International Journal of Public Opinion Research Vol. No. © World Association for Public Opinion Research ; all rights reserved
OUR LONG ‘RETURN TO THE CONCEPT OF POWERFUL MASS MEDIA’— A CROSS-NATIONAL COMPARATIVE INVESTIGATION OF THE EFFECTS OF CONSONANT MEDIA COVERAGE Jochen Peter A B ST RA C T Consonant media coverage, i.e. a very similar presentation and evaluation of issues in all media of a particular country, is usually considered to bring about powerful media effects. Though plausible, this notion has hardly been put to the test. This study investigated whether consonant coverage affects support for European integration. The study linked, at the individual level, content analytic measures to survey data in European countries. If the media in a particular country evaluated EU representatives consonantly, then positive evaluations increased people’s support for further European integration, while negative evaluations decreased it. This effect was boosted by the visibility of EU representatives. Conversely, if the media in a particular country evaluated EU representatives dissonantly, then no effect of the coverage emerged. Cognitive selection mechanisms did not protect EU citizens from the influence of consonant media coverage, which indicates powerful media effects. The findings suggest that media effects research may benefit from the ‘concept of powerful mass media’.
‘Return to the concept of powerful mass media’—this is the programmatic title of an article that Elisabeth Noelle-Neumann published about years ago. Along with Tichenor, Donohue, and Olien’s () study on the knowledge gap hypothesis and McCombs and Shaw’s () article on agenda-setting, NoelleNeumann’s () publication is often considered the turning point in the development from the notion of limited media effects to the rediscovery of powerful mass media (e.g. Donsbach, ; McQuail, ; Severin & Tankard, ). In Noelle-Neumann’s article, a concept called consonance plays a central role for the emergence of powerful media effects. By consonance, Noelle-Neumann () This article was first submitted to IJPOR February , . The final version was received September , . A preparatory version of this article with a somewhat different focus and different analyses has been published in German (Peter, a).
AN INVESTIGATION OF CONSONANT MEDIA COVERAGE
means ‘a large extent of similarity in the presentation of certain material in all the media’ (p. ). The basic idea is that, if all media depict and evaluate an issue similarly, citizens hardly have a chance not to be exposed to that information. According to Noelle-Neumann, this presents an ideal situation for the media to exert powerful effects. However, so far neither Noelle-Neumann herself nor other scholars have adequately tested whether consonant coverage does indeed enhance media effects.1 The few existing studies do show that consonant coverage is conducive to media effects (Noelle-Neumann, ; indirectly also: Knoche & Lindgens, ). However, if one wants to demonstrate clearly that consonant media coverage increases media effects, one needs to compare the effects of consonant coverage with the effects of dissonant coverage (for the general importance of comparison in scientific designs, see King, Keohane, & Verba, ). Only if consonant coverage of a particular issue brings about effects while dissonant coverage of the same issue does not, is there evidence of consonance being conducive to media effects. Thus, the first goal of this study is to investigate, in a cross-nationally comparative setting, whether consonant coverage as opposed to dissonant coverage helps to produce effects of media coverage on citizens’ opinions about a particular issue. More specifically, this study investigates whether the tone of EU coverage in terms of the evaluations of EU representatives affects citizens’ opinions about European integration. Due to its transnational character, EU coverage lends itself to cross-nationally comparative studies. The term EU representatives refers both to EU officials (e.g. members of the European Commission) and to people who are clearly associated with the EU (e.g. national ministers when they are members of the EU’s council of ministers). Research has shown that the coverage of political actors plays an important role in shaping citizens’ opinions about political issues (e.g. Cappella & Jamieson, ; Valentino, Beckmann, & Buhr, ). Intuitively, one would presume that consonantly positive coverage of EU representatives leads to favorable opinions about European integration while consonantly negative coverage leads to unfavorable opinions about European integration. Equally intuitively, however, one would presume that the potential influence of a consonant tone towards EU representatives increases with the visibility of EU representatives. Noelle-Neumann () does not elaborate on whether greater amounts of coverage enhance the impact of consonant reporting. Therefore, the second goal of this study is to test whether the visibility of EU representatives (i.e. the number of EU representatives covered) moderates the 1 This may also be the reason why the concept is not mentioned in a recent publication on ‘powerful media effects’ (Rogers, ). There is, however, some content analytic research on the consonance of various types of coverage (e.g. Eilders, ; Knoche & Lindgens, , Weiß, ). For a reinterpretation of existing media effects studies in the light of the consonsonance concept, see Noelle-Neumann & Mathes ().
