What drives party evaluation? The influence of campaign performance and published poll results

Transcription

1 What drives party evaluation? The influence of campaign performance and published poll results Paper prepared for presentation at the 23 rd IPSA World Congress of Political Science in Montreal, July 19-24, 2014 Thomas De Rocchi* and Georg Lutz # * Department of Political Science, University of Zurich, Switzerland (derocchi@ipz.uzh.ch) # FORS, c/o University of Lausanne, Switzerland (georg.lutz@fors.unil.ch) Abstract: We still know relatively little about short-term influences on party evaluations in multiparty systems. Using a rolling cross-section design, we can show the influence of two shortterm elements: the performance of parties in the campaign and in published pre-election polls. While the voters subjective perceptions of a party s campaign performance has a strong influence for each party, media reports covering the parties campaigns have a relatively small effect on party evaluations. Polls on the other hand show only small effects on party sympathies. Since those effects are negative if a party may lose, and positive if a party is perceived as likely winner, they confirm the existence of a small bandwagon effect. As one would expect, those effects are more pronounced among voters who correctly recall the polls than among those who don t recall the polls correctly. KEYWORDS: pre-election polls, campaign performance, party evaluation, media effects, bandwagon effect, underdog effect, rolling cross-section 1

2 1. Introduction What influences voters evaluations of competing parties during a campaign is still largely a black box. Parties and candidates spend large sums, however, if this actually has an impact on how they are perceived by the voters, is not very well known. In this paper we therefore look at different possible sources that voters might use to evaluate political parties. Using data from a rolling cross-section survey (RCS) conducted in Switzerland prior to the 2011 election, we explore the influence of the presences of parties in the media, the perception of their campaign as well as the subjective and objective performance in opinion polls on the voters sympathy for the different parties. We do not look at voting intentions themselves because we see voting as a two-stage process (see e.g. Bochsler and Sciarini 2010; Steenbergen and Hangartner 2008). In the first step, known as consideration stage, voters develop preferences for parties. Voters consider one or several parties as eligible to vote for and they specifically exclude others. From the resulting set of alternatives, the voters subsequently make in a second step their electoral choices. We look at the first step only in this paper, because we believe that campaigns and polls have an influence on how voters perceive all parties, not only the ones they consider voting for. In our opinion, such an approach seems promising, since it allows analysing possible campaign effects on the citizens voting behaviour in a broader sense. First, we focus on party campaigns. According to Granberg (1993: 70), political perception refers to the process by which people develop impressions of the characteristics and positions of political candidates, parties, and institutions. Following this, the voters' choices might be affected by their perceptions of the candidates' electoral performance, as Ansolabehere and Iyengar (1994: 414) note. In the context of proportional party-list elections, where the importance of the parties outweighs that of individual candidates, the same can be expected with regard to parties. If a voter perceives the way in which a party runs its campaign as particularly successful, it should thus positively affect his evaluation of that party. But where do the voters perceptions of a party s campaign performance come from? Are, for example, parties, whose campaigns are widely covered in the media, also evaluated more positively? And how are such possible effects moderated by the voters party identifications? Second, we analyse if the published results of pre-election polls affect the voters party evaluations in a systematic way. If and how polls influence voters has been subject of a political and public debate for some time. Politicians in different countries repeatedly expressed concerns that poll results might influence the citizens voting behaviour potentially even in a manner that is harmful to democracy (see e.g. Brettschneider 1992; Schuh 2009; Hoffmann 2014). They fear that the predictions have an influence on turnout and likely also on the vote share. As a consequence, they push for clearer regulations of the publication of poll results in the run-up to elections, or even try to ban them completely. This is also true for Switzerland, where, as an example, an MP claimed recently that it is scientifically proven that polls have a mobilizing effect, that they are easy to manipulate, 2

3 and that they may have tangible implications for the outcome of elections or referenda. 1 In addition, there is an ongoing public debate about the quality of poll-based predictions. 2 In contrast, studies so far have not come to any clear indication that polls have such an effect, in particular with regard to multi-party systems. There is still a great deal of uncertainty regarding effects of public opinion polls in general. To this day no study detected significant poll effects in the Swiss context, neither on the mobilization of the electorate nor on the voters party preferences. As de Bock (1976) has shown, voters primarily get mobilized if the outcome of an election is perceived as tight. Swiss elections are not as much about winners and losers, but much more about small shifts in vote shares, though. Therefore it does not come as a big surprise that no effects on the mobilization of the electorate could be detected. 3 Also with regard to the voters actual choices, previous studies found no support for poll effects, neither in the context of elections (Hardmeier and Roth 2003; Hardmeier and Sidler 2003) nor in the context of referenda (Freitag et al. 2010). Is the whole debate thus just much ado about nothing? Not necessarily. Among other things, it might have been the lack of suitable empirical data or inappropriate research designs that lead to those somewhat disappointing results. Using data from Switzerland s first rolling cross-section survey (RCS) conducted prior to the 2011 national election and an associated media content analysis, which were both collected on a daily basis, the study at hand allows a much finer-grained analysis of possible poll effects on the voters party evaluations in the course of an election campaign even if such effects are only of a shorttermed nature. Furthermore, the linkage of survey data and media data also allows differentiating between effects that are based on a voter s subjective perception be it on a party s style of campaigning or its expected success at the ballot box and such that are based on a more objective perception, i.e. reported trends in the mass media. The analyses outlined in this article therefore contribute to a deeper understanding of the reasons why voters evaluate political parties the way they do, and help resolve the question whether the publication of poll results needs to be regulated more clearly, or even banned. 4 1 Verbot von Meinungsumfragen zu Wahlen und Abstimmungen für Radio und Fernsehen, request by Christoph Mörgeli of the Swiss People s Party (Amtliches Bulletin N 2011: 93-95). The request was discussed and rejected by the National Council on 2 March The debate intensified in recent years after a polling company mistakenly indicated the dismissal of a popular initiative to prevent the construction of new minarets, which eventually was approved by the Swiss electorate in November As a consequence, several parties advocate for a reduction of the number of polls to be realised in the run-up to the upcoming elections in 2015, as the NZZ am Sonntag reported on 20 April 2014 ( Bund sucht Alternativen zur Firma von Longchamp ). According to the article, the politicians are hoping that the voters will talk more about content and less about poll results. 3 Apart from this, in Switzerland the importance of elections is comparatively low anyway, due to the various possibilities of direct-democratic participation (see e.g. Klöti and Linder 1998). 4 Currently, there is no legal requirement but only a self-imposed code of conduct within the association of Swiss market and opinion research institutes, not to publish any poll results later than ten days prior to an election. 3

