Intra-Campaign Changes in Voting Preferences: The Impact of Media and Party Communication.

An increasing number of citizens change and adapt their party preferences during the electoral campaign. We analyze which short-term factors explain intra-campaign changes in voting preferences, focusing on the visibility and tone of news media reporting and party canvassing. Our analyses rely on an integrative data approach, linking data from media content analysis to public opinion data. This enables us to investigate the relative impact of news media reporting as well as party communication. Inherently, we overcome previously identified methodological problems in the study of communication effects on voting behavior. Our findings reveal that campaigns matter: Especially interpersonal party canvassing increases voters' likelihood to change their voting preferences in favor of the respective party, whereas media effects are limited to quality news outlets and depend on individual voters' party ambivalence.

Introduction

An increasing number of voters change their party preferences from one election to another, a phenomenon that is commonly described as electoral volatility (e.g., Hobolt, Spoon, & Tilley, 2009; Kramer, 1970; Lazarsfeld, Berelson, & Gaudet, 1948; Pedersen, 1979; Zaller, 2004). Electoral volatility has frequently been observed across countries and been addressed by a growing body of literature (e.g., Dassonneville, 2012; Jennings & Wlezien, 2016; Pedersen, 1979). Within this framework, inter-election vote switching is understood as a long-term process that predominantly affects the formation and evolution of party preferences (e.g., Hobolt et al., 2009).

However, changes in party preferences are also observed in the short term, for example, in electoral campaigns (e.g., Jennings & Wlezien, 2016): Voters may report that they intend to vote for a particular party before the election, but actually cast a ballot for a different party on Election Day. Moreover, undecided voters often develop a particular party preference during the election campaign. While previous research has examined the impact of electoral campaigns on vote choice (e.g., Boomgaarden et al., 2016; Dilliplane, 2014; Druckman, 2004, 2005; Hillygus & Jackman, 2003; Kleinnijenhuis, Van Hoof, Oegema, & de Ridder, 2007; Lengauer & Johann, 2013; Matthews & Johnston, 2010; Schmitt-Beck, 2003), it is yet largely unknown what affects these intra-campaign changes in voting preferences.

This article examines whether and how far political information influences the likelihood of voters to change their party preferences during election campaigns. As political information has been found to shape voters’ political perceptions and attitudes (e.g., Ladd & Lenz, 2009; Lazarsfeld et al., 1948; Lengauer & Johann, 2013; Schmitt-Beck, 2003), we argue that the individual’s information environment provides voters with important cues to make decisions and that it, consequently, has the ability to alter citizens’ voting preferences in the short term (e.g., Gerber & Green, 2000, Green & Gerber, 2015; Jennings & Wlezien, 2016; Karp & Banducci, 2007; van Spanje & de Vreese, 2014). We investigate the effect of communication efforts by the news media and political parties on short-term changes in voting preferences in the weeks before Election Day. While the news media process and present information about political candidates and parties with varying frequency and tone (e.g., Geers & Bos, 2017), party representatives use different direct and indirect channels of communication to canvass citizens (e.g., Gerber & Green, 2000; Green & Gerber, 2015; Lazarsfeld et al., 1948). The type of communication channels used and the extent to which these are used vary by individual, however, making it important to study the individual information context to understand why citizens change their minds (e.g., Evans, 1999; Kritzinger & Johann, 2016).

Even though Lazarsfeld and colleagues (1948) have already noted that both media and party communication are important in shaping citizens’ voting preferences, previous research on campaign effects has predominantly focused on one or the other. By linking high-quality public opinion with media data collected by the Austrian National Election Study (Eberl et al., 2016; Kritzinger et al., 2016a; Kritzinger et al., 2016b), we are able to address how far individual-level media and party communication affect intra-campaign changes in voting preferences.

The article is structured as follows: We begin by reviewing the literature on information environments in the context of electoral campaigns and how they affect changes in voting preferences. Next, we discuss our data and methods before presenting the empirical results. We close with a discussion on the implications of our study for future research.