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
effect of the (consonant) tone of coverage on people’s opinions about European integration. Noelle-Neumann’s () reasoning about powerful media effects of consonant coverage rests upon the assumption that consonant coverage deters people from selectively exposing themselves to media messages. The concept of selective exposure as a protective mechanism against media messages entails a lot of important antecedents and ramifications and is difficult to investigate (for reviews see: Donsbach, ; Zillmann & Bryant, a). However, due to the general lack of research on selective exposure and consonant coverage, it may be worth studying the relationship between the two concepts in the context of media effects on opinions about European integration. Thus, it is the third goal of this study to test whether EU citizens’ protective mechanism of selective exposure is skirted when EU coverage is consonant.
W H A T W E A L W A Y S W A N T E D T O K NO W AB O UT M E D I A E F F E C T S , B U T NE VE R C AR E D T O I NV E S T I GA T E SHORTCOMINGS OF PREVIOUS EVIDENCE To date, there has been little research investigating to what extent consonant media coverage does indeed exert the powerful influence it is presumed to exert. Judged from today, Noelle-Neumann’s () own empirical evidence is methodologically and statistically problematic. Methodologically, content analysis and survey data were loosely linked at the aggregate level. Statistically, no control measures were included. The association found may thus be spurious. Along with the fact that Noelle-Neumann did not investigate media effects when coverage was dissonant, the methodological and statistical shortcomings also impede causal reasoning. The problematic methodological translation of the idea leads to a second, more conceptual problem. The compelling demonstration of media effects requires that the analysis be conducted at the individual level to rule out the possibility of ecological fallacy (Robinson, ). Noelle-Neumann () changes unsystematically between the individual and the collective level—she reasons at the individual level, but analyzes at the collective level. However, if one accepts the primacy of the individual level of analysis in media effects studies, then the concept of consonance has to be slightly specified. No individual is exposed to the entire media coverage in a given country. He/she will only use particular outlets. As a result, the consonance of coverage has to be conceptualized as a ‘surrounding’, contextual factor. The basic question is whether the coverage of a particular outlet, when it is received by an individual, exerts a stronger effect on this individual if the ‘surrounding’ coverage as a whole (i.e. the coverage of all the media in a country) is consonant. More specifically, this study asks whether the particular tone of coverage that individuals get from the outlet(s) they use
AN INVESTIGATION OF CONSONANT MEDIA COVERAGE
affects opinions more strongly if the tone of the coverage in the country as a whole is consonant with the tone of the particular outlet. Tackling this question, this study focuses on the effects of television news coverage as being potentially conditional on the consonance or dissonance of the ‘surrounding’ entire coverage (also including newspaper coverage). Television was chosen because EU citizens consistently name television as the most important source about EU affairs (Eurobarometer –). News coverage was selected because it can be assumed that an institution such as the EU is predominantly covered in the news. A third more theoretical limitation of Noelle-Neumann’s idea may come from conceptually related research on the effectiveness of propaganda and persuasive messages. The power of propaganda is usually rather presumed than empirically demonstrated. Furthermore, the ‘success’ of propaganda may result from the fact that propaganda is typically received in situations of high social control, coercion, or even terror (see Bramsted’s analysis of Nazi propaganda). Empirical studies of propaganda or, more generally, of the effects of persuasive messages on opinions and attitudes have shown that propaganda is ‘contingently effective rather than invariably effective’ (Brown, , p. ). Early studies (e.g. Hovland, Janis, & Kelley, ) as well as more recent investigations (e.g. McGuire, , ; Petty & Cacioppo, ) suggest that the power of media depends on a variety of situational and personal factors. More importantly, the various studies point out that people may learn the basic information presented in the messages, but may not change their opinions, attitudes, or even behavior accordingly (for review, see Perloff, ). T HREE G OALS OF
THE
PRESENT STUDY
