television news, mexico's 2000 elections and media effects...

B.J.Pol.S. 35, 1–30 Copyright © 2004 Cambridge University Press

DOI: 10.1017/S00070123405000013 Printed in the United Kingdom

Television News, Mexico’s 2000 Elections and MediaEffects in Emerging Democracies

CHAPPELL LAWSON A N D JAMES A. M CCANN*

On the basis of an analysis of a four-wave panel survey, we argue that exposure to television news hadsignificant, substantial effects on both attitudes and vote choices in Mexico’s watershed presidential electionof 2000. These findings support the contention, implicit in some research on political communication, that themagnitude of media effects varies with certain features of the political context. In particular, televisioninfluence in electoral campaigns may be substantially larger in emerging democratic systems.

In contrast to ‘small effects’ models of media influence in electoral campaigns, we arguethat television coverage had a powerful impact on Mexico’s 2000 presidential election.Exposure to broadcasts on the Television Azteca network dampened enthusiasm forFrancisco Labastida, nominee of the long-ruling Institutional Revolutionary Party (PRI).Meanwhile, the initially deleterious effects of newscasts from the Televisa network onopinions of Cuauhtemoc Cardenas disappeared when coverage of that candidate improvedmarkedly during the second half of the campaign.

Exposure to television news also affected political behaviour (i.e., vote choice). Forinstance, viewing Television Azteca was associated with substantial increases in thelikelihood of supporting victorious opposition candidate Vicente Fox and substantialdeclines in the likelihood of supporting Labastida. Televisa news coverage, by contrast,had a much more muted effect.

Television coverage was not the only, nor even the most important, influence onMexican voters in 2000; widespread disenchantment with corruption and economicmismanagement, political reforms during the 1990s that levelled the electoral playing field,Fox’s charismatic appeal and Labastida’s lack thereof were also crucial ingredients in thefinal outcome.1 But news coverage was a crucial factor in shaping voter attitudes andbehaviour.

* Department of Political Science at, respectively, Massachusetts Institute of Technology, and PurdueUniversity. The authors are indebted to Chistopher Achen, Steve Ansolabehere, Brandice Canes-Wrone, JorgeDomınguez, Pippa Norris, Alejandro Poire, David Sanders, Jim Snyder, Gadi Wolfsfeld, four anonymousreviewers from the Journal and their collaborators in the Mexico 2000 Panel Study for advice and comments onearlier drafts. Organizers of the Mexico 2000 Panel Study include Miguel Basanez, Roderic Camp, WayneCornelius, Jorge Domınguez, Federico Estevez, Joseph Klesner, Chappell Lawson (Principal Investigator), BeatrizMagaloni, James McCann, Alejandro Moreno, Pablo Paras and Alejandro Poire. Support for the Mexico 2000Panel Study was provided by the National Science Foundation (SES-9905703) and Reforma newspaper. Technicaldetails on the Mexico 2000 Panel Study, as well as copies of the survey instruments, are available at:http://web.mit.edu/polisci/Faculty/C.Lawson.html. The authors are also grateful to Reforma newspaper, theFederal Electoral Institute and Edmundo Berumen for providing data from their analyses of Mexican televisioncoverage, as well as to the Mexican Academy of Human Rights for providing recordings of the news programmeswith which to conduct their own content analysis. The bulk of this analysis was performed by Mariana Sanz.

1 Jorge I. Domınguez and Chappell Lawson, eds, Mexico’s Pivotal Democratic Election: Candidates, Voters,and the Presidential Campaign of 2000 (Stanford and La Jolla, Calif.: Stanford University Press and the Centerfor U.S-Mexican Studies, University of California at San Diego, 2003).

2 L A W S O N A N D McC A N N

These findings have potentially broad implications for the study of elections and politicalcommunication outside of the developed West. Among other things, they suggest thelimited transportability of ‘small effects’ models of campaign influence to emergingdemocracies. They thus lend support to the notion, implicit in some research on politicalcommunication, that the magnitude of media influence varies with certain features of thepolitical context.

The first section of this article briefly reviews the literature on media effects in elections.This section suggests that media influence on voter’s attitudes and behaviour depends onspecific aspects of the political and informational context. Where these aspects are lesspronounced, as in Mexican elections, the potential scope for media effects is substantiallygreater.

The second section evaluates television news coverage in Mexico during the 2000campaign, drawing on data from the Federal Electoral Institute (IFE), Reforma newspaper,and a detailed content analysis of the two leading nightly news programmes. This latterdataset includes measures of visual tone, previously neglected in much of the media effectsliterature.2 All told, content analysis reveals that different television audiences receivedsomewhat different cues at different points in the campaign. In general, the country’slargest network (Televisa) remained more sympathetic towards the ruling party than itsnewer rival (Television Azteca). But campaign coverage, especially Televisa’s coverageof Cardenas, also shifted over the course of the race.

The third section analyses the effects of television coverage on public opinion towardsthe three main candidates. Using data from the Mexico 2000 Panel Study, this sectionshows that large differences in the tone of coverage of the main candidates shaped relativeappraisals of those candidates. Where reporting was balanced, the effects of exposure totelevision news were quite small; however, substantial discrepancies in the tone ofcoverage were typically reflected in public opinion.

The fourth section turns to the impact of television coverage on political behaviour –specifically, self-reported vote choice in the election. Exposure to network news onTelevision Azteca eroded support for Labastida, mainly to the benefit of Fox. The effectsof Televisa coverage, however, were modest.

The fifth section addresses potential objections to our analysis of media effects. Inessence, this section comprises various tests designed to demonstrate the robustness of thefindings summarized above. We conclude that the same results hold when we employdifferent clusters of control variables, apply newer methods for handling missing data,control for measurement error, take into account audience self-selection to ideologicallycongruent media outlets, and allow for switching between television networks over thecourse of the campaign.

The final section briefly considers the implications of these findings for the study ofpolitical communication. For the most part, current arguments about the effects oftelevision news coverage on partisan preferences and voting behaviour are based on theexperience of established democracies in the developed world, such as the United States,

2 See H. M. Kepplinger, ‘Visual Biases in Television Campaign Coverage’, Communication Research, 9(1982), 432–46; H. M. Kepplinger, ‘Content Analysis and Reception Analysis’, American Behavioral Scientist,33 (1989), 175–82; Doris A. Graber, ‘Content and Meaning; What’s It All About?’ American Behavioral Scientist,33 (1989), 144–52; Doris A. Graber, ‘Television News without Pictures’, Critical Studies in Mass Communication,4 (1987), 74–8; S. F. Geiger and Byron Reeves, ‘The Effects of Visual Structure and Content Emphasis on theEvaluation and Memory for Political Candidates’, in Frank Biocca, ed., Television and Political Advertising,Vol. 1 (Hillsdale, N.J.: Erlbaum, 1991), pp. 125–44.

Television News, Mexico’s 2000 Elections and Media Effects 3

Canada, England and Germany.3 Conclusions from such studies may not travel well toemerging democracies in the developing world. In these countries, we argue, the potentialscope for media effects on public opinion and voting behaviour is much greater.

MEXICO AND THE MINIMAL EFFECTS PARADIGM

For almost fifty years, scholars in the United States have maintained a healthy scepticismabout the effects of television news coverage on partisan preferences and voting behaviour.Although media exposure might heighten public consideration of particular issues, helpto frame political debate, and shape the opinions of uninformed or undecided voters, theirdirect impact remains fairly limited.4 In particular, the persuasive influence of televisioncoverage during general election campaigns ranges from modest to minimal.5

3 Angus Campbell, Philip E. Converse, Warren E. Miller and Donald E. Stokes, The American Voter (NewYork: John Wiley, 1960); Thomas E. Patterson and Robert McClure, The Unseeing Eye: The Myth of TelevisionPower in National Elections (New York: Putnam, 1976); William J. McGuire, ‘Theoretical Foundations ofCampaigns’, in Ronald E. Rice and Charles K. Atkins, eds, Public Communication Campaigns (London: Sage,1989), pp. 15–40; Richard Johnston, Andre Blais, Henry E. Brady and Jean Crete, Letting the People Decide:Dynamics of a Canadian Election (Stanford: Stanford University Press, 1992); Holli A. Semetko and KlausSchoenbach, Germany’s Unity Election: Voters and the Media (Cresskill, N.J.: Hampton, 1994); Lawrence LeDuc,Richard G. Niemi and Pippa Norris, eds, Comparing Democracies: Elections and Voting in Global Perspective(Thousand Oaks, Calif.: Sage, 1996); and Pippa Norris, John Curtis, David Sanders, Margaret Scammell and HolliSemetko, On Message: Communicating the Campaign (London: Sage, 1999).

4 Thomas Holbrook, Do Campaigns Matter? (Thousand Oaks, Calif.: Sage, 1996); Larry Bartels, ‘UninformedVoters: Information Effects on Presidential Elections’, American Journal of Political Science, 40 (1996), 194–230;David Domke, David D. Fan, Michael Fibison, Dhavab V. Shah, Steve S. Smith and Mark D. Watts, ‘News Media,Candidates and Issues, and Public Opinion in the 1996 Presidential Campaign’, Journalism and MassCommunication Quarterly, 74 (1997), 718–37; Diana C. Mutz, Paul M. Sniderman and Richard A. Brody, eds,Political Persuasion and Attitude Change (Ann Arbor: University of Michigan Press, 1996), esp. Steven H.Chaffee and Rajiv Nath Rimal, ‘Time of Vote Decision and Openness to Persuasion’, pp. 267–92; StephenAnsolabehere, Roy Behr and Shanto Iyengar, ‘Mass Media and Elections: An Overview’, American PoliticsQuarterly, 19 (1991), 109–13; Stephen Ansolabehere, Roy Behr and Shanto Iyengar, The Media Game: AmericanPolitics in the Television Age (New York: Macmillan, 1993); Jon A. Krosnick and Donald R. Kinder, ‘Alteringthe Foundations of Presidential Support through Priming’, American Political Science Review, 84 (1990),497–512; Dan Drew and David Weaver, ‘Media Attention, Media Exposure, and Media Effects’, JournalismQuarterly, 67 (1990), 740–8; William J. McGuire, ‘Theoretical Foundations of Campaigns’, in Ronald E. Riceand Charles K. Atkins, eds, Public Communication Campaigns (London: Sage, 1989), pp. 43–65; Shanto Iyengarand Donald Kinder, News that Matters: Television and American Opinion (Chicago: University of Chicago Press,1987); Carl R. Bybee, Jack M. McLeod, William D. Luetscher and Gina Garramone, ‘Mass Communication andVoter Volatility’, Public Opinion Quarterly, 45 (1981), 69–90; Steven H. Chaffee and Sun Yuel Choe, ‘Time ofDecision and Media Use in the Ford–Carter Campaign’, Public Opinion Quarterly, 44 (1980), 53–69; Lutz Erbring,Edie Goldenberg and Arthur Miller, ‘Front Page News and Real World Cues: A New Look at Agenda-Settingby the Media’, American Journal of Political Science, 24 (1980), 16–49; Steven H. Chaffee and John L.Hochheimer, ‘The Beginnings of Political Communication Research in the United States: Origins of the “LimitedEffects” Model’, in Everett M. Rogers and Francis Balle, eds, The Media Revolution in America and WesternEurope (Norwood, N.J.: Ablex, 1982), pp. 267–96; Elihu Katz, ‘Platforms and Windows: Broadcasting’s Rolein Election Campaigns,’ Journalism Quarterly, 48 (1973), 304–14; Maxwell McCombs and Donald Shaw, ‘TheAgenda-Setting Function of the Mass Media’, Public Opinion Quarterly, 36 (1972), 176–87.

