1 Issue or Candidate? Mediatized Voting Behaviour in Germany Introduction Personalization is a prominent topic in communication research. Mostly studied in the context of election campaigning and election campaign coverage the concept expands to different domains, like economics (Brettschneider and Vollbracht, 2010). Like Adam and Maier point out, personalization of political communication addresses two separate jet different developments: First a shifting focus from institutions and issues towards political candidates. Secondly a shift in the depiction of politicians from political to personal traits (Adam and Maier, 2010). These developments are joined by the personalization of voting behaviour, meaning that the political personnel and the top candidates in particular play an increasing role in decision making process of the voter. Allegedly this is paired by a decreasing importance of cleavages, partisanship and issue orientation. The debate over the personalization of politics is often embedded in broader trends often with a distinct normative perception. The supposed increasing focus on politicians in European parliamentary democracies is frequently discussed under the label of Americanization, denoting that political coverage outside the US resembles at least in aspects of the US American way to cover politics (and conduct election campaigns) (Negrine and Papathanassopoulos, 1996; Blumler, 1999; Brants, 1999). Personalization meets the prerequisites of the media that is unable or unwilling to cover complex issues(blumler, 1990: p. 104). Like in a system of communicating vessels less space is available to issues, parties and societal institutions when the media attention more and more focus on candidates seems to be a common belief.
2 Numerous studies have been conducted on personalization of political communication on one side and on candidate voting on the other side. Frequently studies of media content in regard to personalization refer to the changes in the electorate as a reason for their undertaking. On the other hand changes in the media coverage of politics often is named as the culprit for the increasing influence of candidates on the vote. This paper endeavors to fill this gap by linking candidate voting to media consumption and thereby at least indirectly to media content, it will not focus on the normative implications of the personalization debate. We therefore look at voting behavior in Germany over a time span of 20 years starting with the re unification election of 1990 as documented in nationwide surveys mainly carried out in the framework of the German Longitudinal Election Studies (GLES). Some remarks on personalization McAllister argues that there is little doubt that politics has become more personalized over the past half century (cf. McAllister, 2007, p. 584). Two sets of reasons can be identified in the literature. Frist the growing dealignment in western democracies, the weakening of traditional bonds to the parties, is met by a stronger focus on the candidates in election campaigns (Dalton et al., 2000). Secondly the increasing importance of television paired with the introduction of commercial channels imposes the modes of presenting politics on the parties (Blumler and Kavanagh, 1999: p. 212), demanding talking heads, sound bites and top rank party representatives as ingredients for the televised news stories. Already Galtung and Ruge have pointed out, that references to elite people and to persons increase the newsworthiness of an event (Galtung and Ruge, 1965: p.70-71), an notion that has been empirically analysed in divergent contexts (Schulz, 1976; Staab, 1990; Ruhrmann et al., 2003). Personalization embodies one symptom of the syndrome that Altheide and Snow named Media Logic (Altheide and Snow, 1979),
3 which describes who media sees and interprets reality comprising the selection of aspects of reality (Mazzoleni, 2008). The media logic therefore becomes more and more basic condition for the parties in their attempts to reach the voters. It seems plausible that politicians gain media visibility and importance in the process of political communication when we take the weakening ties to the parties on the one hand and the growing power of the media and the spreading media logic on the other hand in to account. We should expect parties to concentrate more on their top candidates in order to adapt to the news selection routines and to appeal to the more volatile electorate. The media logic is supposed to more pronounced in commercial television channel (Schulz and Zeh, 2006: p. 279). In order to tackle the growing competition, public broadcasting adopts successful formats and ways of presenting news from their commercial counterparts. Rooted in media logic personalization can be linked to mediatization (Asp, 1986). This process of mediatization entails that media are independent institutions upon which other parts of society depend (Hjarvard, 2008: p 113). Strömbäck and Dimitrova (2011: p. 34 45) as well as Strömbäck and Esser (2009: p. 214 216) distinguish four dimensions of mediatization. The first dimension measures the degree to which media are the most important sources for information, while the second dimension rate the degree of independence of the media from (political) institutions. The third and the fourth dimension are closely linked to media content. Whether the media follow a political logic or their own in covering current affairs relates to the third dimension, that is, it deals with the question of who is in control of the agenda. Following the mediatization hypothesis, one would expect a gradual shift of control from the political arena to the media. Finally, the fourth dimension covers the degree to which other societal subsystems follow this logic replacing their usual rules and ways in which they used to be governed.