4 2. Theoretical expectations Perceived party performance and media coverage of the parties campaigns In order to evaluate a political party, voters may base their verdict on a variety of factors, such as the party s stance on issues perceived as important, its previous performance be it as part of a governing coalition or the opposition or the image and trustworthiness of the party s top candidates. There is an extensive body of literature that deals with the potential influence of each of these factors. An additional but less explored explanation for the voters party evaluations is to be found in the way in which a party runs its campaign. Or in other words: in a party s characteristic style of campaigning. According to Schulz (2008: 270), parties strive after dissociating their own style of campaigning from those of their opponents, with the objective of turning it into a strategic argument in their favour. Provocative campaigns, not only tend to create a high level of media coverage, they may also lead to more positive coverage in the sense that the respective campaign is presented as successful, while the campaigns of competing parties are usually hardly mentioned. In the case of Switzerland, for example, the right-wing populist Swiss People s Party (SVP) clearly benefited from its offensive and confronting style of campaigning in the run-up to the 2007 elections, and, to a lesser degree, again during the 2011 campaign (Udris et al. 2008, 2011). In this sense, a wide coverage of a party s campaign in the media can be seen as a possible key to electoral success. Most electoral campaigns are composed as a series of information-rich events, which are usually brought to the voters attention by means of the mass media. The coverage of a party s campaign is therefore also referred to as metacoverage (see e.g. Esser and D Angelo 2006). The metacoverage encompasses all those political reports that mention one or several parties campaigns as successful or weak in functional terms, e.g. with regard to the level of media attention they attract. Studies in several European countries (for an overview, see Schulz 2008) have shown that a substantial part of the election-related media coverage during a campaign focuses on the campaign as such, rather than on important issues. As far as that goes, it is often the parties themselves that trigger this kind of media coverage, for example by initiating insider reports from their campaign headquarters (Schulz 2008: 270). On the other hand, the mass media also tend to create campaign events on their own, for instance by ordering and publishing pre-election polls, the results of which can then again be covered and commented by other media outlets. 5 One can expect that many voters base their assessment of a party s campaign performance on information obtained from the mass media. However, one cannot expect that every voter perceives the same bits of information or that all voters process the information they receive in the same way. Research in the field of political psychology indicates, the voters attention is rather selective and information tends to reinforce existing believes or predispositions (see e.g. Taber and Lodge 2006). In addition, voters do not necessarily base their evaluations of a party solely on the impressions they receive from the 5 Possible effects of published poll results and the so-called horse race journalism on the voters party evaluations are discussed in more detail in the following chapter. 4

5 mass media. Their assessment may also be influenced by third factors or by information received from other sources (Hoffmann 2014: 70, see also Schmitt-Beck 1996). The voters subjective perceptions are therefore not simply a reaction to the objective evaluation of a party in the media, but potentially an additional and partially independent explanatory factor, whose effect can be tested separately. Based on these theoretical considerations, we hypothesize that: The more frequently a party s campaign is mentioned in the media, the more positive the voters evaluation of this party (metacoverage hypothesis). If a voter subjectively perceives a party s campaign performance as more successful than the performances of the competitors, the evaluation of this party is positively affected (performance hypothesis). In addition, we will analyse how party identification moderates the effects of metacoverage and the perceived campaign performance on a voter s party evaluations. We assume that party identifiers pay higher attention to their parties campaign than those who do not identify with a party. As a consequence we assume that for party identifiers the positive effect of campaign performance on party sympathy is stronger than for those not identifying with a party. Published poll results and horse race journalism Publications of pre-election polls constitute an exceptional form of campaign event, since they not only provide indications for the success or weakness of a party s campaign, but also allow quantifying them with regard to the predicted vote share or the number of seats the party will reach in the upcoming election. Whether poll results have an effect on the citizens voting behaviour and, if yes, in which way has been subject of a long and controversial debate (for recent contributions, see e.g. Irwin and van Holsteyn 2002; Schoen 2002; Faas and Schmitt-Beck 2007; Schuh 2009; Sonck and Loosveldt 2010; Hopmann 2010; Hoffmann 2014). Most of the existing studies have been conducted in twoparty systems, above all in the United States, where, at least in the hot phase of presidential campaigns, polls are published on an almost daily basis. Poll effects in multiparty systems, on the other hand, so far received clearly less attention. Since one cannot assume that the same mechanisms lead to similar effects in both systems, the theoretical expectations with regard to possible effects of poll results on the voters party evaluations will be discussed separately for two- and multi-party systems in the following subsections. 6 Studies focusing on poll effects in two-party systems usually concentrate on the detection of either bandwagon or underdog effects or on both. In case of a bandwagon effect, the predicted winner of the election gains additional votes, since part of the voters prefer being on the wining side (Hoffmann 2012: 2). Their voting behaviour is thus a function 6 The analyses outlined in this paper exclusively focus on effects on the voters party evaluations, not on a possible mobilization of the electorate. For insights on mobilizing effects of pre-election polls, see e.g. De Bock (1976) or Brettschneider (2000); for analyses in the context of Swiss elections, see Hardmeier and Sidler (2003) or Hardmeier (2008). 5

6 of their expectations of the election outcome, as Simon (1954: 245) puts it, whereas the published results of the poll are assumed to influence these expectations. As there are only two competitors in the race, an increase in votes for the predicted winner necessarily goes along with the loss of votes for the predicted loser. Noelle-Neumann (1980: 19) explains such a shift towards the opinion of the majority with the fear of sympathisers of the trailing party of being isolated. However, it is doubtful whether this explanation is useful for the analysis at hand, since the act of voting is confidential (Hopmann 2010: 57) and no voter is obliged to reveal his (true) intention to other voters or to a pollster. An alternative explanation is provided by Schmitt-Beck (1996: 268), according to which particularly less-involved voters without partisan ties tend to use the opinion of the majority as a shortcut that allows reducing the information costs. Following this, a jump on the bandwagon constitutes an act of low information rationality. In case of an underdog effect, on the other hand, the predicted loser of the election gains additional votes, either due to pity, or out of spite, as to prevent an even greater majority for the supposedly winning party (Schoen 2002: 183). Even though both bandwagon and underdog effects are widely discussed in the literature and often used in empirical analyses, some authors claim that none of them could actually have been proven so far (e.g. Brettschneider 2000). While this view is controversial just as some of the methods that were used to confirm the existence of such effects it seems obvious that bandwagon and underdog effects may occur simultaneously. This has the consequence that they potentially cancel each other out in the aggregate, which, depending on the research design, complicates their detection. Beyond that, the strength of such effects might vary among different segments of the electorate (Traugott 1992: 137), since voters are not only characterized by diverse levels of attentiveness and interest, but also by different needs for information at different points in time. The effect of pre-election polls may therefore also depend on temporal aspects, such as the phase of a campaign during which a voter stated his party evaluations. While in a two-party system, it is obvious, which of the competing parties is the winning one as predicted by a poll, in multi-party systems the answer to this question is a bit more tricky. The winner might either be the party with the highest predicted vote share, or the party with the largest gains compared to the last election or the last poll, or even the party that is expected to designate the next head of government (Hopmann 2010: 56). In a multiparty system the increase in votes for a given party does not necessarily go along with an equivalent loss for a specific competitor. Instead, the predicted winner irrespective of how one may define it potentially gets additional votes from former sympathisers of several parties. The loss is thus not confined to a specific one of them. Hardmeier and Sidler (2003: 214) coined the term anti-loser effect to describe this phenomenon. They also introduced the related concept of an anti-winner effect, which takes place, if the predicted loser of an election gets additional votes from former sympathisers of several competing parties, including those of the predicted winner. Analogous to bandwagon and underdog effects, one can expect that anti-loser and anti-winner effects may be effective at the same time, although not necessarily to the same extent. This leads to the assumption that in the aggregate, both the predicted winners and the predicted losers have a good chance of getting additional votes, unless their expected gains are completely cancelled out by their expected losses. Stable parties, on the other hand, may at best hold their vote shares 6

