equation (8), we can see the absolute effect of electoral competition on media bias is smaller in case 3, which will result in proportionally larger media bias in case 3. 1.3.3 Predictions The main prediction from the model is that when the election is more competitive, the journalist will choose higher bias. When deciding the optimal bias, the journalist faces a tradeoff between increased probability of having the preferred candidate elected and the 9 decreased reputation associated with bias. As we can see in the model, when the election is more competitive, the marginal benefit of media bias from the increased probability is larger due to the property of the underlying normal distribution. Therefore, the optimal bias for the journalist will be higher when the election is more competitive. In the above model, there is only one media firm in the market. However, it is not difficult to expect the situation with two media firms. In a market with two media firms with opposite political views, each firm will generate media bias in favor of the preferred candidate. As the number of media firms increases, a single media firm may lose some of their audience and become less influential. The number of media firms may also increase media bias in some firms since firms may use target different groups of audience. Therefore, the number of media firms may have an ambiguous effect on media bias. However, in the equilibrium, the positive effect of electoral competition on media bias will still hold. 1.4 Evidence from Local TV News Empirically, in the presidential election, the motivation of generating media bias is not only affected by state level competitiveness but also affected by national level competitiveness. When the election is competitive nationally, all the battleground states become crucial. Therefore, the presidential election in 2000 provides a good chance to observe the electoral effect on media bias. From the local news archives in 2000 (Kaplan and Hale, 2001), we are able to construct the measure of media bias by using the relative length of the candidates’ speaking time. The data encodes the characteristics of local election news coverage broadcast between 5:00pm and 11:35pm prior to the general election for about one month. Stories not related to the election are not included. The characteristics include the length of stories, the length of each candidates’ sound bites, and the main issue in those stories. The sample consists of 74 television stations from 58 of the top 60 markets across the country. In each market, the station that received the most amount of political advertising revenue during the previous month was chosen. Additional stations were chosen to include more broadcasters that had made commitments to meet the 5/304 standard. 4 White House panel’s recommendation of airing 5 minutes of candidate centered discourse (CCD) a night 10 There are 210 media markets in the United States. Those media markets are defined by the Nielsen Media Research. Some of the media markets cross several states and some of the media markets are within one state. Most are defined by counties. Only a few counties are split into different media markets. County composition and some information for each media market are taken from Broadcasting and Cable Yearbook 2000 and Demographics USA 2000. 1.4.1 Measure of media bias The media bias measure of station i on date t is constructed by the relative length of candidates’ speaking time : biasit = ( 1 Repit − )2 , Repit + Demit 2 where Repit and Demit denote the number of seconds given to the Republican and Democratic president candidates from station i and date t. 1.4.2 Measure of Electoral Competition It is difficult to measure electoral competition for every media market everyday during the campaign period in 2000. One possible source is the polls conducted at the state level. However, those state polls are conducted by various institutions and some states have little polls data available. Therefore, three electoral competition measures are constructed: The electoral competition measure over markets based on the vote return in 1996 election, the daily electoral competition based on the Iowa election market, and a predicted electoral competition measure constructed from the above information under the assumption that people in different states change their political attitude in the same way. A. Measure of electoral competition over markets To construct an electoral competition measure for each media market, we first construct a competition measure for each state. The state competition measure is constructed by the election return data in the previous presidential election. The state competition measure used for the 2000 regression is defined by the square of the difference of vote share in 1996 : Com1996 = −( s V Dem1996 V Rep1996 s s − )2 , 1996 1996 + V Dem1996 V Rep1996 + V Dem V Rep s s s s in the last 30 days of a campaign (1.9) 11 where V Rep1996 and V Dem1996 are the votes for Republican and Democratic candidates in s s state s in 1996. For any media market that is covered by a single state, the above measure is the competition measure for the media market. For the media market that spans across several states, the electoral competition measure is weighted by the total votes in each state within the media market. Consider the media market across n states. The competition measure for the media market m can be expressed as follows: Competition1996 = m n X ws Com1996 , s (1.10) s=1 where ws are the total votes in state s within market m in the 1996 election. B. Daily electoral competition measure The price of the vote share in Iowa election market is used to measure electoral competition over the 30-day period prior to the election. 5 The competition measure on date t is defined by: Competition2000 = −( t P Rep2000 P Dem2000 t t − )2 , P Rep2000 P Rep2000 + P Dem2000 + P Dem2000 t t t t (1.11) where P Rept is the price of the Republican candidate’s vote share at time t from the Iowa election market. C. Predicted electoral competition While Competitionm measures the electoral competition over each market by using the data 4 years before the election, Competitiont measures the electoral competition over the 30-day period prior the election. To get a better competition measurement for each market near the election in 2000, we combine the information from the 1996 election return and the tracking poll in 2000 to construct the variable P redicted Competitionm,t . The first step is to predict the fraction of people who support the Republican candidate in states in 2000 at time t: Reppredicted = V Rep1996 + (P Rep2000 − V Rep1996 s t N ational ), s,t (1.12) 5 The Iowa election market was closed at midnight every day. Therefore, the vote share price on the midnight is matched with the TV news data in the next day. The tracking polls conducted by Gallup group could be also a source to generate the daily electoral competition measure. However, it was a three day average poll and the poll number was released on the day after the three day period. The matching will be less precise in terms of the timing of matching. 12 where V Rep1996 N ational is the national vote share for the Republican candidate in 1996. The second term in the bracket is the national change in Republican supporters. This method implies the assumption that people in different states change their political attitudes in the same direction and by the same magnitude. Then we use the predicted fraction of Republican supporters in state s at time t to construct the competition measure for state s at time t. Compredicted = −( s,t Reppredicted s,t Reppredicted + Dempredicted s,t s,t − Dempredicted s,t Reppredicted + Dempredicted s,t s,t )2 . (1.13) Again, for the media market across several states, we generate the competition measure by weighted state competition measure. Competitionpredicted = m,t ms X ws Compredicted , s,t (1.14) s=m1 1.4.3 Econometric Specification and Results The main prediction of the model is that electoral competition will increase media bias. The baseline specification for the test of the this prediction is: biasit = competitionm β1 + competitiont β2 + Xm γ + αi + it , (1.15) where bit is the media bias measure of station i at time t. Competitionm and competitiont are the electoral competition measure over market and over time constructed by the methods in section 2. Xm include some characteristics of the media markets. αi is the random component for each station. The competition measures increase when the election is more competitive. Therefore, we expect if the supply side factors are dominant, we expect β1 and β2 to be positive, as predicted by the supply-side model. If the demand side factors of media bias are dominant, then we should see negative β1 and β2 . In principle, media bias could also affect electoral competition. If the media generate a large amount of bias in favor of one political party, then the voters’ political preference could change due to the media bias. This could change the electoral competition. Therefore, the competition measure could be endogenous. However, if the media bias makes the election 13 less competitive, the results will favor the story that the bias is generated to cater to the preferences of major consumers. Consider the electoral competition measure over media markets. This measure is driven from the vote return of last presidential election. It could be endogenous if the media bias of the TV stations between two presidential elections are correlated. Since some TV stations may make the election more competitive while others make the election less competitive, overall, the correlation between media bias and electoral competition caused by the endogeneity should be weak. Consider the daily electoral competition measure. Since there are 210 media markets in the U.S, the causality of the media bias from one local TV station on the national electoral competition is even weaker. 1.4.4 Baseline Specification Results Table 1 column 1 presents the result of baseline specification. Column 2 presents the results with time fixed effect. In Column 3, predicted competition measure is used to measure the daily electoral competition of each media market. The results show that media competition increases media bias except for the case of using daily electoral competition measure.6 One possible explanation for the weak effect of daily competition measure is that the change in the national electoral competition does not reflect the change in the electoral competition in every media market. For example, in a Republican stronghold or Democrat stronghold, the election may never be close during the election period. The other possible explanation is that the local TV station or the reporter may not update their attitude everyday. Table 2 presents the result when the electoral competition measure over market is coming from the state polls conducted by the American Research Group in September 2000 instead of the vote return in 1996. Although the size of the state poll was only around 600 people for each state, the time is closer to the election in November 2000. When the vote share of the last presidential election is replaced by the data on state polls in September 2000, the 6 Another possible data source to generate the daily electoral competition measure is from the daily tracking poll conducted by the Gallup group. While using the daily tracking poll to generate daily electoral competition measure, the results depend on the way to deal with the time issue. The daily tracking poll is a three day average poll. If the poll data is matched with the second of three days the poll was conducted, the result will be more media bias as the election is more competitive . However, if the poll data is matched with the date on which the polling results are released, which is the day after the final day of polling, there will be less media bias when the election is more competitive. 14 effect of predicted competition measure becomes more significant while other results remain the same. The variable number of local TV stations is the total number of TV stations in the media markets, including commercial and non-commercial TV stations. It is used as a proxy of competition between local news programs. Intuitively, the influence of one media firm will be lessened when there are more media firms. Therefore, we should see less media bias when there are more TV stations in the media market. However, this variable is not significant in most specifications. The variable total speaking time is the sum of two candidates’ sound bites. This variable has statistical meaning rather than economic meaning. If a journalist is fair to each candidate, when there are more sound bites available in the news stories, we expect to see the fraction of one candidate’s sound bites closer to 1/2. This variable is significant in all specifications. 1.4.5 Potential Problem of the Daily Measure In the election period, candidates could alternate visits to the battleground states, within days of each other, and the result would be a larger variation in the sound bites of one candidate over time. To make sure the competition effect on media bias is not simply reflected by the daily variation in sound bites caused by candidates’ visits, a media bias measure for each station is constructed by the mean of the fraction of Bush’s speaking time: biasi = ( 1 Repit − )2 , Repit + Demit 2 Table 3 column 1 shows that there is more variation in the fraction of Bush’ speaking time caused by the electoral competition. Column 2 and 3 present the OLS results that the stations in the media market where the election is more competitive still have more bias, however, the effect is only 10 percent significant. 1.4.6 Alternative Media Bias Measure While the above media bias measure reflect the relative coverage of candidates, the fraction of reporters’ speaking time measures the possibility of reporters trying to add more personal 15 opinion in the news story.7 While the assumption behind this measure is strong, it has the advantage that it will not be affected by the alternative visits of candidates. The results of using this media bias measure are presented in Table 5. The electoral competition does not affect the variation of this media bias measure and has a positive effect on media bias for this media bias measure for each station. 1.5 Evidence from Media Believability Since electoral competition affects the news coverage in the local TV news programs, it may also affect the way people think about the media if they are aware of the bias. If the media has more bias in the area where the election is more competitive, people who live in that area should be less likely to believe the media. In the believability survey conducted by the Pew Research Center in 2002 and 2004, the survey question asked people to rank the media’s believability. By comparing the effect of electoral competition in the media market on the believability of local TV news, daily newspaper and some other major national news outlets, we can test if the local news outlets are less believable in an area where the election is more competitive. The survey question is “....Please rate how much you think you can believe each organization I name on a scale of 4 to 1. On this four point scale, “4” means you can believe all or most of what the organization says. “1” means you believe almost nothing of what they say....” The news organizations include media outlets such as ABC News, CNN, FOX News, local TV news , USA today, The New York Times, the daily newspaper as well as other news outlets. I use the ordered Probit model to examine the effect of electoral competition on the believability of the media. The dependent variables are the believability of the daily newspaper, local TV news, USA today and ABC news. Dependent variables include competition measure in the market, education, income and city size. The results are presented in Table 4. The believability of daily newspapers is significantly lower in the market where the election is more competitive. However, the believability of the TV news is not significantly 7 Center of Media and Public Affairs analyzed the candidates’ sound bites of ABC, CBS and NBC evening news programs during the presidential election period from 1988. While they did not find any significant difference in the fraction of candidates’ sound bite, they found that the fraction of the speaking time given to the candidates is particularly low in 2000. 