4 Theoretically, it is important to distinguish between the last two dimensions. Changes on the third dimension should be traced by analysing media coverage while changes on the forth dimension affect primarily the behaviour of societal actors such as politicians. In this line of thinking, personalization of media coverage would be an indicator for mediatization connected to the third dimension, whereas the personalization of politics and campaigning would be an indicator for mediatization on the fourth dimension. Wattenberg observed that American newspapers and magazines focused increasingly on politicians to the expense of parties from 1952 till 1980 (Wattenberg, 1994). In Britain Langer observes increasing visibility of the prime minister in the Times over a period from 1945 to 1997(2007). While the high correlation between time and media visibility suggest a clear and linear rise in the focus on the prime minister, one has to note that Harold Wilson was in the 1960ties as visible as Tony Blair 30 years later (cf. Langer, 2007: p. 367). On the other hand she observed a sharp increase in references to the private life of the British prime ministers, something not seen in the Times before John Major (Langer, 2010). German newspapers on the other hand show no clear trend towards more attention to the chancellor candidates in there coverage of elections in analysis of Wilke and Reinemann covering the period from 1949 to 1998 (2001), a picture that does not change extending the timespan over the more recent elections (Wilke and Leidecker, 2010). Looking at television, which is supposed to be more prone to personalize coverage, again no clear tendency toward more candidates can be found. Although Schönbach and Semetko found a rising attention form television news throughout the 1990ties, (2000), longer term studies are less clear in this regard (Schulz and Zeh, 2010).
5 Campus describes Berlusconi and Sarkozy as prime examples of mediatized leadership since their ascent allegedly linked to their mastery of marketing strategies and news management (Campus, 2010: p. 231). But it remains to be seen, whether they really pave the way to a new generation of mediatized leaders. (Campus, 2010p. 232). The rise and fall of Gerhard Schröder in Germany hints, that expertise in political marketing might backfire, when they don t suffice to solve or at least cover up the countries problems. Schröder, who once claimed that he just needs Bild, Bams und Glotze [two tabloids and TV] (Kilz, 1999) to govern was pushed out of the chancellery by Angela Merkel not (yet) known for her ability to spin the news. In an international comparative context, Kreisi finds, that the empirical evidence concerning the personalization of politics thesis is at best mixed. Revealing no trend towards personalization in quality newspapers nor in television news (Kriesi, 2011: p. 17). Nevertheless the lack of prove of a trend does not imply that the concept is useless, when it does not follow a steadily increasing path. The degree to which politics or coverage is personalized varies from year to year, between countries and system levels, on top of that it also depends on the politicians in concern and the circumstances. Under such conditions, one should not expect a simple trend, especially when looking at a longer timespan. Elections with weak candidates are less prone to be personalize by the campaign management and by the media. All things equal, race run under the impression of pressing national problems or even crisis likewise should less personalized. Candidate voting When looking at personalization of coverage or campaigning one might implicitly connect it to the increased importance of candidates in the decision making process of the voters.
6 The Michigan model of voting based on the idea of Campbell et. al. (1960) differentiate between long and short term factors determining the individual voting decision. The most important long term factor is the allegedly declining party identification, sometimes also called psychological partisanship, in theory this factor is deeply rooted in the mind of the voter. Preferences for candidates and the perceived competence in solving the countries (or the respective entity the election is held for) most important problems are factors subjected to more change according to the circumstances of the election in question Figure 1 illustrates this social psychological model of voting. Figure 1 simplified voting model Studies of candidate voting mostly ask whether the influence of candidates becomes more important on the individual voting intention compared to the other factors. Vetter and Gabriel analyzed (West )German voting behavior in the national elections from 1972 1994. In separated regressions for the two major parties CDU (Conservative) and SPD (Social Democrats) they found that party identification and issue competence explained about 70% of the variance in voting intention. The candidate effect was comparatively small, reaching 10 15% in 1994 but not showing a systemic pattern to corroborate the hypothesis of an increased candidate impact on the vote (Vetter and
