Introduction

The number of floating voters in Western democracies has risen over the past decades. Not only do voters change their party preference from election to election, but also during election campaigns. This development has been particularly noteworthy in the Netherlands, where the electorate has changed from one of the least to one of the most volatile electorates in Western Europe since the 1960s (Mair 2008). Understanding the impact of the media on electoral volatility is important in order to judge whether it weakens or rather strengthens democratic processes.

Most research has studied volatility at an aggregate level, looking at the overall shift in party support in society (e.g., Drummond 2006; Mair 2008; Tavits 2008). The studies that address electoral volatility at the individual level are scarce. Yet, it is only at this level that individual predictors of volatility, such as political interest and media use, can be studied. Furthermore, media as a short-term explanation for volatile voting behavior has received little research attention. In this paper, we do study volatility at the individual level, focusing in particular on the influence of the media on vote switching during campaign time.

Moreover, this study distinguishes between different types of volatility that media may affect. One of the first studies on voter behavior already differentiated between different types of vote changes during the campaign (Lazarsfeld et al. 1948): conversion, crystallization, and reinforcement. In this study, we only focus on the first two, since the latter one does not refer to actual change in vote, but to the “effect of reinforcing the original vote decision” (Lazarsfeld et al. 1948, p. 87). Thus, the first type of volatility we will study is conversion, which means “switching from one party to another in response to campaign exposure.” The second is crystallization: when a voter’s latent support for a party changes into an actual vote in response to campaign information. In the current study, we examine how both types of volatility are influenced by exposure to media coverage in the campaign period, which is an important element of campaign exposure. Campaign exposure might also include other aspects, such as exposure to party canvassing, campaign advertising, or exposure to political campaign messages on social media. Yet, mass media exposure is one of the most important elements of campaign exposure.

While Lazarsfeld et al. (1948) already mentioned presumed media influences on voting behavior five decades ago, the studies that have focused on media exposure as a possible explanation of electoral volatility are limited (e.g., Baker et al. 2006; Dassonneville 2011; Van der Meer et al. 2015). In addition, recent research on electoral volatility in Western Europe has not differentiated between the different types of vote change that Lazarsfeld et al. (1948) initially laid out. Scholars either focus solely on conversion (Dassonneville and Stiers 2017; Van der Meer et al. 2015) or study conversion and crystallization together without differentiating between them (Takens 2013). We argue that media may influence crystallization in a different way than it influences conversion, depending on the specific news content that voters are exposed to. However, the influence of specific news content has not been incorporated in previous studies (exceptions included, Adriaansen et al. 2012; Takens 2013). Therefore, we not only study the effect of media exposure on volatility in general, but also study which specific aspects of media content, namely, news on issues and news on polls, induces which type of vote switching. In this way, we are able to examine the link between media and volatility more closely.

Understanding the impact of media content on electoral volatility is important for assessing whether citizens are mostly influenced by media logic or political logic. Poll news, as an important aspect of horse race coverage, is a clear indicator of media logic, which refers to the storytelling techniques media use to compete for people’s attention (Strömbäck 2008). A political (or public) logic, on the other hand, refers to the needs of the political system and is expressed in more substantive issue coverage. Especially the latter is recognized to contribute to a healthy democratic process (Brants and Van Praag 2006). Examining how these types of media content, and thus these logics, influence vote switching, provides important insights into the ability of voters to make a well-informed voting decision (Takens 2013).

This study provides a comprehensive understanding of the effect of media exposure on electoral volatility in two ways. First, this study contributes to the understanding of (limited) campaign effects by distinguishing between two types of vote change that might be influenced by media: conversion and crystallization. Second, it studies the effect of news exposure in an extensive way by incorporating the exposure to specific aspects of media content into the analysis (issue news and poll news), while controlling for other important predictors of volatility.

Media and electoral volatility

Electoral volatility is generally defined as “the changes in party preferences within an electorate” (Crewe 1985, p. 8). Previous studies have explained the rise in volatility by the decline of cleavages and voter loyalties, which implies that the attachment between parties and voters has weakened (Dalton 2000). More recent research has related electoral volatility to the process of voter emancipation (Dassonneville et al. 2015; Van der Meer et al. 2015), implying that voters are making more informed political choices instead of relying on traditional loyalties. Considering the fact that social characteristics as a long-term account for the stability of voter preferences have lost much of their predictive power, one would expect that voting behavior is at least to some degree dependent on short-term factors, such as exposure to media coverage of the election campaign (Dalton 2000). As a consequence, we only focus on vote changes within the election campaign, so-called campaign volatility.