16 affected by the electoral competition, although it is more affected than the national media. These results could be affected by unobserved characteristics of the consumers. For example, people living in the battleground states may be more skeptical or require more accurate news information. However, if that were the case, these unobserved characteristics would also affect their view on national newspapers and TV programs. Since people living in the battleground states do not differ significantly in their opinions about the believability of USA Today and network TV programs, we are able to exclude this possibility. The different results of the TV news and daily newspapers could come from their different market structures. In general, in a media market, there are only one or two newspapers in a media market while there are usually more than two local TV stations with news programs. In a monopoly market structure, the supply side factor should be stronger than it is in an oligopoly market. Also, in a market where there are more choices, consumers are more likely to find the product they prefer. 1.6 Interpretations The empirical results show that media bias increases in the media market where the election is more competitive. The empirical finding is consistent with the supply side story that the media firm or the reporter has its own political preference. From the demand side approach, when there is more than one media firm in the market, the electoral competition effect on media bias is less clear. One the one hand, the major preference still matters; on the other hand, media firms may target different group of people and the news reports will be affected by the audience. The audience may prefer more bias as the election is competitive with the incentive consistent with the spirit of the model. Another possible explanation from the demand side is that people do not care about the election when the election is not competitive. And only when the election is competitive, do people want to watch the election news that share their political views. Therefore, the results are due to the higher demand for news in a competitive election. However, the number of election news stories is not correlated with the electoral competition across markets, which means the demand for news does not increase much in battleground states. However, the empirical findings from the believability survey can not be completely 17 explained by the demand side factors. First, a lot of media markets only have one daily newspaper. In this case, the demand side factors will always force the media firm to have less media bias when the election is more competitive. Second, in the media market where there is more than one daily newspaper, since people are asked to rank the believability of the daily newspaper with which they are most familiar, it is hard to explain why the believability is related to electoral competitiveness from the demand side approach. 1.7 Conclusion In this paper, a simple model is provided to explain how electoral competition affects media bias under the motivation to influence the election outcome. The prediction of this model is that there will be more media bias when the election is more competitive. From the TV news stories prior to the presidential election in 2000 and the believability survey, I find support for the prediction. Rather than catering to the preferences of major consumers, the desire to affect the election outcome remains dominant. While the election is more competitive, information from the media is more important for voters when making voting decisions. However, from this study I find that it is even more difficult to get unbiased information when an election is more competitive. If consumer welfare is related to the believability of newspapers, consumer welfare is weakened when the election is more competitive. This study provides support for the general worries about media bias. If the media bias is generated to cater the majority’s preference during the election, the election outcome is less likely to be affected. However, if media bias is motivated by a desire to affect election outcome, undecided voters may get biased and thus inaccurate information. 18 References Baron, David, (2006) “Persistent Media Bias,” Journal of Public Economics 90(1-2): 1-36. Besley, Timothy and Andrea Prat, (2006) “Handcuffs for the Grabbing Hand? Media Capture and Government Accountability,” American Economic Review, 96(3):720736 Chan, J. and W. Suen, 2004, “Media as watchdogs: The role of news media in electoral competition ,” working paper. Djankov, S., C. Mcliesh, T. Nenova, and A. Shleifer, 2003, “ Who owns the media? ” Journal of Law and Economics, 46: 341-381. Gentzkow, Matthew and Jess M. Shapiro “ Media bias and Reputation,” Journal of Political Economy, April 2006, forthcoming. Groseclose, T. and J. Milyo (2005) “A measure of media bias,” Quarterly Journal of Economics 120(4): 1191-1237. Kaplan, M. and M. Hale (2001) Local TV coverage of the 2000 general election. Norman Lear Center, USC Annenberg School of Communication. Mullainathan, S. and A. Shleifer (2005) “Market for news,” American Economic Review 95(4): 1031-1053. Stromberg, David (2004), “Mass Media Competition, Political Competition, and Public Policy,” Review of Economic Studies 71(1): 265-284. 19 Table 1: Electoral Competition and Media Bias in 2000 --Using Voter Return in 1996 Biasit Dependent variable Competition Measure over (1) (2) 0.4996** 0.4871** (2.03) (2.41) (3) Market Daily Competition Measure 11.2413 (1.41) Predicted Electoral Competition 0.6870* (1.86) Total Speaking Time Number of local TV stations 5/30 stations Date(time trend effect) Log of population Constant -0.0006*** -0.0004*** -0.0006*** (4.54) (3.20) (4.52) -0.0005 -0.0005 0.0006 (0.23) (0.28) (0.26) -0.0285 -0.0284* -0.0340* (1.59) (1.92) (1.88) 0.0014** 0.0010* (2.19) (1.84) -0.0123 -0.0188 -0.0279 (0.49) (0.89) (1.03) 29.8697** -0.59*** 23.73 (2.21) (3.78) (1.36) Date fixed effect Yes Observations 1061 1061 1061 Number of group(station) 74 74 74 Notes: Dependent Variable Biasit is the absolute difference between fraction of Bush’s speaking time and 1/2 in the news stories from station i on date t. The variable Competition measure over market is based on the vote share of two candidates in 1996 Presidential election, as defined in section 4.2. The variable Daily Competition Measure is based on the midnight prices of the vote share of two candidates in the previous day, as defined in section 4.3. The variable Total Speaking Time is the total speaking time of Bush and Gore in the news stories from station i on date t. The variable 5/30 stations is a dummy variable to indicate if the station made the 5/30 commitment according to White House’s suggestion of airing 5 mins of candidates’ coverage in the 30 minutes evening news Absolute value of t statistics in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%. 20 Table 2: Electoral Competition and Media Bias in 2000 --Using State Polls in 2000 Dependent variable Media bias (1) Competition Measure over Market Daily Competition Measure (2) 0.6338** 0.6056*** (2.53) (2.88) (3) 11.1281** (1.39) Predicted Electoral Competition 0.6393 (2.73) Total Speaking Time Number of local TV stations 5/30 stations Date(time trend effect) -0.0006*** -0.0003*** -0.0006*** (4.55) (3.21) (2.73) 0.0005 0.0004 0.0003 (0.21) (0.24) (4.43) -0.0340* -0.0340** -0.0335 (1.95) (2.31) (2.10)** -0.0176 -0.0241 -0.0167 (0.70) (1.13) (0.67) 30.05** 0.6267*** 21.10* (2.23) (3.92) (1.66) 0.0013** (2.20) Log of population Constant Time fixed effect Yes Observations 1061 1061 1061 Number of group(station) 74 74 74 Notes: The variable Biasit is the absolute difference between fraction of Bush’s speaking time and 1/2 in the news stories from station i on date t. The variable Competition measure over market is based on the state polls in September 2000, as defined in section 4.2. The variable Daily competition measure is based on the midnight prices of the vote share of two candidates in the previous day, as defined in section 4.3. Total speaking time is the total speaking time of Bush and Gore in the news stories from station i on date t. 5/30 stations is a dummy variable to indicate if the station made the 5/30 commitment according to White House’s suggestion of airing 5 minutes of candidates’ coverage in the 30 minutes evening news. Absolute value of t statistics in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%. 21 Table 3: The Variance of Media Bias Dependent variable Variance Variance Competition measure 0.205 0.174 0.530 0.453 (2.05)** (1.73)* (1.99)* (1.75)* Number of local TV stations 5/30 stations Log of population Constant Bias Bias -0.000 0.002 (0.15) (0.74) -0.011 -0.015 (1.50) (0.79) -0.006 -0.054 (0.58) (2.09)** 0.044 0.0953 0.077 0.484 (11.22)*** (1.40) (7.40)*** (2.76) Observations 74 74 74 74 Notes: Variance is the variance of the daily fraction of Bush’s speaking time given a station. The variable Bias is the media bias measure on the station basis, defined by the absolute difference of the average fraction of Bush’s speaking time and 1/2. 5/30 stations is a dummy variable to indicate if the station made the 5/30 commitment according to White House’s suggestion of airing 5 minutes of candidates’ coverage in the 30 minutes evening news. Absolute value of t statistics in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%. 22 Table 4: Alternative Media Bias Measure Dependent variable Variance Competition measure 0.0134 0.0140 0.3560** 0.3620* (0.36) (0.36) (2.10) (2.09) Number of local TV Variance bias bias 0.0003 -0.0021 (0.08) (1.41) 0.0006 0.0011 (0.22) (0.10) 0.0003 0.0222 (0.08) (1.27) stations 5/30 stations Log of population Constant 0.0080 0.0072 (5.42)*** (0.27) 0.9949 0.7138 (127.75)*** (6.08)*** Observations 74 74 74 74 Notes: The variable Variance is the variance of the daily fraction of the reporter’s speaking time given a station. The fraction of reporters’ speaking time is measured by 1-(fraction of bush’s speaking time)-fraction of gore’s speaking time. The variable Bias is the average of daily fraction of reporters’ speaking time for each station. 5/30 stations is a dummy variable to indicate if the station made the 5/30 commitment according to White House’s suggestion of airing 5 minutes of candidates’ coverage in the 30 minutes evening news. The variable Log of population is the log of population in the media market. Absolute value of t statistics in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%. 23 Table 5: Media Believability in 2002 and 2004 Dependent variable Competition measure Year2004 Believability Daily Local TV USA Today ABC News newspaper news -8.6956** -4.6866 -6.04 -2.0412 (2.32) (1.23) (1.43) (0.53) -0.1230** -0.0737 -0.063 -0.1301** (2.21) (1.33) (1.03) (2.29) Education categories √ √ √ √ Family income √ √ √ √ √ √ √ √ 1513 1547 1264 1472 categories City size categories Number of Observations Notes: Dependent Variable is the believability of the media ranked by the respondents. “4” means the respondent can believe all or most of what the organizations says. “1” means the respondent believes almost nothing of what they say. Education categories include 7 categories from “none, or grade 1-8” to “post-graduate training”. Family income categories include 8 categories from “less than $10000” to “$100,000 or more”. City size categories include 4 categories from “a large city” to “a rural area”. Absolute value of t statistics in parentheses. *significant at 10%; ** significant at 5%; *** significant at 1%. Chapter 2 Political Differentiation in Newspaper Markets 2.1 Introduction Economists are interested in the role of competition. In many markets, such as automobile or restaurant, competitive environments result in a variety of product characteristics and increase consumer welfare. However, in the media industry, there is less competition and economists disagree about whether media competition will result in differentiation in the political ideologies of media outlets. In the newspaper markets, most daily newspapers enjoy a monopoly, while some daily newspapers compete with others.1 The main question this paper tries to answer is as follows: If some markets have more than one newspaper, will newspapers choose the same popular ideology as each other or will they specialize in different ideologies? The news media plays an essential role in society since voters need information to make decisions. However, news outlets may have their own political ideology which therefore affects the way they select news information (Larcinese, Puglisi and Snyder, 2007). The market structure of the media industry may affect the way newspapers produce ideological contents and opinions. In the literature, theories of media bias have different predictions regarding the effect of media competition. A reason that media competition could induce media firms to select common and popular political ideologies could be the pressure to get 1 The fact that most daily newspapers enjoy a monopoly may be due to the high fixed cost of operating a newspaper; as a result, the number of newspapers is not a linear function of the market size (Berry and Waldfogel, 2006). It could also be due to the fact that advertisers prefer to buy ads in newspapers with more readers, therefore a newspaper with less readers cannot easily survive. 24 25 more readers in order to attract more advertising revenue. Another reason could be that media firms compete on their reputation of truthful and balanced news reporting. In this case, media competition will weaken the connection between the consumers’ beliefs and the ideology of their news sources. However, media firms engaged with price competition could also increase media bias to cater to consumers’ beliefs as a strategy of product differentiation. In this case, the relationship between consumers’ beliefs and their news sources will be stronger under media competition. This paper provides a simple model to predict the effect of newspaper competition. The model assumes that the goal of newspapers is to maximize their readership and that readers may choose not to read a local newspaper. The model predicts that in the case of a monopoly, newspapers will select the median ideology and readers at the extremes will be less likely to read the newspaper. In two-newspaper markets, newspapers will specialize in different ideologies and therefore increases the probability of reading a local newspaper among those with extreme ideologies. While theories in the literature have different predictions about the effect of competition on media ideologies and the information consumers will receive, the related empirical analysis is still limited. The main contribution of this paper is to provide empirical analysis about the effect of newspaper competition on their political ideologies as well as the news consumption of the general public. In this paper, I use the political demographic of newspaper readership in 2000 and 2004 to test the predictions from the simple model. Intuitively, if two newspapers exist in the market at the same time and one has more Republican readers while the other has more Democratic readers, then these two papers probably have different political ideologies. In terms of individual choice, we can test if people at the extremes are more likely to read a local newspaper when there are more than one newspaper in the market. The concern of the comparison of news consumption in different cities is that people in different cities may have different overall demand for news as well as a nonuniform demand among people of different ideologies. I therefore use the variation in newspaper competition from 2000 to 2004 to see how it affects individuals’ news consumption. The identification assumption for the market fixed-effect model is that the change in newspaper competition 26 should not be positively related with the change in the demand of news by people at the ideological extremes. As we will see, the change in newspaper competition from 2000 to 2004 negatively correlated with the growth in the internet but did not correlate with other factors such as the distribution of ideologies. The growth in internet supply is driven by supply side factors and therefore was not likely to relate to the change in the demand for news by people at the extremes. Another way to test if newspaper competition results in the divergence of newspaper ideologies is to estimate the effect of newspaper competition on the incentive to get political information online. Assuming that news sources with similar ideologies are substitutes, newspaper competition will reduce the incentive for people at extremes to get political news online if newspaper competition will make newspapers specialize in different ideologies. The empirical results show that people who identified themselves as politically moderate are more likely to read a local newspaper than those at the extremes. Moreover, newspaper competition encourages more readers with extreme ideologies to read newspapers. The results also show that newspaper competition reduces the incentive for people at the extremes to access political information online. The empirical evidence suggests that readers prefer a newspaper with an ideology that is closer to their own, and that newspaper competition will make newspapers specialize in different ideologies, satisfying the demand for news of people with opposing ideologies. 