5 Paul Lazarsfeld, Bernard Berelson and Hazel Gaudet, The People’s Choice, 2nd edn (New York: ColumbiaUniversity Press, 1948); Bernard Berelson, Paul F. Lazarsfeld and William N. McPhee, Voting: A Study of OpinionFormation in a Presidential Campaign (Chicago: University of Chicago Press, 1954); Campbell et al., TheAmerican Voter; Patterson and McClure, The Unseeing Eye; Thomas E. Patterson and Robert McClure, The MassMedia Election: How Americans Choose Their President (New York: Praeger, 1980); William J. McGuire, ‘TheMyth of Massive Media Impact: Savagings and Salvagings’, Public Communication and Behavior, 1 (1986),


To be sure, a variety of priming, framing and agenda-setting effects have been foundin Western democracies. But precious few studies have demonstrated persuasively thatexposure to conventional news broadcasts during a campaign can lead to significant,substantial changes in public opinion and voting preferences. As one recent summary ofthe literature put it:

Political scientists still routinely attribute electoral outcomes to structural variables – mostnotably, the state of the national economy and the level of the incumbent president’s popularity– giving short shrift to the specifics of day-to-day campaign events. These are generally viewedas having ‘minimal consequences’ … Perhaps the most fundamental obstacle to understandingthe real-world role of political campaigns is a conceptual limitation on what effects are deemedrelevant. Traditional research has looked mainly at persuasion (i.e., the effect of a campaignon voter preference). Within this definition, the law of minimal consequences has some validity… [Moreover] identifiable traces of persuasion are bound to be minimal because mostcampaigns feature offsetting messages.6

The original arguments for ‘minimal’, ‘modest’ or indirect media influence weredeveloped in the United States and (to lesser extent) other affluent, establisheddemocracies. These countries are characterized by several features that would tend to limitmedia effects on voters’ attitudes and decisions. First, pronounced and enduring partisancleavages render most voters resistant to short-term campaign stimuli.7 In addition,widespread access to different sources of information means that media audiences are notheavily dependent on one-sided cues about political alternatives. Instead, voters rely onsources that are either reasonably balanced or that accord with their existing preferences.8

(F’note continued)

173–257; Johnston et al., Letting the People Decide; Steven E. Finkel, ‘Re-examining the “Minimal Effects”Model in Recent Presidential Campaigns’, Journal of Politics, 55 (1993), 1–21; Semetko and Schoenbach,Germany’s Unity Election; David L. Swanson and Paolo Mancini, eds, Politics, Media, and Modern Democracy:An International Study of Innovations in Electoral Campaigning and their Consequences (New York: Praeger,1996); Maxwell McCombs, Juan Pablo Llamas, Esteban Lopez-Escobar and Federico Rey, ‘Candidate Imagesin Spanish Election: Second-Level Agenda-Setting Effects’, Journalism and Mass Communication Quarterly, 74(1997), 703–17; and Timothy Colton, Transitional Citizens: Voters and What Influences Them in the New Russia(Cambridge, Mass.: Harvard University Press, 2000).

6 Shanto Iyengar and Adam F. Simon, ‘New Perspectives and Evidence on Political Communication andCampaign Effects’, Annual Review of Psychology, 51 (2000), 149–69, at pp. 150–1.

7 Lazarsfeld et al., The People’s Choice; Campbell et al., The American Voter; Philip E. Converse and GeorgesDupeux, ‘Politicization of the Electorate in France and the United States,’ Public Opinion Quarterly, 26 (1962):1–23; David Butler and Donald Stokes, Political Change in Britain (New York: St. Martin’s Press, 1969); HerbertB. Asher, ‘Some Consequences of Measurement Error in Surveys’, American Journal of Political Science, 18(1974), 469–85; Christopher H. Achen, ‘Mass Political Attitudes and the Survey Response’, American PoliticalScience Review, 69 (1975), 1218–31; Donald P. Green and Bradley Palmquist, ‘Of Artifacts and PartisanInstability’, American Journal of Political Science, 34 (1990), 872–902; Jon Krosnick, ‘The Stability of PoliticalPreferences: Comparisons of Symbolic and Nonsymbolic Attitudes’, American Journal of Political Science, 35(1991), 547–76; Eric Schickler and Donald Green, ‘The Stability of Party Identification in Western Democracies’,Comparative Political Studies, 30 (1997), 450–83; R. Prislin, ‘Attitude Stability and Attitude Strength: One IsEnough to Make It Stable’, European Journal of Social Psychology, 26 (1996), 447–77; Donald P. Green andDavid H. Yoon, ‘Reconciling Individual and Aggregate Evidence Concerning Partisan Stability: ApplyingTime-Series Models to Panel Survey Data’, Political Analysis, 10 (2002), 1–24.

8 Jack M. McLeod and Lee B. Becker, ‘The Uses and Gratifications Approach’, in Dan D. Nimmo and KeithR. Sanders, Handbook of Political Communication (Beverly Hills, Calif.: Sage, 1981), pp. 67–100; John R. Zaller,‘The Myth of Massive Media Impact Revived: New Support for a Discredited Idea,’ in Mutz, Sniderman andBrody, eds, Political Persuasion and Attitude Change, pp. 17–78; Holli Semetko, ‘Political Balance on Television:Campaigns in the United States, Britain and Germany’, Harvard International Journal of Press/Politics, 1 (1996),51–71; Russell J. Dalton, Paul Allen Beck and Robert Huckfeldt, ‘Partisan Cues and the Media: Information Flows


These features, of course, are not typical of most of the world’s democracies. In manyemerging democratic systems, partisan identifications tend to be weaker and politicalalternatives less familiar.9 As a result, a greater percentage of the population may beavailable for persuasion over the course of the campaign. The informational context in mostdemocracies, too, is quite different. Outside northern Europe and the Anglophoneimmigrant countries, dominant media outlets are typically controlled by private individualswith clear political preferences (Brazil, Mexico, Peru, Venezuela, etc.), by publicauthorities that systematically favour the incumbent party (Hungary, Czech Republic,Taiwan until 2000, etc.), or by some combination of the two (Russia, Italy, etc.).10 Rarelydo viewers have access to balanced reporting on a state-run network, as in Germany andEngland, or substantial choice between different private broadcasting networks, as in theUnited States. Self-selection to ideologically congruent or unbiased media outlets isconsequently more difficult.

Even in developed democracies, scholars have found more pronounced effects when oneor more of the features that normally limit media influence did not hold. For instance, wherepolitical alternatives are less well known and partisan cues cannot be used to distinguishbetween contenders, as with primary elections and independent candidacies, scholars havetended to find larger persuasive impacts.11 Likewise, researchers have found evidence ofsubstantial media influence where audiences depended heavily on biased sources forinformation, or where reporting fell clearly on one side of an issue.12 One interpretation


in the 1992 Presidential Election’, American Political Science Review, 92 (1998), 111–26; and Diana C. Mutzand Paul S. Martin, ‘Facilitating Communication across Lines of Political Difference: The Role of the MassMedia’, American Political Science Review, 95 (2001), 97–114.

9 James A. McCann and Chappell Lawson, ‘An Electorate Adrift: Public Opinion and the Quality of Democracyin Mexico’, Latin American Research Review, 38 (2003), 60–81.

10 Chappell Lawson, Building the Fourth Estate: Democratization and the Rise of a Free Press in Mexico(Berkeley: University of California Press, 2002); Matthew Wyman, Stephen White and Sarah Oates, Electionsand Voters in Post-Communist Russia (London: Edward Elgar, 1998); Ellen Mickiewicz and Andrei Richter,‘Television, Campaigning and Elections in the Soviet Union and Post-Soviet Russia’, in David L. Swanson andPaolo Mancini, eds, Politics, Media, and Modern Democracy: An International Study of Innovations in ElectoralCampaigning and their Consequences (New York: Praeger, 1996), pp. 107–28; LeDuc et al., ComparingDemocracies; Silvio Waisbord, ‘Television and Election Campaigns in Contemporary Argentina’, Journal ofCommunication, 44 (1994), 125–35; Thomas E. Skidmore, ed., Television, Politics, and the Transition toDemocracy in Latin America (Baltimore/Washington, D.C.: Johns Hopkins University Press/Woodrow WilsonCenter Press, 1993); Slavko Splichal, ‘Media Privatization and Democratization in Central-Eastern Europe’,Gazette, 46 (1992), 3–22.

11 Larry Bartels, Presidential Primaries and the Dynamics of Public Choice (Princeton, N.J.: PrincetonUniversity Press, 1988); Samuel Popkin, The Reasoning Voter: Communication and Persuasion in PresidentialCampaigns (Chicago: University of Chicago Press, 1991); John R. Zaller with Mark Hunt, ‘The Rise and Fallof Candidate Perot: Unmediated versus Mediated Politics. Part 1 of a Two-Part Article’, Political Communication,11 (1994), 357–90; Mark Hunt and John Zaller, ‘The Rise and Fall of Candidate Perot: The Outsider vs. TheSystem. Part 2 of a Two-Part Article’, Political Communication, 12 (1995), 97–123; Zaller, ‘The Myth of MassiveMedia Impact Revived’.

12 Sandra J. Ball-Rokeach and Melvin DeFleur, ‘A Dependency Model of Mass Media Effects’, CommunicationResearch, 3 (1976), 3–21; Sandra J. Ball-Rokeach, M. Rokeach and J. W. Grube, The Great American ValuesTest: Influencing Behavior and Belief through Television (New York: Free Press, 1984); L. Becker and D. C.Witney, ‘Effects of Media Dependencies: Audience Assessments of Government’, Communication Research, 7(1980), 95–120; Richard A Brody, Assessing the President: The Media, Elite Opinion and Public Support(Stanford, Calif.: Stanford University Press, 1991); Benjamin I. Page and Robert Shapiro, The Rational Public:Fifty Years of Trends in Americans’ Policy Preferences (Chicago: University of Chicago Press, 1992), pp. 339–47;John R. Zaller, The Nature and Origins of Public Opinion (New York: Cambridge University Press, 1992); Michael


of the scholarly literature, then, is that media effects on partisan preferences and votingbehaviour may be minimal, modest or moderate depending on the political andinformational context.

Mexico shares with developed democracies certain features that tend to limit mediaeffects. Gradual democratization over the last fifteen years has left Mexicans much morefamiliar with their political alternatives.13 Its three main parties are now quite well known,and partisan attachments – while not as entrenched as in most developed democracies –certainly exist.14 The median voter is literate and at least theoretically has access to differentchannels of political communication.15 Moreover, the media environment itself haschanged markedly since the early 1990s, with the relaxation of official controls and theemergence of a second national television network.16

At the same time, several features of the Mexican political context suggest morepronounced media influences. Although all three major parties were well known in 2000,only one of them had ever held power at the national level. Partisan cleavages wererelatively weak at the start of the 2000 campaign, with only 24 per cent of Mexicansidentifying strongly with any party and 30 per cent declining to express a partisanpreference.17 At the same time, Mexicans rely on a relatively limited number of mediaoutlets for information about politics. Broadcast television dwarfs other traditionalinformation sources – such as radio, newspapers and interpersonal communication – in itsreach, and it surpasses those sources in perceived credibility.18 Moreover, television isitself quite concentrated. In February 2000, for instance, approximately 93 per cent ofMexicans who reported watching television news during the campaign watched one of thetwo main networks, and two-thirds watched one of the two main nightly news programmeson these networks.19 There is thus ample reason to anticipate more pronounced media


B. MacKuen, Robert S. Erikson and James A. Stimson, ‘Peasants or Bankers: The American Electorate and theU.S. Economy’, American Political Science Review, 86 (1992), 597–610; Steven H. Chaffee, ed., PoliticalCommunication (Beverly Hills, Calif.: Sage, 1975).

13 Roderic Ai Camp, Politics in Mexico (New York: Oxford University Press, 1999); Wayne Cornelius,Mexican Politics in Transition: The Breakdown of a One-Party-Dominant Regime (La Jolla: Center forU.S.-Mexican Studies/University of California at San Diego, 1996); Jorge I. Domınguez and James A. McCann,‘Shaping Mexico’s Electoral Arena: The Construction of Partisan Cleavages in the 1988 and 1991 NationalElections’, American Political Science Review, 89 (1995), 34–48; Jorge I. Domınguez and James McCann,Democratizing Mexico: Public Opinion and Electoral Choice (Baltimore, Md.: Johns Hopkins University Press,1996); Jorge I. Domınguez and Alejandro Poire, Toward Mexico’s Democratization: Parties, Campaigns,Elections, and Public Opinion (New York: Routledge, 1999); Domınguez and Lawson, Mexico’s PivotalDemocratic Election; Kathleen Bruhn and Daniel C. Levy with Emilio Zepadua, Mexico: The Struggle forDemocratic Development (Berkeley: University of California Press, 2001); Chappell Lawson, ‘Mexico’sUnfinished Transition: Democratization and Authoritarian Enclaves in Mexico’, Estudios Mexicanos/MexicanStudies, 16 (2000), 267–87.