In early studies on the role of media in voting behavior, scholars were convinced that campaigns only have minimal effects (Klapper 1960; Lazarsfeld et al. 1948) and only conversion, due to persuasion by campaign messages, was regarded as an effect. Today, more and more scholars argue that campaigns do matter, in part because they have broadened the definition of campaign effects beyond the focus on persuasive effects (Brady et al. 2006; Farrell and Schmitt-Beck 2002). Besides persuading voters to change their party choice, and convert to a different party, campaigns may have an informational role helping the increasing share of undecided voters to make up their mind and crystallize their vote choice (Arceneaux 2005; Gelman and King 1993; Hillygus and Shields 2009; Hillygus 2010).

The idea of crystallization was already introduced by Lazarsfeld et al. (1948), who stated that “what the campaign does is to activate [voters’] political predispositions” (p. 73). This idea of activation was further developed by Finkel (1993), who suggested that campaign information is more likely to bring voters’ party preferences in line with their own predispositions, rather than changing their attitudes. In line with this, campaign information might activate voters’ ideological predispositions. Since ideological differences between parties in the Netherlands are quite small, multiple parties might match an individual’s activated ideological attitude. Therefore, it is not immediately clear which party is to be preferred at the end of this crystallization process. From this perspective, voters who switch from being undecided to decided, or in other words “crystallize”, can be regarded as volatile voters too.

At present, studies rarely distinguish between these different campaign effects and different types of volatility (exceptions included, Dilliplane 2014; Strömbäck 2008). In this study, we contribute to the understanding of campaign effects by distinguishing between conversion and crystallization and by studying the effect of exposure to media content.

The influence of specific media content on electoral volatility

In Western Europe, the effect of media exposure on electoral volatility has been understudied, let alone the effect of exposure to specific media content. However, for a deeper understanding of why media exposure has an effect on electoral volatility, one should examine the content to which an individual is exposed: Whether voters change their voting intentions, and in what way (i.e., whether they crystallize or convert), is more likely to dependent on the kind of news than merely the degree of news exposure.

To our knowledge, the effect of specific media content on volatility has only been studied by Adriaansen et al. (2012) and Takens (2013). Both studies investigated how exposure to issue news and strategic news can induce volatile voting behavior, but come to diverging conclusions. According to Adriaansen et al. (2012) issue news can induce electoral volatility, whereas strategic news decreases volatile voting behavior. In contrast, the study by Takens (2013) showed that issue news rather leads to stable voting behavior, while contest news increases electoral volatility. These diverging conclusions could be explained by their different approach in operationalizing strategic news as well as volatility. This study bridges the differences between the two studies by first of all departing from the operationalization of strategic news that both previous studies have in common, which is poll news. Secondly, we argue that the distinction between two types of electoral volatility might help us to understand the different results with regard to the effects of issue news.

What both Adriaansen et al. (2012) and Takens (2013) include in their operationalization of strategic news is the media’s coverage of polls. Poll coverage is a key aspect of strategic news (Jamieson 1992) and is closely related to Patterson’s (1993) game frame or a horse race frame. Since one can expect that specifically poll news can induce vote switching, because of the volatile nature of polls, we do not focus on other aspects of strategic news, such as language of war, games, and competition. In this paper, we are particularly interested in issue news and poll news, and will examine their effect on both conversion and crystallization.

Issue news is generally defined as providing information about present and future policies, about political stands of parties, and about ideologies and ideas (Van Praag and Van der Eijk 1998). The basis on which votes are cast depends on the degree of issue news. According to agenda setting and priming theory, issues that are more salient in the news are more central in voters’ considerations when evaluating a party (McCombs and Shaw 1972; Iyengar and Kinder 1987; Zaller 1991). Voters may choose or switch to the party that performed well on the issue in focus, leading to retrospective voting (e.g., Söderlund 2008); or they may vote for the party whose issue position is most in line with their own, leading to prospective voting (Lockerbie 1992). Either way, the party to which a voter switches depends on the media’s coverage of issues and the coverage of parties’ positions and performances on these issues.