2.2 Related Literature Most theories of media bias try to provide a convincing story of where the media bias originates and how the competition between media firms affects media bias. Based on the assumption that consumers value the quality of news by its accuracy, media competition may force media outlets to deliver more accurate information (Besley and Prat, 2002; Anderson and McLaren, 2005). In contrast, media firms may also increase media bias to cater to consumers’ beliefs as a strategy of product differentiation (Mullainathan and Shleifer, 2005). Moreover, sources of revenue and consumers’ preference toward advertisements will affect the ideologies of newspapers under competition pressure (Gabszewicz, Laussel and Sonnac, 2002; Anderson and Gabszewica, 2005). 27 Gentzkow and Shapiro (2006) presented a theory of media bias in which consumers appreciate unbiased news reporting and will choose a media firm based upon the firms’ reputation. Consumers’ judgments about a news outlet depend on their prior beliefs and the feedback they receive after the news report. Media competition will increase the feedback they get and therefore help them to identify which news outlet is unbiased. Therefore, media competition will weaken the connection between consumers’ beliefs and news reports. When beliefs of consumers are heterogeneous, the effect of competition is ambiguous. Both segmented equilibrium and convergence equilibrium can exist. The segmented equilibrium will not exist if the increasing feedback consumers receive caused by competition is high enough. They provide empirical evidence from the local TV market and find that the sound bites of presidential candidates are more balanced in the markets where there are more local TV news programs, suggesting that media competition reduces media bias. Mullainathan and Shleifer (2005) presented a theory on how media competition affects media bias. They assumed that readers hold beliefs which they like to see confirmed and newspapers can slant stories toward these beliefs. Analogous to the standard Hotelling model, when deciding the optimal slant, the price effect dominates the market share effect until firms are very far apart.2 Therefore, duopolist newspapers will differentiate themselves by reporting extreme news so that they can charge higher prices. In this case, if readers do not have access to all news sources, competition will make the relation between readers’ beliefs and the ideologies of their news sources stronger. Considering how sources of revenue affect the ideologies of newspapers under competition pressure, Gabszewicz, Laussel and Sonnac (2002) constructed a model based on a hotelling model with price competition. They show that, as long as the potential advertisement revenue is high enough, advertising may induce newspapers to compete for a maximal audience, and therefore force newspapers to moderate their political messages to the readers.3 2 In the standard Hotelling model with price competition, two firms will perform maximum differentiation in order to have larger monopoly power over consumers (Tirole, 1998). In general, the factors of product differentiation are: the costs of disutility, the demand elasticity, the number of firms, and the density of consumers (Brenner, 2001). 3 The examples provided in their paper are the comparison of newspapers in the US and newspapers in Europe. 28 This paper provides a model to predict the effect of competition on the ideologies of newspapers and consumers’ choice of newspaper. The model shares the assumption with Mullainathan and Shleifer (2005) that readers prefer the newspaper whose slant is consistent with their prior beliefs. The model also uses the idea of Gabszewicz, Laussel and Sonnac (2002) by assuming that newspapers are not engaged in price competition and thus the way for newspapers to maximize profit is by maximizing their readership. However, consumers may choose not read a local newspaper if the utility from outside options is high enough. The model predicts that in a one-newspaper market, outside options for the readers will force the newspaper to cater to the majority. In a two-newspaper market, outside options will force newspapers to specialize in different ideologies. In terms of consumer behavior, people with extreme ideologies will be more likely to buy a local newspaper when there is two newspapers in the market. Those different studies have different points of view on the role of media competition; however, the empirical evidence is relatively scarce. One of the reasons for this lack is probably due to the difficulty in measuring the ideologies of media outlets in an adequate sample size and for more than one year. Groseclose and Milyo (2005) provided a way to measure media bias in the news stories of several major media outlets in the United States. They get a measure by counting the citations of think tanks in the media and then comparing the citations of think tanks by senators. They found that most major media outlets are biased to the left. Gentzkow and Shapiro (2006) constructed a new index of media slant by comparing the language in newspapers and the language used by Republican or Democratic Congressmen. In their study, they were able to apply the method to a larger sample size; however, for most newspapers, they only constructed the measure for newspapers for one year.4 The measure they created correlated with the ratings of political orientation submitted by users to the media directory website Mondo Times.5 They found that the index of media slant can largely explained by consumer preference and can not be explained by ownership preference. They also found that readers respond to the ideology of newspapers since newspapers would 4 For several newspapers with ownership changes, they measure the media slant for more than one year. Every user has the right to rate newspapers, however, a lot of newspapers remained unrated and most newspapers are rated by few people. 5 29 lose readership by deviating from the optimal ideology. Since there is evidence that people are aware of the ideology of newspapers, this study uses readers’ choices of newspapers to measure the ideology and to investigate the effect of competition on it. This measure has two advantages. First, the method is straightforward and not difficult to apply. Second, the measure is meaningful, as whether the ideologies of newspapers are different or not should be decided by the readers. 2.3 A Model of Newspaper Competition The purpose of the model is to investigate the effect of newspaper competition under realistic assumptions and derive testable prediction on newspaper ideologies as well as consumers’ newspaper consumption. There are three key assumptions in the model. First, readers prefer to read a newspaper with an ideology that is closer to their own. Second, newspapers are not engaged in price competition. Therefore, newspapers simply choose the ideology to maximize the market share.6 Third, there exists outside options other than newspapers for consumers. Therefore consumers can choose not to read a newspaper. 2.3.1 Model Setup In the model, there are one or two newspapers and the population size of consumers is normalized to 1. The unit cost per copy is c and newspapers are sold at fixed price p. The goal of newspapers is to maximize profit. Consumer i has ideology ci , which is symmetric distributed with mean 0 and cumulative density function F(.). The utility of reading a newspaper j with ideology nj to consumer i is given by Uij = a − t|ci − ej | − p. (2.1) Consumer i will read newspaper j if and only if Uij = max{Uij , Ui0 }, (2.2) where Ui0 is the utility from outside options if consumer i is not reading a local newspaper. 6 Consider the fact that more than 60 percent of the revenue of newspapers come from advertisement, it is reasonable to assume that newspapers are trying to maximize the market share, rather than the revenue from selling newspapers. One piece of evidence to support this assumption is that the New York Times canceled the fee for the editorial section in 2007 in order to maximize the market share. 30 2.3.2 Case I: Monopoly Newspaper We first consider the case when there is only one newspaper in the market. The demand for the newspaper is: N where l = = P rob(a − t|ci − e| > Ui0 ) (2.3) = F (e + l) − F (e − l), (2.4) a−Ui0 t . Since the goal of the newspaper is to maximize its market share, the optimal ideology e∗ will satisfy the following first order condition: f (e∗ + l) = f (e∗ − l). (2.5) Therefore, the optimal choice of ideology, e∗ is 0 when the ideological distribution of consumers is symmetric with mean 0. If the distribution is skewed to the right, then the optimal choice will be greater than the mean. 2.3.3 Case II: Two Newspapers Now we assume that there are two newspapers, 1 and 2, with ideologies n1 and n2 in the market. Without loss of generality, we assume that n1 < n2 . Let x be the ideology of the consumer who is indifferent to reading newspaper 1 and not reading a newspaper, x be the ideology of the consumer who is indifferent to reading newspaper 2 and outside options, and x∗ be the ideology of the consumer who is indifferent to reading newspaper 1 and newspaper 2. The demand for newspaper 1 and newspaper 2, respectively, can be represented as N1 = F (x∗ ) − F (x) (2.6) N2 = F (x) − F (x∗ ). (2.7) If the utility from outside options is extremely low, then almost every consumer will read a local newspaper. Thus we will get a result similar to the median voter theorem. As long as the utility from outside options is high enough, newspapers will face the tradeoff between losing some of the median consumers or losing some of the extreme consumers when deciding the ideology. Thus we have the following proposition. The detailed proof is 31 given in the appendix. Proposition. Let l be a− Ui0 t . When there are two newspapers in the market, there are two equilibriums depending on Ui0 . (i) If Ui0 < a − t ∗ f −1 (0.5 ∗ f (0)), both newspapers will choose the median ideology of potential consumers. (ii) If Ui0 > a − t ∗ f −1 (0.5 ∗ f (0)), newspapers will choose ideologies (e1 , e2 ) = (l − f −1 (0.5 ∗ f (0))), f −1 (0.5 ∗ f (0)) − l). The proposition says that whether newspapers will have different ideologies depends on the parameters a, t, Ui0 , and the distribution of consumers’ ideologies. When two newspapers have different ideologies, the distance between the ideologies of two newspapers will be larger if the consumers are more sensitive to the ideologies of newspapers, the utility from outside options is higher or the distribution of consumers’ ideologies is flatter. 2.3.4 Predictions The model has the assumptions that the goal of newspapers is to maximize their readership and consumers may choose not to read a local newspaper.7 Based on the assumptions, the model predicts that, in the case of a monopoly, newspapers will select the median ideology. As a result, readers whose ideologies are away from the median ideology are less likely to read a newspaper. The second testable predictions from the model is that newspapers will specialize in different ideologies when facing competitors. In terms of consumers’ behavior, consumers with extreme ideologies are more likely to read a local newspaper when there is newspaper competition. These predictions will be tested by the data illustrated in the next section. 2.4 Data and Context This section illustrates the data sources, definitions of key variables, and a test of the second prediction by using aggregate readership data. 7 The model does not consider the possible influence of owner preferences. Since newspapers in a monopoly market have a larger monopoly power, the influence of owner preference on newspapers’ ideologies could be larger in single-newspaper markets. 32 2.4.1 Newspaper Competition In order to investigate the effect of newspaper competition on the ideologies of newspapers, it is necessary to define the market area and measure newspaper competition. In the U.S., most daily newspapers are city newspapers, and a few newspapers are county specific. Since the circulation area of a daily newspaper is usually larger than a city, the newspaper market is defined as either MSA (Metropolitan Statistic Area) or county. Newspaper competition is measured by the number of newspapers in the city. The Editor and Publisher International Year Book listed multiple-newspaper cities in the U.S. and the newspapers in those multiple-newspaper cities. In 2000, there were 59 cities that had more than one newspaper. From 2000 to 2004, ten cities became single-newspaper cities and two cities became multiple-newspaper cities. For the market that did not have any multiplenewspaper cities, I assume the number of newspapers in the market is one. However, some counties could have no newspaper in the U.S. Therefore I dropped individuals who lived in a market with a population of less than 15,000 to reduce possible measurement error. Since one MSA area could have several cities with one or zero newspapers, the newspaper competition measure does not account for all available newspapers in the market. The variation of newspaper competition relies on whether there is any city with more than one newspaper in the market area. In some multiple-newspaper cities, newspapers have the same owners or have an agreement to cooperate in printing and distribution. However, since those newspapers still have different editorial staffs, those newspapers are treated the same as other newspapers with competition. Business newspapers and foreign language newspapers are excluded when counting the number of newspapers in the market. 2.4.2 Newspaper Reading and Political Demographics The data about individuals’ news consumption and political demographic derives from the 2000 and 2004 National Annenberg Election Surveys (NAES) conducted by the Annenberg Public Policy Center of the University of Pennsylvania. The institution did a rolling crosssection survey from the end of 1999 to the end of 2000, and performed a similar rolling crosssection survey from the end of 2003 to the end of 2004. The survey data included information 33 on people’s media consumption, ideology, political affiliation, and voting behavior in the current year and in the previous election year. Local Newspaper Reading The survey asked respondents which newspaper they had read most in the past week. If the newspaper they had read most was a national newspaper or an international newspaper, the individual is defined as not reading a local newspaper. However, there are several major daily newspapers circulated outside their MSA area. For example, the Boston Globe is also circulated in the Providence MSA area. Since the newspaper competition measure does not consider all the newspapers available in the market, the readers of the major daily newspapers are not defined as reading a local newspaper if they are living outside of the MSA where the newspaper is published. Political Ideology of the Respondents When the distribution of consumers’ ideologies is not symmetric, the model predicts that when there is only one newspaper in the market, the newspaper will choose the ideology that is between median and mean. In terms of consumers’ choice, people whose ideologies are far from the ideology of the newspaper will be less likely to read a local newspaper. The political ideology of the respondent is self-reported by categories as very liberal, liberal, moderate, conservative, or very conservative. I recoded the ideologies from -2 to 2; very liberal is coded as -2 and very conservative is coded as 2. In every market, we can obtain the median ideology and construct an index to measure the relative extremity of the respondents’ politics. Specifically, the relative extreme index of individual i in market m is defined as: ExtremeM edian = |ideoim − medianm |, (2.8) where medianm is the median ideology of the market m, and ideoim is the ideology of individual i in the market m. Similarly, we can also construct an extreme index relative to the mean ideology of the market. 