7 according to this logic, since they are expected to lose votes to both the predicted winners and losers, without having a real chance to win additional votes themselves. Concepts outlined in the previous two subsections were developed to explain potential poll effects on vote intentions or a voter s actual party choice. In this study we use a measure where voters evaluate each of the competing parties separately and these evaluations are not necessarily dependent on one another anymore. This means that there are still plausible explanations for possible effects on the evaluation of the predicted poll winners and losers, however, a positive evaluation of one party is no longer linked to a negative evaluation of another one. On the contrary, it might very well be that a given voter evaluates both a winning and a losing party in a positive way, though not for the same reasons. The anti-winner and anti-loser effects introduced by Hardmeier and Sidler (2003) are therefore only of limited use when focusing on party evaluations. Schmitt-Beck s interpretation of a bandwagon effect, on the other hand, which in contrast to the traditional understanding only takes into account a voter s decision to join the predicted winner s camp, regardless of whether he initially intended to support another party or whether he was still undecided (on this point, see Hoffmann 2014: 38), is directly applicable. 7 The same is true with regard to underdog effects understood as a reaction out of pity. That s why we will stick to the terms bandwagon and underdog effect, notwithstanding that we analyse poll effects in the context of a multi-party election. Based on the theoretical considerations outlined above, we further hypothesize that: If a party belongs to the predicted winners of the election according to the published results of the most recent poll, the evaluation of this party is positively affected. If it belongs to the predicted losers, this has a negative effect on the party s evaluation (bandwagon hypothesis). If a party belongs to the predicted losers of the election according to the published results of the most recent poll, the evaluation of this party is positively affected (underdog hypothesis). Polls can only have the hypothesised effects on the voters party evaluations if the main messages the predicted results for the major parties eventually are perceived. This can happen in three different ways: 1. A voter directly perceives the predictions when a poll is published. As a first analysis of the campaign dynamics in the run-up to the 2011 elections in Switzerland has shown, the share of respondents that declares to have seen a poll in the days before the interview constantly oscillates around 20 per cent (Lutz 2012: 59). This is interesting, since there were only very few polls published during the last weeks of the campaign (see also Chapter 3). 2. A voter perceives the poll results via subsequent media reports that incorporate the respective predictions. This type of media coverage is also known as horse race 7 Based on similar considerations, it may be rational for a voter to leave the camp of a predicted loser too. A bandwagon effect should therefore be understood as a two-dimensional concept, rather that a unidimensional one. 7

8 journalism and usually the respective reports have a narrow focus on the assessment of the competing parties based on their chances of success at the ballot box, irrespective of the characteristics of candidates or a party s issue positions (see e.g. Udris et al 2008). 3. The predicted results for the major parties reach a voter through interpersonal communication, also known as buzz market. In such a case, there is no need that a voter has seen any polls at all (Hoffmann 2012). Regardless of the channel through which a voter receives the main messages of a preelection poll, the question arises whether he understands them correctly. Or, in other words, whether a voter s subjective perception of the winners and losers according to the polls is consistent with the published predictions. This is not necessarily the case, since voters might have been distracted, and the messages might have been distorted before they finally reached the voter. We will therefore test possible effects of both the subjective and objective perceptions of the likely winners and losers of the election on the evaluation of the competing parties, and analyse how such effects vary depending on the channel through which a voter received the respective messages. Overall we assume that the effect of polls is strongest among those who have learned about the pools directly and among those who can recall the polls correctly. 3. Data, measurement, and analysis strategy For the analysis we use date from a rolling cross-section survey (RCS) and an associated media content analysis that were in the field in the run-up to the Swiss 2011 national elections. Both data collections cover the last six weeks prior to the election day, in the German- as well as in the French-speaking part of Switzerland. In total, 4002 interviews were conducted within 41 days. 8 The dependent variable is a voter s evaluation of a given party measured on an 11-point sympathy scale, ranging from No sympathy at all to A lot of sympathy. 9 Compared to vote intentions which are usually expressed for a single party, the use of party evaluations has the advantage that the voters not only assess one party, i.e. the one they intend to vote for, but several parties at the same time. Tactical arguments therefore only play a marginal role. Beyond that, the characteristic small-n problems with regard to smaller parties can be avoided. The voters evaluations are measured separately for the seven largest parties in Switzerland, the Swiss People s Party (SVP), the Conservative Democratic Party (BDP), the Liberals (FDP), the Christian Democratic People s Party (CVP), the Swiss Green 8 For a detailed description of design, fieldwork, and data quality of the Swiss RCS, see Lutz et al. 2013; for a general overview of rolling cross-section surveys, see Johnston and Brady More information on the Swiss Electoral Studies as well as the data sets used for the analyses presented in this paper can be found on the project website: 9 The sympathy score is used as an alternative to the propensity to vote (PTV) for the respective party, which was not directly collected during the survey. Although the two concepts are, of course, not perfectly identical (see e.g. van der Eijk et al. 2006), a voter s sympathy for a party can safely be expected as a main factor if not the principal one in determining the propensity to vote for that party. In compliance with most of the literature, we assume that our sympathy score is metrically scaled (see e.g. Bochsler and Sciarini 2010: 384). 8