As explicated above, previous research on the effect of issue news on electoral volatility has yielded inconclusive findings. Adriaansen et al. (2012) found that issue coverage can induce voter uncertainty, which was particularly the case for highly sophisticated voters. Contrarily, Takens (2013) found a negative effect from two types of issue coverage on volatility. This suggests we should study the relationship between issue news and volatility more closely and focus on the mechanism at work. The distinction between crystallization and conversion volatility is useful in that regard. Issue news might especially have an informational function for undecided voters, helping them to make up their mind and eventually crystallize their vote choice. Based on motivated reasoning literature, we assume that voters without a prior preference are guided by different motivational goals than voters with a prior preference when processing information (Kunda 1990; Nir 2011). Undecided voters are more likely to be driven by accuracy goals than directional goals, because of an absent or weaker prior preference, and thus invest more effort in processing issue-relevant information (Kunda 1990). They may use issue news to learn about parties’ performances and stances on issues, to get their party choice in line with their pre-existing attitudes (Arceneaux 2005) and as a consequence crystallize their vote choice. On the other hand, voters with a prior preference are more likely to be driven by directional goals and thus use issue information to confirm their prior view, which reinforces their original preference instead of converting to another party. We, therefore, expect that:

H1a

Exposure to issue news has a positive effect on crystallization, in that voters without a prior party preference acquire a party preference in response to issue news exposure.

H1b

Exposure to issue news has a negative effect on conversion, in that voters with a prior party preference stick to that party preference in response to issue news exposure.

Poll news may lead to vote switching in several ways. Extant research has shown that poll news can lead to a so-called bandwagon effect (e.g., Farnsworth and Lichter 2006; Kleinnijenhuis et al. 2007a, b). The bandwagon effect refers to the tendency of voters to vote for successful parties. Furthermore, polls provide voters with information about possible future coalitions, which may prompt voters to cast a strategic vote (Meffert and Gschwend 2011). Regardless of whether poll exposure invokes a bandwagon effect or prompts strategic voting, in both cases it may lead to more volatility due to the dynamic nature of polls. Parties’ stances in the polls fluctuate over the course of the campaign, and media’s coverage on potential winners and losers of the election is often based on these polls. This fluctuation in who is winning or losing according to the media possibly also leads to fluctuations in a voter’s party preference (Ansolabehere and Iyengar 1994; Kleinnijenhuis et al. 2007a, b). Accordingly, it is imaginable that voters with a prior party preference will convert to another party in response to poll news. Yet, poll news might also help individuals without a prior preference to crystallize their vote choice. For instance, undecided voters might wait for additional information, like poll news, until the last moment, in order to cast an informed strategic vote (Irwin and Van Holsteyn 2008). We thus argue that poll news might lead to both types of electoral volatility.

H2a

Exposure to poll news has a positive effect on crystallization, in that voters without a prior party preference acquire a party preference in response to poll news exposure.

H2b

Exposure to poll news has a positive effect on conversion, in that voters with a prior party preference end up with a preference for another party in response to issue news exposure.

Method

The Dutch case

The Dutch situation is an interesting case to study in this regard, as the Netherlands has had some of the most volatile elections within Western Europe since the 1960s (Mair 2008). Previous research has shown that Dutch voters tend to switch to ideologically similar parties within one of the two party blocks: a block of left-wing parties and a block of right-wing parties (Van der Meer et al. 2012). Furthermore, the Netherlands is a multi-party system with a high number of parties and small ideological differences between parties, making voters more likely to rely on the media when they change their vote intention. Hence, in this study, we focus on the Dutch 2012 elections. At the start of the campaign of the Dutch 2012 elections, the Socialist Party and the Liberal Party were the largest parties in the polls. However, in the last weeks of the campaign the support for the Labour Party increased. Eventually, the Liberal Party won the elections, and the Labour Party became second at the expense of the Socialist Party. This change suggests that a share of the voters indeed switched, most notably within the block of left-wing parties (Van der Meer et al. 2012).