7 Gabriel, 1998). Using partially the same survey data, Ohr came to the conclusion that the impact of the candidate preference on the German vote has risen from 1961 till 1998, especially when looking at the elections of 1994 and 1998 (2000). The different assessment based on essentially the same data more than else might be attributed to the fact, that 1998 marked the and of Ohr s analysis. The 1998 was highly personalized in two ways, first it was a plebiscite against the long term chancellor Kohl and secondly he was challenged by the allegedly telegenic and comparatively young opponent Gerhard Schröder. Comparing USA, Great Britain and Germany, Brettschneider detected no support for expanding candidate effects on individual voting intentions, whereas the institutional setting mattered (2002). Naturally the US American presidential elections were more personalized, than general elections in Germany and Britain, but again, party identification and issue competence accounted for most of the variance in the reported voting behavior. Mughan, looking at the media visibility of British party leaders and it s impact on voting, finds that from 1960ies onwards the general election has become more presidentialized (2000). Meaning, that party leaders gained media visibility over the parties themselves, and that increased candidate voting can be observed. To sum up, there seems to be a considerable degree of confusion on the question of candidate voting. Analyses based sometimes on the same data seem to differ in results, even when using the more or less the same indicator. Several explanations might be given for muddle. First different time frames are applied. When candidate voting increased in two subsequent elections at the end of a time frame, it is often greeted as a prove for the increased impact hypothesis. Adding an additional election might change the overall
8 picture again, since we are dealing with five measurement points (=elections) even when we cover a time span of 20 years. Secondly the modeling differs: Sometimes OLS Regressions are used (Gabriel and Vetter, 1998), sometimes logistic regressions (Ohr and Klein, 2001), sometimes models are calculated separately for the major parties, sometimes for major parties combined. Thirdly, multicollinearity is a big issue, although it seems plausible that the individual voting decision is strongly influenced by candidate preference, assessment of competence and party identification, empirically these indicators are highly correlated. This does not affect the overall fit of the models but renders it impossible to calculate the individual impact of the predictors on the vote (Rottmann and Auer, 2010: p. 503 515; Urban and Mayerl, 2011). Depending on the order in which the predictors are entered into the model, the effect of candidate preference might be obscured, since there is no or little variance left to be explained by it, when party identification and issue preference are already controlled for. Research questions, data and design The aim of this paper is not add just a few additional measuring points to the candidate voting question. Rather it seeks to connect the impact of candidate preference to media use and subsequently to the amount of personalization of the election coverage. Hence the first research question reads as follows: RQ1 Is there a link between media use and candidate voting? Since the media logic is most pronounced in television (news) two hypotheses can be connected to this research question: H1 Watching TV news increases the impact of the candidate preference on the voting intention.
9 As we also expect newspapers to focus more on issues, reading them might serve as an antidote to candidate voting: H2 Reading quality newspapers reduces the impact of the candidate preference on the voting intention. If media uses should have an impact on how voters form there decision, it should be linked to media content. Therefore a second research question is asked: RQ2 Is there a link between media content and amount of candidate voting? The first research question is directed to survey data that can be retrieved form the central archive of social sciences in Germany (Gesis). Corresponding to the focus of this paper, the Bundestag elections from 1990 are analyzed. For each election a roughly comparable set of data is available. The data sets used are summarized in table 1. Table 1 Data for the analyses 1 Election Survey Nr. Sample Size Wave 1990 ZA1919 1451 3rd wave (only West Germany) 1994 ZA3066 2070 pre election 1998 ZA3067 1633 pre election 2002 ZA3861 1665 pre election 2005 ZA4322 2540 post election 2009 GLES1103 1.3 2173 pre election 2013 ZA5700 2003 pre election Apart from 2005, which was in several ways a special case, the surveys conducted prior to the election day were used. In 2005, due to the premature call for an election no preelection data was available. Table 2 summarizes the variables used in the following regression analyses. Only for the media uses variable we observe a disturbing 1 The data and detailed study descriptions (in English) are available online: http://zacat.gesis.org/webview/ and http://gles.eu/index.en.htm