34 ExtremeM ean = |ideoim − meanm |. 2.4.3 (2.9) High-Speed Internet Supply The quality and the price of the internet may directly affect the utility of accessing information from the internet and therefore provide consumers with more news sources. As a result, the growth in the high-speed internet supply may directly affect the readership of newspapers. In this paper, the number of high-speed internet providers is used as a proxy for the speed and quality of internet connection. The Federal Communications Commission (FCC) provides information on the number of high-speed internet providers by zip code.8 The mean of the number of high-speed internet supplier in the sample is 3.57 in 2000 and 8.44 in 2004. 2.4.4 A Test Using Aggregate Readership Data Graph 1 presents the ideological distribution of residents in a one-newspaper market, Hartford, and the ideological distribution of readers of the Hartford Courant. Comparing theses two distributions, the distribution of the readers of the Hartford Courant is more concentrated in the middle than are the residents of the Hartford market. This comparison indicates that readers at the extremes are less likely to read a local newspaper. Graph 2 presents the ideological distribution of readers of the Boston Globe and the Boston Herald in this two-newspaper market. Those two distributions suggests that the Boston Herald has relatively more conservative readers. This suggests that Boston Globe and the Boston Herald might have different ideologies. For each newspaper, we can calculate the mean ideology of its readers as the ideology measure of it. Furthermore, we can use this information to derive an F-statistic to test if newspapers in a two-newspaper market, relative to one-newspaper market, are targeting different groups of readers. The detail of the derivation is provided in Appendix 2. The null hypothesis is that all newspapers target the middle of the ideological distribution in the 8 The data is available from 1999 for every half year. 35 market. The alternative hypothesis is that newspapers in a two-newspaper market target different groups of readers. The F-statistic, 1.9, is greater than the critical value at 5 percent significance, 1.42, and therefore rejects the null hypothesis that all newspapers target the majority of the readers in the market. This provides evidence to support the prediction that newspapers will specialize in different ideologies in two-newspaper markets. The next section provides empirical analysis based on individuals’ choice of newspaper. 2.5 Empirical Analysis This section presents the empirical analysis to test the predictions from the theoretical model, examining the relationship between the structure of media market and the consumption of news. The model predicts that in a market with only one newspaper, the newspaper chooses the ideology near the median resident and people whose ideologies differ from the median resident are less likely to read a local newspaper. However, when there is more than one newspaper in the market, newspapers will specialize in different ideologies, and the probability of reading a local newspaper for those who are at the extremes will increase. The empirical analysis has several parts. The first part restricts the sample in respondents in one-newspaper markets and tests whether respondents with extreme ideologies are more likely to read a local newspaper. The second part investigates whether newspaper competition increases the probability of reading a local newspaper. To exclude alternative interpretations, market fixed-effect specifications, instrument variable method, and falsification test for consumption of non-political news are conducted. The third part tests whether newspaper competition reduces the incentive to get political information online. 2.5.1 The Case of a One-Newspaper Market The first prediction of the model on newspaper consumption is that people at the extremes are less likely to read a local newspaper in a single-newspaper market. To test this prediction, the simplest specification is a probit or linear probability specification. The dependent variable is a dummy variable that indicates if the respondent read a newspaper or not. Independent variables are variables that represent the ideologies of respondents and other 36 control variables that may affect the probability of reading a local newspaper, including the possible impact from the internet supply. Table 2 presents the results when the sample is restricted to the respondents in the market with only one newspaper. Column (1) shows that people who are very conservative or very liberal are less likely to read a local newspaper, relative to people who are moderate. Column (2) uses an extreme index instead of ideology dummy variables.9 In column (3) and (4), after controlling for respondents’ ideology, results show that individuals whose ideology varies from the median or mean ideology in the market are less likely to read a local newspaper. On average, people who are at the extremes are 4.1 percent less likely to read a local newspaper. These results could mean that very liberal people in a relatively conservative town are less likely to read a local newspaper because the ideology of the newspaper is more distant from their own. It could also mean that, in general, those at the extremes in the market have less of a demand for local news. However, if people at either end of the political spectrum simply have less of a demand for local news, newspaper competition should not have larger effects on the probability of reading a local newspaper for people at the extremes in the market. Therefore, analyzing the effect of competition will help us to distinguish between those two possible reasons. 2.5.2 The Effect of Newspaper Competition Table 3 column (1) and column (2) presents the result when the sample is restricted to the respondents in multiple-newspaper markets. While consumers who are more liberal are more likely to read a local newspaper, the difference between moderate readers and readers with extreme ideologies are not as clear as the results for single-newspaper markets. This means that in competitive markets, readers with extreme ideologies are more likely to read a newspaper, relative to the monopoly case. Results from column (3) and column (4) of Table 3 further confirm this pattern. Using the relative extreme index, Table 4 estimates the effect of newspaper competition on the probability of reading a local newspaper. Column (1) of Table 4 presents the results 9 The extreme index is 2 if very conservative and very liberal, 1 if conservative or liberal, and 0 if moderate. 37 from the pooled cross-section regression. The results show that newspaper competition has a larger effect on newspaper reading for people who are at the extremes. These results support the hypothesis that readers prefer a newspaper with an ideology closer to their own and that newspaper competition will make newspapers specialize in different ideologies. In terms of magnitude, on average, 6 newspapers will diminish the gap of local newspaper readership between people with different ideologies. 2.5.3 Alternative Explanation – Heterogenous News Demand The number of newspapers may also be related to an unobservable news demand. Specifically, the number of newspapers may be positively correlated to a demand for news among people at the extremes. In this case, we will observe a spurious correlation between newspaper competition and newspaper readership among those at the ideological extremes. A. Market Fixed-Effect Specification One way to account for the concern is to rely on the change in competition over time by including market fixed effects. In the market fixed-effect specification, the variation of newspaper competition comes from the change from 2000 to 2004. Column (3) and column (4) of Table 4 present the results with market fixed effects. Compared with the results from the pooled cross-section results in column (2), the differential effects of newspaper competition are similar, suggesting that potential bias caused by heterogenous news demand in different cities is not serious. The identification assumption for the market fixed-effect specification to be unbiased is that the change in newspaper competition should not be related to a change in demand for news among people at the extremes. Table 5 column (2) tries to capture the factors that caused the change in newspaper competition from 2000 to 2004. The result finds that the change in newspaper competition from 2000 to 2004 is affected by the growth of high-speed internet providers in the market. As Greenstein and Prince (2006) demonstrated, in the earliest years of broadband internet access (prior to 2003), it was very much supply-driven in the sense that supply-side issues were the main determinants of broadband internet availability and hence adoption. 38 Specifically, highly populated areas were more profitable due to economies of scale and lower last mile expenses.10 While the early stage of high-speed internet focused on metro areas and areas with higher income, the diffusion process makes it such that the growth rate is unlikely to be positively correlated with a change in news demand. The Internet may directly affect an individual’s incentive to read a local newspaper. If the Internet decreases the newspaper competition as well as reduces more readership of those at the extremes, relative to those in the middle, the effect of newspaper competition will be confounded with the direct effect of the Internet. However, the possible impact can be controlled by using the number of high-speed internet providers at the zip-code level. Moreover, the negative effect of the Internet is smaller for people at the extremes. Therefore, while the growth in internet supply decreased newspaper competition, it did not reduce more readership of people at the extremes, relative to people whit more ideologies. This suggests that the direct effect of the Internet will not confound the results. B. Instrument Variable Another way to address the concern is to use the number of professional sport team in the market as an instrument variable. The number of professional sport teams in a market is related to a demand for sport news, and therefore related to the number of newspapers in the market. However, the number of professional sport teams should not relate to the relative demand for news between those at extremes and those with moderate ideologies since the demand for sport news should not relate to individuals’ ideologies. Therefore, using the number of professional sport teams in the market can help us identify the differential effect of newspaper competition for people with different ideologies.11 First stage results based on the instruments appear in Table 6. Second stage results appear in Table 7. Compared with the results from pooled cross-section specifications and market fixed-effect specifications, the coefficient of the interaction term is larger. However, similar to the previous results, 6.2 newspapers will diminish the gap of local newspaper readership between people with different ideologies. 10 Greenstein and Prince(2006) found that the diffusion of dial-up spread quickly since it relied on the existing phone line infrastructure. 11 The number of professional sport teams includes teams in four major leagues: National Football League, Major Baseball League, National Basketball Association and National Hockey League. 39 Since the number of sport teams could relate to the demand for non-political news and therefore relate to the demand for newspaper, distinguishing demand for political news and non-political news will help us understand the change in newspaper demand caused by competition. C. Political News vs. Non-Political News In 2004, in addition to asking respondents which newspaper they read the most in the past week, the ANES survey also asks respondents if they get political information about the presidential campaign from the newspaper they read. The exact survey question is the following: “During the past week, how much attention do you pay to newspaper articles about campaign for president - a great deal of attention, some, not too much, or no attention at all.” Therefore we can split local newspaper readers into political information readers and non-political information readers by their answers. Local newspaper readers who answered “a great deal of attention” and “some” are coded as political readers and those who answered “not too much” or “no attention” are coded as non-political readers. Table 7 column (2) and column (3) shows that the effect of newspaper competition has a larger effect on the probability of being a political newspaper reader among those at the extremes. However, newspaper competition does not have a differential effect on the probability of being a non-political newspaper reader. The results show that people at the extremes increase their demand for political news as there are more newspapers in the market, but their demand for non-political news does not have the same significant change. This suggested that the differential effect of newspaper competition we observed from the OLS, market fixed-effect, and IV specifications are most likely caused by the change in political news. If newspaper competition makes newspapers differentiate in other aspects, this comparison can help us rule out the case that people at the extremes are more sensitive to changes other than political ideologies, such as the amount of sport news or newspaper quality. 40 2.5.4 Reverse Causality So far in this paper, the ideologies of individuals are seen as not affected by the newspaper they read. This section provides discussion about the related literature and how this would change the interpretation of empirical results in this paper. There is some evidence that shows that the media may affect voting behavior (DellaVigna and Kaplan, 2007 ; Gerber, Karlan, and Bergan, 2006); however, there is limited evidence that media will affect the political ideologies of individuals. Bernhardt, Krasa and Polborn (2006) argue that the increase in partisan behavior we observe is not due to a fundamental change of voters’ political preference. Rather, they argue it is due to the different information voters received since the ideological distribution is stable in the recent years, but political opinions become more diverse. Moreover, they argued that in 2004, Bush and Kerry supporters held vastly different beliefs about facts influenced by the media, relative to their differences in core beliefs(eg. abortion), which are less influenced by the media. If newspaper will affect readers’ ideologies, the OLS estimates in the last section could be caused by reverse causality. The OLS results still show that, in the equilibrium, newspapers under competition have different ideologies. However, we can not make an inference about whether readers prefer a newspaper consistent with their prior beliefs. The variation of competition in the fixed effect specification is from the change in competition from 2000 to 2004. Since the ideologies are less likely to be affected by newspapers significantly in four years, the results from the fixed-effect specification is less sensitive to the possible effect of newspapers on ideologies. Since the estimates from the OLS and fixedeffect specifications are similar to each other, the potential problem of reverse causality is not serious. 