9 Liberal Party (GLP), the Swiss Social Democratic Party (SP), and the Swiss Green Party (GPS). The voters subjective perceptions of a party s campaign performance are measured on the basis of a survey question asking about the party that runs the best campaign. A dummy variable indicates if a party performed best in the view of the respective voter. To measure the frequency in which a party s campaign is mentioned in media reports, we make use of a media content analysis. In parallel to the RCS, researchers at the Institute of Mass Communication and Media Research at the University of Zurich analysed the 28 most important newspapers and online news pages in the German- and French-speaking parts of Switzerland. They coded all front-page articles as well as all the articles with a teaser on the front page, if they were somehow related to the upcoming election. Altogether, the media content analysis includes 1922 articles. 178 of them explicitly cover the campaign of at least one of the seven largest Swiss parties. In total, these articles 335 times mention one of the parties campaign performances. As Table 1 shows, 57 of these mentions indicate a positive assessment of a party s campaign and 79 a negative one, while the vast majority only refer to a party s campaign without a valuation. [TABLE 1 ABOUT HERE] To have any effect on the voters party evaluations, media reports need to be perceived. The RCS does not contain questions concerning the specific newspapers or online news pages a voter usually reads. Instead we measure the information context as the sum of the reports published in newspapers and on online news pages on a certain day. Because some interviews took place early in the morning, part of the respondents very likely had no chance to read the reports published on that day. Therefore we included for each respondent the articles published on the day of the interview, as well as on the day before 10. In a second step we excluded all articles that were published in a language other than the one the respondent has chosen for the interview and all articles published in a media format (quality newspaper, daily free newspaper, or online news page) towards which the respondent explicitly stated that he did not pay any attention in the 24 hours prior to the interview. The remaining articles are weighted according to the circulation numbers of the media outlets in which they were published, since trans-regional media can be expected to reach significantly more voters than regional or even local ones. 11 Of course, the individual information context narrowed down this way cannot perfectly mirror a respondents media consumption. However, the weighted number of articles a respondent potentially perceived and thus the amount of coverage of every party s campaign is undoubtedly more realistic than simply working with all the articles published up to the day a respondent was interviewed. Since the differences with regard to the weighted number of articles the 10 This approach also takes into account the findings of Krause and Gehrau (2007) that suggest that the voters memories of the content of news broadcasts only recede after two to three days (cited in: Hoffmann 2014: 191). Although their analysis focused on television news only, it seems plausible that a comparable duration of media effects can also be expected with regard to newspapers and online news. 11 For a more detailed description of the construction of the individual information context, see De Rocchi (2013). 9

10 respondents potentially perceived are rather small, though (between 0 and 4.98 articles, depending on the party), we use a binary version of the variable, indicating if a voter potentially perceived more media reports on the parties campaigns than the average of the respondents. To determine the most recent poll a voter has seen, the day on which a respondent was interviewed serves as an anchor. The dates as well as the results of all polls that were published in the six weeks prior to the 2011 elections are recorded in the media content analysis. Only two of the polls were based on samples that are representative for Switzerland as a whole: one published in the Sonntagszeitung on 25 September 2011, as well as the seventh and last wave of the SRG Wahlbarometer published on 12 October To test both the bandwagon and the underdog hypothesis, two sets of dummies are used that indicate if a party ranks among the winners or the losers of the election according to the most recent poll, compared to the parties vote shares in the 2007 elections. 13 As shown in Table 2, these predictions are rather stable from one poll to the next. Only with regard to the two large parties at the poles of the political spectrum, the Swiss People s Party and the Social Democrats, the predictions slightly vary over the course of the campaign. 14 This has consequences for the analysis strategy, as we will discuss in more detail at the end of the chapter. [TABLE 2 ABOUT HERE] To analyse the influence of a voter s subjective perception of the likely winners, all respondents who declared that they have seen a poll in the days before they were interviewed (907 out of 4002 respondents) had the chance to name up to two parties that were mentioned as winners in the most recent poll according to their memory. A dummy for each of the seven major parties indicates if a respondent perceives the party as a likely winner. The same procedure was applied as well with regard to the predicted losers of the election. To test possible effects of the voters objective perceptions, on the other hand, the parties named as winners and losers according to the polls are compared with the predictions that were disseminated by the media, as presented in Table 2. If 50 per cent or more of a respondent s mentions are in line with the published results of the most recent poll, the 12 In addition to that, several regional polls (one for the Canton of Grisons on 8 October 2011, one for the Canton of Berne on 19 October, as well as one for the Canton of Argovia on 23 September, focussing on the election to the Council of States) were published during the field time. Of course, they have the potential to influence the voters evaluations of the competing parties as well. Due to their limited geographical scope they are nevertheless excluded from the analysis, since one can expect that their effect if there is one is also limited to respondents from the respective three Cantons. 13 A party was coded as winner or loser (see columns Correct perception in Table 2) if the media presented that party as a predicted winner or loser in the context of the publication of the results of a poll. Differences up to ±0.4 per cent in comparison with the 2007 vote shares were therefore de facto treated as stable, since the media in their articles did likewise. In case of respondents interviewed prior to 25 September 2011, the 2007 vote shares are compared with the results of the last nation-wide poll that was published before the beginning of the RCS (the sixth wave of the SRG Wahlbarometer, 9 September 2011). 14 In case of the BDP, the stable prediction as a likely winner is explained by the fact that the party did not yet run for office on the national level in 2007, since it only seceded from the SVP that year. 10

11 respondent is treated as one that has saw a poll and understood it correctly. This applies to 576 out of the above-mentioned 907 respondents. If less than 50 per cent of the mentions are in line with the objective predictions, we assume that the respondent has not seen any polls at all. 15 As Figure 1 shows, the daily share of voters who saw a poll and understood it correctly (darker grey area) is always seven to twelve per cent lower than the share of voters who just declared that they have seen a poll (lighter grey area). The two curves, however, are largely identical. This leads to the conclusion that almost half of the constant share of respondents, that declares to have seen a poll, even if there was no one published (see Chapter 2), can thus be explained if one controls for the correct perception of the poll results. 16 [FIGURE 1 ABOUT HERE] We included an objective measure of poll-related media coverage, using all the reports about polls in our collection of articles. To determine the amount of poll-related media coverage a voter potentially perceived, the respective articles are singled out and weighted according to the same procedure outlined above for the campaign performance. In total, 96 articles contain at least one reference to a pre-election poll. In 40 of them, polls and their results are central, which means that they are mentioned in the header and/or the lead of the article. In another 22 of the articles, polls are mentioned in a prominent way, i.e. in a section or a subhead. The remaining 34 articles only contain peripheral references to preelection polls. For the analyses at hand, we concentrate on articles in which polls are either central or at least prominent, since the other ones hardly qualify as horse race journalism. As shown in Figure 1, the number of poll-related articles potentially perceived by the voters clearly increases after the publication of the second poll on 12 October To a lesser extent, this also happened after the first poll on 25 September Unsurprisingly, voters who saw a poll and understood it correctly (dashed line) also tend to be a little more receptive to media coverage. However, their pattern differs only slightly from that of voters who did not see any polls at all (dotted line). 17 One may therefore conclude, that there is a good chance that the main messages of a poll also reached a voter who declared that he has not seen any polls even without active knowledge about it. 15 Alternatively we could have used a stricter rule, i.e. only treating a respondent as one that has saw a poll and understood it correctly if all his mentions are in line with the published results of the most recent poll. However, it happens quickly that a respondent mixes up two parties, or that he perceives a party as likely winner (or loser) while it was presented as just stable in the media. This is especially true with regard to respondents that made use of the possibility to mention several poll winners and losers according to their memory. The application of the stricter rule would therefore most probably have led to an underestimation of the effect of a correct poll perception. Therefore we have opted for a more flexible rule, which, however, is still strict enough to separate out those voters, who simply mentioned their favorite parties as the likely winners, irrespective of what the polls actually predicted. 16 Beyond that, part of the remaining share of incorrect perceptions can be explained by the fact that some voters might have seen one or several of the regional polls we excludes from this analysis (see footnote 12). 17 The same can be observed with regard to all the voters that declared that they saw a poll (solid line in Figure 1). 11