If one transfers these findings to the idea of powerful media effects of consonant coverage, it becomes clear that such effects should not be taken for granted. Although Noelle-Neumann () does not elaborate on it, it is important to point out that her idea implies that people do not only learn from the media, but also adjust their opinions to the tone of coverage. Concerning potential effects of the tone of coverage of EU representatives on opinions about European integration, this means that it is crucial to demonstrate two points. First, the tone of television coverage of EU representatives influences citizens in countries where coverage of EU representatives as a whole is consonant, while the tone has no or very little influence on citizens in countries where coverage is dissonant. Second, citizens surrounded by consonant coverage adjust their opinion about European integration to the tone of coverage of EU representatives. If the coverage is consonantly positive, opinions about European integration will be more positive than in countries where coverage is consonantly negative. Technically speaking, I expect an interaction between the (contextual) consonance/dissonance of coverage as a whole and the tone of coverage in the particular outlets that individuals are exposed to.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
The second goal of this study centers upon the question of whether higher visibility of EU representatives enhances the effect of the tone of coverage. Because there is no research to build expectations upon, I start with some basic assumptions. If the tone of coverage of EU representatives positively affects people’s opinions in countries with consonant media coverage, then this effect will become stronger as EU representatives become more visible. However, the less visible EU representatives are, the weaker the effect of the tone on opinions will become. Put differently, there is no homogenous effect of the tone of coverage of EU representatives on opinions about European integration. The impact of the tone of coverage of EU representatives rather depends on the visibility of EU representatives. Greater visibility of EU representatives will boost the effect of the tone of coverage, while lower visibility will impede this effect. Technically speaking, this presents a three-way interaction effect between the individually received tone of coverage of particular outlets, the consonance/dissonance of the coverage as a whole, and the visibility of EU representatives. The third goal of this study is to investigate, in a tentative first step, whether consonant coverage circumvents selective barriers (Noelle-Neumann, , a, b). Noelle-Neumann’s reasoning concerning the power of consonant television coverage is very much linked to the particular situation in Germany in the late s and early s with only two public-broadcasting networks and highly politicized public debates (see especially her reasoning in a). Moreover, the barriers of selective processes are more complex than conceptualized by Noelle-Neumann (e.g. a, b), ranging from selective exposure via selective attention/perception and selective comprehension/interpretation to selective retention (e.g. Donsbach, ; Zillmann & Bryant, b). The variety of selective barriers renders it difficult for messages to get through unfiltered. Given, however, the striking lack of research, NoelleNeumann’s basic idea nevertheless deserves attention. It will hence be investigated whether the selective barrier of an individual’s support for the EU is surmounted by media messages in a consonant media environment. Support for the EU is only a proxy for the more complex selective mechanisms described above, but studies from diverse research areas consistently document that support for politicians, parties, or political groups determines whether and how media messages are received (e.g. Moy, Pfau, & Kahlor, ; Vallone, Ross, & Lepper, ). Generally, people seek information consistent with their opinions (for review see: Cotton, ). As a result, one would expect that EU supporters are influenced by a positive tone toward the EU, while EU opponents are affected by a negative tone. However, if the tone of EU coverage is consonantly negative or consonantly positive across media, EU supporters may permanently be confronted with negative messages and EU opponents may permanently be confronted with positive messages. Therefore, EU supporters and EU opponents will not differ in their reaction to the particular tone of coverage. Both EU supporters and EU opponents will show increasing support for EU
AN INVESTIGATION OF CONSONANT MEDIA COVERAGE
integration if the tone of coverage is consonantly positive. Both EU supporters and EU opponents will show decreasing support for EU integration if the tone of coverage is consonantly negative. Technically speaking, I expect that individual EU support will not moderate the cross-level interaction between the tone of EU coverage of particular outlets that individuals are exposed to and the consonance of the coverage as a whole. In other words, there will be no significant three-way interaction between these two variables and individual EU support. METHOD PROCEDURE AND DATA As a secondary analysis, this study links surveys carried out in all EU member states immediately after the European elections to a content analysis of the television news coverage of the June European election campaign.2 The content analysis was conducted for a two-week period preceding election day. Units of investigation were, per country, the main evening news outlet of both the most widely-watched public broadcasting and private networks. Because Belgium consists of Dutch-speaking Flanders and French-speaking Wallonia, the content analysis included the evening news of the two most widelywatched Dutch- and French-speaking Belgian channels. Private channels do not exist in Austria or were insignificant in