2.5.5 Effect of Newspaper Competition on Getting Information Online If news sources with similar ideologies are substitutes and newspaper competition will make newspapers specialize in different ideologies, then newspaper competition may affect the incentive for people at the extremes to get political information online. Table 8 presents the results regarding the effect of newspaper competition on political news consumption from the internet. The dependent variable is the number of days people accessed political 41 information in the past week.12 The result shows that people who are at the extremes access political information from the internet less frequently in the presence of newspaper competition. This result also provides evidence to support the hypothesis that newspapers under competition have different ideologies. Regarding other factors such as as income, education, or age, people with higher income or higher education are more likely to read a local newspaper and get political information on the internet. Older people are more likely to read a local newspaper; in addition, they are less likely to get information from the internet. 2.5.6 Influence of the Internet The growth of the high-speed Internet supply may increase the quality or the speed of the Internet, which increases the utility for people to get information online in general. There are two reasons why the quality of in Internet connection could have a positive effect on local newspaper readership. First, since now a lot of newspapers have online versions, people with internet access have a lower cost to read a local newspaper. Second, the news information available on the internet and local newspaper may compliment each other for the readers. However, it also may have a negative effect on local newspaper readership since the internet make more choices available. The results in Table 4 find that the internet has a predominantly negative effect on reading a local newspaper. The results also show that the negative effect is stronger for people with moderate ideologies. There are two possible explanations. First, the ideological distribution of news outlets available from the internet may be similar to the ideological distribution of the general public; therefore, people who are moderate are more likely to find something that they like on the internet. Second, information on the internet could be a poor substitute for a local newspaper for readers who are at the extremes. While the internet provides consumers with more choices, the internet may provide consumers with a wider range of news sources. According to Gentzkow and Shapiro (2006), when there are more news sources available, consumers will get more feedback regarding 12 The exact survey question is: ”How many days in the past week did you access information about the campaign for president online?” Some respondents are asked in different wording: ”How many days in the past week did you read information about the campaign for president online?” 42 news reports. As a result, consumers will put less weight on their prior beliefs when choosing their news sources. Since the internet provides more news sources for consumers, we can test the hypothesis by looking at changes in news consumption caused by the growth in the high-speed internet supply. Table 9 shows how the internet changes the readership of The New York Times and The Wall Street Journal. Column (1) of Table 9 shows that the readers of The New York Times are more liberal and the readers of The Wall Street Journal are more conservative. The growth in the internet supply makes the liberal more likely to read The New York Times and makes the conservative more likely to read The Wall Street Journal. The results do not support the theory that more news sources will weaken the connection between consumers’ prior beliefs and news reports. In contrast, the internet helps consumers obtain the information that is more consistent with their own ideologies. 2.6 Conclusion The news media provide people with information to make different decisions in daily life, from choices of investments to the way they vote in elections. News information may have an impact on politics since it may change collective decisions by affecting individuals’ decisions. If media competition makes media outlets specialize in different ideologies, it would change the incentive for voters to get information and the type of information voters receive. This paper analyzes how media competition affects the political ideologies of media outlets as well as the news consumption of the general public. Using the NAES in 2000 and 2004, the empirical results show that the moderate are more likely to read a local newspaper than are people at the ideological extremes. However, newspaper competition has a larger positive effect on newspaper reading for people at the extremes. The results also show that newspaper competition reduces the incentive for people at the extremes to access political information online. These results suggest that readers prefer a newspaper with an ideology that is closer to their own, and that newspaper competition will make newspapers specialize in different ideologies, satisfying the news demand for people with opposing ideologies. This paper also finds that the Internet makes the readership of the New York Times more liberal and the readership of the Wall Street Journal more conservative. This suggests that with more choices available, readers do not have the tendency to obtain information 43 from news outlets that is not consistent with their own ideologies. 44 References Anderson, Simon P. and Jean J. Gabszewicz, (2005) “The media and Advertising: a Tale of Two-sided Markets,” CORE Discussion Paper 2005/88. Baron, David (2006) “Persistent Media Bias,” Journal of Public Economics 90(1-2): 1-36. Bernhardt, Dan, Stefan Krasa, and Mattias Polborn (2006), “Political Polarization and the Electoral Effects of Media Bias,” CESifo Working Paper Series No. 1798. Berry, Steven and Joel Waldfogel, (2006) “Product Quality and Market Size,” NBER Working paper. Besley, Timothy and Andrea Prat, (2006) “Handcuffs for the Grabbing Hand? Media Capture and Government Accountability,” American Economic Review, 96(3):720736 Brenner, Steffen, “Differentiated Customers and Firms,” Working paper. DellaVigna, Stefano. and Ethan Kaplan (2007) “The Fox News Effect: Media Bias and Voting,” Quarterly Journal of Economics, 122 :1187-1234. Gabszewicz, Jean J., Didier Laussel and Nathalie Sonnac (2005), “Press Industry and the political differentiation of newspapers,” Journal of Public Economic Theory 4(3): 317-334. Gentzkow, Matthew and Jess M. Shapiro (2006), “Media Bias and Reputation,” Journal of Political Economy, 114(2):280-316. Gentzkow, Matthew and Jess M. Shapiro (2006), “What Drives Media Slant? Evidence from U.S. Daily Newspapers.” Working paper. Gerber, A. and Karlan, D. and Bergan, D., “Does The Media Matter? A Field Experiment Measuring the Effect of Newspapers on Voting Behavior and Political Opinions,” Yale Economic Applications and Policy Discussion Paper No. 12. 45 Groseclose, T. and J. Milyo (2005), “A Measure of Media Bias,” Quarterly Journal of Economics, 120(4): 1191-1237. Greenstein, Shane and Jeff Prince (2006) “The Diffusion of the Internet and the Geography of the digital deivide in the United States,” NBER Working Paper. Larcinese, Valentino and Riccardo Puglisi and James M. Snyder (2007), “Partisan Bias in Economic News: Evidence on the Agenda-Setting Behavior of U.S. Newspapers,” NBER Working Paper: No. 13378. Mullainathan, Sendhil and Andrei Shleifer (2005), “Market for News,” American Economic Review 95(4): 1031-1053. Prieger. James E. (2003), “The Supply Side of the Digital Divide: Is There Equal Availability in the Broadband Internet Access Market?” Economic Inquiry 41 (2): 346363. 46 Appendix 1 The appendix gives the proof for proposition 1. If Ui0 < a − t ∗ f − 1(0.5 ∗ f (0)), the 2 market share for newspaper 1 will be F ( n1 +n 2 ) and the market share for newspaper 2 will 2 be 1-F ( n1 +n 2 ). The Nash equilibrium in this case will be (0,0). Let l be a− Ui0 t . If Ui0 > a − t ∗ f − 1(0.5 ∗ f (0)), the market share for newspaper 1 2 will be F ( n1 +n 2 ) − F (n1 − l). we will have the best response function for newspaper 1 by differentiating the market share of newspaper 1 with respect to n1 , f( n1 + n2 1 ) ∗ − f (n1 − l) = 0. 2 2 (2.10) Similarly, the best response function for newspaper 2 will be: f (l + n2 ) − f ( 1 n1 + n2 )∗ =0 2 2 (2.11) Therefore, we will have (n1 , n2 ) = (l − f −1 (0.5 ∗ f (0))), f −1 (0.5 ∗ f (0)) − l) to be the equilibrium. 47 Appendix 2 From the ANES survey, we can calculate the mean ideology of the readers for each newspaper. We are interested in testing whether ideologies of newspapers in a two-newspaper city target different groups of readers, as the model in this paper predicts. The null hypothesis is that newspapers in two-newspaper markets target the middle of the ideological distribution in their markets. The alternative hypothesis is that newspapers in two-newspaper markets target different groups of readers. Assume that every newspaper has the same ability to target the majority of readers in its market. Under the null hypothesis, ideologies of newspaper i in market m can be represented as follows: eim = µm + εim , (2.12) where µm is the mean ideology of the market, and εim is i.i.d. normally distributed with mean 0 and variance σ 2 . Suppose that there are N1 newspapers in a one-newspaper market and N2 newspapers in a two-newspaper market. For any newspaper in a one-newspaper market, we will have Σ (eim − µm )2 ∼ χ2N1 . σ2 (2.13) Similarly, for newspaper i and j in a two-newspaper market, Σ (eim − ejm )2 ∼ χ2N2 . 2 ∗ σ2 (2.14) Therefore, we can derive a statistic that is distributed as F-distribution with degree of freedom N2 , and N1 (eim −ejm )2 /N2 2σ 2 2 m) Σ (eim σ−µ /N1 2 Σ ∼ FN 1,N 2 . (2.15) 48 Graph 1: One-Newspaper Market – Hartford, CT Graph 2: Two-Newspaper Market – Boston, MA .5 Readers of Boston Globe Readers of Boston Herald .4512 .4 .3 .3 .4 .4384 .2857 .2 .2 Density Density .2871 .1724 .1 .1 .1484 .085 .0591 .0443 0 0 .0283 -3 -2 -1 0 1 2 -2: very liberal; 2: very cons ervative -3 -2 -1 0 1 2 -2: very liberal; 2: very conservative 49 Table 2.1: Summary Statistics Sample: Number of Newspapers Read a local Newspaper (1 or 0) Wall Street Journal( 1 or 0) USA Today(1 or 0) Ideology Relative extreme index Absolute extreme index Internet supply at market level Internet supply at zipcode level College or more High school Logincome Age Frequency of getting information online Log (population) Sample: All Sample Obs 112111 112111 112111 112111 112111 112111 112111 112111 112111 112111 112111 112111 112111 Mean 1.517 0.712 0.013 0.022 0.152 0.720 0.727 6.849 6.422 0.671 0.259 10.704 46.299 Std. Dev. 1.093 0.453 0.113 0.146 0.973 0.669 0.664 2.695 4.024 0.470 0.438 0.773 16.138 Min 1.0 0.0 0.0 0.0 -2.0 0.0 0.0 0.0 0.0 0.0 0.0 9.0 18.0 Max 5.0 1.0 1.0 1.0 2.0 3.0 2.0 11.7 19.0 1.0 1.0 12.0 97.0 96898 1.129 2.147 0.0 7.0 112111 13.686 1.769 9.6 16.7 One Newspaper Market Market Multiple Newspaper (Obs: 31465) (Obs: 80646) Number of Newspapers Read a local Newspaper (1 or 0) Wall Street Journal( 1 or 0) USA Today(1 or 0) Ideology Relative extreme index Absolute extreme index Internet supply at market level Internet supply at zipcode level College or more High school Logincome Age Frequency of getting information online log (population) Mean Std. Dev 1.000 0.000 0.727 0.445 0.011 0.102 0.024 0.152 0.204 0.969 0.725 0.672 0.733 0.666 6.003 2.417 5.741 3.805 0.652 0.476 0.275 0.446 10.653 0.762 46.681 16.260 Obs 31465 31465 31465 31465 31465 31465 31465 31465 31465 31465 31465 31465 31465 Std. Mean Dev 2.841 1.348 0.674 0.469 0.019 0.136 0.017 0.127 0.018 0.969 0.709 0.660 0.709 0.660 9.016 2.091 8.168 4.042 0.720 0.449 0.217 0.412 10.833 0.785 45.320 15.779 1.071 2.098 27395 1.278 2.258 13.025 1.505 31465 15.380 1.164 50 Table 2.2: One-Newspaper Markets (1) (2) (3) (4) Probit Probit Linear Prob. Linear Prob. Absolute Extreme Index -0.0206*** (0.0024) Extreme Index (Median) -0.0141** (0.0063) Extreme Index (Mean) Very Conservative Conservative Liberal Very liberal -0.0574*** (0.0065) -0.0152*** (0.0038) -0.0081 (0.0051) -0.0338*** (0.0086) Internet Supply Internet Supply*Extreme (Median) -0.0412*** (0.0121) -0.0090 (0.0060) 0.0019 (0.0076) -0.0139 (0.0146) -0.0020** (0.0008) 0.0010* -0.0278*** (0.0105) -0.0259 (0.0163) -0.0052 (0.0064) 0.0129 (0.0110) 0.0082 (0.0206) -0.0023** (0.0009) (0.0006) Internet Supply*Extreme (Mean) Log pop 0.0013* (0.0007) -0.0141*** (0.0024) -0.0237*** (0.0033) -0.0140*** (0.0024) Year2004 -0.0240*** -0.0185*** -0.0185*** (0.0033) (0.0044) (0.0044) Constant 0.0492* 0.0553* (0.0294) (0.0295) Market Fixed Effects No No Yes Yes Observations 80646 80646 80646 80646 R-squared . . 0.0518 0.0518 Note: Marginal effects are reported in the Probit Specifications. Absolute Extreme Index is 2 if the ideology of the respondent is very conservative or very liberal, 1 if liberal or conservative, and 0 if moderate. Standard errors are adjusted for market-level clustering and appear in parentheses. A* indicates significance at 10%; ** significance at 5%; *** significance at 1%. 51 Table 2.3: Markets in Competition Multiple-Newspaper Markets (1) (2) Probit Probit Number of Papers Very Conservative Conservative Liberal Very Liberal -0.0452*** (0.0160) -0.0197** (0.0080) -0.0123 (0.0079) -0.0219 (0.0152) Number of Papers* Very Con. Number of Papers* Conservative Number of Papers* Liberal Number of Papers* Very Liberal Absolute Extreme Index Number of Papers* Absolute Extreme Index -0.0162*** (0.0052) All Respondents (3) Probit 0.0066 (0.0056) -0.0656*** (0.0090) -0.0173*** (0.0061) -0.0137* (0.0071) -0.0527*** (0.0107) 0.0084*** (0.0030) 0.0013 (0.0031) 0.0021 (0.0024) 0.0128*** (0.0039) (4) Probit 0.0054 (0.0055) -0.0262*** (0.0031) 0.0048*** (0.0011) Log Income 0.0085 0.0083 0.0267*** 0.0266*** (0.0170) (0.0169) (0.0059) (0.0058) High School 0.1322*** 0.1324*** 0.1248*** 0.1249*** (0.0136) (0.0136) (0.0057) (0.0057) College 0.1309*** 0.1321*** 0.1347*** 0.1354*** (0.0167) (0.0168) (0.0072) (0.0072) Log pop -0.0045 -0.0043 -0.0198*** -0.0197*** (0.0111) (0.0112) (0.0056) (0.0056) Male 0.0155*** 0.0150*** 0.0018 0.0015 (0.0053) (0.0052) (0.0032) (0.0032) Age 0.0032*** 0.0032*** 0.0034*** 0.0034*** (0.0002) (0.0002) (0.0001) (0.0001) Year 2004 -0.0235*** -0.0234*** -0.0226*** -0.0228*** (0.0065) (0.0065) (0.0032) (0.0032) Observations 31465 31465 112111 112111 R-squared 0.02 0.02 0.02 0.02 Note: Marginal effects are reported, rather than coefficients. Standard errors are adjusted for market-level clustering and appear in parentheses. A* indicates significance at 10%; ** significance at 5%; *** significance at 1%. 52 Table 2.4: Effect of Newspaper Competition on Reading a Local Newspaper (1) (2) (3) (4) Probit Probit Linear Prob. Linear Prob. Number of Papers 0.0056 -0.2137 -0.0111 -0.0118 (0.0053) (0.1542) (0.0106) (0.0107) Extreme(Median) -0.0178*** -0.0178*** -0.0191*** (0.0066) (0.0066) (0.0064) Number of Papers*Extreme 0.0033*** 0.0035*** 0.0032** (Median) (0.0012) (0.0012) (0.0013) Log(pop)*Extreme (Median) 0.0134 (0.0095) Extreme (Mean) -0.0308*** (0.0102) Number of Papers*Extreme (Mean) 0.0039*** (0.0014) Internet Supply -0.0034*** -0.0033** -0.0021* -0.0024* (0.0013) (0.0013) (0.0013) (0.0013) Internet Supply*Extreme (Median) 0.0013** 0.0013** 0.0011** (0.0005) (0.0005) (0.0005) Internet Supply*Extreme (Mean) 0.0013** (0.0006) Very conservative -0.0446*** -0.0452*** -0.0396*** -0.0281* (0.0130) (0.0130) (0.0118) (0.0147) Conservative -0.0123** -0.0125** -0.0108* -0.0078 (0.0062) (0.0062) (0.0059) (0.0059) Liberal -0.0059 -0.0059 -0.0001 0.0079 (0.0084) (0.0084) (0.0071) (0.0095) Very liberal -0.0229 -0.0233 -0.0157 0.0003 (0.0143) (0.0144) (0.0135) (0.0184) Log pop -0.0169*** -0.0306** (0.0048) (0.0138) Year2004 -0.0103 -0.0115 -0.0187*** -0.0187*** (0.0073) (0.0070) (0.0060) (0.0060) Constant 0.1142*** 0.1203*** (0.0433) (0.0431) Market Fixed Effect No No Yes Yes Observations 112111 112111 112111 112111 R-squared 0.03 0.03 0.0655 0.0656 Note: Standard errors are adjusted for market-level clustering and appear in parentheses. A* indicates significance at 10%; ** significance at 5%; *** significance at 1%. 