14 Domınguez and Lawson, Mexico’s Pivotal Democratic Election; Joseph L. Klesner, ‘The Structure of theMexican Electorate: Social, Attitudinal, and Partisan Bases of Vicente Fox’s Victory’, in Domınguez and Lawson,Mexico’s Pivotal Democratic Election, pp. 91–122; McCann and Lawson, ‘An Electorate Adrift?’; Domınguezand Poire, Toward Mexico’s Democratization; Domınguez and McCann, Democratizing Mexico.

15 James A. McCann, ‘The Changing Mexican Electorate: Political Interest, Expertise, and Party Support inthe 1980s and 1990s’, in Monica Serrano, ed., Governing Mexico: Political Parties and Elections (London:Institute of Latin American Studies, 1998), pp. 15–37.

16 Lawson, Building the Fourth Estate.17 Mexico 2000 Panel Study; McCann and Lawson, ‘An Electorate Adrift?’; Klesner, ‘The Structure of the

Mexican Electorate’.18 Chappell Lawson, ‘Television Coverage, Vote Choice, and the 2000 Campaign’, in Domınguez and Lawson,

eds, Mexico’s Pivotal Democratic Election, pp. 187–209.19 Mexico 2000 Panel Study.


effects in Mexico than have so far been found in the United States, Canada, Australia andWestern Europe.

TELEVISION COVERAGE IN 2000

Where audiences are exposed to roughly the same signals, media effects may be difficultto detect. Where different segments of the electorate receive divergent cues, however, largeeffects may be more readily apparent.20 One important step in analysing media effects,therefore, is to find adequate variation in media exposure.

In Mexico, the long-dominant Televisa network has frequently been criticized for biasin favour of the ruling party.21 News on the smaller, newer Television Azteca network, bycontrast, is generally seen as more balanced, at least with regard to the conservativeNational Action Party (PAN). Both networks are regarded as more hostile towards theleftist Party of Democratic Revolution (PRD), although the extent of this hostility hasvaried over time. Finally, coverage over the course of the 2000 campaign appears to haveshifted following Labastida’s poor performance in the first televised debate in late April.In the wake of that event, the PRI put on intense pressure to accord greater coverage toFox’s rivals, and some broadcasters capitulated to these pressures.22 Assuming that theseimpressions of television coverage are correct, it should be possible to identify segmentsof the electorate who received quite different media cues, either chronically or at specificperiods, during the 2000 campaign.

Perhaps the simplest way to evaluate coverage on Mexico’s main networks is bymeasuring the volume of airtime devoted to each candidate. Table 1 presents two differentinventories of television news coverage for the three major candidates – Labastida (PRI),Fox (of the Alliance for Change, dominated by the PAN), and Cardenas (of the Alliancefor Mexico, dominated by the PRD) – on the two national networks from the officialbeginning of the campaign in February 2000 to the end of June that same year.23 The firstset of columns shows IFE figures for the major nightly news programme on each network.The second shows data from Reforma newspaper’s monitoring of all daily coverage on the

20 Zaller, ‘The Myth of Massive Media Impact Revived’.21 Lawson, Building the Fourth Estate; Mony de Swaan, Carolina Gomez and Juan Molinar-Horcasitas,

‘Medios y objetividad’, Milenio Diario, 25 June 2000; Luis Guillermo Hernandez, ‘Piden equidad en medios’,Reforma, 4 May 2002; Guadalupe Irızar, ‘Es equidad tema central en Televisa’, Reforma, 9 May 2000; DanielC. Hallin, ‘Media, Political Power and Democratization in Mexico’, in James Currant and Myung-Jin Park, eds,De-Westernizing Media Studies (London: Routledge, 2000), pp. 97–110; William A. Orme Jr, ed., A Culture ofCollusion: An Inside Look at the Mexican Press (Miami, Fla.: North-South Center, University of Miami and TheCommittee to Protect Journalists, 1997); Ilya Adler, ‘The Mexican Case: The Media in the 1988 PresidentialElection’, in Skidmore, ed., Television, Politics, and the Transition to Democracy in Latin America, pp. 145–74;Pablo Arredondo-Ramırez, Gilberto Fregoso-Peralta and Raul Trejo-Delarbre, Ası se callo el sistema:Comunicacion y elecciones en 1988 (Guadalajara: Universidad de Guadalajara, 1988); Raul Trejo-Delarbre, ed.,Las redes de Televisa (Mexico City: Claves Latinoamericanas, 1988); Fernando Mejıa-Barquera and RaulTrejo-Delarbre, Televisa: el quinto poder (Mexico City: Claves Latinoamericanos, 1985).

22 First author’s interviews with journalists, academics, electoral authorities and political advisers in MexicoCity, May–June 2000; Ramon Sevilla-Turcios, ‘Alertan sobre favoritismo en televisoras’, Reforma, 27 April 2000;Mark Stevenson, ‘Pressure Increases on Mexican Media’, Associated Press, 20 June 2000; Sallie Hughes andChappell Lawson, ‘Propaganda and Crony Capitalism: Partisan Bias in Mexican Television News’, LatinAmerican Research Review (forthcoming).

23 The table excludes data for the three minor presidential candidates – Manuel Camacho, Porfirio Munoz Ledoand Gilberto Rincon Gallardo – who collectively garnered less than 3 per cent of the vote.


TABLE 1 Volume of Television Coverage, by Network over Time

IFE Reforma

Candidate (coalition/party) Televisa Azteca Televisa Azteca

Labastida (PRI), Feb–April 33% 38% 36% 38%Labastida (PRI), May–June 32% 33% 30% 34%Labastida (PRI), total 33% 36% 34% 35%

Fox (AC/PAN), Feb–April 38% 42% 39% 38%Fox (AC/PAN), May–June 37% 41% 33% 37%Fox (AC/PAN), total 37% 40% 37% 39%

Cardenas (AM/PRD), Feb–April 29% 20% 25% 24%Cardenas (AM/PRD), May–June 31% 26% 37% 30%Cardenas (AM/PRD), total 30% 24% 30% 26%

Total seconds, Feb–April 17,568 22,455 55,433 56,181Total seconds, May–June 22,020 22,474 54,914 62,181Total seconds 39,588 44,929 110,347 118,362

two major networks.24 Because of differences in the programming analysed, the figuresare not identical. Both, however, reveal the same basic tendency: major-party candidatesreceived roughly comparable shares of airtime throughout the campaign. Fox received themost coverage, followed by Labastida and then Cardenas, on both networks throughoutthe campaign. Discrepancies between the two measures tend to cancel each other out; theonly clear trend in both is increased attention to Cardenas during the second half of therace. Thus, considering only the volume of coverage, one would not perceive significantbiases in television coverage nor expect major differences in patterns of attitude changeacross viewers of the two networks. Nor would one anticipate major changes in attitudesover time, with the possible exception of slightly increased support for Cardenas towardsthe end of the campaign.

Of course, focusing exclusively on time may obscure important differences in the toneof news reports. To assess tone in a systematic way, we conducted a more detailed analysisof campaign coverage in a randomly selected sample of forty broadcasts from the two mainnightly news programmes (twenty from each) during the period 18 February to 1 July2000.25 In addition to measures such as time devoted to different candidates and verbalcharacterizations of the main contenders, our content analysis devoted substantial attentionto visual tone of campaign coverage. To this end, we recorded and evaluated a separate‘image’ each time the camera angle changed.26 Each image was classified into one ofthirteen categories: (1) pan or aerial shots of a large supportive crowd; (2) the candidate

24 Reforma’s monitoring was conducted for twelve hours every day for three different periods: early morning(6 through 10 a.m.), early afternoon (1 through 4 p.m.), and evening prime time (7 through 11 p.m.). Every typeof programme broadcast in the corresponding time and channels was monitored: news programmes, sports, soapoperas, entertainment shows, etc.

25 This period was selected to coincide with the timing of surveys for the Mexico 2000 Panel Study. Daysselected included: 25 February, 3, 13, 17, 20, 22 and 27 March, 13, 17 and 25 April, 2, 4, 8 and 12 May, 1, 7,13, 14, 28 and 30 June. Full results from the content analysis are available from the first author upon request.

26 All ‘images’ were treated the same, regardless of length. The number of different images per story rangedfrom one to forty.


with a supportive crowd; (3) the candidate accompanied by a popular personality, such asan entertainer, clergyman or athlete; (4) an interview with the candidate by an agreeableinterviewer; (5) an interview with a supporter of the candidate; (6) the candidatesurrounded by reporters; (7) the candidate alone; (8) the candidate speaking to a passiveaudience, typically in a formal setting; (9) a disturbance at the candidate’s event; (10) thecandidate with an unsupportive crowd or heckler; (11) an interview with the candidate byan aggressive or hostile interviewer; (12) an interview with a critic of the candidate; and(13) other shots during campaign coverage.27 These types of images were then combinedinto positive, negative and neutral categories. Pan or aerial shots of a large supportivecrowd, shots of the candidate with a supportive crowd, images of the candidateaccompanied by a popular personality, interviews of the candidate by an agreeableinterviewer and interviews with a supporter of the candidate were coded as positive images.The reverse of these categories were coded as negative images (the exception being thatthere were no images of candidates posing with generally detested or unpopular figures).Images of the candidate alone, the candidate surrounded by reporters, or passive audienceslistening to the candidate’s speech were coded as neutral; ‘other’ images were excluded.The summary measure ‘net positive images’ represented positive images minus negativeimages, divided by total images (including neutral images but excluding ‘other’ images).28

Although this measure is hardly comprehensive, it captures a crucial element oftelevision coverage. This measure of tone also confirms anecdotal impressions about therelative treatment of different candidates on Mexican television at different stages in thecampaign. For instance, Television Azteca was widely regarded as more sympathetictowards Fox than was Televisa, which was generally seen as more sympathetic towardsthe ruling party. Likewise, various observers noted that coverage of Labastida andCardenas improved during the second half of the race.29 All of these patterns show up inour data.

Trends in visual tone of coverage largely paralleled those in verbal tone – that is, positiveand negative references to the main candidates.30 Although verbal characterizations were

27 ‘Other’ images included shots of a landscape, an unknown speaker, the president (not with the candidate),etc. Collectively, ‘other’ images accounted for approximately 16 per cent of the total, most of which were suigeneris.

28 To maximize the validity and reliability of the data, the coding scheme was piloted three times – and amendedeach time to clarify coding criteria – on news broadcasts not included in the sample. The principal coder and thefirst author then separately recorded the number of positive, negative and neutral images for each of the three maincandidates in a randomly selected news broadcast. Inter-coder reliability was measured by the following formula:[1 � [(number of images coded by first coder) � (number of images coded by second coder)]/[(number of imagescoded by both coders)/2]. In this test, average inter-coder reliability across all nine categories (positive imagesof Labastida, negative images of Labastida, neutral images of Labastida, positive images of Fox, negative imagesof Fox, neutral images of Fox, positive images of Cardenas, neutral images of Cardenas, negative images ofCardenas) exceeded 90 per cent. After the coding was complete, two individuals who had not participated in theoriginal coding process were given basic instructions and asked to record the number of positive, negative andneutral images of each candidate in the same randomly selected broadcast. Inter-coder reliabilities between thetwo new coders averaged 75 per cent across all categories. Reliabilities between each new coder and the primarycoder were higher than reliabilities between the new coders, as were reliabilities for the summary measure ‘netpositive images’ for each candidate.

29 Hughes and Lawson, ‘Propaganda and Crony Capitalism’; Lawson, ‘Television Coverage, Vote Choice, andthe 2000 Campaign’.

30 Our analysis coded included all positive and negative references to the three main candidates in the courseof a broadcast, regardless of the source, other than references to the candidates by anchors or reporters that weredevoid of editorial content (for instance, when they were introducing a candidate’s campaign stop that day).


