Portal de Periódicos da CAPES

Effects of Election News Coverage: How Visibility and Tone Influence Party Choice

2010; Taylor & Francis; Volume: 27; Issue: 4 Linguagem: Inglês

10.1080/10584609.2010.516798

1091-7675

David Nicolas Hopmann, Rens Vliegenthart, Claes H. de Vreese, Erik Albæk,

Media Studies and Communication

Abstract Previous research has shown that a party's election results can depend on visibility and tone in the media. Using daily content data from the major news bulletins and daily survey data from the 2007 national election campaign in Denmark (N = 5,083), our analysis improves upon two central aspects of this earlier research. First, the effects on vote choice of the parties' visibility and tone in the media are studied concurrently in the same model. Second, a distinction is made between the effects of direct exposure to specific news content and the effects of the cumulative information environment created by the media. Overall, it is found that the more visible and the more positive the tone toward a given party is, the more voters are inclined to vote for this party. The effects are primarily effects of the information environment, not effects of direct exposure, though undecided voters are also directly affected. In the discussion, central conditions for the strength of media effects are identified. Keywords: public opinionmedia effectselectionssurveyscontent analysis Acknowledgments For their helpful comments, the authors are grateful to Sara Hobolt, Robert Klemmensen, and Rüldiger Schmitt-Beck, as well as Shanto Iyengar and the anonymous reviewers. Notes 1. It should be acknowledged that Jerit and colleagues look at the effects of the information environment on the individual, in their case on citizens' political knowledge, but they do not distinguish between indirect and direct effects. In general, they devote little attention to the mechanisms by which media can have effects on individuals' attitudes and behavior. 2. Preliminary analyses based on data from the Danish Election Study indicate that the proportion of late deciders was slightly higher in 2007; about two-fifths decided during the campaign. 3. In a rolling cross-sectional design, if a respondent is not reached on the first day, several attempts are made during the consequent days to contact that person. The strategy pursued in our case is slightly different, as new attempts to call a respondent are made only during the same day (up to seven times). 4. Comparing the reported prior vote choices in 2005 (including blanks and abstention) with the actual result from the 2005 election yields the following results: all daily samples correlate (Pearson r) at least .93 with the election outcome of 2005 (p < .001; unweighted data: .92, p < .001; n = 13 equivalent to the number of choices voters had, i.e., to vote for one of the parties or candidates running in the campaign, to abstain, or to cast a blank vote). Of all respondents, 52.3% were women (51.0% in the adult population) with an average age of 49.9 years (female adult population: 47.6), whereas men's mean age was 49.3 (male adult population: 49.7). Of the respondents, 43.3% had completed a higher education, which is higher than the proportion of the adult population aged 20–69 (27.6%). There are no statistically significant differences with respect to the daily mean age. On 2 of the 18 days, there was a slightly larger proportion of men. 5. Of the 32,563 randomly chosen telephone numbers called over 18 days, 81 were out of order, 20,528 were unanswered, 2,395 were answered but ineligible (e.g., business or government offices), 4,476 refused (including break-offs), and 5,083 accepted participation. To ensure the representativeness of the daily samples, Capacent Epinion assigned weights based on gender, age, education, and prior vote choice to all respondents, varying from .32 to 3.16 with a mean of 1 (e.g., using the weights, the percentage with higher education decreases to 25.3%, which is very close to the actual percentage in the population). In addition, the analyses presented here were also conducted with unweighted data, yielding no substantial differences. More generally, we note that a high response rate is not necessarily a prerequisite for representativeness (CitationKrosnick, 1999). 6. Question wordings were as follows: "Whom will you vote for in the general election November 13?" (68.8% reported a preference, 25.2% were in doubt, 1.3% were not entitled to vote, 4.7% refused to answer) and "It can be difficult to remember whom one voted for in the last general election, or if one voted at all—can you remember whom you voted for in the last election?" (87.0% reported a choice, 5.8% were not entitled to vote, 7.2% did not remember or refused to answer). 7. Question wording was as follows: "Although you are in doubt whom to vote for, I'd like to ask you if you are leaning toward a party?" (66.5% reported a preference, 33.5% reported no preference). 8. Question wording was as follows: "Did you watch news on DR or TV2 yesterday?" (see Table 1). 9. Of all media appearances by actors with party affiliations, 10% were coded as valenced (note that numerous actors typically appear in a news story). This raises the question as to whether the applied coding captured all leanings; possibly some instances of unidirectional, but not strong, leaning were not captured due to the application of three codes only. The initial coding during pretesting included a more detailed coding scheme, but the analyses suggested that the overall story tendency yielded reliable results and were in line with previous similar studies. The tests of intercoder reliability yielded a Krippendorff alpha of .96 for affiliation and .75 for media tone toward actors (CitationKrippendorff, 2004). 10. For visibility, the correlation coefficients for the daily relative visibility of parties in the four news bulletins range between .329 and .531 (Pearson r). The correlation coefficients for tone are not significant except for one case (DR 21.00 and TV2 19.00), where a minor negative correlation is found. 11. Without doubt, it is hard to disentangle effects attributable to the information environment from those attributable to accumulated direct news exposure. The analyses presented below are replicated by replacing the information environment variable by a variable that is constructed on the basis of an extrapolation of yesterday's media exposure to the preceding days, assuming that media exposure in the 4 preceding days is similar to yesterday's media exposure. Results indicate a less good model fit in the latter instance, suggesting that information from media content to which the respondent is not directly exposed adds to the explanation of political preferences. 12. However, one can also think of the data set as having a multilevel structure: individuals nested within parties nested within days. Therefore, the following analyses were also estimated as three-level models using MCMC estimations in MLwiN. First, these analyses yielded essentially the same results as reported here. Second, calculating an intraclass correlation coefficient as suggested by CitationSnijders and Bosker (1999, p. 224) for the highest level (i.e., days) yields a very small coefficient (< .001), indicating very little variance at the day level. 13. Detailed results of our robustness checks can be downloaded at http://www.rensvliegenthart.com/polcom. 14. Adding different background characteristics does not change the results of the models presented here. This also makes sense intuitively: those background variables might influence the probability that a respondent indicates a preference for a certain party, but when taking all parties together, as is done here, background characteristics can affect only the probability of indicating a preference for any party. 15. We also tested for possible interaction effects, for example between the party tone in the information environment (Partytone_env) and day of campaign (Tday). Including these interactions did not significantly improve the explanatory power of the model. 16. For the first two hypotheses, one can also expect a reversed causal effect, where high standings in the polls contribute to more visibility and more favorable coverage of the party. While these are plausible expectations, here we focus on the reverse causal relationship. Our research design is especially suitable to do so: by taking a low level of analysis (the unit of analysis is the respondent's position with regard to a party) and only considering media content that the respondent is (likely to be) exposed to before being asked vote intention, we believe claims about causal mechanisms based on the results are persuasive. 17. Following the reasoning on the importance of parties' ownership of particular issues, one wonders whether issue ownership should have been included in our models. The basic idea of the concept of issue ownership is that certain parties "own" certain issues and that increased media attention to those issues will produce a gain for their owners (CitationBudge & Farlie, 1983; CitationGreen-Pedersen, 2007). However, no data on issue ownership at the party level is available for the 2007 election campaign. Modeling issue ownership on the basis of parties' press releases (data kindly made available by Christian Elmelund-Præstekær) and including these data in our models does not change the substantial findings presented here (for similar results, see CitationKleinnijenhuis et al., 2007).