10 discontinuity, in stead of directly asking the amount of days per week a certain media is used an ordinal measure has been applied. All political variables are recoded according to the same logic, that was already implemented in a previous study by Ohr and (2000p. 282) and by Bartels (Bartels, 2000: p. 38) (see table 2). Table 2 Operationalization Variable Operationalization Coding Voting Intention Party Vote, next Sunday Question 1= Cons; 0 = Other; 1 = Soc* Candidate Preference Which candidate preferred as Chancellor 1= Cons; 0 = Other/none; 1 = Soc Issue Preference Which parties will solve the most important issue 1= Cons; 0 = Other; 1 = Soc Media use TV News/ National 0=never to 7= every day Newspaper except 1994: 0=never to 5=very often *Cons = Conservative Party/Candidate (CDU/CSU)Soc = Social Democratic Party/Candidate (SPD) To test our assumptions several multinominal logistic regressions were calculated for each election year. The reported voting intention serves as the dependent variable in each model, the category other party is used as the reference category for the voting intention for the mayor parties. First we assess to which degree candidate voting play role compared to issue competence. We deliberately did not entered party identification into the models for two reasons: First, we are interested in predictors, that themselves are subject to short term change, partisanship is supposed to be rather stable as mentioned above. Secondly, we fear that by controlling for party identification we would obscure our findings due to the additional multicollinearity we add to the models. To answer the first research question, we are not interested in the overall effect size of candidate voting, but in the amount it is boosted (or hampered) by media use. More technically, we are testing for the impact of the interaction between candidate preference and the quantitative measure for news consumption. Entering interaction effects into a regression is a secure source for multicollinearity, when at least one term of the interaction effect enters the
11 model. Yet by centering the predictors around their grand mean, this kind of multicollinearity can be leveled out (Urban and Mayerl, 2011: p. 238 240). Adding to that, a direct effect between a news consumption and a preference for a certain party should not be expected. Table 2 also implies, that we are concentrating on the two major parties in Germany, the conservative CDU/CSU and the Social democrats (SPD). Up to the current election, they still attracted a vast part of the total vote and they are the only parties, that run for the chancellor office. 2 Substantial candidate effects on voting therefore should be observable only for the major parties. To tackle the second research question, we draw on data from content analyses of the television news coverage of the hot campaign phase, starting usually four weeks prior to the election day. The main evening news bulletins of the two nationwide public broadcasting channels (ARD and ZDF) and the two major commercial channels (RTL and SAT.1) were screened for news items at least mentioning one of the two chancellor candidates. The data have continuously been gathered after each election campaign. The analyses are based on the same set of indicators. Originally, they were collected for the study of political balance in the election news coverage of German chancellor candidates and its impact on voting (Kindelmann, 1994; Zeh, 2005; Schulz and Zeh, 2007, 2010). Since only a crude indicator for the degree of personalization of the coverage is derived from this data, we abstain to deliver detailed methodological information already published elsewhere. The indicator consist of the relative proportion of news bulletins 2 The German Chancellor is elected by the parliament. Still both major parties nominate a candidate who will usually get elected, according to which of the two party gains the relative majority in the election. 2002 was a particularity, when the liberal party appointed a third chancellor candidate. Since there were not in the position of
12 mentioning at least one of the two chancellor candidates in relation to the total number of news bulletins. The data is highly aggregated for each election, thus reliability should not be an issue anyway. Findings Clearly, the power of the two major parties is declining steadily in Germany, as can be seen in Figure 2. In 2009 only 58% of the respondents reported, that they intend to vote for one of the two catch all parties. Yet, since none of the other parties was close to the big two, it seems to be legitimate to concentrate on them, modeling the vote in Germany basically as a left vs. right decision. Recent state elections in Germany suggest, that the country might be shifting to a three party system, wherein the Green party might become the third pole. 3 This development is paralleled by the reduction in explanatory power of the voting models as expressed through the models pseudo R² seen in table 3 5. Some doubt has to be cast on the quality of the 2005 survey, since it yields a ten percent lead of the Social democrats over the Conservative party, that actually won the election held only a few days before field start of the survey. 4 3 The state of Baden Württemberg is now governed by a green prime minister. Whether this is just an effect of the nuclear disaster in Japan or a sustainable development, remains to be seen. 4 Even when applying the representative weights on the data, the lead of the losing party holds substantially.