A longitudinal linkage study

Most campaign studies are based on cross-sectional data (Iyengar 2001), which inhibit conclusions about causal order since they measure both cause and effect at one point in time (Bartels 2006). In order to properly study media effects on vote switching, panel surveys are required, since they allow for the measurement of changes in voting behavior. Furthermore, in contrast to experimental studies, panel survey studies are conducted in a real-world setting, leading to a higher external validity and generalizability (Kinder 2007). Even considering the trade-off with a panel survey approach, such as less internal validity in comparison to an experimental approach, the nature of a survey approach is more suitable for this study. Whereas experimental studies involve forced exposure to specific news content, news exposure in the real world varies considerably among individual citizens, depending on the specific media outlets that citizens use. Therefore, exposure to the available media content should be analyzed at the individual level, which “entails content analyses of media outlets and media specific measures of exposure” (Druckman 2005, p. 517). Especially, since linking content analysis data to individual media exposure measures is the state-of-the art in media effects research (Scharkow and Bachl 2017). Therefore, we use a panel dataset and link this to a substantive content analysis of campaign news on television and in newspapers during the Dutch 2012 election campaign. In this way we are able to assess the impact of exposure to media content on individual-level volatility.

Panel data

The panel survey dataset we used was collected by TNS NIPO in collaboration with University of Amsterdam and de Volkskrant using computer-assisted self-interviewing. These data were gathered in the campaign period of the 2012 Dutch parliamentary elections of September 12. The first respondents were approached on May 17, 2012 (t-4: N = 1537), and recontacted on June 21 (t-3: N = 1239; recontact rate: 81%), August 16 (t-2: N = 1206; recontact rate: 97%), August 30 (t-1: N = 1187; recontact rate: 98%), and September 14 (t: N = 1162; recontact rate: 98%). In this study, we only included those respondents that have participated in all waves (N = 765).Footnote 1 We only used the data of last two waves (t-1 and t), since we are interested in the influence of the media’s campaign coverage which only started after t-2. The media coverage on the election campaign is limited in the period before t-2. Therefore, content analysis data before t-2 are not available in this study. Since the influence of media content on electoral volatility is the main interest of this study, we only use the panel survey data collected in the last two waves for which also content analysis data are available.Footnote 2 Our data are by and large representative of the Dutch population.Footnote 3

Measures

The dependent variable is based on one variable in the panel dataset measured at two points in time.Footnote 4 At t-1, respondents were asked which party they would vote for if elections were held today. At t, the post-election wave, respondents were asked which party they ended up voting for in the elections. We constructed a dependent variable with four possible outcomes: (1) stable: staying loyal to same party between t-1 and t, (2) abstention: abstaining from voting at t, (3) crystallization: changing from not voting or being undecided to a party choice between t-1 and t, (4) conversion: changing party choice between t-1 and t.Footnote 5 Of the 765 respondents, 66% stayed with the same party in both waves, 8% crystallized their vote choice, 16% switched their party preference, and 10% eventually did not turn out on Election Day.

We also included several control variables, starting with the usual socio-demographic variables, measured at t-4: age (M = 51, SD = 17), sex (49.7% male, 50.3% female), education (measured in seven categories ranging from “no education” to “bachelor degree or higher,” M = 4.07, SD = 1.81). In addition, we controlled for various individual predispositions measured at t-2.Footnote 6 First, political interest, which is measured with an item that asked respondents how interested they are in politics on a 7-point scale (1 = not at all interested and 7 = very interested, M = 4.35, SD = 1.66). Second, ideology, which is measured with a variable tapping left–right placement (1 = left and 10 = right, M = 6.39, SD = 2.28). Third, ideological extremity, by recoding ideology 1 through 5, where “1” denotes being in the middle of the political spectrum, and “5” being either at the left or right extreme end.Footnote 7

Content analysis

We used a content analysis of the last 3 weeks of the election campaign for the television programs and newspapers (August 22 to September 12, 2012).Footnote 8 All items with political content were coded in collaboration with the Dutch public broadcasting agency (NPO), by a team of four coders. In this study, we only included those media outlets for which media exposure was tapped in the panel dataset (i.e., the most used media outlets in the Netherlands). Those are the news programs of the public broadcaster NOS Journaal, and two commercial stations RTL Nieuws and Hart van Nederland, the current affairs programs Eén Vandaag and Nieuwsuur, the talk show Knevel and van den Brink (all public broadcasts), and the infotainment programs De Wereld Draait Door, Pow News, and RTL Boulevard (only the latter is a commercial broadcast). For the newspapers, we included two broadsheet/elite newspapers: de Volkskrant and NRC Handelsblad, two semi-tabloid newspapers: de Telegraaf and Algemeen Dagblad, and two (popular) free dailies: Sp!ts and Metro. Items were coded that satisfied the conditions of campaign news, in the sense that the story was about the elections, party leaders, or about the government. Items were identified based on content and form.Footnote 9 Information about the coding procedures is included in Online Appendix A.