53 Table 2.5. Change in Newspaper Competition (1) (2) COEFFICIENT OLS Fixed Effects Avg. Income -0.006 -0.013 (0.028) (0.011) Avg. Education -0.013 -0.001 (0.008) (0.004) Avg. Internet Supply 0.004 -0.006*** (0.006) (0.002) Frac. Very Liberal -0.106 -0.022 (0.073) (0.029) Frac. Very Conservative 0.043 -0.001 (0.105) (0.041) Log (pop) 0.082*** (0.006) year2004 -0.013 0.013* (0.020) (0.007) Constant 0.322 1.201*** (0.270) (0.116) Observations 1926 1926 R-squared 0.140 0.970 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 54 Table 2.6. First Stage Regression (1) Number of Papers (2) Number of Papers*Extreme(Mean) Number of Sport Teams 0.410*** 0.032 (0.086) (0.023) Extreme(mean) -0.152** 0.925*** (0.074) (0.061) Number of Sport Teams*Extreme(Mean) -0.012** 0.035*** (0.0053) (0.075) Very Conservative 0.067* 0.049 (0.040) (0.041) Conservative 0.067** 0.060*** (0.039) (0.020) Liberal 0.159** 0.072*** (0.079) (0.024) Very Liberal 0.030** 0.129** (0.146) (0.063) Log (income) -0.012 -0.001 (0.009) (0.008) College 0.004 -0.000 (0.018) (0.016) High School 0.003 -0.004 (0.018) (0.014) Age -0.000 -0.000 (0.000) (0.000) Male 0.008 0.007 (0.007) (0.003) Internet Supply -0.037*** -0.012*** (0.010) (0.004) Internet Supply*Extreme(mean) 0.004*** -0.017*** (0.001) 0.006 Log pop -0.072 -0.054 (0.064) (0.048) Market Fixed Effect No No Observations 112111 112111 Partial R-squared 0.42 0.51 F-Statistic 65.04 11.08 Note: Standard errors are adjusted for market-level clustering and appear in parentheses. A* indicates significance at 10%; ** significance at 5%; *** significance at 1%. 55 Table 2.7: Political News v.s. Non-political News (1) (2) (3) (4) (5) IV Probit Probit IV IV Read a local Political Non-Political Political Non-Political paper Number of Papers 0.0018 0.0015 0.0045 -0.0040 0.0057 (0.0098) (0.0034) (0.0044) (0.0062) (0.0061) Number of Papers* 0.0067** 0.0061*** -0.0006 0.0075** 0.0029 Extreme(mean) (0.0027) (0.0024) (0.0020) (0.0031) (0.0046) Extreme(mean) -0.0108 -0.0444** 0.0339*** -0.0475*** 0.0288** (0.0164) (0.0190) (0.0115) (0.0167) (0.0121) Very Conservative -0.0567** 0.0261 -0.0777*** 0.0282 -0.0833*** (0.0234) (0.0271) (0.0149) (0.0249) (0.0175) Conservative -0.0190** -0.0035 -0.0138* -0.0024 -0.0143* (0.0091) (0.0104) (0.0078) (0.0099) (0.0080) Liberal -0.0123 0.0561*** -0.0663*** 0.0567*** -0.0684*** (0.0180) (0.0184) (0.0099) (0.0169) (0.0106) Very liberal -0.0360 0.0947*** -0.1127*** 0.0955*** -0.1259*** (0.0304) (0.0331) (0.0167) (0.0304) (0.0216) Log income 0.0281*** 0.0340*** -0.0050 0.0329*** -0.0046 (0.0059) (0.0048) (0.0031) (0.0047) (0.0030) High school 0.1414*** 0.1020*** 0.0342*** 0.0931*** 0.0349*** (0.0064) (0.0097) (0.0077) (0.0085) (0.0076) College 0.1405*** 0.1356*** -0.0062 0.1292*** -0.0054 (0.0076) (0.0087) (0.0077) (0.0080) (0.0076) Age 0.0033*** 0.0044*** -0.0011*** 0.0044*** -0.0011*** (0.0001) (0.0002) (0.0001) (0.0001) (0.0001) Internet Supply -0.0031** -0.0008 -0.0021** -0.0009 -0.0019** (0.0012) (0.0011) (0.0008) (0.0011) (0.0008) Internet Supply* 0.0005 -0.0004 0.0011* -0.0004 0.0010 Extreme(mean) (0.0006) (0.0008) (0.0007) (0.0008) (0.0007) Year2004 -0.0094 (0.0076) Male 0.0019 0.0348*** -0.0330*** 0.0341*** -0.0329*** (0.0032) (0.0046) (0.0036) (0.0045) (0.0036) Logpop -0.0160*** -0.0064* -0.0113*** -0.0044 -0.0129*** (0.0039) (0.0035) (0.0021) (0.0032) (0.0021) Constant 0.3794*** -0.1620** 0.5636*** (0.0995) (0.0749) (0.0451) Observations 112111 66024 66024 66024 66024 R-squared 0.03 0.02 0.01 0.03 0.01 Note: Standard errors are adjusted for market-level clustering and appear in parentheses. A* indicates significance at 10%; ** significance at 5%; *** significance at 1%. 56 Table 2.8: Effect of Newspaper Competition on Getting Information Online (1) (2) (3) Frequency of getting political information online OLS Fixed effects Fixed effects Number of Papers 0.0032 0.0249 0.0135 (0.0155) (0.0507) (0.0495) Extreme (Median) -0.0322 -0.0398 (0.0331) (0.0337) Number of Papers*Extreme(Median) -0.0172** -0.0164** (0.0072) (0.0071) Extreme (Mean) -0.0965* (0.0501) Number of Papers*Extreme(Mean) -0.0134* (0.0077) Very conservative 0.2152*** 0.2278*** 0.2927*** (0.0535) (0.0551) (0.0677) Conservative 0.0208 0.0288 0.0347 (0.0274) (0.0283) (0.0290) Liberal 0.1426*** 0.1439*** 0.2194*** (0.0309) (0.0315) (0.0420) Very liberal 0.5229*** 0.5282*** 0.6497*** (0.0602) (0.0620) (0.0774) Internet Supply 0.0250*** 0.0214*** 0.0213*** (0.0038) (0.0037) (0.0040) Internet Supply*Extreme(Median) 0.0050 (0.0034) Internet Supply*Extreme(Mean) 0.0042 (0.0039) Log income 0.3031*** 0.2970*** 0.2581*** (0.0127) (0.0128) (0.0128) High school 0.0672*** 0.0722*** 0.1061*** (0.0213) (0.0217) (0.0221) Age -0.0107*** -0.0106*** -0.0101*** (0.0006) (0.0006) (0.0006) College 0.5700*** 0.5636*** 0.6039*** (0.0231) (0.0237) (0.0241) year2004 -0.7868*** -0.7627*** -0.7545*** (0.0207) (0.0212) (0.0210) Male 0.4210*** (0.0179) Constant -1.7746*** -1.7334*** -1.5395*** (0.1302) (0.1577) (0.1596) Observations 96898 96898 96898 R-squared 0.0722 0.0822 0.0914 Note. Standard errors are adjusted for market-level clustering and appear in parentheses. A* indicates significance at 10%; ** significance at 5%; *** significance at 1%. 57 Table 2.9: Readership of the New York Times and Wall Street Journal (1) (2) (3) (4) Wall Street NY Times Wall Street NY Times Journal Journal COEFFICIENT Probit Probit Probit Probit Conservative Moderate Liberal Very Liberal -0.0885** (0.0401) -0.1809*** (0.0396) -0.3958*** (0.0466) -0.4464*** (0.0727) 0.155*** (0.054) 0.476*** (0.052) 0.764*** (0.052) 1.110*** (0.056) Liberal Index -0.003*** (0.000) 0.000 (0.000) -0.001*** (0.000) 0.012*** (0.001) -0.004** (0.001) 0.003** (0.001) -0.001*** (0.000) 0.000*** (0.000) 0.012*** (0.001) 0.000 (0.000) 0.001*** (0.000) 0.013*** (0.001) 0.001 (0.002) 0.017*** (0.002) -0.001*** (0.000) 0.000*** (0.000) -0.001** (0.001) 9 -8.2846*** -7.922*** -0.090*** (0.2357) (0.166) (0.006) 120989 120989 120989 0.2 0.2 0.2 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 0.004*** (0.001) 9 -0.140*** (0.008) 120989 0.2 Internet Internet* liberal index Log income High school College Age Age Squared Log pop Year2004 Fixed Effects Constant Observations R-squared 0.4998*** (0.0190) 0.1481 (0.1116) 0.6209*** (0.1067) -0.0305*** (0.0039) 0.0003*** (0.0000) 0.0641*** (0.0069) -0.0711*** (0.0221) 0.276*** (0.013) 0.072 (0.066) 0.634*** (0.062) -0.028*** (0.003) 0.000*** (0.000) 0.177*** (0.006) 0.068*** (0.017) Chapter 3 Media Bias and Influence: Evidence from Newspaper Endorsements1 3.1 Introduction The provision of information by the media to imperfectly informed voters has traditionally been viewed as a key ingredient in the development of a well-functioning democracy. This view was central to the U.S. Bill of Rights providing for freedom of the press, as expressed by Jefferson: ”I am persuaded that the good sense of the people will always be found to be the best army. They may be led astray for a moment, but will soon correct themselves. The people are the only censors of their governors, and even their errors will tend to keep these to the true principles of their institution. To punish these errors too severely would be to suppress the only safeguard of the public liberty. The way to prevent these irregular interpositions of the people is to give them full information of their affairs through the channel of the public papers, and to contrive that those papers should penetrate the whole mass of the people.” –Thomas Jefferson to Edward Carrington, 1787. If voters rely on such information when making decisions over candidates and policies, however, partisan editors may have an incentive to manipulate media content in order to 1 This is joint work with Brian Knight. 58 59 improve the electoral prospects of their favored party. Rational voters may then attempt to filter out such media bias in a variety of ways, including reducing their reliance on biased news sources. This idea is expressed quite simply by Jefferson in the following passage: ”The press is impotent when it abandons itself to falsehood.” –Thomas Jefferson to Thomas Seymour, 1807. Due to this political manipulation of news and the associated behavioral response, voters receive less information from the media, potentially resulting in poor decisions when choosing candidates and policies. While this drawback of political manipulation of the media, a reduction in the degree of information provided to voters, has long been considered a key issue in debates over government regulation of the media, there is little direct evidence that voters respond to media bias in this manner. There are several obstacles to the empirical implementation of such a test. First, one would need variation in the degree of bias across media outlets or across time. Second, one would need a measure of voter information, which is difficult to gauge. Third, one would need to account for potential self-selection into media outlets. For example, if uniformed voters tend to access biased media outlets, then a correlation between the degree of media bias and voter information may represent this self-selection rather than the hypothesized effect. In this paper, we attempt to overcome these hurdles via an examination of the effect of newspaper endorsements on voting in the U.S. Presidential elections of 2000 and 2004. We begin by developing a simple econometric model in which voters have incomplete information and attempt to gather information over candidate quality, which can be interpreted in a variety of ways, including candidate experience, integrity, or competence. Newspapers have better information over quality than do voters but are potentially biased in favor of one of the candidates and may thus endorse candidates of relatively low quality if the bias is severe. Voters are rational and attempt to filter out any bias on the part of the media. In this way, the influence of an endorsement depends upon its credibility, which in turn depends upon the ideology of the newspaper. For example, endorsements for the Democratic candidate from a left-leaning newspaper are less credible and thus have less influence than a similar endorsement from neutral or right-leaning source. 60 In terms of empirical implementation, we use information from daily opinion polls, which have individual-level data on voting intentions as well as newspaper readership. We combine these data with information on the date and identity of newspaper endorsements. Our measures of credibility are derived from an analysis of the role of group ownership of newspapers and reader preferences over candidates prior to the publication of endorsements. Using these credibility measures, we show that endorsements are influential in the sense that readers are more likely to support the recommended candidate after publication of the endorsement. Importantly, however, the credibility of the endorsement is the most important determinant of its influence. In particular, we show that endorsements from extremely biased newspapers have no influence and that influence is increasing in the credibility of the endorsement. This result is robust to several alternative specifications. We also find that endorsements have an effect on voter evaluations of candidate favorability that is similar to the effect on voting intentions. Taken together, these results suggest that rational voters attempt to discount politically-manipulated information and that their reliance on the media in choosing candidates in elections is increasing in its credibility. 3.2 3.2.1 Related Literature Sources and Measure of Media Bias Political endorsements from interest groups can be seen as a way to convey information on party policies to their imperfectly informed group members(McKelvey and Ordeshook, 1985; Grossman and Helpman (1998)). Newspaper endorsements share similar characteristics with interest group endorsements in that newspapers provide information to readers through endorsements. However, while group leaders are more likely to maximize the utility of their group members, newspapers and their readers may have different political preferences. With a few exceptions, such as Kim (2008) and Ansolabehere et al. (2006), there is little research into the motivation behind political endorsements made by newspapers. More generally, however, there is an emerging literature that investigates the sources of media bias. While newspaper endorsements may be affected primarily by editorial boards, executives and owners, newspaper endorsements and media bias in news reporting may share common factors. Some researchers explain media bias as deriving from the demand for news: Under 61 the assumption that consumers like to see news that confirms their beliefs, competition forces newspapers to differentiate themselves by reporting extreme news (Mullainathan and Shleifer, 2005). Even if consumers wish to choose media outlets that deliver unbiased information, their choices may be still affected by their prior beliefs, therefore inducing media outlets to slant their news reports (Gentzkow and Shapiro, 2006). In addition to the influence of demand, media bias may reflect the preferences of journalists (Baron, 2006), editors, owners (Djankov et al., 2003) or governments (Besley and Prat, 2006). Due to the increasing-return-to-scale technology and their dependence on advertising revenue, media outlets may deliver more news to large groups and groups that are valuable to advertisers (Stromberg, 2002). Media sends out information via news stories, editorials, advertisements, or other formats. It is often more difficult to detect media bias from news stories than from other information sources since news stories are usually not slanted in an obvious way. Groseclose and Milyo (2005) provided a way to measure media bias in the news stories of several major media outlets in the US. They arrive at their measure by counting the citations of think tanks in the media and then comparing the citations of think tanks by the Republican or Democratic senators. Gentzkow and Shapiro (2006) constructed another index of media slant by comparing the language in newspapers and the language used by Congressmen. The measure they created correlates with the ratings of political orientation submitted by users to the media directory website Mondo Times.2 They found that the index of media slant can largely be explained by consumer preferences rather than by ownership preferences. The model in this paper predicts that the influence of newspaper endorsements depends on their ideological bias. In the empirical analysis, we account for the possible influence from both the demand side and the supply side to predict the editorial positions of newspapers. 3.2.2 Media Influence Theoretically, media may have persuasive effects, confirmative effects or no effect on political behavior. According to Bray and Kreps (1987), on average, voters can filter out bias without being persuaded. However, recent research is not so optimistic. Even when readers 2 On the website, every user has the right to rate newspapers. However, a lot of newspapers remained unrated and most newspapers are rated by few people. 