Ireland in . As a result, only the public broadcasting channel with the largest reach was included in these two countries. Because only a minority watches the Greek public broadcasting channel, ET, a second private channel was analyzed in Greece. Due to its limited reach in comparison to networks in other countries, the Luxembourg channel was not part of the analysis. Portuguese coding of a crucial category deviated in a non-explainable way from coding in other countries. As a precautionary measure, Portugal was hence excluded from the analysis.3 For further information on the outlets investigated see Table . The unit of analysis was the single news story (defined as semantic entity with at least one topic delimited from another story by a change of topic). Overall, , stories were screened; , of them were political stories (defined as stories explicitly mentioning politics and/or sufficiently depicting political actors). Of the political stories, , were analyzed with respect to visibility and evaluation of EU representatives.4
2 The surveys are part of the European Election Study (EES). The original collectors of the data bear no responsibility for the analyses or interpretations published here. 3 Including Portugal in the analyses presented in this study would render the effects found stronger. 4 Because stories about the Kosovo war heavily dominated the period of investigation and because the focus of the content analysis was on the coverage of the European election campaign, Kosovo stories were only coded if they were linked to the EU. This explains the difference between the total sum of political stories (n = ,), which includes all Kosovo stories, and the reduced sample of , stories, which includes Kosovo stories only if they clearly referred to the EU. The figures refer to the sample without Portugal.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
This study focuses on the effects of television coverage. To receive a more encompassing notion of the dissonance or consonance of the coverage as a whole, it also includes the coverage of the most prestigious newspapers.5 For each country, the front-page of the most prestigious newspaper was analyzed. The front-page presents the most important part of each newspaper and gives a good overview of what a particular newspaper considers important. The most prestigious newspaper was chosen because it can to some extent be seen as representative of a country’s newspaper coverage and may moreover affect the coverage of other newspapers (Dearing & Rogers, ). The newspapers analyzed are documented in Table . The news stories were coded by native speakers, who were trained during the six weeks before coding, tested for inter-coder reliability, and supervised throughout the whole coding period. For each country, the stories were randomly assigned to the coders. The majority of the coding was done at the University of Amsterdam to keep the coding as comparable as possible. For the reliability test, coders of all country groups had to code at least randomly selected television stories per channel. Further information on the coding procedure can be found in Peter (b). To assess citizens’ support for further European integration along with a number of control variables, the (identical) surveys carried out in the EU member states immediately after the European elections were used. The computer-assisted telephone interviews were conducted between June and July , and were based, in each country, on a nationally representative random sample of people older than years of age. The sample size was at least , respondents in Denmark, France, Germany, Italy, the Netherlands, Spain, and the UK and at least respondents in Austria, Belgium, Finland, Greece, Ireland, and Sweden. The response rates varied between percent in Greece and percent in Denmark (see Table for detailed information).6 These low response rates may not necessarily bias the results, as Keeter, Miller, Kohut, Groves, and Presser () have recently demonstrated. MEASURES —DEPENDENT AND KEY I NDEPENDENT V ARIABLES People’s opinion about European integration was gauged with the question ‘Some say European integration should be pushed further. Others say that it has 5 In contrast to the impact of television coverage, the effects of the tone in newspaper coverage were not investigated for two reasons. First, television is the leading medium for information about EU affairs (Eurobarometer –). Second and more importantly, the potentially moderating impact of visibility of EU representatives cannot be investigated for both television and newspapers. The appearance of EU representatives in television bulletins means something else than their mentioning in a newspaper article. The effects reported in this article remain the same if the analyses in which the visibility variable is not involved include newspaper coverage (Peter, a). 6 In Italy, an internet panel with , respondents was used. The Spanish sample was a quota sample. The computation of the response rates is based on a conservative definition of the net sample, which includes the relatively high amount of losses of respondents who could not be contacted at all. The fact that no contact at all was made with the person to be interviewed might, however, also indicate corporate lines, fax numbers etc., i.e. quality-neutral losses.