TABLE 2 Tone of Coverage, by Network over Time

Labastida Fox Cardenas

Net positive coverage on Azteca, February–April 27% 41% 31%Net positive coverage on Azteca, May–June 41% 50% 52%Net positive coverage on Azteca, total 36% 46% 44%

Net positive coverage on Televisa, February–April 37% 45% 7%Net positive coverage on Televisa, May–June 74% 25% 71%Net positive coverage on Televisa, total 59% 36% 42%

Total images on Azteca, February–April 91 111 88Total images on Azteca, May–June 155 147 114Total images on Azteca 246 258 202

Total images on Televisa, February–April 57 82 56Total images on Televisa, May–June 81 71 68Total images on Televisa 138 153 124

Note: Days included in the February–April sample were: 25 February, 3, 13, 17, 20, 22 and 27 March, 13, 17 and25 April. Days included in the May–early June period were 2, 4, 8 and 12 May, 1, 7, 13, 14, 28 and 30 June.Uncategorized (‘other’) images are excluded.

substantially less positive towards all of the candidates than were visual images, the relativetone of coverage for the main candidates was similar for both measures.31 We thus feelcomfortable treating visual cues as a rough indicator of relative coverage of the maincandidates, which could be used to predict changes in viewers’ relative opinions of thecandidates.

Overall tone of coverage was similar for the three candidates: the percentage of netpositive images over the course of the campaign was 44 per cent for Labastida, 42 per centfor Fox, and 43 per cent for Cardenas. These aggregate data, however, mask importantdifferences in coverage between the two networks. The visuals that supported Labastida,it turns out, came disproportionately from Televisa. By contrast, Television Aztecacoverage remained more favourable towards Fox than towards the PRI candidate.Coverage of the Left, meanwhile, was approximately the same on both networks.

Also noteworthy are changes in the tone of coverage over the course of the campaign.On Television Azteca, coverage of all three candidates shifted slightly, but the differencewas not dramatic. From February through April, coverage was 27 per cent net positive forLabastida, 41 per cent net positive for Fox, and 31 per cent net positive for Cardenas; thecorresponding figures for May–June were 41 per cent, 50 per cent and 52 per cent. Changes

31 For instance, Labastida fared poorly compared to Fox on Television Azteca in the first half of the campaign,and he fared poorly compared to both opposition candidates on that network in the second. Likewise, verbalcharacterizations of Cardenas were particularly harsh on Televisa during the first half of the race but moderatedsubstantially in the second half. Had we used only verbal cues as our measure of media content, we would stillhave found a media-content-basis for the large, significant attitude changes that we document in our analysis ofthe panel data. In theory, we could use either verbal or visual measures. In practice, however, there were far fewercoded verbal references to the candidates in our sample than there were coded images – only 191, in contrast to1,121 images. Because these references were divided across three candidates, two networks and two time periods,there were an average of approximately sixteen references per ‘cell’, with the smallest cells having just eightreferences. Consequently, we felt that visual characterizations offered equally valid and more reliable estimatesof tone.


in the tone of coverage on Televisa, however, were pronounced. Net positive images ofLabastida on the country’s largest network soared from a respectable 37 per cent to awhopping 74 per cent. The shift in favour of Cardenas was even more marked, from 7 percent net positive to almost 71 per cent. Meanwhile net positive images of Fox on Televisa’snightly news programme dropped from 45 per cent to 25 per cent.32 These findings areshown in Table 2.

In summary, content analysis reveals fairly even-handed coverage of the three maincandidates on television overall. Biases in the volume and tone of coverage were modestover the course of the 2000 presidential campaign. Nevertheless, it is possible to identifydifferences in coverage of the main candidates on different networks and at different timesduring the race.

MEDIA EFFECTS IN 2000

Attitudes Towards the Main Candidates

We now turn to whether the patterns identified by content analysis were reflected in theattitudes expressed by television viewers. As indicated in Table 2, Televisa’s coverage ofCardenas in the early campaign period was harsh in comparison to coverage of Fox andLabastida; this may have hurt the PRD leader’s standing relative to that of his rivals. InJune, however, the network reversed course: Cardenas received both more and bettercoverage – roughly equivalent to the coverage received by Labastida – while coverage ofFox deteriorated. If media coverage had an impact on public opinion, Cardenas (andLabastida) should have enjoyed a boost relative to Fox among Televisa viewers in thesecond half of the race. On Television Azteca, meanwhile, coverage was less favourabletowards Labastida compared with both his main rivals. This was especially true in thesecond half of the race. Assuming that this pattern of coverage was reflected at the masslevel, Television Azteca viewers should have downgraded their assessments of Labastidarelative to Fox and Cardenas.

To assess changes in viewers’attitudes, we use survey data from the first three wavesof the Mexico 2000 Panel Study.33 The first wave was administered in February 2000 toa randomly selected national sample of 2,400 Mexicans; it was timed to coincide with theofficial start of the general election campaign. The second wave was conducted in lateApril–early May, on a subset of respondents chosen at random (N � 952). In June, withjust two weeks left in the campaign, a third questionnaire was administered to a randomlychosen group of participants from the second survey wave plus all available respondentsfrom the initial sample who had not been interviewed in the second wave (a total of 974).34

32 These shifts and the relative tone of coverage towards the main candidates are directionally similar if weexclude the second half of June from our sample, though the magnitude of the changes in coverage on Televisais somewhat smaller.

33 Approximately half of the respondents in the first round of the panel reported watching Televisa news at leastoccasionally, the bulk of whom watched three to five days per week. Just over one third of the sample reportedwatching some news on Television Azteca. Thirteen per cent of the respondents watched news programming onboth networks, and approximately 30 per cent never watched either. Average exposure levels in later waves wasslightly higher on Television Azteca and about the same on Televisa.

34 By varying the number of times respondents were contacted, the study sought to cut down oninterview-induced bias and panel attrition. On the potential for interviewing to cause changes in respondentattitudes and behaviour, see R. Gary Bridge, Leo G. Reeder, David Kanouse, Donald R. Kinder and Vivian Tong


By using panel data, we take into account the fact that audiences might self-select todifferent media outlets. For instance, it may be that opponents of the PRI disproportionatelyfavoured Television Azteca news, while sympathizers with the ruling party disproportion-ately remained with Televisa.35 If so, cross-sectional data alone would not allow us todistinguish between initial audience predispositions and the effects of exposure to differentnews programmes. Panel data, however, allow us to assess changes over time in the sameindividual. We can thus control for respondents’ views towards the main candidates (andother factors that might influence their vote) at the beginning of the period in question.Largely for this reason, panel studies are the preferred method for analysing media effectsin the mass public.36

To examine the effect of the two networks on relative evaluations of the main candidates,we calculated attitude differentials based on eleven-point ‘feeling thermometer’ scores.37

Thus we subtracted feeling thermometer ratings of Fox from analogous ratings ofLabastida to produce a Fox–Labastida score that theoretically ranged from � 20 to � 20.38

We then regressed these attitude differentials on self-reported television exposure from thesame wave,39 plus lagged attitude differentials from the previous wave.40 The addition ofthe lagged dependent variable on the right-hand side makes the model explicitly dynamic,allowing us to ascertain whether certain factors changed citizens’ initial judgements.

In theory, the use of panel data with lagged dependent variables should control forrespondents’ attitudes at the start of the campaign. It is possible, however, that respondentsnot only held different initial attitudes towards the main candidates, but also had differentpredispositions to change their attitudes about these candidates. If respondents’ pro-


Nagy, ‘Interviewing Changes Attitudes – Sometimes’, Public Opinion Quarterly, 41 (1977), 56–64; and JamesDowning, Charles M. Judd and Markus Brauer, ‘Effects of Repeated Expressions on Attitude Extremity’, Journalof Personality and Social Psychology, 63 (1992), 17–29.

35 On the other hand, it is possible that choice of news programmes was primarily determined by preferencefor entertainment programmes that preceded and followed the news, rather than by political preferences. In thatcase, we would expect relatively little self-selection to ideologically congruent broadcasts.

36 See Jeffrey Wooldridge, Introductory Econometrics: A Modern Approach (Mason, Ohio: South-WesternCollege, 2000), chap. 9.

37 Feeling thermometer ratings were obtained from the following item: ‘[Show card] I am going to ask youropinion about political parties and candidates for President. On this scale, zero indicates that your opinion is verybad and ten that your opinion is very good. If you don’t have an opinion, just tell me and we’ll go on to the nextone. What is your opinion of …? [READ AND ROTATE].’ Options were ‘PRI, PAN, PRD, Francisco Labastida,Vicente Fox, Cuauhtemoc Cardenas’.

38 These attitude differentials better anticipate our subsequent model of vote choice, which involves relativecomparisons between the three main candidates. Mean feeling thermometer differentials ranged from � 0.30(Labastida minus Fox in June) to 1.82 (Fox minus Cardenas in February); standard deviations ranged from 4.58(Labastida minus Fox in June) to 3.43 (Fox minus Cardenas in April/May).

39 Television news exposure was measured using the following item: ‘Do you watch any television newsprogram? [IF YES] Which one? Approximately how many days a week do you watch it?’ Interviewers wereinstructed to mark up to two programmes. Respondents who reported watching a particular programme six or sevendays a week were treated as watching five days a week. Those who reported watching two programmes on thesame network were treated as having been doubly exposed; i.e., their self-reported exposure to one programmewas added to their self-reported exposure to the other programme. For both networks in all three waves, over 93per cent of respondents were coded as watching five or fewer days per week. Mean exposures for Television Aztecaranged from 1.43 in February (with standard deviation of 2.41) to 1.86 in April/May (with standard deviation of2.56); mean exposures for Televisa ranged from 1.90 in July (with standard deviation of 2.54) to 2.34 in April/May(with standard deviation of 2.57).

40 In formal terms, the model is: Candidate Attitude Differentialt � � � �1 Televisa Exposuret � �2 AztecaExposuret � �3 Candidate Differentialt - 1 � �4 – 21 Controls.


pensities to alter their relative assessments of the candidates were correlated with newsexposure, excluding them could lead us to make spurious inferences about the effects oftelevision.41 For instance, in comparison to the Televisa audience, Television Aztecaviewers are somewhat better educated, more affluent and more likely to live in urban areas.More educated, affluent urbanites may not only be more inclined to favour Fox overLabastida initially; they may also be inclined to favour Fox over Labastida increasinglyduring the course of the campaign, regardless of their exposure to television. To take intoaccount such crystallization effects, we include the main demographic variables thoughtto affect partisan preferences in Mexico:42 age, gender, education, socio-economic status,43

union membership, church attendance, region of the country in which the respondentresided and whether the respondent lived in a city.

A similar point could be made with respect to partisanship. For this reason, we alsoincluded dummy variables for the following categories: PRI identifiers or leaners, strongPRI identifiers, PAN identifiers or leaners, strong PAN identifiers, PRD identifiers orleaners and strong PRD identifiers.44 Finally, in order to isolate the effects of television,we added controls for other types of campaign cues that might be correlated with bothtelevision exposure and attitude change. These include respondents’ self-reported attentionto the campaign and the frequency with which they reported discussing politics.45

Table 3 shows the results of all three regressions on attitude differentials in lateApril–early May. Table 4 reports the same results for attitude differentials in June. Theresults for television exposure conform with expectations from content analysis. Early inthe campaign, those who watched newscasts on Televisa grew more negative towards theLeft relative to both of the other main candidates; this effect was not evident for exposureto Television Azteca, which accorded Cardenas coverage that was roughly equivalent intone to that of the other two contenders. Different patterns of coverage during the secondhalf of the campaign produced the opposite results. Television Azteca’s coverage ofCardenas in May–June was similar to coverage of Fox, and Televisa’s coverage ofCardenas during that period was similar to its coverage of Labastida; in neither of thesecases did exposure exercise a clear effect on public opinion. By contrast, Television Aztecaclearly favoured Cardenas over Labastida, and Televisa’s coverage clearly favoured himover Fox; in both of these cases, the influence of television exposure proved statisticallysignificant.

One puzzling finding from this analysis concerns the lack of a clear effect from exposureto Televisa on relative attitudes towards Fox and Labastida in the second half of the race.

41 Measurement error could also produce what appear to be crystallization effects. The issue of measurementerror is discussed further below.

42 Domınguez and Lawson, Mexico’s Pivotal Democratic Election; Domınguez and Poire, Toward Mexico’sDemocratization; Monica Serrano, ed., Governing Mexico: Political Parties and Elections (London: Institute ofLatin American Studies, 1998); Domınguez and McCann, Democratizing Mexico.

43 We measured socio-economic status through a factor score based on: (1) interview coding of the respondent’sdwelling on a five-point scale; and (2) number of light bulbs in the respondent’s house. Using different measuresof socio-economic status does not materially change our findings about media influence.