12 To analyse the influence of poll-related media coverage on the voters party evaluations, we use another binary variable, indicating if a voter potentially perceived more poll-related media reports than the average of the respondents. Together with the above-mentioned distinction between voters who saw a poll and understood it correctly, and voters who did not see any polls at all, it allows us to draw conclusions on the moderating effect of the channel, through which a voter received the main messages of a pre-election poll, on the strength of potential bandwagon and underdog effects. To estimate the influence of the explanatory and moderating variables discussed in this chapter on the voters party evaluations, two sets of OLS models were used: one to test possible effects of the voters subjective perceptions, and another one to test the influence of the voters objective perceptions as well as combined effects of both their subjective and objective perceptions. 18 The performance hypothesis, as well as the influence of the voters subjective perceptions of a party as likely winner or loser according to the polls, are tested separately for each of the seven parties. Beyond that, possible moderating effects of a voter s party identification are examined as well. As shown in Table 2, the predictions of the likely winners and losers of the election are rather stable, if one compares the published results of one poll with those of the following one. A separate analysis of the effects of these predictions on a voter s evaluation of each of the parties therefore does not seem to be rewarding, due to the obvious lack of variance. The influence of the voters objective perceptions in terms of the parties campaign performances and their expected success at the ballot box can therefore only be estimated globally, i.e. with regard to their overall effect on the voters party evaluations. Among other things, this allows to examine, if the mention of a party as likely winner in the published results of a poll generally leads to an increase (or a decrease) in the voters evaluation of the party, regardless of which party it is. To enable such a comprehensive analysis, the data set had to be stacked, in order to obtain separate observations for a voter s evaluation of each of the seven parties. The stacked data set thus contains a total of observations. As a consequence, we use OLS models with robust standard errors that are clustered by the respondents of the survey, since one cannot assume that a given voter s evaluations of the different parties are independent from each other. To disentangle possible moderating effects of the correct perception of the most recent poll and the related media coverage, we use three-way interactions, since it must be assumed that such effects are, at least partly, interlinked. 18 Given that the data we use has a hierarchical structure respondents are nested within campaign days multilevel analyses, as the ones applied by Hoffmann (2014), constitute a viable alternative. However, since the construction of the individual information context allows to integrate both the relevant media coverage (campaign day level) as well as a voter s media consumption (individual level) we decided to refrain from them. 12

13 4. Results Subjective perceptions The results of the first series of regression models are presented in Table 3. They show that the perception of a party as the one with the best campaign performance always has a significant positive effect, even if a respondent does not identify with the respective party. The effects are rather strong, but vary from party to party. In case of the SVP, the perception as the party that runs the best campaign (variable: bestcamp) leads to an increase of units on the sympathy scale. This is the weakest effect. In case of the BDP, on the other hand, the increase amounts to units. If a respondent does identify with a party (pi), this not only has a strong and positive independent effect on the evaluation of this party, in three cases it also leads to strong and positive effects in combination with the perception of the party as the one that runs the best campaign. The respective interaction terms (bestcamp*pi) indicate significant effects that are clearly stronger than the ones of the voter s campaign perceptions alone. In case of the GLP, the combined effect leads to an increase of units on the sympathy scale, in case of the CVP to units, and in case of the SP the increase even amounts to units. However, as Table 3 clearly shows, the larger part of these combined effects are explained by a respondent s party identification. Concerning the voter s subjective perceptions of the likely winners and losers of the election according to the polls the results in Table 3 indicate bandwagon effects for some parties. These effects, however, are not very strong, and they do not show one unique pattern, but rather two complementary ones: In case of the GPS, the GLP, and the CVP, the perception as a likely winner (pollwin) leads to increases between and units on the sympathy scales. At the same time, the perception as a likely loser (polllose) has a negative effect on the evaluation of the GLP, the SP, and the CVP. The strength of these effects varies between and units on the sympathy scales. For the SVP, FDP and BDP there are no significant effects. [TABLE 3 ABOUT HERE] The analyses of the voters subjective perceptions provide ample evidence for the expected positive effect of the perception of a party as the one that runs the best campaign, as formulated in the performance hypothesis. With respect to the effects of poll results, the analyses further provide some evidence for the existence of bandwagon effects, these effects are not very strong, though, and they can only be identified with regard to four out of the seven major parties. For the hypothesised underdog effects, the analyses provide no evidence. Having said this, it must be stressed that these interpretations are only accurate for a part of the electorate, namely the respondents who declared that the have seen a poll in the days before they were interviewed (irrespective of whether they understood those polls correctly). Only these respondents were asked to name the likely winners and losers mentioned in the respective poll according to their memory. As a consequence, the numbers of observations are rather small in the analyses shown in Table 3, compared to the total of 13

14 4002 respondents that took part in the survey. For a general examination of the hypothesised effects, one therefore cannot focus on the subjective perceptions of the voters only. Instead one needs to take into account more objective perceptions, to allow for conclusions that are accurate for the whole group of respondents and with regard to all of the parties. Objective perceptions In order to estimate the influence of the voters objective perceptions on their party evaluations, as well as the combined effects of subjective and objective perceptions (these will be discussed in the following section), we ran a series of regression models, whose results are shown in Table 4. The models estimate the effects for all parties simultaneously, using the stacked data set mentioned above. [TABLE 4 ABOUT HERE] Model 1 in Table 4 only contains the voters objective perceptions. The estimated results suggest that the fact, that a voter potentially perceived more media reports on a party s campaign than the average of the respondents (meta), leads to an increase of units on the sympathy scale, even if that the voter does not identify with the respective party. As expected, the identification of a voter with the party also has a strong positive effect ( units). The interaction of the two variables (meta*pi), however, does not lead to an additional increase on the sympathy scale, at least not a significant one. To estimate the effect of a party s mention as a likely winner or loser of the election (obj_polls), as well as of the way in which such an effect is moderated by the channel through which a voter received the respective information, a three-way interaction has been integrated into the model. It allows to distinguish between 1) voters that have seen a poll and understood it correctly (correct), 2) voters that have not seen a poll directly, but potentially perceived more poll-related media reports than the average of the respondents (polls_context), as well as 3) voters that have not seen a poll and only perceived a belowaverage amount of poll-related media reports. In case of the latter, one can assume that the main messages of a poll reach a voter primarily by means of interpersonal communication if they reach him at all. As the estimated results of Model 1 suggest, the mention of a party as winner (obj_polls, winner) has no significant effect on the evaluation of the party, whereas the mention as a loser (obj_polls, loser) leads to a significant decrease of units on the sympathy scale. If a voter has seen the most recent poll and understood it correctly, the mention of a party as winner leads to an increase of units (obj_polls*correct, winner). In case of the mention of a party as loser, on the other hand, the correct perception of the poll does lead to an additional significant effect. Also in the case of voters that have not seen a poll directly, but potentially perceived an above-average amount of poll-related media reports, the results only indicate a significant effects if a party was mentioned as a winner (obj_polls*polls_context, winner). In this case, it leads to an increase of units on the sympathy score. 14