Media and party communication effects on changes in voting preferences

Scholars commonly acknowledge that citizens’ level of knowledge is positively linked to a better understanding of political processes and policy content (e.g., Delli Carpini & Keeter, 1996; Jerit, Barabas, & Bolsen, 2006). Prior work has further indicated that electoral choices may be influenced by the way in which political information is gathered, processed, and presented in the weeks leading up to Election Day (e.g., Arcuri, Castelli, Galdi, Zogmaister, & Amadori, 2008). For instance, informed voters perceive the costs of voting as lower and the benefits as higher, which increases turnout among this group (Karp & Banducci, 2007). In the following, we discuss the impact of (a) news media reporting and (b) party communication on short-term changes in party preferences in more detail. We also focus (c) on potential individual-level differences in campaign communication.

News Media Reporting

An increasing body of literature indicates that news media reporting affects citizens’ political perceptions and behavior (e.g., Iyengar & Kinder, 1987; Scheufele, 2000; Scheufele & Tewksbury, 2007; Zaller, 1992). The news media provide citizens with information on politics, political events, political candidates, and political parties’ policy positions, especially during electoral campaigns (e.g., Aalberg, van Aelst, & Curran, 2010; Banducci, Giebler, & Kritzinger, 2017; Barabas & Jerit, 2009; Kleinnijenhuis et al., 2007; Druckman, 2004, 2005; Norris, 2002). However, not all political parties or candidates will be treated in the same way by all media outlets: Some will receive more attention than others (visibility bias); others will be criticized more severely for their actions and plans (tonality bias; see Eberl, Boomgaarden, & Wagner, 2015; Hopmann, van Aelst, & Legnante, 2011).

As previous research has shown, a greater visibility in the news media enhances the general saliency of a topic among the public as well as the popularity of the political actors in question and their vote share (Chong & Druckman, 2007; Lengauer & Johann, 2013). This effect should also be found with respect to changes in voting preferences during an electoral campaign. A more frequent exposure to information about a particular party will increase citizens’ knowledge of this party (Banducci et al., 2017; Banducci & Semetko, 2003). However, it may also make this party more accessible in citizens’ memories (so-called priming; Chong & Druckman, 2007; Zaller, 1992) and trigger changes in voting preferences. Changes between parties may occur without any significant shifts in the underlying ideological or issue preferences in multiparty democracies, where citizens have the choice between several parties offering similar political platforms. Our “media visibility hypothesis” thus posits:

The more visible a particular party is in the news media an individual is exposed to, the more likely it is that the individual will change his or her party preference in favor of this party.

Regarding tonality bias, prior research indicates that a favorable, neutral, or unfavorable tone regarding political actors, parties, or policies may prime individuals in thinking about these actors in a certain way (e.g., Scheufele, 2000; Scheufele & Tewksbury, 2007; Zaller, 1992). Moreover, the tone of news media reporting affects citizens’ party choices: Voters are more likely to vote for a candidate or party if the news media report favorably about them (e.g., de Vreese & Semetko, 2004; Geers & Bos, 2017; Hopmann, Vliegenthart, de Vreese, & Albæk, 2010; Lengauer & Johann, 2013; Norris, 1999; Vliegenthart, Schuck, Boomgaarden, & de Vreese, 2008; but see Dassonneville, 2012; Giebler, Kritzinger, Xezonakis, & Banducci, 2017).

Dual processing theories (Petty & Cacioppo, 1986) suggest two alternative mechanisms responsible for such effects: Positive (or negative) reporting may make the party itself, but also the positive (or negative) evaluations, more accessible in people’s memories (heuristic processing or priming). Alternatively, positive reporting might actually persuade voters by presenting convincing arguments for a certain candidate, especially if voters are motivated to process the information more thoroughly (systematic processing). Both effects should also be important for intra-campaign changes in voting preferences. Heuristic processing may trigger a decision for the party that has been presented more favorably in individuals’ news media environment (e.g., Fournier, Nandeau, Blais, Gidengil, & Nevitte, 2004). Systematic processing may persuade voters to change their vote preference as a result of more careful considerations of the information provided about the different parties. Our “media tone hypothesis” thus postulates:

The more positive the tone of news media reporting about a particular party in an individual’s news diet is, the more likely it is that individual will change his or her party preference in favor of this party.

Furthermore, the news media also vary in the quality and depth of the political information provided: Citizens who rely on quality media outlets are more likely to be exposed to policy-relevant news than those consuming the tabloid press (Aalberg et al., 2010; de Vreese, Banducci, Semetko, & Boomgaarden, 2006). Even though persuasion effects based on heuristic processing might occur for any media outlet, those triggered by systematic processing of information should be more likely to occur among citizens relying on a quality news media diet that provides them with substantive arguments for changing their voting preference to a different party. Our “quality media hypothesis” suggests:

The impact of news media reports concerning a particular party on changes in voting preference for this party is more pronounced among voters consuming quality news outlets.