44 Partisan identification was measured using the following item: ‘In general, do you consider yourself priısta,panista or perredista? [If R has affiliation:] Do you consider yourself very priısta/panista/perredista or somewhatpriısta/panista/perredista? [If R has no affiliation]: Toward which party do you most lean?’

45 Some of the demographic controls mentioned above might also be construed as proxies for certain campaignsignals. For instance, church attendance is an indicator of exposure to electoral cues from religious authorities,and regional dummies may capture differences in partisan mobilization and canvassing. See Domınguez andLawson, Mexico’s Pivotal Democratic Election, pp. 34–5, 81–3, 200.


TA

BL

E3

Eff

ects

onC

andi

date

Att

itud

eD

iffe

rent

ials

,A

pril

/May

Lab

asti

da–F

oxL

abas

tida

–Car

dena

sF

ox–C

arde

nas

Var

iabl

e�

s.e.

�s.

e.�

s.e.

Con

stan

t1.

230.

939

1.17

0.91

0�

0.14

50.

797

Nor

ther

nre

gion

�0.

082

0.39

10.

091

0.37

90.

165

0.33

2M

etro

regi

on0.

023

0.40

90.

399

0.39

60.

319

0.34

6C

entr

alre

gion

�0.

166

0.38

20.

081

0.37

00.

288

0.32

4

Age

�0.

024*

0.01

0�

0.01

8†0.

009

0.00

70.

008

Mal

e�

0.15

60.

267

�0.

440†

0.25

9�

0.26

90.

227

Urb

anre

side

nt�

0.09

00.

313

0.08

00.

303

0.20

10.

266

Edu

cati

on�

0.30

9*0.

136

�0.

121

0.13

20.

187

0.11

5R

elig

iosi

ty0.

133

0.11

50.

045

0.11

1�

0.07

30.

098

Uni

onm

embe

r0.

422

0.40

30.

203

0.39

0�

0.22

90.

342

Inco

me

prox

y�

0.04

60.

143

0.02

40.

139

0.05

30.

121

Wea

kP

RI

part

isan

orle

aner

(Feb

.)1.

30**

0.40

01.

43**

0.39

00.

047

0.33

2W

eak

PA

Npa

rtis

anor

lean

er(F

eb.)

�1.

14**

0.41

2�

0.42

50.

388

0.85

4*0.

349

Wea

kP

RD

part

isan

orle

aner

(Feb

.)�

0.01

50.

544

�1.

24*

0.53

4�

1.19

*0.

470

Str

ong

PR

Ipa

rtis

an(F

eb.)

2.05

**0.

467

2.11

**0.

455

�0.

081

0.37

7S

tron

gP

AN

part

isan

(Feb

.)�

1.33

*0.

633

�0.

306

0.59

51.

21*

0.53

4S

tron

gP

RD

part

isan

(Feb

.)0.

233

0.83

0�

1.82

*0.

834

�2.

14**

0.72

9

Att

itud

edi

ffer

enti

al(F

eb.)

0.40

5**

0.04

10.

376*

*0.

036

0.37

1**

0.03

7T

alk

abou

tpo

liti

cs(A

pril

/May

)�

0.08

40.

135

�0.

077

0.13

1�

0.00

50.

114

Cam

paig

nat

tent

ion

(Apr

il/M

ay)

0.08

60.

176

�0.

046

0.17

0�

0.14

10.

149

Azt

eca

expo

sure

(Apr

il/M

ay)

0.04

70.

056

�0.

033

0.05

4�

0.07

60.

048

Tel

evis

aex

posu

re(A

pril

/May

)0.

040

0.05

40.

129*

0.05

20.

091*

0.04

5

N74

673

474

1A

djus

ted

R2

0.34

0.35

0.26

F-s

tati

stic

19.3

19.8

13.3

Not

e:P

airw

ise

dele

tion

empl

oyed

.��

Est

imat

edco

effi

cien

ts;

s.e.

�st

anda

rder

rors

.†S

igni

fica

ntat

0.10

.*S

igni

fica

ntat

0.05

.**

Sig

nifi

cant

at0.

01.


TABLE 4 Effects on Candidate Attitude Differentials, June

Labastida–Fox Labastida–Cardenas Fox–Cardenas

Variable � s.e. � s.e. � s.e.

Constant 1.54 1.35 1.10 1.14 � 0.323 1.25

Northern region � 0.078 0.561 � 0.913† 0.473 � 0.795 0.521Metro region 0.485 0.586 � 0.785 0.495 � 1.23* 0.544Central region � 0.315 0.547 � 0.837† 0.462 � 0.518 0.508

Age � 0.002 0.014 0.009 0.012 0.010 0.013Male � 0.019 0.385 0.101 0.325 0.104 0.357Urban resident � 0.814† 0.452 � 0.445 0.381 0.382 0.419Education � 0.101 0.203 0.258 0.171 � 0.157 0.189Religiosity � 0.042 0.165 � 0.199 0.139 � 0.153 0.153Union member 0.686 0.578 0.285 0.487 � 0.322 0.536Income proxy � 0.277 0.206 � 0.220 0.173 0.060 0.191

Weak PRI partisan orleaner (Feb.) 0.185 0.572 0.711 0.486 0.662 0.521

Weak PAN partisan orleaner (Feb.) � 1.07† 0.583 0.687 0.484 1.64** 0.541

Weak PRD partisan orleaner (Feb.) � 0.908 0.780 � 1.70* 0.665 � 0.962 0.733

Strong PRI partisan (Feb.) 2.08** 0.667 2.08** 0.568 0.228 0.596Strong PAN partisan (Feb.) � 2.50** 0.895 0.416 0.747 2.84** 0.835Strong PRD partisan (Feb.) � 2.80* 1.19 � 4.16** 1.02 � 1.54 1.13

Attitude differential (April/May) 0.544** 0.050 0.598** 0.044 0.546** 0.056

Talk about politics (June) 0.002 0.185 0.141 0.157 0.148 0.172Campaign attention (June) 0.049 0.260 0.267 0.220 0.181 0.242

Azteca exposure (June) � 0.080 0.082 � 0.124† 0.069 � 0.041 0.076Televisa exposure (June) 0.088 0.071 � 0.062 0.060 � 0.160* 0.067

N 316 315 316Adjusted R2 0.49 0.57 0.36F-statistic 15.5 21.2 9.54

Notes: Pairwise deletion employed; � � estimated coefficients; s.e. � standard errors.† Significant at 0.10.* Significant at 0.05.** Significant at 0.01.

To be sure, the coefficient has the expected sign; although it is not significantly differentfrom zero (p � 0.22), it is significantly different from the effect of exposure to TelevisionAzteca. Nevertheless, it appears that even quite biased coverage on Televisa had only alimited impact on relative evaluations of the ruling party’s candidate. (We return to thisissue when discussing the effects of Televisa exposure on vote choice.)

Coefficients for the control variables in Tables 3 and 4 suggest that attitudes tended tocrystallize along expected lines over the course of the campaign. The more educatedincreasingly favoured the opposition, especially the PAN, over the ruling party. Likewise,initial supporters of the PRI liked Labastida increasingly more relative to his main rivals.Nevertheless, these crystallization effects attenuated substantially as the campaignprogressed. By June, even partisan predispositions exercised only a limited effect on manyattitude differentials.


The effects of television news exposure were quite powerful relative to the impact ofother variables. Standardized coefficients for news exposure were consistently larger thanthose for generalized attention to the campaign or interpersonal communication aboutpolitics (which were never significant) and for the bulk of the demographic controls. Ingeneral, television effects approximated the crystallizing influence of education andreached about half those of partisan predispositions. Moreover, the effects of televisioncoverage remained strong in the second half of the race, after crystallization effects beganto fade.

All told, these results demonstrate that television coverage had substantial andsignificant effects on attitudes towards the main candidates in Mexico’s 2000 presidentialrace. There was thus an important linkage between the tone of news coverage to whichrespondents were exposed (on the one hand) and respondents’ opinions (on the other).Candidates who received relatively favourable coverage tended to prosper in relativeterms; candidates who did not tended to suffer compared with their main competitors.

Vote Choice

So far, our analysis has focused on political attitudes: specifically, relative assessments ofthe main candidates at different points in the campaign. We have not yet directly addressedthe issue of political behaviour – in this case, how respondents actually voted on electionday. Although it seems likely that attitude change would find expression in votingbehaviour, it is also conceivable that the effects of media exposure on mass opinion couldhave failed to shape voting patterns. For instance, changes in relative candidate evaluationsmight not have provoked corresponding shifts in electoral outcomes if these changes wereconfined to non-voters or to voters who had already made up their minds. It is also possiblethat other factors (such as election-day mobilization) could have reversed the effects oftelevision coverage on candidate preference. Finally, factors like vote-buying, coercionand heavy-handed clientelism could have prevented the translation of attitude change intoshifts in electoral behaviour.

To assess the impact of television coverage on how citizens actually voted, we makeuse of the fourth wave of the Mexico 2000 Panel Study, collected in the week followingthe 2 July election. Among those respondents who reported voting, 47 per cent chose Fox,38 per cent backed Labastida and 14 per cent favoured Cardenas. This breakdown reflectsfairly well the official tally (43 per cent, 36 per cent, 17 per cent, respectively).46 Thus,although levels of self-reported turnout were substantially higher in the post-electoral waveof the panel, the partisan division of the vote was similar to that of the broader electorate.47

Table 5 presents the results from a multinomial logistic regression, where vote choicewas modelled as a function of presidential voting preference in the first survey wave

46 Vote in July was measured by the following question: ‘Did you vote in the elections of July 2? [If yes:] Couldyou mark on this ballot for whom you voted in the elections for President of the Republic? [Hand ballot and requestthat R deposit it in the box.]’ Vote preference in February was measured using the following item: ‘Now for thepurposes of this survey let’s suppose that today is election day and you are going to vote for President of theRepublic. For this I am going to give you a piece of paper where you can mark your response without me seeingit and afterwards deposit it in this box. If the elections for President of the Republic were today, for whom wouldyou vote?’ [Respondents were then handed a sample ballot.] In both February and July, those who voted for oneof the tiny leftist parties were grouped with Cardenas.

47 As is typical in post-electoral surveys, self-reported voter participation was inflated in our data. The actualrate reported by the IFE was 64 per cent; in the Mexico 2000 Panel Study, it was 81 per cent.


TABLE 5 Television Exposure and Vote Choice in July

Fox v. Cardenas/ Non-voterLabastida Other v. Labastida v Labastida

Variable � s.e. � s.e. � s.e.

Constant 0.420 0.703 � 0.892 0.939 3.18** 0.877

Northern region � 0.008 0.263 � 0.445 0.378 0.832* 0.327Metro region � 0.190 0.289 0.232 0.367 0.319 0.367Central region 0.024 0.261 � 0.448 0.362 0.671* 0.323

Age � 0.011 0.007 0.001 0.010 � 0.045** 0.010Male 0.024 0.197 0.342 0.262 0.402† 0.237Urban resident 0.407† 0.221 � 0.155 0.300 0.275 0.259Education � 0.070 0.108 0.092 0.146 � 0.221 0.138Religiosity 0.026 0.082 0.117 0.108 0.067 0.097Union member � 0.640* 0.283 � 0.077 0.371 � 0.780* 0.387Income proxy 0.095 0.115 � 0.094 0.165 0.048 0.144

Vote for Labastida (Feb.) � 1.02** 0.308 � 1.25** 0.422 � 1.29** 0.353Vote for Fox (Feb.) 1.54** 0.371 0.484 0.470 � 0.212** 0.445

Vote for other candidate(Feb.) 0.257 0.451 1.53** 0.476 � 0.772 0.539

PRI partisan or leaner (Feb.) � 1.17** 0.255 � 1.32** 0.369 � 0.921** 0.313PAN partisan or leaner (Feb.) 0.171 0.366 � 0.536 0.493 0.067 0.449PRD partisan or leaner (Feb.) � 0.748† 0.453 0.396 0.450 0.018 0.510

Talk about politics (Feb.–July) 0.071 0.137 0.070 0.183 � 0.166 0.168Campaign attention (Feb.–

July) 0.008 0.174 � 0.056 0.236 � 0.571** 0.218Azteca exposure (Feb.–July) 0.129* 0.053 0.137* 0.069 0.079 0.066Televisa exposure (Feb.–July) 0.009 0.042 0.006 0.057 0.019 0.051

Valid N 1,062Pseudo-R2 (Cox and Snell) 0.26Chi-squared statistic (60 d.f.) 704

Notes: � � estimated coefficients; s.e. � standard errors.† Significant at 0.10.* Significant at0.05.** Significant at 0.01.