13 Figure 2 Share of vote for the two major parties In the first set of regressions it was tested whether issue or candidate voting had the bigger impact on the voting intention reported. Not surprisingly, we don t find a trend or pattern along the timeline. In four of the six elections analyzed, issue voting supersedes candidate voting. Again, not surprisingly these models again confirm the strong impact of the candidates in the 1998 election. The candidate preference worked from both directions: a referendum against the incumbent Helmut Kohl and a positive preference for the Social democrat Gerhard Schröder (cf. Gabriel and Brettschneider, 1998). Table 3 Multinominal regression: candidate vs. issue voting Likelihood Coefficient 2 Log McFadden Candidate Likelihood Pseudo R² Preference Issue Competence N 1990 1432,4 0,48 240,44 281,41 1374 1994 1537,2 0,50 266,88 358,08 1405 1998 1375,0 0,43 173,82 154,74 1121 2002 1379,4 0,43 117,29 224,47 1117 2005 2582,9 0,38 350,01 268,20 1872 2009 1704,5 0,43 156,80 468,62 1399 2013 All coefficients are significant on the p < 0.001 level Reference category for the dependent variable: other parties
14 The strong impact of candidate voting in 2005 is less obvious, since the defeated Chancellor Schröder led over Angela Merkel in the voters preference for chancellor even in other polls, that reported voting intention much closer to the real outcome of the 2005 election (cf. Forschungsgruppe Wahlen, 2005). Table 4 Multinomial regression: candidate voting & TV news consumption 1990 1994 1998 2002 2005 Conservative Vote Social democratic Vote B SE B SE Likelihood ratio Cand. pref 2,038*** 0,158 1,052*** 0,101 1068,2*** Candpre*TVnews ( 0,048) 0,055 0,114** 0,046 10,2** Cand. pref 2,871*** 0,168 0,998*** 0,105 1191,9*** Candpre*TVnews 0,198# 0,103 ( 0,074) 0,079 (4,0) Cand. pref 2,154*** 0,150 1,521*** 0,144 920,4*** Candpre*TVnews (0,05) 0,062 0,199*** 0,053 13,6*** Cand. pref 1,585*** 0,108 1,792*** 0,192 840,8*** Candpre*TVnews (0,024) 0,054 0,148** 0,054 (7,3) Cand. pref 1,623*** 0,097 1,769*** 0,111 1281,6*** Candpre*TVnews ( 0,044) 0,040 (0,036) 0,045 (2,2) McFadden Pseudo R².39***.37***.37***.33***.30*** 2009 Cand. pref 2,498*** 0,168 1,277*** 0,103 863,7***.28*** Candpre*TVnews 0,104* 0,049 0,097** 0,042 10,9** Cand. pref 2013 Candpre*TVnews *** p <.001; ** p <.01; * p <.05; # p <.10; Coefficients in brackets are not significant Reference category for the dependent variable: other parties Table 4 and 5 address the first research question. As said before to avoid the natural multicollinearity, just candidate preference was used as a predictor for the voting intention. Surely we risk to overestimate candidate voting, but the main aim of these models is to test the interaction effect between candidate preference and news consumption on the reported voting intention. Not as a surprise, these parsimonious models explain much less of individual voting behavior. The interaction term has a statistically significant overall impact on the vote in four out of six elections, as tested trough the likelihood ratio. H1 postulates, that watching more TV news increase leads to
15 an additional impact of candidate preference. Looking at the signs of the regression coefficient, this holds true in every case, at least when the likelihood ratio proves to be significant. The sign for the regression coefficient for the Conservative party has to be negative, likewise positive for the Social democratic vote. 5 The results strongly support H1. Leaving 2005 out of the picture four out five pre election surveys deliver the expected results. The models that incorporate the interaction term between candidate preference and reading quality newspapers yield ambiguous results. We expect, that the quality newspapers report substantially more on issues and party platforms than TV news. Therefore we should expect those frequently reading these paper to rely less on candidate preferences for the electoral decision making. 6 5 At first this seems puzzling, but one was to reckon, that the negative sign of the coefficient means, that the odd ratio for voting conservative is reduced, when the candidate preference (or the interaction term) moves towards the social democratic candidate. 6 The obvious notion, that reading might boost issue voting is not reported in this paper.
16 Table 5 Multinomial regression: candidate voting & newspaper reading 1990 1994 1998 2002 2005 Conservative Vote Social democratic Vote B SE B SE Likelihood Coefficient Cand. pref 2,015*** 0,159 1,023*** 0,101 1019,3*** Candpre*TVnews (0,318) 0,266 (0,108) 0,236 (1,4) Cand. pref 2,877*** 0,168 0,984*** 0,104 1186,0*** Candpre*TVnews 0,457# 0,270 (0,312) 0,216 6,1* Cand. pref 2,139*** 0,150 1,560*** 0,145 923,0*** Candpre*TVnews ( 0,252) 0,278 1,018*** 0,234 18,4*** Cand. pref 1,570*** 0,109 1,791*** 0,192 821,1*** Candpre*TVnews 0,500# 0,285 ( 0,437) 0,282 4,9# Cand. pref 1,601*** 0,099 1,789*** 0,115 1206,7*** McFadden Pseudo R² 2009 Candpre*TVnews ( 0,164) 0,212 (0,273) 0,216 (2,5) Cand. Pref 2,507*** 0,168 1,310*** 0,102 903,3*** Candpre*TVnews ( 0,058) 0,295 ( 0,243) 0,261 (0,9) Cand. Pref 20123 Candpre*TVnews *** p <.001; ** p <.01; * p <.05; # p <.10; Coefficients in brackets are not significant Reference category for the dependent variable: other parties.38***.37***.37***.33***.30***.27*** Only two models show a significant overall impact of for the interaction term. According to the logic of the multinomial logistic regression, we should expect a negative regression coefficient for the social democratic vote (and a positive one for the conservative vote). This proves to be the case only in 1998. Those how frequently read newspapers in that election, were less likely to rely on their preference for Gerhard Schröder. Apart from this singular finding nearly all other coefficients point against the hypothesized direction, i.e. to an increase in candidate voting 7. Different form the models with the TV news interaction, we do not find a systematic and hypothesized pattern. Therefore we have to reject H2. 7 If we lower the threshold of significance to the.10 level, 1994 and 2002 deliver additional candidate voting effects among newspaper readers, but just for the conservative candidate.