Interpersonal and Impersonal Party Canvassing

Direct and indirect party canvassing is also deemed important. Prior research on party canvassing has suggested that local canvassing has a mobilizing effect on voters: Citizens are more likely to turn out in an election if they have been approached by a party representative during the campaign (e.g., Caldeira, Clausen, & Patterson, 1990; Fisher, Fieldhouse, Johnston, Pattie, & Cutts, 2015; Gerber & Green, 2000; Green & Gerber, 2015; Marsh, 2004; Wielhouwer & Lockerbie, 1994). However, a lot remains to be learned about the relationship between parties’ campaign activities and individuals’ party preferences (e.g., Hillygus & Jackman, 2003, but Matthews & Johnston, 2010). Empirical studies have provided evidence that increased party contact may influence citizens’ vote choices (e.g., Fisher et al., 2015; Górecki & Marsh, 2012, 2014; Johnston & Pattie, 2003; Lefebvre, 2014). The underlying mechanism triggering this effect is believed to be similar to the priming argument regarding media coverage: More frequent exposure to information provided by a particular party should increase the popularity of this party among the public by increasing the accessibility of its political platforms (see Chong & Druckman, 2007).

Political parties predominantly rely on two channels of communication (for a review, see Fisher et al., 2015; Gerber & Green, 2000; Green & Gerber, 2015): (a) more traditional, labor-intensive campaigns in which party members and volunteers approach citizens interpersonally (directly) at the doorstep, at rallies, or other campaign events introducing the core policy proposals of the respective party; and (b) modern, less expensive mass campaigns in which parties contact voters impersonally (indirectly) by distributing letters and leaflets, sending text messages or e-mails, or by using other means of indirect communication to present their policy proposals.

Previous studies on voter turnout and party canvassing highlight that interpersonal campaign efforts are more effective in mobilizing citizens to vote than impersonal party contact (e.g., Fisher & Denver, 2009; Fisher et al., 2015; Gerber & Green, 2000; Green & Gerber, 2015; Marsh, 2004). However, the findings on the impact of canvassing on voting preferences are rather contradictory. Some studies conclude that neither interpersonal nor impersonal canvassing affects voting preferences (e.g., Denver & Hands, 1997; Kramer, 1970; Magalhães, 2007). Yet, other research suggests that citizens indeed respond to the respective parties’ canvassing efforts by voting for these parties, especially when they have been contacted interpersonally (e.g., Fisher & Denver, 2009; Fisher et al., 2015; Górecki & Marsh, 2012, 2014; Johnston & Pattie, 2003; Lefebvre, 2014; Pattie & Johnston, 2003).

Our study aims to further disentangle the influence of interpersonal and impersonal party canvassing on intra-campaign changes in voting preferences. The role of direct social contact as an effective mean to influence voting decision has long been recognized (Arceneaux, 2009; Berelson et al., 1954; Katz, 1957). Messages delivered personally are harder to ignore, as a connection between the provider and the receiver of the information is established (Arceneaux, 2009; Niven, 2004). Interpersonal contact may also reduce prejudice (Christ et al., 2014) against parties that citizens had not initially considered as a viable option. Furthermore, interpersonal contact is more likely to motivate citizens to process and consciously evaluate information, i.e., systematic processing (Chong & Druckman, 2007; Katz, 1957).

By contrast, impersonal means of communication may reach voters simply by the volume and frequency with which they are sent (Arceneaux, 2009; Fieldhouse, Cutts, Widdop, & John, 2013), but they lack the social contact component and allow voters to rationally select and process information (e.g., Arcuri et al., 2008). Consequently, we would expect voters who are exposed to interpersonal interaction with a particular party representative to be more likely to change their preferences in favor of this party than those contacted by impersonal means. Our “party canvassing hypothesis” posits:

The impact of interpersonal party contact on changes in voting preferences is more pronounced than the effect of impersonal canvassing.