(February) and average exposure to network news over the course of the campaign.48 Asin our analysis of attitude differentials, we include a full complement of control variables.To make the main campaign controls comparable to television exposure, attention to thecampaign and interpersonal communication about politics were averaged over all threepre-election waves.

The results in the first column show the impact of early presidential preferences andtelevision news on choosing Fox over Labastida; those in the second column show the sameresults for the choice of another opposition candidate; and those in the last column showthe results for not voting (versus voting for Labastida). As might be expected, there wasa great deal of continuity in candidate support between February and July; citizens who

48 The Hausman test supports the assumption of ‘independence from irrelevant alternatives’ (IIA) for themultinomial logistic regression model.


TABLE 6 Effects of Television Exposure on Vote Choice in July

Fox Labastida Cardenas/Other None

Television Azteca0 days 44% 33% 10% 14%1 day 46% 30% 10% 14%2 days 48% 28% 11% 14%3 days 50% 25% 11% 14%4 days 51% 23% 12% 14%5 days 53% 21% 12% 13%

Televisa0 days 47% 29% 11% 14%1 day 47% 29% 11% 14%2 days 47% 29% 11% 14%3 days 47% 29% 11% 14%4 days 47% 28% 11% 14%5 days 47% 28% 11% 14%

Notes: Figures represent the marginal effect of television exposure on probability of votingfor any one candidate, with all other variables held at their mean values. Simulations usingClarify software show that the difference in support for Fox between minimum and maximumexposure to Television Azteca is significant at the 10 per cent level and that the difference forLabastida is significant at the 5 per cent level; other differences are not significant. See MichaelTomz, Jason Wittenberg and Gary King, ‘Clarify: Software for Interpreting and PresentingStatistical Results’; this program is freely available at http://gking.harvard.edu. For technicaldetails regarding estimation routines, see Gary King, Michael Tomz and Jason Wittenberg,‘Making the Most of Statistical Analyses: Improving Interpretation and Presentation’,American Journal of Political Science, 44 (2000), 347–61.

supported Labastida in February were inclined to remain loyal to him on election day.Several other control variables were also significant, operating in the expected direction.Nevertheless, television exposure clearly had important electoral consequences, con-trolling for respondents’ initial dispositions, exposure to other campaign cues and variousdemographic factors. Those watching Television Azteca became significantly more likelyto support the main opposition candidates, especially Fox, over Labastida. Exposure toTelevisa, by contrast, exercised no discernible effect.

These effects can be seen more clearly in Table 6, where the probability of voting foreach candidate is mapped as a function of television exposure levels in June. Holding allother predictors at their mean values, individuals who did not see any Television Aztecanews programming in the weeks leading up to election day were nearly as likely to chooseLabastida as Fox (33 per cent versus 44 per cent). These probabilities diverged rapidly,however, as news exposure increased. Citizens who reported seeing five days worth ofnewscasts overwhelmingly flocked to Fox. Television Azteca also benefited the Left,although this difference was much more modest. Clearly, though, Television Aztecaexerted far more than ‘minimal effects’.

One puzzling finding – which also surfaced in analysis of attitude differentials –concerns the failure of sympathetic coverage on the Televisa network to generate supportfor Labastida. If audiences responded mechanically and automatically to media messages,Televisa viewers should have upgraded their assessments of Labastida compared with Foxduring the second half of the race, and they should also have become increasingly inclined


to vote for the ruling party over the course of the whole campaign. Our analysis uncoveredonly weak evidence for the former and no evidence for the latter. Labastida even appearedvulnerable when he and Cardenas received apparently comparable coverage, as onTelevision Azteca during the beginning of the campaign and Televisa at the end (thoughneither effect reached statistical significance). It thus appears that television exercisedasymmetric effects on attitudes towards the ruling party: unfavourable coverage helped toerode support for Labastida, but even highly favourable coverage could not do much tostimulate it.

One possibility is that Labastida was an inherently unappealing candidate relative to theother main contenders and thus tended to lose ground when coverage was balanced.Another possibility is that some Mexican voters disliked the PRI but initially feltinsufficiently informed about the opposition parties to support them; as they learned moreabout both groups, even when what they learned about the opposition was not uniformlypositive, they tended to shift away from Labastida.49 A third possibility is that other aspectsof television coverage on both networks – such as attention to issues like crime, corruptionand drug trafficking – may have reinforced negative attitudes towards the ruling party.50

Consequently, what appeared to be favourable coverage toward Labastida was actuallyrelatively balanced coverage overall, while what appeared to be only slightly favourablecoverage of the opposition was actually quite favourable.51

TESTS OF ROBUSTNESS

Because the persuasive effects we report are considerably larger than previous researchon television and campaigns might suggest, it is important to demonstrate that our resultsare not flukes or artefacts of a particular statistical approach. To this end, we explicitlyconsider whether the results reported above might be the product of (1) our treatment ofmissing data; (2) measurement error on the key variables; (3) omitted variable biasstemming from left-out demographic factors or campaign stimuli; or (4) audienceself-selection to particular news networks.

Missing Data

Our data are characterized by two broad types of missingness. First, and mostconventionally, we have missing answers to particular questions for certain respondentsin each panel wave. The extent of this sort of missingness is actually quite small, and usingdifferent methods to correct for it – for example, listwise vs. pairwise deletion in ourordinary least squares (OLS) regressions – does not materially change our findings. Asecond type of missingness results from the panel structure of our data, in which certainrespondents were interviewed in some waves but not in others. Some of this missingness

49 Alejandro Moreno, ‘The Effects of Negative Campaigns on Mexican Voters’, in Dominguez and Lawson,Mexico’s Pivotal Democratic Election, pp. 243–68.

50 Our content analysis suggests that issue coverage was similar across the two networks but that crime,corruption and drug trafficking received substantial attention (317 minutes in our sample) compared to coverageof the president (79 minutes), the economy (113 minutes) or education (87 minutes).

51 One other possible explanation is that Televisa was regarded as less credible than Television Azteca and thusexercised less of an effect on attitudes and behaviour. Although this explanation seems plausible, we found thatexposure to Television Azteca was not significantly associated with greater confidence in the credibility oftelevision news than exposure to Televisa.


was the product of panel attrition, as respondents from the first wave could not be locatedin subsequent waves. In the fourth wave, where interviewers attempted to contact allrespondents from the February wave, attrition accounted for all missing respondents. Inthe second and third waves, however, interviewers consciously attempted to contact onlya randomly pre-selected sample of respondents from the first wave in order to minimizecost and attrition. Thus, most ‘missing’ respondents in the second and third waves wereintended to be missing.

Analysis undertaken as part of the Mexico 2000 Panel Survey suggests that panelattrition introduced only very small biases in the composition of the sample.52 Moreover,the potential directional consequences of any sample biases for our analysis areindeterminate, and the fact that these individuals were randomly selected implies that theirexclusion should not affect our analysis. Nevertheless, the availability of new methods fordealing with missing data allows us to address this possibility directly. To this end, were-ran our analyses using multiple imputation to fill in missing data for all respondents.53

The use of multiple imputation substantially enhanced the statistical significance of ourfindings.

Measurement Error

Survey data are often riddled with measurement error. Measurement of our keyindependent variable, network news exposure, is likely to be particularly unreliable.54

Random measurement error on this variable might likely bias our results towards zero,making media effects look smaller than they really are. Meanwhile, measurement error onour dependent variable (attitude differentials) or on other independent variables coulddistort our results in ways we have not anticipated. To guard against these eventualities,and any spurious inferences that might follow from them, we factored out measurementerror for media exposure and attitude differentials using the Wiley–Wiley method.55 Theeffects of news exposure in this more elaborate model closely parallel those reported inTables 3 and 4. (These findings are available upon request.)

Omitted Variable Bias

As discussed above, our analysis controls for obvious confounding variables: demographicfactors, the strength and direction of partisan affiliations, and certain non-televisioncampaign influences. In theory, however, the attitude change that we ascribe to televisioncould be the result of some omitted variable that is correlated with exposure to newsprogramming and with changes in attitudes towards the main candidates. To address this

52 Domınguez and Lawson, Mexico’s Pivotal Democratic Election, pp. 14, 347–8.53 Gary King, James Honaker, Anne Joseph and Kenneth Scheve, ‘Analyzing Incomplete Political Science

Data: An Alternative Algorithm for Multiple Imputation’, American Political Science Review, 95 (2001), 49–69.54 Vincent Price and John Zaller, ‘Who Gets the News? Alternative Measures of News Reception and Their

Implications for Research’, Public Opinion Quarterly, 57 (1993), 133–64. See also Larry Bartels, ‘MessagesReceived: The Political Impact of Media Exposure’, American Political Science Review, 87 (1993), 267–85.

55 Following Wiley and Wiley, we assume that: (a) the residual term in a particular regression is not correlatedwith residuals in any other wave, the latent attitude variables, or any measurement errors in the survey instruments,(b) measurement error variance is constant for each item throughout the panel, and (c) these errors in measurementare not correlated with latent attitudes or measurement errors for survey items in any other panel wave. See DavidE. Wiley and James A. Wiley, ‘The Estimation of Measurement Error in Panel Data’, American SociologicalReview, 35 (1970), 112–17.


possibility, we re-ran our analyses with various combinations of control variables. Theseincluded:

—A bloc of demographic variables, including education, rural residency, unionmembership, socio-economic status, church attendance, gender and age.

—Campaign engagement: An index of frequency with which respondents discussedpolitics over all four waves of the panel, campaign attention over all four waves ofpanel, and newspaper readership over first three waves of panel. (This last item wasnot asked in the fourth wave.)

—Alternative campaign engagement: All elements of the index of campaignengagement, plus interest in politics, averaged over all four waves of panel.

—Debate exposure: An index of whether respondents saw all, part or none of the firstpresidential debate; saw all, part or none of the second debate; discussed the firstdebate with others; discussed the second debate with others; saw comments about thefirst debate; and saw comments about the second debate – based on items from thesecond, third and fourth waves of panel.

—Attitude differential (AD) in February: Differences in feeling thermometer ratingsbetween the main candidates in February.

—Party attitude differential (AD): Differences in feeling thermometer ratings betweenthe main parties in February.

—1994 vote: Presidential vote in 1994, coded as Zedillo (PRI), Fernandez (PAN), othercandidate and non-voter/not sure.

—1999 primary vote: Vote in 1999 PRI primary, coded as Labastida, losing candidateand did not participate.

—Partisan identification (PID): strong PRI identification, weak PRI identification,independent leaning to PRI, strong PAN identification, weak PAN identification,independent leaning to PAN, strong PRD/other identification, weak PRD/otheridentification, independent leaning to PRD/other, and independent non-leaners, asreported in February.

Reporting tests of robustness across different combinations of these controls, different timeperiods, different networks and different candidate dyads would fill hundreds of tables. Foreconomy of presentation, we simply summarize the effects of exposure to the two mainnetworks on differences in feeling thermometer ratings of the three main candidates in thethird wave of the panel (June), controlling for different combinations of variables. Theresults of OLS regression are reported in Table 7. The first two columns of Table 7 reportthe effects of exposure to Televisa news and Television Azteca news on differencesbetween Labastida and Fox. The next two columns report the effects of exposure toTelevisa news and Television Azteca news (respectively) on differences betweenLabastida and Cardenas. As the table shows, the effects of television news exposureattenuate with the addition of control variables, but they remain impressive. Addingattitudinal controls tends to moderately diminish their influence, especially for TelevisionAzteca. Adding controls for demographics and other campaign-related stimuli weakens theeffect of Television Azteca exposure but enhances the effects of exposure to Televisa.Regardless, however, significant effects from exposure to television news persist whendifferent combinations of controls are added.