17 The reason why we do find rather strong effects for television and not for newspapers might primarily the question of reach. While the two major newspapers are each read by round about 1 million people daily, the news show of the first public channel has an audience of over 8 million viewers (Zubayr and Gerhard, 2011). The last research question turns the attention again towards TV news. From the content analyses a crude indicator for the personalization of the coverage was aggregated. It expresses the ratio between news stories with at least on reference to a candidate and the total number of stories in the four bulletins analyzed. In order to assess whether the results shown in table 4 are related to the actual coverage, the personalization indicator is plotted against the average regression weight for the interaction term. 8 Figure 3 Personalization and candidate voting The graphic inspection supports the notion that personalization of coverage and has an impact voting behavior. With the exception of 2002 the two lines are nearly parallel. If the coverage of a campaign is more personalized, it will increase the amount of candidate 8 Actually, the comparing the unstandardized B is in general problematic, but all the models are calculated on the same variables with identical coding. On Top of that the sample size and the R² of the models are similar, we dare to use this as an indicator for the total effect size. If the coefficient was negative, the sign was omitted to assess the size of the effect (and not it s direction)
18 voting among those voters that use television more frequently. Whether the degree of personalization of the coverage is media driven or campaign driven is not relevant in this respect. Voters get in touch with the it mainly through media and especially through television, this still holds true for the 2009 election, where TV was the most use channel of information (Geese et al., 2009: p. 638). Discussion The absence of a trend does not necessarily imply the uselessness of concept. Neither personalization nor candidate voting have followed the hypothesized path. The degree to which an elections personalized for the degree of candidate voting is still largely a matter of circumstance. But this does not mean that the hypothesis is falsified. Obviously reality is more complex than a linear trend. Sometimes an election is personalized and the voters care more for issues, sometimes the voters care for certain candidates without the additional help of the campaigners or the media. Therefore it does make sense to look for effects. According to a large body of research personalization cannot be seen as the direct cause for candidate voting. But as we have shown here, it is still connected. It depends on the individual level of television news consumption whether personalized coverage has an impact on voting behavior. The interaction effect basically shows that candidate preference becomes more important for the voting decision when more television news is watched. The effect is small, but it is large enough to be traced in our analysis and it is stable over time. Above all it plausibly relates to an indicator from the coverage. Yet it is also conceivable that the effect are found mainly for television, since it is the most important media channel, still. In 2009 young voters used the internet their prime source of information for the election nearly as often as they used television (Geese et al., 2009).
19 Adding to that we would like to point out to some restrictions of our study. Strict sociodemographic controls have not been applied, mainly to for the sake of parsimonious models. But media use might be correlated with certain demographics. Newspaper readers differ in age an education from the average population. At least the television news viewership more heterogeneous thus we could argue that the interaction effect is more than a spurious correlation. Another limitation addresses the party setup analyzed. Basically Germany is treated as a two party system, this is in fact less and less true as discussed with figure 2 above. Therefore the results are limited to the two big parties and it remains unclear, to which degree personalization or candidate voting is an issue for small parties. Furthermore we think that our paper has shown a fruitful way of studying media effects where the quest for direct effects has failed. In analogy with Cohen s dictum we argue that the media may not be successful, much of the time in telling people what to think, but it is stunningly successful, in telling its users how to think (cf. Cohen, 1963). References Adam S and Maier M (2010) Personalization of Politics. Communication Yearbook 34: 213 257. Altheide DL and Snow RP (1979) Media logic. Beverly Hills, CA: Sage. Asp K (1986) Mäktige massmedier Studier i politisk opinionsbildning. Stockholm: Förlaget Akademilittertur AB. Bartels LM (2000) Partisanship and Voting Behavior, 1952 1996. American Journal of Political Science 44(1): 35 50. Blumler JG (1990) Elections, the media and the modern publicity process, pp. 101 113. in Ferguson M (ed) Public communication: The new imperatives. London: Sage. Blumler JG (1999) Political Communication Systems All Change.: A Response to Kees Brants. European journal of communication 14(2): 241 249. Blumler JG and Kavanagh D (1999) The third age of political communication: Influences and features. Political Communication 16(3): 209 230.