Measures

In the content analysis, we coded indicators of issue news and poll news for each item. Issue news was coded with the following dummy variables: “Is the story mainly about substantial policy issues, problems and solutions?,” “Does the story describe the content or details of (proposed) legislation or other government policy?,” “Does the story describe the position or standpoints of the actor on substantial policy issues?,” “Does the story describe the consequences or effects of (proposed) legislation for the public?” Poll news was coded with “Does the story pay attention to (the results of) polls?” and “Does the story pay attention to the position of politicians or parties in the polls?” In both cases “1” indicated the presence of the type of news. The intercoder reliability was sufficient, with an average Krippendorff’s Alpha of 0.74 for both the coding of issue and poll news. We checked whether the different items indeed constitute a scale by using a Mokken scale. The Mokken scale is a probabilistic version of the better-known Guttmann scale (Mokken 1971) and is used for dichotomous items. The Mokken scale analysis showed that the four items measuring issue news together form a strong scale (H = 0.748, p = 0.000), and the two items tapping poll news also form a strong scale (H = 0.564, p = 0.000). We used an average score to tap the presence of each type of news in each item or article.

Linking survey data to content data

To link media content to the individual-level data, we asked respondents about their exposure to the various media outlets included in the analysis: “Can you indicate how often you read the following newspapers?,” “Can you indicate how often you watch the following television programs?” These media exposure variables were measured on a 5-point scale ranging from never (0) to (almost) daily (4).Footnote 10

For each respondent, exposure to media content was weighted on the basis of the issue news and poll news variables, computing individual exposure to the two types of news, see Eq. 1 (following e.g., Schuck et al. 2016).

$${\text{Type }}\;{\text{of}} \;{\text{news}}_{i} = \frac{{\mathop \sum \nolimits {\text{Type }}\;{\text{of}} \;{\text{news}}_{{{\text{medium}}_{i} }} \times {\text{Media}} \;{\text{exposure}}_{{{\text{medium}}_{i} }} }}{{\mathop \sum \nolimits {\text{medium}}}}.$$
(1)

These weighted media exposure variables were thus determined by the media outlets each respondent uses and the average attention to issue news and poll news in each outlet.Footnote 11 The total exposure to issue news and poll news in newspapers and television programs was divided by the total amount of media outlets included in the analysis.Footnote 12 For example, a respondent reads the newspaper NRC Handelsblad on a daily basis (media exposure = 4) and the average presence of issue news in NRC Handelsblad is 0.34. We link this presence of issue news to the respondent by multiplying the issue news score with the respondent’s media exposure score for NRC Handelsblad, resulting in an issue news exposure score of 1.36. This step is repeated for each newspaper (total of 6 outlets) and each television program (total of 9 outlets). Subsequently, we computed an average issue news exposure score for newspapers and television separately.Footnote 13 For newspapers, the issue news exposure score is first summed for all 6 newspaper outlets and then divided by 6 to obtain an average issue news exposure score. For television, the issue news exposure score is first summed for all nine television outlets and then divided by 9. A similar procedure is conducted to obtain average poll news exposure scores.

Results

Table 1 shows the average amount of issue coverage and poll coverage in television programs and newspapers, as well as the average media consumption per media outlet. In general, media cover more on issues than on polls. However, there are some differences between the various media outlets. Overall, newspapers pay more attention to issue news than television programs; and on television, current affair programs and talk shows feature more issue news than infotainment programs. Conversely, television programs pay more attention to poll news than newspapers. Infotainment programs generally pay little attention to both issue and poll news, suggesting that other (campaign) news is probably more present in these programs.