15 To sum up, one can conclude that Model 1 provides evidence for both the metacoverage and the bandwagon hypothesis. Apart from the combined effect of a voter s identification with a party and an over-average amount of media reports on that party s campaign (bestcamp*meta), these effects are not very strong, though. As far as the bandwagon effects are concerned, the analysis has shown that the positive influence of a party s mention as winner only is significant if a voter has seen a poll and understood it correctly or if he perceived more poll-related media reports than the average of the respondents. The negative effect of a mention as loser, on the other hand, is only significant if a voter has not seen any polls at all and only perceived less poll-related media reports than the average. Analogous to the case of the voters subjective perceptions, no indications for underdog effects were detected. Combined effects of objective and subjective perceptions In addition to the variables tested in Model 1, the second model contains an indicator for a voter s subjective perception of a party as the one with the best campaign performance (bestcamp), as well as three interaction terms that allow to test for combined effects of this additional variable with 1) the voter s identification with the party (bestcamp*pi), 2) the fact, that he potentially perceived an above-average amount of media reports on the party s campaign (bestcamp*meta), and 3) with both of them (bestcamp*pi*meta). The estimated results suggest that the perception of a party as the one that runs the best campaign has a significant and positive effect on its evaluation (increase of units), even if a voter does not identify with the party and only received an amount of media reports on that party s campaign that is below the average. When it comes to the independent effects of those two variables (meta and pi), the results are by and large identical with the ones shown in Model 1. Beyond that, the positive effect of the subjective perception of a party s campaign performance is significantly reinforced, if a voter identifies with that party. The combined effect (bestcamp*pi) leads to an increase of units on the sympathy scale. To the greatest extent, however, this strong effect can again be explained with the voter s identification with the party. With respect to the influence of the objective perception of poll results on the voter s party evaluations, the first two models show very similar results. In Model 2, however, the mention of a party as likely winner now also has a significant effect (+0.07 units on the sympathy scale) if a voter has not seen a poll and only perceived less poll-related media reports than the average (obj_polls, winner). In Model 3, an additional indicator for the voters subjective perceptions of the winners and losers mentioned in the polls according to their memory (sub_polls) is included. The number of observations is therefore considerably smaller than in the two previous models because we only included respondents that had seen the polls. In contrast to the following models, the voters subjective and objective perceptions are not yet interacted in Model 3, since the model was only included to demonstrate that the amount of poll-related media reports no longer matters, once the analysis is limited to the group of respondents that declared to have seen a poll. The respective variable, as well as the related interaction terms, are therefore excluded with regard to the following models. 15

16 In Model 4, the voters subjective and their objective perceptions with respect to the predictions contained in the most recent poll are eventually interacted. Instead of the interaction between the published results of a poll and its correct perception we used so far (obj_polls*correct), for the final two models a single simplified variable is used. It directly indicates the correct perception of the published poll results (obj_polls_correct), since, with regard to the interaction of subjective and objective perception, it is only relevant to know if a respondent understood the poll results correctly or not. Before we get to the effects of poll results on the voters party evaluations, however, we will discuss the results with regard to the parties campaign performances. As shown in Table 4, the perception of a party as the one that runs the best campaign only has a significant effect, if a respondent also identifies with that party (bestcamp*pi). If this is the case, the sympathy scale increases by units. If a voter does not identify with a party, Model 4 no longer provides evidence for the effects postulated in the performance as well as in the metacoverage hypothesis. One must therefore conclude that the respective effects that we found in the first two models disappear, once the voters subjective perceptions of the likely winners and losers according to the most recent polls are integrated into the analysis. As for the bandwagon hypothesis, the estimated results shown in Model 4 predominantly meet the expectations. They show that if a voter understood the most recent polls correctly and if he did not state a subjective perception with regard to a party, the mention of that party as winner (obj_polls_correct, winner) has a positive effect on the evaluation of this party ( units). 19 If the voter did state a subjective perception that is in line with the objective perception of the party as a likely winner, the increase even amounts to units on the sympathy scale (obj_polls_correct*sub_polls, winner*winner). 20 A similar pattern can be observed if a party is mentioned as a likely loser in the published results of a poll. In absence of a subjective perception, the results indicate a decrease of units (obj_polls_correct, loser). If a voter s subjective and objective perceptions are in line, the decrease amounts to units (obj_polls_correct*sub_polls, loser*loser). If a voter, on the other hand, did not understand the main messages of the most recent poll correctly, the predictions as stated in the published results of the poll logically have no significant effects. But and this is in line with the expectations the subjective perceptions do have an effect in such a case. If a party was mentioned as a likely loser in the most recent poll according to a voter s memory (sub_polls, loser), this leads to a significant decrease of units on the sympathy scale. However, if a voter perceives a party as likely winner and this now runs contrary to our expectations this also leads to a significant decrease (-0.42 units) according to the results shown in Model 4. A possible 19 Due to the fact that the respondents had the possibility to name a maximum of two parties they remember being mentioned as winners in the most recent poll, it is possible that a voter correctly remembered a third party as winner, but did not get the chance to name it. 20 As a possible consequence of the way in which the correct perception of a poll was coded (see footnote 15), it may occur that a voter s subjective and his objective perceptions are not in line (e.g.. that a party is mentioned as winner in the published results of a poll, but the voter perceives that party as a likely loser). These occasions are extremely rare, though (only 23 out of the 5688 observations), and the respective effects are not significant. 16