Individual-level Differences in Campaign Communication Effects

Voters may differ in how firmly or loosely they feel about the party choices on offer. Previous studies have linked volatile voting choices to weakening party attachments, suggesting that apartisanship and indecisiveness increase the likelihood for changes in voting preferences (e.g., Dalton, 2012, 2013; Gramberg & Holmberg, 1990). More recent studies have indicated that party ambivalence—that is, voters are attracted to several parties at the same time (Schmitt-Beck & Partheymüller, 2012)—also influences vote choices and makes voters postpone their final vote decisions. We expect that ambivalent voters are more attentive to and thus also more susceptible to campaign messages by both the political parties and the news media (Basinger & Lavine, 2005; Chong & Druckman, 2007; Petty & Briñol, 2008) as they feel pressured to come to a decision. Our “ambivalence hypothesis” states:

Ambivalent voters are more likely to change their voting preferences in response to the media’s and parties’ campaign communication.

Data and methods

The Case

Austria is an ideal case to study our hypotheses empirically: It is characterized by a multiparty context, in which changes in party preferences during the electoral campaign are more likely to occur, as citizens can switch to a different party in the same ideological camp (see Schmitt-Beck & Partheymüller, 2016 for empirical evidence on Germany). The Austrian electoral system further encourages interpersonal party contact as voters cast party votes, but they may also indicate their preferred candidate on the respective party list. Thus, the political system fosters interpersonal canvassing activities by all parties and candidates and at all levels. Austria is also characterized by a generally high electoral turnout, which is ideal to test whether parties pursue persuasion strategies by contacting potential voters (e.g., Karp & Banducci, 2007). Finally, the Austrian media system is characterized by a comparatively high level of political parallelism (Udris & Lucht, 2014) and a very strong competition for audiences (Karmasin, Kraus, Kaltenbrunner, & Bichler, 2011). As a result of this, even though most media outlets can be attributed to a certain ideological camp (Magin, 2012), they may change their preferences in favor of parties within this camp to distinguish themselves more clearly in their political reporting from their competitors.

Data

Many empirical studies in the field of communication effects suffer from a variety of methodological problems (e.g., Druckman, 2005). First, they either attempt to draw causal conclusions from cross-sectional data (for a critical discussion, see Dilliplane, 2014) or lack concrete media content data and base their conclusions merely on media usage. The integrative design employed by the Austrian National Election Study (AUTNES) overcomes these problems: it allows us to connect high-quality public opinion data (Kritzinger et al., 2016a; Kritzinger et al., 2016b) to the media content that respondents consume (Eberl et al., 2016).

Second, by focusing on changes in party preferences during the electoral campaign as our dependent variable, we can overcome endogeneity concerns with which previous studies on canvassing may have been plagued. As parties are more likely to contact their supporters, it may be that observed effects on vote choice or turnout do not really originate from parties’ canvassing effort but from voters’ general party preferences. We study the impact of canvassing on voters who either had not reported a party preference in the pre-election survey or had a different party preference. Our findings with regard to campaign effects should thus be more robust.

Public Opinion Data

Public preferences and attitudes were collected in a pre- and post-election panel survey, with the pre-election survey conducted according to a rolling cross-sectional (RCS) design (N = 4,011). A total of 2,607 respondents also participated in the post-election questionnaire. Voter preferences were surveyed between August, 5, and September, 27, 2013, using Computer Assisted Telephone Interviews (CATI). The panel design allowed us to track the proportion of voters who developed and changed their party preferences over the campaign. We compared their vote intention to their reported vote choice after Election Day (September 29, 2013).

Media Content Data

Our media data include articles published in the six most relevant Austrian newspapers (Der Standard, Die Presse, Kronen Zeitung, Österreich, Heute, and Kleine Zeitung; N = 4,265) and TV newscast Zeit im Bild 1 (N = 273) issued between August, 19, and September, 28, 2013. Der Standard, Die Presse, and Zeit im Bild 1 represent quality media outlets, and Kronen Zeitung, Österreich, Heute, and Kleine Zeitung are tabloids (Bakker & Seethaler, 2009).

The media content was manually coded. A standardized content analysis was conducted on the basis of all media reports referring to an Austrian political candidate or party in the six weeks leading up to Election Day. For each media report, the evaluations of political actors or parties in the title as well as in the lead and the first paragraph (total N = 34,485) were coded by a team of six coders. We obtained satisfactory results in our reliability test for both variables (evaluation object: Krippendorff’s α = .74; tone: Krippendorff’s α = .76, based on a random sample of 1,123 sentences).