20 Brants K (1999) A Rejoinder to Jay G. Blumler. European Journal of Communication 14(3): 411 415. Brettschneider F (2002) Spitzenkandidaten und Wahlerfolg. Personalisierung Kompetenz Parteien. Ein internationaler Vergleich. Wiesbaden: Westdeutscher Verlag. Brettschneider F and Vollbracht M (2010) Personalisierung der Unternehmensberichterstattung, pp. 133 158. in Eisenegger M and Wehmeier S (eds) Personalisierung der Organisationskommunikation: VS Verlag für Sozialwissenschaften. Campbell A, Converse PE, Miller WE and Stokes DE (1960) The American Voter. New York. Campus D (2010) Mediatization and Personalization of Politics in Italy and France: The Cases of Berlusconi and Sarkozy. International Journal of Press/Politics 15(2): 219 235. Cohen B (1963) The Press and Foreign Policy. Princeton, N.J.: Princeton University Press. Dalton RJ, McAllister I and Wattenberg MP (2000) The consequences of partisan dealignment, pp. 37 63. in Dalton RJ and Wattenberg MP (eds) Parties without partisans. Political change in advanced industrial democracies. Oxford: Oxford University Press. Forschungsgruppe Wahlen. (2005). Politbarometer September II 2005. ZDF Pressemitteilung KW 36 Retrieved 10th October, 2011, from http://www.forschungsgruppe.de/umfragen/politbarometer/archiv/politbaro meter_2005/september_ii/ Gabriel OW and Brettschneider F (1998) Die Bundestagswahl 1998: Ein Plebiszit gegen Kanzler Kohl? Aus Politik und Zeitgeschichte : Beilage zur Wochenzeitung Das Parlament(B52/98): 14 32. Gabriel OW and Vetter A (1998) Bundestagswahlen als Kanzlerwahlen?, pp. 505 536. in Kaase M and Klingemann H D (eds) Wahlen und Wähler : Analysen aus Anlaß der Bundestagswahl 1994. Opladen: Westdeutscher Verlag. Galtung J and Ruge MH (1965) The Structure of Foreign News.: The Presentation of the Congo, Cuba and Cyprus Crises in Four Foreign Newspapers. Journal of Peace Research 2: 64 91. Geese S, Zubayr C and Gerhard H (2009) Berichterstattung zur Bundestagswahl 2009 aus Sicht der Zuschauer. Ergebnisse einer Repräsentativbefragung und der AGF/GfK Fernsehforschung. Media Perspektiven(12): 637 650. Hjarvard S (2008) The Mediatization of Society: A Theory of the Media as Agents of Social and Cultural Change. NORDICOM Review 29(2): 105 134. Kilz HW. (1999, 27.11.1999). Stehaufmann Schröder, Süddeutsche Zeitung, p. 4. Kindelmann K (1994) Kanzlerkandidaten in den Medien.: Eine Analyse des Wahljahres 1990. Opladen: Westdeutscher Verlag. Kriesi H (2011) Personalization of national election campaigns. Party Politics. Langer AI (2007) A Historical Exploration of the Personalisation of Politics in the Print Media: The British Prime Ministers (1945 1999). Parliamentary Affairs 60(3): 371 387. Langer AI (2010) The Politicization of Private Persona: Exceptional Leaders or the New Rule? The Case of the United Kingdom and the Blair Effect. The International Journal of Press/Politics 15(1): 60 76. Mazzoleni G (2008) Media Logic. in Donsbach W (ed) The International Encyclopedia of Communication. Malden, MA: Blackwell. McAllister I (2007) The Personalization of Politics, pp. 571 589. in Klingemann H D and Dalton RJ (eds) The Oxford Handbook of Political Behavior. Oxford: Oxford University Press.