62 are rational and understand the nature of bias, important information can be still lost and affect readers’ action (Baron, 2006; Bernhardt, Krasa, and Polborn, 2007; Anderson and McLaren, 2007). The effect of media bias may depend on the ideological positions of newspapers. Ideologically extreme newspapers may have no effect on voters’ decisions if voters all choose the newspaper that is closest to their own ideologies (Chan and Suen, 2004). If voters are not rational, media will have persuasive effects on voting behavior (De Marzo, Vayanos, and Zwiebel, 2003; Lakoff, 1987). Bernhardt, Krasa and Polborn (2007) and Anderson and McLaren(2007) assume that newspapers manipulate news information by suppressing some information from readers in their models. Thus, for example, readers may be informed about negative news of one candidate without knowing if there is more negative news of the other candidate. Baron (2006) assumes that reporters receive binary signal about the true state and do not always report the signal truthfully. Readers are rational, update their beliefs according to the information from the news report. In this paper, we do not consider suppression of information. Instead, we consider the binary messages of newspapers in the form of endorsements when the information newspapers receive is continuously distributed. We assume that voters are rational, and understand that newspapers are biased to cater to readers’ preferences and affected by their owners’ preferences. Similar to Baron (2006), the model predicts that an endorsement for a Democratic candidate from a liberal newspaper is still informative and influential in the sense that it means the Democratic candidate can be a credible one. However, endorsements for Democrats from non-partisan or right-leaning sources are more informative and therefore more influential than are those from left-leaning sources. In general, non-partisan newspapers are more informative and can lead to fewer electoral mistakes. Empirically, many studies have observed associations between media exposure and political behavior and attitudes(Brians and Wattenberg, 1996; Hibbing and Theiss-Morse 1998). However, consumers’ tendencies to choose news outlets that share similar political perspectives make it difficult to identify a causal effect. Several studies have made efforts in different ways to deal with the potential selection bias. Kaplan and Della Vigna (2007) identified the effect of FOX News on voting behavior by looking at the introduction of FOX 63 News Channel in some areas at a town-by-town level. They found that FOX News convinced 3 to 28 percent of its viewers to vote Republican. Gerber, Karlan, and Bergan (2007) conducted a field experiment and found that random free subscriptions of the Washington Post increased the probability of voting for the Democratic candidate by eight percentage points in the 2005 Virginia gubernatorial election. Other related research includes studies on voter turnout (Green and Gerber, 2004) and on political knowledge (Zukin and Snyder, 1984). It is more difficult to identify the effect of newspaper endorsements than media exposure. Aside from unobservable differences between readers of different newspapers, unobserved factors in a market that affect endorsements and political behavior at the same time can make it hard to identify the causal effects of endorsements. However, studies of newspaper endorsements do not account for readers’ selection of particular newspapers or rely on information on voting behavior after endorsements. Erikson (1976) estimated the impact of newspaper endorsements on vote share at county level in the 1964 presidential elections by ordinary least square (OLS) and two stage least square (TSLS) methods. When implementing TSLS, newspaper endorsements were predicted by newspapers ideologies in 1959.3 Kahn and Kenney (2002) found that the endorsement decision changes the tone of coverage at incumbent Senator races. They also found significant positive effect of endorsement on the comparative feeling thermometer score by using 1988-92 NES/SES data. Similar to Kahn and Kenney (2002), Druckman and Parkin (2005) first identified the effect of endorsements on the quantity and tone of campaign coverage of two competing newspapers and then estimated the effect of endorsements on voting decision using exit polls on the election day. While most recent studies use self-reported party identification/political attitudes to control for potential difference between readers of different newspapers, some unobservable factors may cause upward bias. This paper contributes to the empirical literature of newspaper endorsements in several ways. We deal with the potential endogeneity problems by using readership data and collecting the information on the timing of endorsements. Moreover, we test the idea that the influence of newspaper endorsements should depend on their ideological positions. 3 Newspaper ideologies are measured by columnists’ political ideologies. 64 3.3 Model The model consists of two candidates (c ∈ {D,R}) competing for election, a set of voters, indexed by v, and a set of newspapers, indexed by n. Voters have ideological preferences over candidates but also care about the quality of the candidates, which is assumed to be unknown before the election. During the campaign, newspapers receive information regarding candidate quality and make endorsements based on this information as well as their ideological positions. Readers attempt to learn about candidate quality from endorsements but this inference is potentially complicated by the ideological position of newspapers. 3.3.1 Voter behavior More formally, we assume that voter v receives the following payoff from candidate D winning the election: Uv = q + iv (3.1) where q represents the quality of candidate D (relative to candidate R) and iv represents the ideology of voter i. Voters are uncertain over relative candidate quality, and initial priors are normally distributed with mean µ and a variance σq2 . Voters are assumed to support the candidate who maximizes their expected utility. Voters are associated with a single newspaper and potentially observe an endorsement from newspaper n for either the Democrat (en = 1) or for the Republican (en = 0). Before observing an endorsement, voter i supports the Democrat if the following condition holds: E(Uv ) = µ + iv > 0 (3.2) After observing the endorsement, voter i supports the Democrat if the following condition holds: E(Uv |en ) = E(q|en ) + iv > 0 (3.3) In order to understand how voters update following the endorsement, we next present a framework for newspaper endorsement decisions. 65 3.3.2 Newspaper Endorsements Similarly to the preferences of voters, the editor of newspaper n is assumed to receive the following payoff from candidate D winning the election: Un = q + in (3.4) where in is the ideology newspaper n. Again, candidate quality is unknown, and initial priors over quality are normally distributed with mean equal to µ and a variance of σq2 . Prior to making an endorsement, newspapers receive an unbiased signal over candidate quality: θ n = q + εn . (3.5) where εn is the noise in the signal and is assumed to be normally distributed with mean zero and variance σε2 . After observing the signal, newspaper editors update over quality as follows: E(q|θn ) = (1 − α)µ + αθn (3.6) where the weight on the signal is given by: α= σq2 σq2 + σε2 (3.7) Reflecting well-known results, the weight on the signal is increasing in the degree of initial uncertainty over quality and is decreasing in the degree of noise in the signal. Assuming sincere endorsements, newspaper n endorses the Democratic candidate if the following condition holds:   −in − (1 − α)µ en = 1[E(Un |en ) > 0] = 1 θn > α 3.3.3 (3.8) The Influence of Endorsements Returning to voter behavior, we can now evaluate how individuals attempt to infer quality from newspaper endorsements. In particular, voters incorporate the newspaper endorsement rule given by equation (8) and update as follows in response to an endorsement for the Democratic candidate: 66   √ −in − (1 − α)µ E(q|en = 1) = E q|θn > = µ + ασq λnd α (3.9) where λnd is the Mills ratio: λnd √ φ[(−in − µ)/ ασq ] √ = 1 − Φ[(−in − µ)/ ασq ] (3.10) and φ and Φ are the Normal density and distribution function, respectively. Thus, the influence of an endorsement for the Democrat can be expressed as follows: E(q|en = 1) − E(q) = √ ασq λnd (3.11) To provide further interpretation of this influence expression, consider first the special case of (ex-ante) equally qualified candidates (µ = 0) and an unbiased newspaper (in = 0). √ In this case, the influence of a Democratic endorsement equals 2φ(0) ασq and can be interpreted as the influence of an unbiased endorsement. Influence from such an unbiased source is increasing in α,which is the weight that newspapers place upon the signal and hence reflects the informational content of such an endorsement. In addition, influence is increasing in σq , which reflects the degree of uncertainty in voter’s perceptions of candidate quality and hence their reliance on endorsements in evaluating candidate quality. Returning to the more general case of influence, we have that the influence of an endorsement for the Democrat is increasing in λnd , which can be interpreted as the credibility of the endorsement for the Democrat. It can be shown that λnd is monotonically decreasing in the ideology of the newspaper (in ) and approaches zero (infinity) as newspaper ideology converges to the extreme left (right).4 Said differently, an endorsement for the Democrat from a relatively left-leaning newspaper, such as the New York Times, provides less information to voters than does an endorsement from a relatively right-leaning newspaper, such as the Washington Times. Voters update in an analogous manner upon observing a Republican endorsement, and the influence of such an endorsement can be written as follows: 4 See, for example, Heckman (1979). 67 √ E(q|en = 0) − E(q) = − ασq λnr (3.12) where the credibility of a Republican endorsement can be written as follows: λnr √ φ[(−in − µ)/ ασq ] √ = . Φ[(−in − µ)/ ασq ] Similarly to the discussion of the credibility of Democratic endorsements, the credibility of Republican endorsements is increasing in newspaper ideology and such an endorsement from a left-leaning source provides more information to voters than does an endorsement from a right-leaning source. To provide further interpretation of the effects of media bias on the quality of information received by voters, we define the ex-ante influence of a newspaper’s endorsement as follows: δn ≡ Pr(en = 1) × |E(q|en = 1) − E(q)| + Pr(en = 0) × |E(q|en = 0) − E(q)| √ Using the fact that Pr(en = 1) = 1 − Φ[(−in − µ)/ ασq ] and Pr(en = 0) = Φ[(−in − √ µ)/ ασq ], we can then say that: √ √ δn = 2 ασq φ[(−in − µ)/ ασq ] Thus, expected influence is maximized at in = −µ. In the special case of ex-ante equally qualified candidates (µ = 0), expected influence is thus maximal for unbiased sources. More generally, expected influence converges to zero as ideology grows large in either direction. This notion of expected influence captures the idea that media bias may reduce the quality of information provided to voters. 3.4 Data In order to measure the influence of endorsements, we use voter reactions to endorsements as captured in opinion polls. Individual-level data is from the National Annenberg Election Surveys of 2000 and 2004. The Annenberg survey is a rolling cross-section survey that polled hundreds of people on a daily basis during the 2000 and 2004 election years. It 68 includes information on voting intentions, favorability of candidates and newspapers read most often. Information about the dates of newspaper endorsements and candidates endorsed comes from several different sources, including the collection done by Eric Appleman, newspaper archives in Lexis-Nexis and Factiva, Associated Press...etc. Graph 1 shows that there is substantial variation in the timing of newspaper endorsements. Kim (2008) provided data on group ownership. Combining the Annenberg data with information on newspaper endorsements, we are able to control for newspaper fixed effects and time fixed effects when estimating the effect of newspaper endorsements. The sample is restricted to those respondents who read a newspaper with available endorsement information. Those respondents who read a newspaper that made a non-endorsement are also excluded. In the sample, we have 28,345 respondents before newspaper endorsements and 2,141 respondents after newspaper endorsements. The summary statistics of variables is presented in Table 1. 3.5 Empirical Implementation We estimate the differential effect of newspaper endorsements by three-stage estimation. According to the model in section 3, voters are assumed to support the candidate who maximizes their expected utility. The utility of having a Democratic candidate elected depends on the ideology of voter v, and the candidate’s relative quality, q. E(Uv ) = E(q) + iv . (3.13) We consider that voters may receive some common information about the quality of candidates on a daily basis and that ideologies of voters may be affected by their background variables and related to the ideology of the newspaper they read. We can then express the expected utility of voter v voting for the Democratic candidate before observing a newspaper endorsement as: E(Uvt ) = βXv + ηt + ηn + ηvt , (3.14) where ηt is date-fixed effects, ηn is newspaper-fixed effects, and Xv includes respondents’ gender, education, race, and income. We assume that ηvt is uniformly distributed; therefore 69 the probability of voting for a Democratic candidate can be expressed as: Pr(Vitn = 1) = βXi + ηt + ηn . (3.15) In the first stage, we use all individuals before observing newspaper endorsements to estimate β, ηt and ηn in the above equation by a linear probability model. In the second stage, we estimate λnr and λnd by the following probit model. Pr(en = 1) = Φ(γDn + Igroup ), (3.16) where Dn is readers’ political preference, measured by the average predicted probability of its readers to choose the Democratic candidate from the first stage, Pr(Vitn ). Igroup represents group owner-fixed effects, which control for the influence from a group owner, which is defined as a company that owns more than four daily newspapers in the sample. With estimates of γ and group owner-fixed effects, we can then infer the credibility of endorsements for each newspaper, represented by λrep and λnd .5 In the third stage, we use respondents after endorsements to estimate the effect of endorsements on voting intentions. Based on the estimates in the first stage, we can predict each respondent’s probability of voting for the Democratic candidate without observing \ endorsements, Pr(V i = 1). From the second stage estimation, we have estimates of λnr and λdem for each newspaper. The model predicts that the influence of newspapers’ endorsements on the expectation of q depends on the credibility of their endorsements , represented by λnr and λnd . Therefore, we can estimate the differential effect of newspaper endorsements by the following linear probability model. \ Pr(Vi = 1|en ) − Pr(V i = 1) = βd en × λnd − βr × (1 − en ) × λnr (3.17) The model in section 3 predicts that newspaper endorsements are influential and endorsements for Republicans from a left-leaning or non-partisan newspaper are more influential. Since λnr and λnd capture the differential effects affected by the newspapers’ ideological positions, we expect βd and βr to be significant. Also, we expect that the empirical model with differential effects will have a higher explanatory power than an empirical model that does not include those impact measures, λnr and λnd . 5 √ Note that the probability of a Democratic endorsement is given by: Pr(en = 1) = 1−Φ[(−bn −µ)/ ασq ]. 70 3.6 Results We use the bootstrap method to estimate all coefficients. Table 2 presents the results of the three-stage estimation. Table 2 panel A presents the result of the first stage. Voters who are older or black are more likely to vote for Democrats and voters who attend religious services or consider themselves born-again Christians are more likely to vote for the Republican. Table 2 panel B presents the result of the second stage. Newspaper endorsements reflect their readers’ political preferences, as measured by the average predicted probability of voting for Democratic candidates. Newspapers that are owned by Advance Publications, Cox Newspapers, Gannett, Hearst Newspapers, Knight Ridder, McClatchy Newspapers, and the New York Times Co. are more likely to endorse Democratic candidates. The average predicted probability of endorsing a Democratic candidate is 0.52. Among newspapers predicted to make Democratic endorsements, 74% actually made Democratic endorsements. Table 2 panel C presents the result of the third stage under different specifications. In column 1, the basic specification includes impact measures and constraints Democratic and Republican endorsements to have equal and opposite effects. The result indicates that an unbiased newspaper will increase or decrease the probability of voting for the Democrat by 3% by making a Democratic or Republican endorsement, respectively. Since the average λrn among those newspapers with Republican endorsements is 0.65 and the average λdn among those newspapers with Democratic endorsements is 0.61, a Democratic or Republican endorsement on average will increase or decrease the probability of voting for the Democratic candidate by 0.37 or 0.32. The second specification in column 2 estimates the effect of endorsements without considering the possible differential effects reflected by impact measures. Column 3 includes both endorsement effect and impact measures. While the endorsements effect and impact measures are both significant in the first two specifications, the impact measure is more significant. When both are included in the third specification, only the credibility measure is significant. This is consistent with the expectation that the model with impact measures will fit the data better. Column 4 presents the results of the specification without the constraint of having the 71 effects of Democratic endorsements and Republican endorsements to be equal and opposite. The result shows that the effect of Democratic endorsements is not significant, while the effect of Republican endorsements is negatively significant. Lastly, we present the results in 2000 and 2004 separately. The credibility of endorsements has slightly weaker effect in 2004 than in 2000. In addition to vote intention, the survey also asked respondent to rate both candidates’ favorability. Table 3 presents the result when the dependent variable is replaced by the difference of candidates’ favorability. Overall, the results are similar but weaker. The difference in favorability should capture more variation even for voters who are very Republican or Democratic. However, the way people evaluate candidates in favorability may be different from person to person and result in larger noise than vote intention. We can then define a new dummy variable ”Prefer Democratic Candidate” as 1 when the respondent give the Democratic candidate higher favorability. Overall, the result is similar, but the credibility of endorsements is not significant when estimate the effect separately in 2000 and 2004. 