17 explanation might be, that in case of parties a voter does not have lot of sympathy for i.e. the ones he evaluated poorly the indication of an expected success at the ballot box may have led to an even lower degree of sympathy. To test if this ad hoc explanation is plausible, we will therefore estimate a final model. It will further allow to answer the more general question, whether the effects we found eventually have the potential to affect the composition of a voter s choice set. Or, in other words, if they are also effective with respect to the parties a voter ranked highest on the sympathy scale, since, realistically, only these parties represent eligible alternatives with regard to the voter s final electoral choice. To this end, Model 4 was estimated again, using a squared version of the sympathy scales as dependent variable this time. As a consequence of this transformation, positive party evaluations become more important, negative ones analogously less important. If the effects discussed in the previous sections remain consistent, this can be seen as an indication that these effects cannot be limited to parties with low ratings on the sympathy scales only the parties a voter will never cast his ballot for, but that they may also have an influence on the parties that actually have a chance to get a respondent s vote. As the results in Model 4 squared show, the effect of the subjective perception of a party as likely winner (sub_polls, winner) on its evaluation although still negative no longer is significant, if more weight is attached to the parties the voters evaluated best. The analysis thus provides at least some evidence that our ad hoc explanation in this regard might be plausible. Beyond that, Model 4 squared reveals an additional significant effect: the perception of a party as the one that runs the best campaign (bestcamp) now has a positive effect on the evaluation of that party, even if a voter does not identify with it. This again seems plausible since one can expect that voters care much more about the campaigns of the party they actually have a lot of sympathies for. All things considered, the estimated results of the final model can therefore be seen as encouraging with respect to the actual relevance of our findings. 5. Conclusions In this paper we explored the influence of the voters subjective and objective perceptions of the parties campaign performances as well as the influence of poll results on party sympathy scores. The analysis provides clear evidence for our performance hypothesis: if respondents think that a party runs the best campaign, they express higher sympathies for this party. How often a party s campaign was mentioned in the media, however, is not a very strong predictor of a voter s sympathy for a party. The effects are rather small and they disappear once we only look at the sub-sample that declared to have actually seen a poll. The perception of polls on the other hand only has a limited effect on the ratings on the sympathy scales. If only the voters subjective perception is taken into account i.e. the parties they remember being mentioned as winners or losers in the most recent poll they have seen for three parties the ratings on the sympathy scales increase among those who saw the party as a winner, while for another three parties the ratings decrease for those who indicated that this party was among the losers. Both the positive effects for winning parties as well as the negative effects for losing parties point into the direction of a bandwagon 17

18 effect (and not an underdog effect). For the remaining parties the effects were not significant. For all parties the effect was rather small if significant, also taking into account that the majority of voters do not remember having seen any poll results at all. The voters objective perceptions of the predictions actually contained in the published results of a poll, on the other hand, proved to be more important, especially with regard to voters that interpreted those poll results correctly. The analyses have shown, that the ratings of parties that appear to lose decrease on the sympathy scales, while the ratings of those parties that appear to win increase, which is in line with our predictions. These effects remain important and tend to become even stronger once the squared values of the party sympathies are used as dependent variable. This provides some evidence that the poll effects not only matter with regard to parties a voter has little sympathies for, but also with regard to the parties a voter rated high on the sympathy scale. This is important, since one can expect that only these parties realistically have a chance to become part of a voter s choice set. While the study design we used here a rolling cross section survey has some clear strengths, the study also has its limitations. We can show that only a minority of voters has an active memory of a poll and many of them don t get the winners and losers according to the polls right. There are also relatively few media reports, both in relation to polls as well as on the parties campaigns. Nevertheless the existing reports do affect the voters. Voters have higher sympathies for parties that are present in the media and lower sympathies for parties that do bad in the polls, independent from the subjective feelings they may have with regard to a party s campaign or with regard to how the party does in the polls. A further limitation is that we don t look at preference or electoral choice in this paper but at sympathies expressed for all major parties. Given that the effects of both the voters evaluations of the parties campaigns as well as of poll results on party sympathies are rather weak, we would not expect a major effect on party preferences either. 18

19 References Ansolabehere, S. and S. Iyengar (1994). Of horseshoes and horse races: Experimental studies of the impact of poll results on electoral behaviour. Political Communication 11 (4): Bochsler, D. and P. Sciarini (2010). So Close But So Far: Voting Propensity and Party Choice for Left-Wing Parties. Swiss Political Science Review 16 (3): Brettschneider, F. (1992). Der taktische und rationale Wähler. Über den Einfluss von Wahlumfragen auf das Wählerverhalten bei den Bundestagswahlen 1983 bis Politische Vierteljahresschrift 33 (1): Brettschneider, F. (2000). Demoskopie im Wahlkampf Leitstern oder Irrlicht? In Klein, M., W. Jagodzinski, E. Mochmann, and D. Ohr (eds.), 50 Jahre empirische Wahlforschung in Deutschland. Entwicklung, Befunde, Perspektiven, Daten. Wiesbaden: Westdeutscher Verlag ( ). De Bock, H. (1976). Influence of In-State Election Poll Reports on Candidate Preference in Journalism Quarterly 53 (3): De Rocchi, T. (2013). How citizens respond to media content and how this affects their vote intentions. An analysis of the interplay between issue salience and issue ownership in the dynamic context of an election campaign. Paper presented at the workshop The 2011 Swiss elections, Solothurn, 25 October Esser, F. and P. D Angelo (2006). Framing the Press and Publicity Process in U.S., British, and German General Election Campaigns: A Comparative Study of Metacoverage. The Harvard International Journal of Press/Politics 11 (3): Faas, T. and R. Schmitt-Beck (2007). Wahrnehmung und Wirkung politischer Meinungsumfragen. Eine Exploration zur Bundestagswahl In Brettschneider, F., O. Niedermayer, and B. Wessels (eds.), Die Bundestagswahl Analysen des Wahlkampfes und der Wahlergebnisse. Wiesbaden: VS Verlag für Sozialwissenschaften ( ). Freitag, M., A. Vatter, and T. Milic (2010). Der Einfluss von Umfragen auf den direktdemokratischen Meinungsbildungsprozess. Studienbericht im Auftrag der Chefredaktorenkonferenz der SRG. Konstanz/Bern, August Granberg, D. (1993). Political Perception. In Iyengar, S. and W. McGuire (eds.), Explorations in Political Psychology. Durham: Duke University Press (70-112). Hardmeier, S: (2008). The Effects of Published Polls on Citizens. In Donsbach, W. and M. Traugott (eds.), Handbook of Public Opinion Research. London: Sage ( ). Hardmeier, S. and H. Roth (2003). Die Erforschung der Wirkung politischer Meinungsumfragen: Lehren vom Sonderfall Schweiz. Politische Vierteljahresschrift 44 (2): Hardmeier, S. and A. Sidler (2003). Meinungsumfragen und Wahlentscheidung: Nutzung und Wirkung im empirischen Test. In Sciarini, P., S. Hardmeier, and A. Vatter (eds.), Schweizer Wahlen Elections fédérales Bern: Haupt ( ). Hoffmann, H. (2012). Effects of Opinion Polls and Media Reporting on Voters. A Multilevel Analysis. Paper prepared for the XXII nd IPSA World Congress of Political Science in Madrid, July 8-12, Hoffmann, H. (2014). Wahlumfragen und Wähler. Analysen zum Einflusspotential veröffentlichter Umfrage-ergebnisse bei den Bundestagswahlen 2009 und Frankfurt am Main: Peter Lang. Hopmann, D. (2010). Vom emotionalen Underdog zur bewussten Strategie: Wie Meinungsumfragen die Parteipräferenz der Wähler beeinflussen. In Faas, T., K. Arzheimer, and S. Roßteutscher (eds.), Information - Wahrnehmung - Emotion. Politische Psychologie in der Wahl- und Einstellungsforschung. Wiesbaden: VS Verlag für Sozialwissenschaften (51 69). Irwin, G. and J. van Holsteyn (2002). According to the Polls. The Influence of Opinion Polls on Expectations. The Public Opinion Quarterly 66 (1): Johnston, R. and H. Brady (2002). The rolling cross-section design. Electoral Studies 21 (2): Klöti, U. and W. Linder (1998). Vergleichende Perspektiven. In Kriesi, H., W. Linder, and U. Klöti (eds.), Schweizer Wahlen Bern: Haupt ( ). Krause, B. and V. Gehrau (2007). Das Paradox der Medienwirkung auf Nichtnutzer. Eine Zeitreihenanalyse auf Tagesbasis zu den kurzfristigen Agenda-Setting-Effekten von Fernsehnachrichten. Publizistik 52 (2): Lutz, G. (2012). Eidgenössische Wahlen Wahlteilnahme und Wahlentscheid. Lausanne: Selects FORS. Lutz, G., T. De Rocchi, and N. Pekari (2013). The Swiss 2011 Rolling Cross-Section Study: Design, field work, and data quality. FORS Working Paper Series, paper Lausanne: FORS. Noelle-Neumann, E. (1980). Die Schweigespirale. Öffentliche Meinung - unsere soziale Haut. München: R. Riper Verlag. 19