Table 1 Overview issue coverage, poll coverage, and media use per medium
Full size table
Table 2 The impact of media content on crystallization and conversion
Full size table

The effect of issue news and poll news on conversion and crystallization is tested with multinomial logistic regression analyses, since the dependent variable has four possible outcomes. In multinomial logistic regression, the impact of predictors on the outcome is compared relative to the impact of the predictors on the base category. For theoretical reasons, the impact of the variables on the outcome “conversion” is compared relative to the base category “stable.” The impact of the variables on the outcome “crystallization” is compared relative to the base category “abstention.”Footnote 14

Hypothesis 1a stated that exposure to issue news has a positive effect on crystallization. Indeed, we find a significant positive effect of issue news in newspapers on crystallization. However, the effect of televised issue news on crystallization is insignificant.Footnote 15 Hence, hypothesis 1a is partly confirmed. Figure 1 shows the predicted probability of crystallization for different values of issue and poll news exposure. The graph illustrates that among respondents with an average level of exposure to issue coverage (M = 0.21, SD = 0.21) the predicted probability of crystallization was 0.10. By comparison, among respondents whose exposure to issue news was two standard deviations above the mean, the predicted probability of crystallization was 0.28. This means that more exposure to issue news in newspapers leads to a higher chance that people eventually crystallize their vote choice. So, for voters who are undecided in the last weeks of the election campaign, exposure to issue coverage helps them make up their mind. Hypothesis 1b stated that exposure to issue news has a negative effect on conversion. We indeed find a significant negative effect of issue news on conversion. Figure 2 illustrates the predicted probability of conversion for different values of issue and poll news exposure. Voters who are exposed to issue news are less likely to convert, i.e., switch between parties, but rather remain stable. Hence, hypothesis 1b is supported. We also find effects for political interest and age on crystallization. The more politically interested are more likely to crystallize than to abstain from voting. And older voters are more likely to abstain from voting than to crystallize their vote choice.

Fig. 1
figure 1

Note: The graphs show the predicted probability for respondents with average level of news exposure and the predicted probability for respondents with one and two standard deviations above and below the mean. Estimates are calculated while keeping the other variables at observed values

The predicted probability of crystallization for different values of issue and poll news exposure.

Full size image
Fig. 2
figure 2

Note: The graphs show the predicted probability for respondents with average level of news exposure and the predicted probability for respondents with one and two standard deviations above and below the mean. Estimates are calculated while keeping the other variables at observed values

The predicted probability of conversion for different values of issue and poll news exposure.

Full size image

Hypothesis 2a stated that exposure to poll news has a positive effect on crystallization. We actually find the opposite effect: exposure to poll news in newspapers decreases the chance of crystallization, and leads undecided voters to abstain from casting a vote. Hence, hypothesis 2a is not supported. Hypothesis 2b stated that exposure to poll news has a positive effect on conversion. Indeed, we find a significant positive effect of exposure to televised poll news on conversion. However, the effect of exposure to poll news in newspapers on conversion is insignificant. Hypothesis 2b is thus partly supported. Figure 2 illustrates that voters who are exposed to poll news on television are more likely to switch between parties.Footnote 16 For the other predictors, we find an effect of education, ideology, and ideological extremity on conversion. Voters who are lower educated, left-wing, and ideologically less extreme are more prone to convert from one party to another during the campaign than to remain stable.Footnote 17

Discussion

The increase of electoral volatility over the past decades has inspired many scholars to explain this phenomenon. However, little research has focused on short-term explanations of electoral volatility. Especially the effect of specific media coverage on vote switching has been understudied. This study examined how exposure to specific media content, issue news and poll news, affects vote change at the individual level. In addition, we distinguished between different types of volatility (conversion and crystallization). The study showed that different kinds of news can lead to different voting decision patterns.

First of all, we find an effect of issue news for both types of vote switching. Whereas issue news exposure decreases conversion, it increases crystallization. Thus, voters with an existing preference rather remain stable than switch their preference when exposed to issue news. Undecided voters eventually crystallize their vote choice in response to issue news exposure. It thus seems that exposure to issue news has a reinforcing role for voters with an existing preference and informational role for undecided voters. Voters who are undecided at the start of the campaign learn about parties’ performances and stances on issues in order to get their party choice in line with their attitudes (Arceneaux 2005; Gelman and King 1993). However, we should note that the effects for issue and poll news are not across the board for the different types of media. For example, the positive effect of issue news on crystallization is found for newspapers, but not for television. This can partly be explained by the fact that the differences in type of media content are related to the medium, in the sense that Dutch newspapers pay more attention to issue news than Dutch television news programs (Kleinnijenhuis et al. 2007a). Nevertheless, the results provide support for the argument that scholars should adopt a broader definition of campaign effects, including also the informational role of media, when studying the impact of media on voting behavior (Hillygus 2010).