AN INVESTIGATION OF CONSONANT MEDIA COVERAGE
already gone too far. What is your opinion?’ Response categories ranged from (unification has already gone too far) to (unification should be pushed further). The average support for European integration in the various countries investigated can be found in Table . More information about the political situation and the contentiousness of the EU in the countries is available in Peter (b). Visibility of EU representatives was measured as the number of their appearances in different stories. Up to six actors could be coded per story, but the same actor was coded only once per story. To qualify as a story actor, a person, group, or institution had to be either depicted and mentioned at least once, or quoted and verbally mentioned, or mentioned verbally at least twice. If more than six actors qualified in a news story, the actors to be coded were selected in terms of their importance for the story (operationalized as amount of information given about a particular actor, frequency of being mentioned, visibility, and quotes by a particular actor). For each actor, it was coded whether the actor was, as an individual or group, related to the EU on any other level (e.g. regional, domestic, or world level). For the analysis, the individual and the group code for both EUrelated and EU-unrelated actors were collapsed. For the coding of the actors as EU-related or EU-unrelated, the average inter-coder agreement was .. Evaluation of EU representatives and of other actors was assessed by coding explicit judgments of story actors. The coding categories were (neutral), (unfavorable), (mixed), and (favorable). I computed, for every outlet, the difference between the number of favorable and the number of unfavorable evaluations of EU representatives. The evaluation differential is based upon absolute figures because the absolute number of evaluations was sometimes very low and percentages would grossly distort the outcomes. The evaluation differential was also preferred to the mean evaluation. Kepplinger and Maurer () have recently demonstrated that the evaluation differential is more comparable to people’s perceptions of evaluative tendencies in the media than the mean evaluation and seems, thus, more appropriate for effect-oriented studies. The inter-coder reliabilities, measured as the average agreement between the coders, was .. Both measures, the tone of coverage of EU representatives and the number of EU representatives covered, were centered around their mean to avoid multicollinearity problems in the analysis of interaction effects (Aiken & West, ; Jaccard, Turrisi, & Wan, ). People’s general support for the European Union as proxy for more complex selective mechanisms was assessed with the question ‘Generally speaking, do you think that membership of the European Union is a good thing, a bad thing, or neither good nor bad?’ Recoding of the response categories led to a scale where equals a bad thing, equals neither good nor bad, and equals a good thing. For the analysis of whether consonance skirts selective barriers, the support variable was centered around its mean. Whether the coverage was consonant or dissonant in its tone towards EU representatives was assessed by checking, per country, the direction of the tone in
− − − − − − + − − − −
Greece
Germany
France
Finland
Denmark
+ − −
Antenna ‘Ta Nea Tou’ (pr) Mega ‘Kentriko deltio’ (pr) ET ‘News’ (pb)
Belgium– Flanders Belgium – Wallonia
VTR ‘Het Journaal’ (pb) VTM ‘Nieuws’ (pr) La Une ‘JT Meteo’ (pb) RTL ‘Le Journal’ (pr) TV ‘TV-Avisen’ (pb) TV ‘Nyhederne’ (pr) Yle ‘Finish News’ (pb) MTV ‘News’ (pr) TF ‘Le Journal’ (pr) F ‘Le Journal’ (pb) ARD ‘Tagesschau’ (pb) RTL ‘RTL Aktuell’ (pr)
ORF ‘ZiB’ (pb)
Evaluation Number
EU representatives in TV news
+
TV Channels analyzed
Austria
Country
TABLE Background information and figures
−
Frankfurter Allgemeine Zeitung Kathimerini
−
−
−
No
No
Yes (–)
No
Yes (–)
No
Yes (–)
No
.