20 Schmitt-Beck, R. (1996). Mass Media, the Electorate, and the Bandwagon. A Study of Communcation Effects on Vote Choice in Germany. International Journal of Public Opinion Research 8 (3): Schoen, H. (2002). Wirkung von Wahlprognosen auf Wahlen. In Berg, T. (ed.), Moderner Wahlkampf. Blick hinter die Kulissen. Opladen: Leske + Budrich ( ). Schuh, S. (2009). Umfragen als Anker? Studien zur Wirkung rezipierter Umfrageergebnisse. Wiesbaden: VS Verlag für Sozialwissenschaften. Schulz, W. (2008). Politische Kommunikation. Theoretische Ansätze und Ergebnisse empirischer Forschung. Wiesbaden: VS Verlag für Sozialwissenschaften. Simon, H. (1954). Bandwagon and Underdog Effects and the Possibility of Election Predictions. The Public Opinion Quarterly 18 (3): Sonck, N. and G. Loosveldt (2010). Impact of Poll Results on Personal Opinions and Perceptions of Collective Opinion. International Journal of Public Opinion Research 22 (2): Steenbergen, M. and D. Hangartner (2008). A Manski-Type Choice Set Model with Screening Rules. Paper prepared for the Annual Meeting of the American Political Science Association, Boston, August 27-31, Taber, C. and M. Lodge (2006). Motivated Skepticism in the Evaluation of Political Beliefs. American Journal of Political Science 50 (3): Traugott, M. (1992). The Impact of Media Polls on the Public. In Mann, T. and G. Orren (eds.), Media polls in American politics. Washington D.C: Brookings Institution ( ). Udris, L., J. Lucht, and K. Imhof (2008). Resonanz- und Reputations-Politikanalyse (REREPA): Nationalratswahlen fög research papers Zürich: fög Forschungsbereich Öffentlichkeit und Gesellschaft. Udris, L., K. Imhof, and P. Ettinger (2011). Die Medienleistung aus Sicht der Öffentlichkeitsforscher: Themen und ihre Dynamiken. Presentation at SRG Mediensymposium Wahlen 2011: Medien auf dem Prüfstand, Zürich, 3 September Van der Eijk, C., W. Van der Brug, M. Kroh, and M. Franklin (2006). Rethinking the dependent variable in voting behaviour: On the measurement and analysis of electoral utilities. Electoral Studies 25 (3):

21 Table 1: Media coverage regarding success and weaknesses of the parties campaigns SVP FDP BDP CVP GLP SP GPS total Positive mentions of party's campaign 28 (24) 4 (06) 2 (22) 2 (06) 2 (15) 14 (23) 5 (13) 57 (17) Negative mentions of party's campaign 19 (17) 29 (45) 0 (00) 8 (24) 8 (62) 10 (16) 5 (13) 79 (24) Neutral mentions of party's campaign 68 (59) 32 (49) 7 (78) 23 (70) 3 (23) 37 (61) 29 (74) 199 (59) total 115 (100) 65 (100) 9 (100) 33 (100) 13 (100) 61 (100) 39 (100) 335 (100) Note: The numbers represent the total number of articles published. 21

22 Table 2: Winning and losing parties according to the most recent polls compared to 2007 vote shares Time of interview before 25 September between 25 September and 12 October after 12 October Elections 2007 Wahlbarometer 9 September Difference Correct perception Sonntagszeitung 25 September Difference Correct perception Wahlbarometer 12 October Difference Correct perception SVP loser loser stable FDP* loser loser loser BDP** winner winner winner CVP stable stable stable GLP winner winner winner SP winner winner stable Greens stable stable stable * FDP and Liberal Party combined (15.8 per cent plus 1.9 per cent) ** BDP seceded from SVP in 2007 and did not yet run in that year's election Datasources: BFS 2007, Sonntagszeitung/Isopublic, SRG/gfs.bern. 22

23 Table 3: Subjective perceptions of metacoverage and poll results SVP FDP BDP CVP GLP SP Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. GPS Constant * * * *** ** Metacoverage bestcamp ** *** ** *** *** *** * pi *** *** *** *** *** *** *** bestcamp*pi ** ** * Polls pollwin * *** * polllose * * ** Controls women education level age * ** *** *** *** ** * *** *** * * N R * p<0.05; ** p<0.01; *** p<

24 Table 4: Objective perceptions of metacoverage and poll results and combined effects (stacked data) Model 1 Model 2 Model 3 Model 4 Model 4 squared Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Coef. Robust S.E. Sig. Constant *** *** ** *** *** Metacoverage bestcamp ** ** meta *** *** pi *** *** *** *** *** Polls (bestcamp*pi) *** ** ** ** (meta*pi) (bestcamp*meta) (bestcamp*pi*meta) sub_polls winner ** loser * ** obj_polls winner * loser *** *** correct polls_context (obj_poll*correct) winner *** *** ** loser (obj_polls_correct) winner ** ** loser *** *** (obj_polls*polls_context) winner * * loser (obj_polls*correct*polls_context) winner loser (obj_polls_correct*sub_polls) winner*winer *** winner*loser loser*winner loser*loser *** *** Controls high awareness women education level age *** *** *** *** *** * * R Observations Clusters * p<0.05; ** p<0.01; *** p<

25 Figure 1: Perception of pre-election polls and poll-related media coverage (smoothed) Note: The left Y-axis denotes the share of respondents that declared that they saw a poll in the days before they were interviewed (lighter grey area) as well as the share of respondents that saw a poll and understood it correctly (darker grey area). The right Y-axis denotes the weighted number of articles covering polls potentially perceived by respondents that declared that they saw a poll (solid line), by respondents that saw a poll and understood its results correctly (dashed line), as well as by respondents that did not see any polls (dotted line). All curves depict smoothed data (5-day prior moving averages). 25