21 Mughan A (2000) Media and the presidentialization of parliamentary elections. Houndmills: Palgrave. Negrine R and Papathanassopoulos S (1996) The Americanization of Political Communication. The Harvard International Journal of Press/Politics 1(2): 45 62. Ohr D (2000) Wird das Wählerverhalten zunehmend personalisierter oder: Ist jede Wahl anders Kandidatenorientierungen und Wahlentscheidung in Deutschland von 1961 bis 1998, pp. 272 308. in Klein M, Jagodzinski W, Mochmann E and Ohr D (eds) 50 Jahre empirische Wahlforschung in Deutschland. Entwicklung, Befunde, Perspektiven, Daten. Wiesbaden: Westdeutscher Verlag. Ohr D and Klein M (2001) Das Wölfchen und die fünf Clementinen: Die Wahrnehmung des Provatlebens der beiden Spitzenkandidaten bei der Landtagswahl in Nordrheinwestfalen vom 14.: Mai 2000. ZA Information 48. Rottmann H and Auer B (2010) Statistik und Ökonometrie für Wirtschaftswissenschaftler. Wiesbaden: Gabler. Ruhrmann G, Woelke J, Maier M and Diehlmann N (Eds.). (2003). Der Wert von Nachrichten im Deutschen Fernsehen.: Ein Modell zur Validierung von Nachrichtenfaktoren. Opladen: Leske & Budrich. Schönbach K and Semetko HA (2000) "Gnadenlos professionell". Journalisten und die aktuelle Medienberichterstattung in Bundestagswahlkämpfen 1976 1998, pp. 69 78. in Bohrmann H, Jarren O, Melischek G and Seethaler J (eds) Wahlen und Politikvermittlung durch Massenmedien. Wiesbaden: Westdeutscher Verlag. Schulz W (1976) Die Konstruktion von Realität in den Nachrichtenmedien.: Analyse der aktuellen Berichterstattung. Freiburg/München: K. Alber. Schulz W and Zeh R (2006) Die Kampagne im Fernsehen Agens und Indikator des Wandels. Ein Vergleich der Kandidatendarstellung, pp. 277 305. in Holtz Bacha C (ed) Die Massenmedien im Wahlkampf. Die Bundestagswahl 2005 Wiesbaden: VS Verl. für Sozialwiss. Schulz W and Zeh R. (2007). Changing Campaign Coverage of German Television: A Comparison of Five Elections 1990 2005. Paper presented at the 57th Annual Conference of the International Communication Association, San Francisco, CA. Schulz W and Zeh R (2010) Die Protagonisten in der Fernseharena. Merkel und Steinmeier in der Berichterstattung über den Wahlkampf 2009, pp. 313 338. in Holtz Bacha C (ed) Die Massenmedien im Wahlkampf. Das Wahljahr 2009. Wiesbaden: VS Verl. für Sozialwiss. Staab JF (1990) Nachrichtenwert Theorie.: Formale Struktur und empirischer Gehalt. Freiburg (Breisgau), München: K. Alber. Strömbäck J and Dimitrova DV (2011) Mediatization and Media Interventionism: A Comparative Analysis of Sweden and the United States. The International Journal of Press/Politics 16(1): 30 49. Strömbäck J and Esser F (2009) Shaping Politics: Mediatization and Media Interventionism, pp. 205 223. in Lundby K (ed) Mediatization. Concept, changes, consequences. New York, NY: Peter Lang. Urban D and Mayerl J (2011) Regressionsanalyse: Theorie, Technik und Anwendung. Wiesbaden: VS Verlag für Sozialwissenschaften. Vetter A and Gabriel OW (1998) Candidate evaluations and party choice in Germany, 1972 94: Do the candidates matter?, pp. 71 98. in Anderson CJ and Zelle C (eds) Stability and change in German elections. How electorates merge, converge, or collide. Westport, Conn.: Praeger. Wattenberg MP (1994) The decline of American political parties, 1952 1992. Cambridge: Harvard University Press.
22 Wilke J and Leidecker M (2010) Ein Wahlkampf, der keiner war? Die Presseberichterstattung zur Bundestagswahl 2009 im Langzeitvergleich Die Massenmedien im Wahlkampf, pp. 339 372. in Holtz Bacha C (ed): VS Verlag für Sozialwissenschaften. Wilke J and Reinemann C (2001) Do the candidates matter?: Long term trends of campaign coverage A study of the German press since 1949. European Journal of Communication 16(1): 291 314. Zeh R (2005) Kanzlerkandidaten im Fernsehen: Eine Analyse der Berichterstattung der Hauptabendnachrichten in der heißen Phase der Bundestagswahlkämpfe 1994 und 1998. München: Verlag Reinhard Fischer. Zubayr C and Gerhard H (2011) Tendenzen im Zuschauerverhalten Fernsehgewohnheiten und Fernsehreichweiten im Jahr 2010. Media Perspektiven(3): 126 138.