We repeated this analysis for vote choice in July, substituting vote preferences inFebruary (Labastida, Fox, Cardenas/other, and none/undecided) for attitude differentials


TABLE 7 Effects of Network News Exposure on Evaluation of the MainCandidates in June

Labastida Labastidaminus Fox minus Cardenas

Controls Televisa Azteca Televisa Azteca

AD 0.14* � 0.39** 0.14* � 0.26**AD, campaign engagement 0.16* � 0.35** 0.16** � 0.24**AD, campaign engagement, debate

exposure 0.18* � 0.30** 0.17** � 0.19**AD, campaign engagement, debate

exposure, demographics 0.26** � 0.26** 0.19** � 0.16*

AD, PID 0.13* � 0.23* 0.11* � 0.21**AD, party AD, PID, ’94 vote, PRI

primary vote 0.13* � 0.17** 0.01 � 0.11**AD, demographics, PID 0.16** � 0.17** 0.14** � 0.15**AD, campaign engagement,

demographics, PID 0.16** � 0.17** 0.14** � 0.15**AD, campaign engagement, debate

exposure, demographics, PID 0.16** � 0.17* 0.15** 0.15*AD, party AD, PID, campaign

engagement, debate, demographics 0.16** � 0.13* 0.12* � 0.13*AD, campaign engagement,

demographics, ’94 vote, primaryvote 0.18** � 0.19** 0.16** � 0.14*

** Significant at 0.01.* Significant at 0.05.† Significant at 0.1.

in February. For economy of presentation, Table 8 presents only the results for TelevisionAzteca news. (Televisa effects were consistently not significant.)

As Table 8 indicates, the coefficient for exposure to Television Azteca news remainssignificant at the 5 per cent level regardless of which controls are included, and itsmagnitude changes little. The effects of Television Azteca news on support for the Leftare somewhat less robust, although they are normally significant as well.

Self-selection to Media Outlets

Related to left-out variable bias is the issue of self-selection to ideologically congruentmedia outlets – i.e., the fact that Fox supporters might prefer Television Azteca andLabastida supporters might lean towards Televisa. In general, the panel nature of our datashould ensure that this sort of self-selection does not drive our results.56 However, it isconceivable that initial measures of partisan predispositions do not adequately capturepropensities for attitude change, and that those propensities are somehow correlated withnetwork preference. If so, what we have interpreted as media-induced attitude changecould be nothing more than attitude crystallization over the course of the campaign.

If self-selection were driving our results, we would expect relatively consistent impactsof exposure to different networks on attitudes towards the main candidates (and vote

56 See Wooldridge, Introductory Econometrics, chap. 9.


TABLE 8 Effect of Television Azteca Exposure on Vote in July (Base is Vote forLabastida)

Controls Fox Other Non-voter

February vote 0.16** 0.14* � 0.06February vote, campaign engagement 0.16** 0.15* 0.02February vote, demographics, region, campaign

engagement 0.16** 0.13* 0.03February vote, demographics, region, alternative

campaign engagement, debate exposure 0.14** 0.10 0.02

February vote, PID 0.14** 0.13* � 0.07February vote, campaign engagement, PID 0.12* 0.13* � 0.00February vote, alternative campaign engagement, PID 0.12* 0.12† � 0.00February vote, demographics, campaign

engagement, PID 0.13* 0.13* 0.01February vote, demographics, alternative campaign

engagement, PID 0.13* 0.13† 0.01February vote, demographics, region, alternative

campaign engagement, debate exposure, PID 0.13* 0.09 0.01February vote, demographics, campaign

engagement, 1994 vote, 1999 primary vote 0.14** 0.09 0.01

** Significant at 0.01.* Significant at 0.05.† Significant at 0.1.

choice). That is, watching Televisa in February should improve assessments of Labastidarelative to Fox in April–May; watching Television Azteca should have the opposite effect.Similarly, we would expect exposure to different networks to exercise roughly the samesort of effect on Cardenas at different points in the campaign. In both cases our data showthe opposite. Watching Televisa (and Television Azteca) was not a significant predictorof attitude change towards the two main candidates during the first half of the race, whencoverage of the candidates was similar on both networks. However, television newsexposure was associated with attitude change later in the race, when the tone of newscoverage on the two networks was quite different. Self-reported exposure to network newsmattered when the content of that news was different; it did not matter when networksaccorded the candidates similar coverage.

It should also be noted that self-selection only makes sense with respect to the effectsof Television Azteca on comparisons between Fox and Labastida. It could not be expectedto explain attitudes towards Cardenas, given that the Left had historically received poorcoverage on both networks. If anything, there was less animosity between the PRD andTelevisa than between the PRD and Television Azteca, which had been involved in aprotracted battle against the PRD’s administration in Mexico City. Nevertheless, Televisacoverage was associated with diminished support for Cardenas during the first half of thecampaign, whereas Television Azteca exposure was not.

It is theoretically possible, of course, to devise a direct test of the possibility thatself-selection to ideologically congruent media is driving our results. This approach hingeson identifying a variable other than news exposure that captures underlying propensitiesto prefer one network over another, and thus hidden dispositions to change one’s opinionsof the candidates in a way correlated with network news exposure. We chose networkpreference in February, based on which television news broadcast a respondent reported


watching first, regardless of whether the respondent also reported watching anotherbroadcast and regardless of how frequently she reported watching television news. We thenre-ran our analyses with both network preference and network news exposure. Includingthis variable attenuated certain effects but neither altered the direction the coefficientsreported above nor rendered them statistically insignificant.

There remains the possibility that our results might be the product of viewers switchingfrom one network to another over the course of the campaign in response to politicalcoverage. For instance, Fox voters may have switched away from Televisa and towardsTelevision Azteca in response to perceived differences in coverage on the two networks.If so, Fox supporters would wind up becoming disproportionately reliant on TelevisionAzteca, while Labastida supporters would wind up relying more on Televisa. Suchswitching by committed partisans from one network to another (or changes in the amountof television news they watched on each channel) could be responsible for what we haveinterpreted as attitude change brought on by exposure to media messages.

In the Mexican context, this possibility seems extremely remote. Because biases werequite modest compared to past elections, viewers who were inclined to change their choiceof news outlet based on their political attitudes would presumably either have switchedbefore the 2000 campaign or remained unmotivated to switch in the run-up to the elections.Moreover, as an empirical matter, several of our results are the opposite of what wouldbe expected from self-selection during the campaign. For instance, if strong Cardenassupporters who watched Televisa had switched away from that network duringFebruary–April, Televisa exposure in May–June should be correlated negatively withattitudes towards the Left. In fact, the reverse was true.

It is nevertheless possible to address the issue of endogeneity bias explicitly by fittingsimultaneous equations, where candidate differentials affect television viewing in a givensurvey period while at nearly the same time are affected by the content of the news.57 Informal terms, these equations would be specified as follows:

Candidate differentialt � b1 (candidate differentialt - 1) � b2 (Aztecaexposuret) � b3 (Televisa exposuret) � b4 - k (controls)

Azteca exposuret � b1 (Azteca exposuret - 1) � b2 (Televisaexposuret - 1) � b3 - k (controls)

Televisa exposuret � b1 (Televisa exposuret - 1) � b2 (Aztecaexposuret - 1) � b3 - k (controls)

The assumption underlying this model is that lagged values of television exposure areexogenous to changes in feeling thermometer scores from one survey wave to the next.58

The independent effect of media exposure on candidate differentials when endogeneity ispurged, shown in Tables 9 and 10, corroborates the findings from Tables 3 and 4. In theApril/May wave, exposure to the Televisa network drove respondents away from Cardenasand towards Fox; to a lesser extent, individuals who saw programming on TelevisionAzteca moved closer to Fox relative to Labastida. In the June wave, Television Aztecaviewers became less supportive of Labastida relative to Cardenas. All of these findingswere statistically significant.

57 This procedure would also correct for any tendency for respondents to overreport or underreport their levelof exposure to news broadcasts as their candidate evaluations changed over the course of the campaign.

58 Steven E. Finkel, Causal Analysis with Panel Data (Thousand Oaks, Calif.: Sage, 1995).


Based on these findings, we conclude that our substantive inferences are not the productof our treatment of missing data, measurement unreliability on the key variables, omittedvariable bias, self-selection to ideologically congruent news broadcasts or relatedconfounding factors.

DISCUSSION

A decade ago, John Zaller argued that ‘massive’ media effects could be observed whereanalysts developed good measures and found adequate variation on key independentvariables.59 Although our operationalization differs from Zaller’s, we reach a similarconclusion: in Mexico’s 2000 race, we find evidence for substantial, if not ‘massive’, mediaeffects. Television Azteca tended to hurt the ruling party relative to its main rivals,especially Vicente Fox. Meanwhile, Cuauhtemoc Cardenas was initially harmed bynegative coverage on Televisa but then rebounded when coverage of him on that networkimproved markedly in the second half of the race. The scope of these effects on bothattitudes and voting behaviour was quite large.

At first glance, it seems conceivable that these findings could be the product of moreaccurate measures of media content and exposure, rather than differences in politicalcontext between the United States and Mexico. For instance, our measure of tone ofcoverage is arguably more sophisticated than the purely verbal indicators employed inmany analyses of television influence. We also explicitly take advantage of the fact thatcoverage may differ substantially across networks, a source of variation not examined inmost earlier studies of effects. It might be that other researchers using these same methodswould uncover comparable levels of persuasive influence in, say, American or Britishelections. If so, our findings suggest a new direction for research on television and electoralcampaigns in established democracies.

That said, we are sceptical of the notion that researchers would regularly have foundlarge effects in established democracies if only they had looked harder. First, despiteseveral decades of research, few scholars have found large-scale effects of television newscoverage on voters’ partisan preferences in general election campaigns. Secondly, thechanges in attitudes and behaviour that we have documented were produced by relativelymodest biases on television. Although these biases were somewhat larger than thosedocumented by content analyses of television news in the United States, they were notdramatically so.60 Nevertheless, the effects that we report on attitudes and behaviour aresubstantially greater than those found in most general election campaigns in establisheddemocracies. We thus lean strongly towards the conclusion that television influence onvoters during Mexico’s 2000 campaign was the product of particular features of theMexican political context. These features include not only a different media environmentbut also greater susceptibility to media influence.

59 Zaller, The Nature of Public Opinion.60 See Doris A. Graber, ‘Framing Election News Broadcasts: News Context and its Impact on the 1984

Presidential Election’, Social Science Quarterly, 68 (1987), 552–68; Doris A. Graber, ‘Kind Pictures and HarshWords’, in Kay Schozman, ed., Elections in America (Boston, Mass.: Unwin-Hyman, 1987), pp.115–41;Kepplinger, ‘Visual Biases in Television Campaign Coverage’; C. Richard Hoffstetter, Bias in the News: NetworkTelevision Coverage of the 1972 Election (Columbus: Ohio State University Press, 1976), pp. 119–30; MichaelRobinson and Margaret A. Sheehan, Over the Wire and on TV: CBS and UPI in Campaign ’80 (New York: RussellSage, 1983); Ansolabehere et al., ‘Mass Media and Elections’, p. 64.


TA

BL

E9

Eff

ects

onC

andi

date

Att

itud

eD

iffe

rent

ials

,A

pril

/May

(Tw

o-St

age

Mod

el)

Lab

asti

da–F

oxL

abas

tida

–Car

dena

sF

ox–C

arde

nas

Var

iabl

e�

s.e.

�s.

e.�

s.e.

Con

stan

t1.

281.

211.

011.

190.

055

1.02

Nor

ther

nre

gion

�0.

224

0.50

30.

249

0.49

30.

523

0.43

7M

etro

regi

on�

0.09

00.

482

�0.

068

0.46

30.

141

0.41

1C

entr

alre

gion

0.23

30.

501

�0.

206

0.47

2�

0.27

40.

421

Age

�0.

006

0.01

3�

0.01

50.

013

�0.

010

0.01

1M

ale

0.17

40.

300

0.01

90.

291

0.15

10.

257

Urb

anre

side

nt0.

285

0.38

30.

460

0.36

80.

114

0.32

5E

duca

tion

�0.

004

0.18

10.

086

0.17

60.

108

0.15

5R

elig

iosi

ty�

0.07

50.

131

0.07

20.

126

0.12

90.

111

Uni

onm

embe

r0.

862†

0.44

00.

207

0.42

3�

0.49

20.

376

Inco

me

prox

y�

0.16

50.

170

�0.

170

0.16

50.

035

0.14

8

PR

Ipa

rtis

anor

lean

er(F

eb.)

1.45

**0.

492

1.65

**0.

472

0.24

90.

416

PA

Npa

rtis

anor

lean

er(F

eb.)

�1.

10*

0.50

4�

0.19

10.

476

0.99

5*0.