Secondly, we find a positive effect of exposure to poll news on conversion and a negative effect on crystallization. Thus, exposure to poll news either induces voters to switch parties or to abstain from voting. The finding that poll news increases the chance of conversion is in line with work by Takens (2013), who found that strategic news enhances electoral volatility. Future, experimental research should examine whether this can be explained by the bandwagon effect or by strategic voting (Meffert and Gschwend 2011). As for the impact of poll news on crystallization, our results show that poll news does not provide undecided voters with additional information helping them to crystallize their vote choice by casting a strategic vote. Instead, poll news leads to abstaining from voting. One explanation for this result is that voters might decide not to turn out when their favorite party is performing poorly in the polls, and will not have an impact on the election results anyway. Another possible explanation is the fact that poll news, as a part of strategic and game news, might induce political cynicism (Adriaansen et al. 2010), which in turn leads to demobilization (Cappella and Jamieson 1997) or frustrated vote switching (Zelle 1995). Future research should further explore these findings, by examining the underlying mechanisms of media effects on electoral volatility.

We also find an effect of political interest on crystallization. The more politically interested voters are more likely to crystallize, and are not more likely to convert. Yet, we did find that highly educated voters are more likely to remain loyal to the same party than to convert to another party. These results are in line with Zaller (1991), who argues that highly politically sophisticated individuals are more apt and motivated to absorb information during campaigns, but not very likely to accept the new information and to consequently change their existing beliefs and preferences. Yet, highly interested voters who are still undecided at the start of the campaign do crystallize their vote choice. They might be searching for additional information until the last moment in order to cast an informed vote (Irwin and Van Holsteyn 2008).

Understanding the impact of media content on electoral volatility is important for assessing whether it is positive or negative for democracy. It is often argued that electoral volatility leads to an unstable democracy and complicates governability. However, volatility can also be positively interpreted, as a sign of voter emancipation (see Van der Meer et al. 2015). On the one hand, vote switching could be positive if it is based on an informed decision driven by substantive considerations: when voters crystallize in response to issue news. On the other hand, volatility motivated by peripheral cues, such as polls, could be a signal of indecisive and uninformed voting, leading to arbitrary election outcomes. Still, it is difficult to judge whether the effect of polls, used as a peripheral cue, on vote switching is inherently positive or negative. The influence of polls is often regarded as the latter, as it induces “heuristic” information processing leading to a bandwagon effect based on irrational considerations (see Hardmeier 2008) or induces political cynicism leading to frustrated vote switching (Zelle 1995). However, poll reports can also be used as a guidance for strategic considerations to cast an informed vote. This may lead to volatile voting behavior, which is not per se negative, but also implies voter emancipation.

This study is obviously not without a few shortcomings. First, we only focus on a sample of the information sources available, excluding online sources (although we do include exposure to newspaper websites). Yet, most people tend to visit only the online news sources of the traditional news media, often combined with using the offline counterpart (Trilling and Schoenbach 2015). We are, therefore, confident that including online sources would lead to similar conclusions, because of similar news content. Second, we are restricted to self-reported measures of news exposure, of which accuracy has been questioned (e.g., Prior 2009). Yet, our study is one of the few that measures news exposure on an individual level, differentiating between different media outlets, which has shown to be a reliable and valid approach (see Dilliplane et al. 2013).

Of course, future longitudinal and comparative research is important to substantiate the conclusions of this study. Meanwhile, this study provides support for the idea that scholars should go beyond estimating the effect of campaigns and instead investigate in what ways, which campaign news influences vote choice or vote change (Hillygus 2010). In particular, this study shows that different kinds of campaign news can have different effects on different types of voting behavior. Whereas some campaign coverage, i.e., poll news, might persuade voters to alter their party choice, other campaign content, i.e., issue news, can affect voting behavior in a more indirect way. Our findings also demonstrate that volatile voters are not necessarily uninterested and ignorant. Electoral volatility might also be a result of a continuing process off voter emancipation, with voters using campaign news as input for substantial deliberation to come to an informed vote choice.