.
.
.
.
.
.
.
Mean Evaluation Consonant coverage support for of EU European representatives integration in newspaper
Le Monde
Morgenavisen. Jyllandsp Helsingin Sanomat
La Libre Belgique
De Standard
Die Presse
Newspaper analyzed
.
.
.
.
.
.
.
.
Response rate survey
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
Note: (pb) = public broadcasting; (pr) = private
UK
Sweden
Spain
Netherlands
TVE ‘Telediario’ (pb) Tele ‘Telecino’ (pr) TV ‘Rapport’ (pb) TV ‘Nyheterna’ (pr) BBC ‘Nine o’clock news’ (pb) ITV ‘News at .’ (pr)
RTE ‘News (.)’ (pb) Rai Uno ‘TG’ (pb) Canale ‘TG’ (pr) NOS ‘Het Journaal’ (pb) RTL ‘Nieuws’ (pr)
Ireland
Italy
TV Channels analyzed
Country
TABLE (continued)
100
− − + + + − + +
Evaluation Number
EU representatives in TV news
Guardian
Dagens Nyheter
El Pais
NRC Handelsblad
Corriere della Sera
Irish Independent
Newspaper analyzed
Yes (+) No No
− −
No
No
Yes (–)
+
−
−
.
.
.
.
.
.
Mean Evaluation Consonant coverage support for of EU European representatives integration in newspaper
.
.
Quota sample
.
. Internet panel
Response rate survey
AN INVESTIGATION OF CONSONANT MEDIA COVERAGE
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
three outlets—in the main evening news of the most widely-watched publicbroadcasting and private networks and in the most prestigious newspaper. If the direction in all three outlets was identical (i.e. either positive, neutral, or negative), the coverage in the particular country was defined as consonant. If one of the three outlets deviated in its tone from the tone of the remaining two outlets (e.g. two outlets are negative, one is neutral), the coverage in the particular country was defined as dissonant (see Table for further information). CONTROL VARIABLES AND MISSING V ALUES A number of control measures were included in the analysis to rule out the possibility of spurious media effects on citizens’ opinions on European integration. Research on the antecedents of opinions about European integration and on alternative explanations of media effects suggest that the following variables should be controlled for: utilitarian motives (e.g. Gabel, ); party cues (e.g. Franklin, Marsh, & McLaren, ); left–right position (e.g. Budge, Robertson, & Hearl, ); satisfaction with domestic democracy (e.g. Anderson, ); political interest (e.g. Anderson, ; Holtz-Bacha, ); interpersonal communication (e.g. Lazarsfeld, Berelson, & Gaudet, ; Schmitt-Beck, ); media exposure and media attention (e.g. McLeod & McDonald, ); and demographics (e.g. Anderson, ; Janssen, ; Gabel, ). As outlined above, the evaluation of EU representatives is computed with absolute figures, and this operationalization might be affected by how frequently particular outlets cover EU representatives. In order to preclude spurious effects, the number of EU representatives covered was controlled for in the models in which visibility of EU representatives did not present a moderating variable. Additionally, I controlled for EU support in the models in which this variable does not have a moderating function because opinions about European integration may not present consistent entities (Saris, ). Because EU citizens typically lack knowledge about EU affairs (e.g. Anderson, ; Janssen, ), they may rationalize and render consistent the little they know about European integration (for general information on this problem, see e.g. Zaller & Feldman, ). The operationalization of the control variables is available from the author upon request. In order to prevent a large number of cases being lost due to missing data (when using listwise deletion), I replaced missing values by valid scores derived from other, related, responses.7 Remaining missing values were replaced either 7 For example, if a respondent had not placed himself/herself on the left–right scale, but had indicated the likelihood of voting particular parties, his/her left–right position could be concluded from where he/she had placed the particular party on the left–right scale. Similarly, political interest was concluded from people’s participation in political events, attention to European news was deduced from people’s exposure to European election news, news exposure from media use, etc.