418

PR

Dpa

rtis

anor

lean

er(F

eb.)

0.18

00.

619

�1.

16†

0.61

9�

1.48

*0.

549

Str

ong

PR

Ipa

rtis

an(F

eb.)

2.90

**0.

600

2.90

**0.

567

�0.

018

0.51

0S

tron

gP

AN

part

isan

(Feb

.)�

1.36

†0.

718

0.14

70.

694

1.79

**0.

618

Str

ong

PR

Dpa

rtis

an(F

eb.)

0.09

31.

07�

1.09

1.08

�1.

500.

942

Att

itud

edi

ffer

enti

al(F

eb.)

0.38

1**

0.04

70.

396*

*0.

047

0.32

9**

0.05

0T

alk

abou

tpo

liti

cs(A

pril

/May

)0.

499

0.79

7�

0.04

00.

774

�0.

580

0.67

6C

ampa

ign

atte

ntio

n(A

pril

/May

)�

0.84

40.

997

�1.

050.

962

�0.

153

0.85

0

Azt

eca

expo

sure

(Apr

il/M

ay)

�0.

326

0.24

40.

114

0.23

20.

396†

0.20

9T

elev

isa

expo

sure

(Apr

il/M

ay)

�0.

233

0.18

50.

227

0.17

90.

411*

*0.

156

N69

067

768

4A

djus

ted

R2

0.34

0.34

0.24

F-s

tati

stic

17.8

17.9

11.1

Not

es:

Pai

rwis

ede

leti

onem

ploy

ed;�

�es

tim

ated

coef

fici

ents

;s.

e.�

stan

dard

erro

rs.†

Sig

nifi

cant

at0.

10.*

Sig

nifi

cant

at0.

05.*

*S

igni

fica

ntat

0.01

.


TA

BL

E1

0E

ffec

tson

Can

dida

teA

ttit

ude

Dif

fere

ntia

ls,

June

(Tw

o-St

age

Mod

el)

Lab

asti

da–F

oxL

abas

tida

–Car

dena

sF

ox–C

arde

nas

Var

iabl

e�

s.e.

�s.

e.�

s.e.

Con

stan

t1.

281.

49�

1.40

1.44

�1.

911.

40

Nor

ther

nre

gion

0.11

00.

583

0.16

40.

562

0.17

30.

549

Met

rore

gion

0.60

00.

542

�0.

461

0.51

2�

1.12

*0.

505

Cen

tral

regi

on0.

018

0.51

8�

0.14

00.

499

�0.

165

0.48

8

Age

0.01

10.

015

0.01

80.

014

0.00

90.

014

Mal

e�

0.66

70.

420

�0.

796*

0.40

10.

065

0.39

4U

rban

resi

dent

�0.

252

0.47

90.

164

0.46

40.

305

0.45

5E

duca

tion

�0.

326

0.28

7�

0.24

80.

277

0.14

90.

271

Rel

igio

sity

�0.

259

0.17

6�

0.14

30.

167

0.03

20.

167

Uni

onm

embe

r0.

397

0.59

20.

547

0.54

70.

069

0.54

6In

com

epr

oxy

0.31

70.

251

�0.

087

0.24

1�

0.36

10.

234

PR

Ipa

rtis

anor

lean

er(F

eb.)

1.04

†0.

598

1.69

**0.

571

0.69

00.

562

PA

Npa

rtis

anor

lean

er(F

eb.)

�1.

21†

0.62

11.

05†

0.58

42.

20**

0.57

2P

RD

part

isan

orle

aner

(Feb

.)0.

196

0.75

8�

0.40

50.

734

�0.

177

0.73

2S

tron

gP

RI

part

isan

(Feb

.)1.

86**

0.66

12.

82**

0.64

21.

03†

0.58

9S

tron

gP

AN

part

isan

(Feb

.)�

1.97

*0.

855

0.29

30.

818

�2.

40**

0.82

4S

tron

gP

RD

part

isan

(Feb

.)�

2.14

†1.

130

�3.

49**

1.11

�0.

972

1.10

Att

itud

edi

ffer

enti

al(A

pril

/May

)0.

480*

*0.

055

0.46

6**

0.05

30.

520*

*0.

063

Tal

kab

out

poli

tics

(Jun

e)�

0.06

10.

615

0.06

80.

596

�0.

176

0.59

3C

ampa

ign

atte

ntio

n(J

une)

0.68

00.

675

0.83

60.

647

0.24

50.

653

Azt

eca

expo

sure

(Jun

e)�

0.33

8†0.

202

�0.

461*

0.19

1�

0.12

10.

190

Tel

evis

aex

posu

re(J

une)

�0.

121

0.18

10.

019

0.18

30.

049

0.18

1

N29

629

529

5A

djus

ted

R2

0.51

0.54

0.38

F-s

tati

stic

15.5

17.3

9.5

Not

es:

Pai

rwis

ede

leti

onem

ploy

ed;�

�es

tim

ated

coef

fici

ents

;s.

e.�

stan

dard

erro

rs.†

Sig

nifi

cant

at0.

10.*

Sig

nifi

cant

at0.

05.*

*S

igni

fica

ntat

0.01

.


From a theoretical perspective, our findings suggest a re-reading of the literature oncampaign effects. Rather than presume that the persuasive effects of television exposureon partisan preferences and voting behaviour are minimal, researchers might insteadassume that media effects vary with the political and informational context. Where partisanidentifications are strong, political alternatives well known, media coverage balanced andaudiences free to choose between a range of outlets, effects may well be modest. Bycontrast, where partisan identifications are weak, political alternatives more unfamiliar,media coverage less balanced and audiences dependent on a small number of sources forpolitical information, effects may be pronounced. Such an interpretation of campaigneffects is in keeping with recent scholarship on political communication, which hasidentified various caveats to the ‘minimal effects’ model.

The magnitude of television influence in 2000 can only lead us to speculate about theimpact of television news on past presidential campaigns in Mexico. During the 2000 race,both networks accorded the three main candidates roughly the same amount of coverage,and residual biases in tone were relatively subtle. In the presidential campaigns of 1988and 1994, by contrast, both the volume and tone of coverage were grossly skewed in favourof the ruling party, and audiences were even more dependent on Televisa news forinformation about the main candidates.61 It seems likely that television bias preventedsubstantial defections from the ruling party in those contests.

Empirically, our ability to generalize beyond Mexico is limited by the paucity of panelsurveys with adequate indicators of media exposure. Theories of media influence,however, offer grounds for suspecting pronounced effects. In terms of the relativeweakness of partisan attachments, the comparative unfamiliarity of political alternatives,and the extent of dependence on biased outlets, most of the world’s electoral contests lookmore like Mexico’s than they look like those in the United States. Consequently, limitedeffects models developed in the United States may substantially understate media-inducedattitude change in much of the world.

Indeed, by global standards, Mexico is really an intermediate case; the factors that appearto permit substantial media effects in that country are even more pronounced elsewhere.In Brazil, for instance, partisan attachments at the mass level are extremely weak; citizensrely overwhelmingly on television for information about candidates; and bias on televisioncan be striking.62 One might expect even greater media influence in such contexts. Suchdeveloping democracies clearly represent promising territory for research on the politicalimpact of television.

61 Lawson, Building the Fourth Estate; Murray Fromson, ‘Mexico’s Struggle for a Free Press’, in Richard R.Cole, ed., Communication in Latin America: Journalism, Mass Media, and Society (Wilmington, Del.: ScholarlyResources Books, 1996), pp. 115–38; Adler, ‘The Mexican Case’; Arredondo-Ramirez et al., Ası se callo elsistema; Victor Manuel Bernal-Sahagun and Eduardo Torreblanca-Jacques, eds, Espacios de silencio (MexicoCity: Editorial Nuestro Tiempo, 1988).

62 Roberto Amaral and Cesar Guimaraes, ‘Media Monopoly in Brazil’, Journal of Communication, 44 (1994),26–38; Mauro Porto, ‘Making Sense of Politics: TV News and the Interpretation of Politics in Brazil’ (paperpresented at the XXII Conference of the Latin American Studies Association, Miami, 2000); Joseph D. Straubhaar,Organ Olsen and Maria Cavaliari Nunes, ‘The Brazilian Case: Influencing the Voter’, in Skidmore, ed., Television,Politics, and the Transition to Democracy in Latin America, pp. 118–36.


APPENDIX

QUESTION WORDING AND CODINGS

Television news exposure: ‘Do you watch any television news program? [IF YES] Which one?Approximately how many days a week do you watch it?’ Interviewers were instructed to mark upto two programs. Respondents who reported watching a particular program six or seven days a weekwere treated as watching five days a week. Those who reported watching two programs on the samenetwork were treated as having been doubly exposed; i.e., their self-reported exposure to oneprogram was added to their self-reported exposure to the other program. For both networks in allthree waves, over 93 per cent of respondents were coded as watching five or fewer days per week.

Feeling thermometers: ‘[SHOW CARD] I am going to ask your opinion about political parties andcandidates for president. On this scale, zero indicates that your opinion is very bad and ten that youropinion is very good. If you don’t have an opinion, just tell me and we’ll go on to the next one.What is your opinion of …? [READ AND ROTATE].’ Options were ‘PRI, PAN, PRD, FranciscoLabastida, Vicente Fox, Cuauhtemoc Cardenas’. These feeling thermometer scores were thendifferenced to create attitude differentials.

Vote preference (February): ‘Now for the purposes of this survey let’s suppose that today is electionday and you are going to vote for President of the Republic. For this I am going to give you a pieceof paper where you can mark your response without me seeing it and afterwards deposit it in thisbox. If the elections for President of the Republic were today, for whom would you vote?’Respondents were then handed a sample ballot.

Vote choice (July): ‘Did you vote in the elections of July 2? [IF YES:] Could you mark on this ballotfor whom you voted in the elections for President of the Republic? [HAND BALLOT ANDREQUEST THAT R DEPOSIT IT IN THE BOX.]’

Partisan identification: ‘In general, do you consider yourself priısta, panista or perredista? [IF RHAS AFFILIATION:] Do you consider yourself very priısta/panista/perredista or somewhatpriısta/panista/perredista? [IF R HAS NO AFFILIATION]: Toward which party do you most lean?’

Campaign attention: ‘And in particular how much attention are you paying to the presidentialelection campaign this year – a lot, some, a little, or none?’

Talks about politics: ‘How often do you talk about politics with other people – daily, a few timesa week, a few times a month, rarely, or never?’

Debates: ‘Did you see the televised debate last week/month between …? [IF YES] Did you see allof it or just a part? [FOR ALL RESPONDENTS] From which of the following have you heardcommentary about the debate?’ [Options were ‘Television reports, close family members, friendsor acquaintances, radio reports, any other?’].

Credibility of television news: ‘When they talk about the candidates for the presidency, how muchdo you believe the news on television – a lot, a little, some, or none?’


TABLE A1 Responses and Distributions for Main Variables

Variable Minimum Maximum Mean St. dev. N

TV Azteca exposure (February) 0 10 1.43 2.41 2,355TV Azteca exposure (April/May) 0 10 1.86 2.56 952TV Azteca exposure (June) 0 10 1.70 2.46 974TV Azteca exposure (July) 0 10 1.76 2.56 1,254

Televisa exposure (February) 0 10 2.39 2.75 2,355Televisa exposure (April/May) 0 10 2.34 2.57 952Televisa exposure (June) 0 10 2.38 2.72 974Televisa exposure (July) 0 10 1.90 2.54 1,254

Labastida–Fox (February) � 10 10 � 0.26 3.90 1,982Labastida–Fox (April/May) � 10 10 � 0.05 4.31 851Labastida–Fox (June) � 10 10 � 0.30 4.58 901

Labastida–Cardenas (February) � 10 10 1.55 4.19 1,966Labastida–Cardenas (April/May) � 10 10 1.30 4.17 847Labastida–Cardenas (June) � 10 10 0.60 4.20 900

Fox–Cardenas (February) � 10 10 1.82 3.47 1,965Fox–Cardenas (April/May) � 10 10 1.35 3.43 852Fox–Cardenas (June) � 10 10 0.90 3.80 901

Note: Respondents scored 10 on network exposure if they reported watching two differentshows on the same network every day. Fewer than 7 per cent of respondents scored over 5on any of the network exposure variables listed above.

television news, mexico's 2000 elections and media effects...

Documents