3.6.1 Persuasion Rate and Counterfactuals Based on the estimation in the second stage, we can predict the probability of endorsing Democrats for each newspaper. Taking the estimate of β from the basic specification in the third stage, we can predict the influence of the endorsement made by each newspaper. Table 5 listed the predicted probability of endorsing Democrats, actual endorsements, and persuasion rates of newspapers with circulations in the top 20 in 2000.6 The endorsement for Gore made by the New York Times only convinces 0.9% of its readers to vote for Gore, while the Chicago Sun Times convince 3% of its readers to vote for Bush.7 The higher probability of endorsing Democrats of the New York Times means that the standard of the relative quality of Democratic candidates for the New York Times to make an endorsement is lower. Therefore, the endorsement for Gore made by the New York Times is not very informative and has relatively small influence. On the other hand, the 6 Note that the predicted effects are based on owners’ preference and readers’ preference with the assumption that the influence from the readers are the same across newspapers. 7 If we take the estimates of βrep and βdem from the third specification, the endorsement for Gore from the New York Times has small and non-significant effect, while the Chicago Sun Times convince 2.6% of its readers to vote for Bush. 72 endorsement made by the Chicago Sun Times in 2000 are relatively influential. The Chicago Sun Times was predicted to be more likely to endorse Democrats but made a Republican endorsement. This is very informative to its readers since it revealed that the information the Chicago Sun Times received indicate that the quality of Bush as a presidential candidate is higher than Gore. We can then present the counterfactual case of when all newspapers in the sample made Democratic endorsements in 2000 and 2004. As presented in Table 6, vote share of Gore will increase by 3% in 2000 and vote share of Kerry will increase by 2% in 2004 across all respondents in the sample. Similarly, if all newspapers made Republican endorsements, vote share of Gore will decrease by 5% in 2000 and vote share of Kerry will decrease by 4% in 2004. To get a bigger picture of how such changes in newspaper endorsements can influence election outcome in US presidential elections, we can multiply changes in vote share by the fraction of voters who read a newspaper. From the Annenberg Survey, around 75% of voters said that they read a newspaper at least one day in the past week.8 The predicted changes in vote share are listed in Table 6 row 3 and 5. The empirical findings indicate that the effect of newspaper endorsements on the vote share in the U.S. is around 2.51% to 6.31% in an election. This means that the influence of newspaper endorsements can be critical in a close election. For example, the predicted increase in votes for Gore is 22897 votes if Orlando Sentinel in Florida endorsed Gore, which is enough to change the election outcome. 3.7 Conclusion In this paper, we have investigated the influence of newspaper endorsements on voting patterns in the 2000 and 2004 U.S. Presidential Elections. We first develop a simple econometric model in which voters are uncertain over candidate quality and turn to newspaper endorsements for information about the candidates. Newspapers, however, are potentially biased in favor of one of the candidates and voters thus rationally account for the credibil8 In the survey done by Pew Research Center, 65% of respondents said that they read newspapers in the average week. According the circulation data published in Editor and Publisher Year Book, circulation rate is around 20% in whole population. Although the circulation rate is low, consider the fact that there are three people per household on average, the actual fraction of people who read a newspaper should be not far from 60%. 73 ity of any endorsements. Our primary finding is that endorsements are influential in the sense that voters are more likely to support the recommended candidate after publication of the endorsement. The degree of this influence, however, depends upon the credibility of the endorsement. In this way, endorsements for the Democratic candidate from leftleaning newspapers are less influential than are endorsements from neutral or right-leaning or newspapers and likewise for endorsements for the Republican. These findings suggest that voters do rely on the media for information during campaigns but that media bias reduces the quality of this information and thus ultimately may lead to mistakes by voters in choosing candidates. 74 References Ansolabehere, S.,(2006) “The Orientation of Newspaper Endorsements in US Elections, 1940–2002,” Quarterly Journal of Political Science, 1(4): 393-404. Baron, David, (2006) “Persistent Media Bias,” Journal of Public Economics 90(1-2): 1-36. Bernhardt, Dan, Stefan Krasa, and Mattias Polborn (2006), “Political Polarization and the Electoral Effects of Media Bias,” CESifo Working Paper Series No. 1798. Besley, Timothy and Andrea Prat, (2006) “Handcuffs for the Grabbing Hand? Media Capture and Government Accountability,” American Economic Review, 96(3):720736 Bray, Margaret and David M. Kreps, (1987)“ Rational Learning and Rational Expectations,” Arrow and the Ascent of Modern Economic Theory, 597-625. DellaVigna, Stefano. and Ethan Kaplan (2007) “The Fox News Effect: Media Bias and Voting,” Quarterly Journal of Economics, 122: 1187-1234. DeMarzo, Peter M., Dimitri Vayanos and Jefferey Zweiebel (2003) “Persuasion Bias, Social Influence, and Unidimensional Opinions,”Quarterly Journal of Economics, 118(3):909968. Erikson, R.S., “The Influence of Newspaper Endorsements in Presidential Elections: The Case of 1964,” American Journal of Political Science, 20(2): 207-233. Grossman, Gene M. and Helpman, Elhanan, “Competing for Endorsements,” American Economic Review, 89(3): 501–24 Gabszewicz, Jean J., Didier Laussel and Nathalie Sonnac (2005), “Press Industry and the political differentiation of newspapers,” Journal of Public Economic Theory 4(3): 317-334. Gentzkow, Matthew and Jess M. Shapiro (2006), “Media Bias and Reputation,” Journal of Political Economy, 114(2): 280-316. 75 Gentzkow, Matthew and Jess M. Shapiro (2006), “What Drives Media Slant? Evidence from U.S. Daily Newspapers.” Working paper. Gerber, Alan, Karlan, Dean S. and Bergan, Daniel, “Does The Media Matter? A Field Experiment Measuring the Effect of Newspapers on Voting Behavior and Political Opinions,” Yale Economic Applications and Policy Discussion Paper No. 12. Groseclose, Tim and Jeffrey Milyo (2005), “A Measure of Media Bias,” Quarterly Journal of Economics, 120(4): 1191-1237. Kim, Chisoo (2008), “Group Ownership and Newspaper Endorsements.” Working paper. McKelvey, Richard and Peter Ordeshook (1985), “Elections with limited information: A Fulfilled Expectations Model Using Contemporaneous Poll and Endorsement Data as Information Sources,” Journal of Economic Theory, 36(1): 55-85. Mullainathan, Sendhil and Andrei Shleifer (2005), “Market for News,” American Economic Review 95(4): 1031-1053. Stromberg, David (2004), “Mass Media Competition, Political Competition, and Public Policy,” Review of Economic Studies 71(1): 265-284. 76 0 number of endorsements 50 100 150 Graph 3.1. Dates of Newspaper Endorsements in 2000 and 2004 14oct2000 21oct2000 28oct2000 endorsment date in 2000 04nov2000 0 number of endorsements 50 100 150 07oct2000 09sep2004 23sep2004 07oct2004 21oct2004 endorsment date in 2004 04nov2004 77 Table 3.1 Summary Statistics Variable Obs Mean Std.Dev. Min Max Intend to vote for Democratic Candidate 30,486 0.544 0.498 0 1 Difference in Favorability 29,939 0.338 6.003 -10 10 Have a high school degree, no college 30,486 0.217 0.412 0 1 Have some college or higher 30,486 0.741 0.438 0 1 Male 30,486 0.473 0.499 0 1 Black 30,486 0.093 0.291 0 1 Age 30,486 47.391 16.054 18 97 Born-again Christian 30,486 0.313 0.464 0 1 Attend religious services The newspaper read by the respondent made a democratic endorsement 30,486 0.388 0.487 0 1 30,486 0.593 0.491 0 1 78 Table 3.2: Effect of Newspaper Endorsements on Vote Intention Panel A: First Stage; Dependent Variable: Vote for Democratic Candidate(VD) High school -0.044*** (0.015) College -0.008 (0.015) Male -0.089*** (0.006) Black 0.435*** (0.008) Age 0.003*** (0.001) Age Squared -0.000** (0.000) Born Again Christian -0.157*** (0.007) Attend Religious Services -0.121*** (0.006) Constant 0.700*** (0.281) Income Categories 9 Newspaper Fixed Effects 9 Time Fixed Effects 9 Observations: 28345 Sample: Newspaper Readers before endorsement Panel B: Second Stage; Dependent Variable: Endorse Democratic Candidate Readers Preference (Avg. Predicted Vote dem) Group owner effect 1 Advance Publications Inc. 2.860*** (0.344) 0.230*** (0.060) Cox Newspapers 1.096*** (0.069) E W Scripps Co. -0.095 (0.095) Gannett Co. Inc 0.897*** (0.052) Hearst Newspapers 0.518*** (0.056) Knight Ridder 1.079*** (0.052) Lee Enterprises Inc. -0.121 (0.124) McClatchy Newspapers 1.564*** (0.070) New York Times Co. 0.906** (0.076) Constant -1.729*** (0.175) Observations :376 Sample: Newspapers made endorsements in 2000 or 2004 79 Note. 1. Default Category: Newspapers not owned by group owners. Companies own more than 4 newspapers are defined as group owner of newspapers. 2. Newspapers with the same name in different years are treated as different newspapers. Panel C: Third Stage Dependent Variable: VD-Predicted VD I β (βdem= -βrep) 0.056*** (0.020) Endorsement II βrep Observations 0.016 (0.031) 2141 2000 2004 0.045* (0.032) 0.043* (0.024) 0.043 (0.049) -0.065* (0.043) 0.024 0.022 -0.043 (0.033) (0.034) (0.057) 2141 2141 742 0.049 (0.041) 1399 0.096** (0.039) 0.025** -0.030 (0.014) (0.027) βdem Constant III 0.011 (0.031) 2141 IV Note. Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1 80 Table 3.3: Effect of Newspaper Endorsements on Candidates’ Favorability Panel A: First Stage Dependent Variable: Difference in Favorability (Dem-Rep) High school -0.053** (0.020) College -0.007 (0.019) Male -0.121*** (0.006) Black 0.482*** (0.011) Age 0.003** (0.001) Age Squared -0.000** (0.000) Born Again Christian -0.199*** (0.008) Attend Religious Services -0.150*** (0.007) Constant 0.267 (0.484) Income Categories 9 Newspaper Fixed Effects 9 Time Fixed Effects 9 Observations: 27814 Sample: Newspaper Readers before endorsement Panel B: Second Stage Dependent Variable: Endorse Democratic Candidate Readers Preference (Avg. Predicted favorability) Group owner effect 1 Advance Publications Inc. 2.527*** (0.027) 0.352*** (0.060) Cox Newspapers 1.144*** (0.070) E W Scripps Co. -0.078 (0.101) Gannett Co. Inc 0.875*** (0.058) Hearst Newspapers 0.538 (0.058) Knight Ridder 1.077*** (0.051) Lee Enterprises Inc. -0.119 (0.147) McClatchy Newspapers 1.535*** (0.069) New York Times Co. 0.900*** (0.077) Constant -0.263*** (0.021) Observations :376 Sample: Newspapers made endorsement in 2000 or 2004 81 Note. 1. Default Category: Newspapers that are not owned by groups. Companies own more than 4 newspapers are defined as group owner of newspapers. 2. Newspapers with the same name in different years are treated as different newspapers. Panel C: Third Stage Dependent Variable: Diff in Favorability-Predicted diff in Favorability I β (βdem=βrep) II III 0.032** (0.016) 0.054 (0.050) 0.001 (0.032) 0.056** (0.024) Endorsement βdem βrep Constant Observations -0.003 (0.039) 2141 -0.005 (0.039) 2141 Range of Dependent variable: -1 to 1. IV 2000 2004 0.044 (0.042) 0.046 (0.033) -0.034 (0.064) -0.128*** (0.048) -0.002 0.046 -0.069 (0.040) (0.042) (0.054) 2141 2141 742 0.040 0.052 1399 82 Table 3.4: Effect of Newspaper Endorsements on Voters’ Preference Panel A: First Stage Dependent Variable: Prefer Dem (Dem-Rep>0) High school -0.040*** (0.017) College -0.003 (0.017) Male -0.087*** (0.006) Black 0.422*** (0.001) Age 0.003** (0.001) Age Squared -0.000** (0.000) Born Again Christian -0.163*** (0.007) Attend Religious Services -0.123*** (0.007) Constant 0.700 (0.494) Income Categories 9 Newspaper Fixed Effects 9 Time Fixed Effects 9 Observations: 27814 Sample: Newspaper Readers before endorsement Panel B: Second Stage Dependent Variable: Endorse Democratic Candidate Readers Preference (Avg. Predicted favorability) Group owner effect 1 Advance Publications Inc. 2.667*** (0.315) 0.350*** (0.056) Cox Newspapers 1.144*** (0.073) E W Scripps Co. -0.101 (0.089) Gannett Co. Inc 0.867*** (0.050) Hearst Newspapers 0.540*** (0.051) Knight Ridder 1.060*** (0.051) Lee Enterprises Inc. -0.185* (0.132) McClatchy Newspapers 1.565*** (0.076) New York Times Co. 0.959*** (0.070) Constant -1.623*** (0.160) Observations :376 Sample: Newspapers made endorsement in 2000 or 2004 83 Note. 1. Default Category: Newspapers that are not owned by groups. Companies own more than 4 newspapers are defined as group owner of newspapers. 2. Newspapers with the same name in different years are treated as different newspapers. Panel C: Third Stage Dependent Variable: Prefer_Dem - Predicted Prefer_Dem I β (βdem=βrep) 0.048** (0.019) Endorsement II III 0.115*** (0.041) 0.016 -0.050** (0.012) (0.026) βdem βrep Constant Observations -0.017 (0.033) 2141 IV -0.005 (0.039) 2141 0.014 (0.040) 2141 0.029 (0.049) -0.063* (0.040) 0.026 (0.035) 2141 2000 2004 0.037 (0.037) 0.036 (0.025) -0.043 (0.055) 742 0.048 0.040 1399 84 Table 3.5: Influence of Top 20 Newspapers in 20001 Newspaper Name Readers' Predicted Prob. (VD=1) New York Times Washington Post New York Daily News Chicago Tribune Newsday Houston Chronicle Dallas Morning News Chicago Sun Time 0.77 0.65 0.69 0.53 0.54 0.39 0.37 0.67 Boston Globe San Francisco Chronicle Arizona Republic New York Post Rocky Mountain News Denver Post Philadelphia Inquirer San Diego UnionTribune 0.71 Group Owner2 New York Times Co. ----Hearst Newspapers --New York Times Co. Predicted Probability of Endorsing Gore Endorsed Gore or Bush Predicted Influence on Prob. (VD=1) 0.92 0.55 0.60 0.41 0.42 0.46 0.25 0.57 1 1 1 0 1 0 0 0 0.009 0.040 0.036 -0.037 0.052 -0.041 -0.023 -0.051 0.89 1 0.012 0.77 Hearst Newspapers 0.44 -0.47 -0.48 -0.52 -0.62 Knight Ridder 0.84 0.32 0.35 0.36 0.40 0.87 1 0 0 0 1 1 0.016 -0.030 -0.032 -0.033 0.054 0.014 0.48 0.36 0 -0.033 -- Note. 1. USA Today, Wall Street Journal and LA Times are not in this table because those newspapers did not make an endorsement or made a non-endorsement in 2000. 2. Missing (--) means that the newspaper is not owned by a group owner. Group owner is defined as a company that owns more than four daily newspapers in the survey. 85 Table 3.6: Counterfactuals Year 2000 Year 2004 Vote share of the Democratic candidate in sample 53.04% 55.56% Predicted change in Democratic vote share in sample if all newspapers made Democratic endorsements +4.23% +3.35% Predicted change in Democratic vote share in the U.S. if all newspapers made Democratic endorsements +3.17% +2.51% Predicted change in Democratic vote share in sample if all newspapers made Republican endorsements -5.40% -6.31% Predicted change in Democratic vote share in the U.S. if all newspapers made Republican endorsements -4.05% -4.73%