AN INVESTIGATION OF CONSONANT MEDIA COVERAGE
by mean substitution (for metric variables) or by the modal value (for dichotomies). In order to minimize the danger of arbitrary data modification, the dependent variable and the measures of general support for the European Union and satisfaction with domestic democracy were excluded from this procedure because they are hard to impute. DATA A NALYSIS Because the survey sample sizes varied across countries, the data had to be weighted. The total sample (without Luxembourg and Portugal) was , respondents. Due to its bicultural character, Belgium was split in its Flemish and Wallonian part resulting in an overall figure of ‘systems’ for analysis. The samples of each of the systems were weighted such that each system had the same sample size while the original total sample size was preserved. Per system, the particular evaluation and visibility of EU representatives were added to each respondent who regularly watched one or both of the two news outlets that were content analyzed. For example, in Britain, those respondents who watched the BBC o’clock news, or ITN’s News at :, or both, were assigned the EU coverage measures. This means that, per system and per measure, three different values were assigned. Not all respondents regularly watched one or both of the outlets. As a result, the overall sample sized reduced to , respondents. However, the selected sample did not meaningfully deviate from the original sample. Additionally, each respondent in a particular system received a value indicating whether he/she received information in a media environment where the tone towards EU representatives in the coverage as a whole was consonant (coded as one) or dissonant (coded as zero). In regular OLS regressions, the standard error of cross-level interactions is underestimated. In this study, this implies an increased risk of a type-I error with respect to the expected cross-level interaction between the country-level factor consonance/ dissonance and the individually received tone of coverage. Therefore, the standard error was corrected with the so-called ‘sandwich’ estimation of the standard error (Huber, ; White, ). This correction entails that the critical t-values are assessed on the basis of degrees of freedom (i.e. number of systems minus one; for a more detailed explication of this problem and for a discussion why multi-level modeling is not appropriate for the data at hand, see Peter, b). Besides the correction of the standard error, the data analysis differed from ordinary regression analyses in a second aspect. The expected two-way and three-way interactions were post hoc probed for significant difference of the simple slopes from zero.8 This 8 By simple slope the slopes of the focal independent variable (i.e. tone) as conditional on the values of the first-order moderating variable (i.e. consonance/dissonance) and, in a three-way interaction, of the values of the second-order moderating variable (i.e. visibility) are meant.
INTERNATIONAL JOURNAL OF PUBLIC OPINION RESEARCH
procedure has been suggested by Jaccard et al. (, pp. –, –) and by Aiken and West (, chs and ). R E S UL T S DOES CONSONANCE ENHANCE MEDIA EFFECTS? The first goal of this study was to establish whether media effects are more powerful if coverage is consonant than if coverage is dissonant. More specifically, the tone of television coverage of EU representatives was expected to affect people’s opinions about European integration only if the overall tone of coverage was consonant with the specific tone of coverage an individual receives. It was further presumed that, in this case, citizens would adjust their opinion about European integration to the tone of coverage. Model in Table shows a significant interaction between the tone of coverage of EU representatives and consonance/ TABLE Impact of various cross-level interactions on support for European integration
Key variables Tone of coverage Visibility of EU representatives EU support Consonance Two-way interactions Tone X Consonance Visibility X Consonance Tone X Visibility Tone X EU support EU support X consonance Three-way interactions Tone X Consonance X Visibility Tone X Consonance X EU support Constant R square
Model (n = ,)
Model (n = ,)
Model (n = ,)
−. (.)
−. (.) . (.)
−. (.)
−. (.) . (.)*
. (.)** . (.) . (.) # . (.) −. (.) #
−. (.) . (.)* . (.) . (.)
. (.)* −. (.) −. .
−. .
−. .
# = p