Changes in Voting Preferences

As our analyses focus on intra-campaign changes in voting preferences, we first identified all respondents who had not yet decided which party to support in the pre-election survey, but reported that they voted for a party on Election Day (= late deciders), and those who had indicated a vote intention in the pre-election survey which was different from their final vote choice in the post-election survey (= vote switchers). Next, we determined these voters’ final party preferences by looking at the traditional vote choice question in the post-election survey. We considered the four biggest parties represented in parliament: The Social Democrats (SPÖ) and the People’s Party (ÖVP), as well as two opposition parties, radical right Freedom Party (FPÖ) and the Greens. Due to low case numbers, we had to exclude smaller parties from the analyses.

Visibility and Tone of News Media Reporting

To measure media effects we looked both at the visibility and tone of news media reporting on political parties (see Lengauer & Johann, 2013). Visibility was measured by counting the evaluations for each party and media outlet per day and weighting them by the party with the most evaluations on that day (Relative Media Visibility; see Table 1). A 3-point scale was employed to capture the tone of the actors’ evaluations (–1 = negative, 0 = neutral, +1 = positive). The evaluations were averaged for each day and then weighted by the overall tone of the media outlet on that day across all parties (Relative Media Tone; see Table 2) .

Table 1

Average relative media visibility by media outlet and party

	De Der Standard	Die Presse	Zeit im Bild 1	Heute	Kronen Zeitung	Österreich	Kleine Zeitung	Total
SPÖ	.10	.13	.10	.09	.13	.14	.13	.82
ÖVP	.09	.11	.08	.09	.12	.11	.12	.73
FPÖ	.05	.04	.04	.06	.05	.07	.07	.37
Greens	.04	.04	.03	.05	.06	.05	.05	.32
N	2,096	2,488	273	1,665	3,337	5,219	3,614	18,419

Note. Relative Media Visibility (RMV): Count of evaluations of a party in a specific media outlet (for each day), weighted by the party with the highest number of evaluations (for that day). RMV takes values between 0 and 1, with 1 denoting the party with the highest visibility in all outlets.

Table 2

Average relative media tone by media outlet and party

	De Standaard	Die Presse	Zeit im Bild 1	Heute	Kronen Zeitung	Österreich	Kleine Zeitung	Total
SPÖ	–.19	–.25	–.23	–.23	–.14	–.07	–.13	–.17
ÖVP	–.29	–.13	–.24	–.13	–.22	–.16	–.13	–.20
FPÖ	–.12	–.25	–.07	–.24	–.18	–.10	–.12	–.15
Greens	–.10	–.10	.00	–.19	–.14	–.02	.00	–.07
N	2,096	2,488	273	1,665	3,337	5,219	3,614	18,419

Note. Relative Media Tone (RMT): Average evaluation of a party in a specific media outlet (for each day), weighted by the overall tone of the media outlet. RMT takes values between –1 and +1, with +1 denoting the party with only positive evaluations.

We calculated the Media Visibility Index (MVI) for each party and respondent by weighting the visibility of a party for an outlet by the individual’s weekly media use of the particular outlet. These values were then combined for all six newspapers outlets and the main newscast. Lower values of the MVI indicate that the respondent was not or only infrequently exposed to coverage about a particular party; higher values of the MVI indicate that the respondent was frequently exposed to coverage about a particular party.

Similarly, we calculated the Media Tone Index (MTI) for each party and respondent by weighting the tone of each outlet for that party by the individual’s weekly media use of the particular outlet and by combining these values for all six newspaper outlets and the main newscast. Lower values of the MTI indicate that respondents were predominantly exposed to a negative tone about a specific party in their consumed media outlets; higher values suggest that they were exposed to a more positive tone.

Impersonal and Interpersonal Party Canvassing

Party canvassing effects are indicated by survey questions capturing whether or not respondents were contacted by the respective parties as well as by what means they were contacted. In total, we derive two measures of party canvassing distinguishing interpersonal (e.g., canvassing at the doorstep or at a rally) from impersonal ways of party contact (e.g., by mail, e-mail, or text message). We achieve this by generating dummy variables for each party and type of contact. The variable interpersonal contact takes a value of 1 if a party canvassed respondents at a campaign stand or approached them at the doorstep in direct face-to-face interaction; it is coded as 0 otherwise. Similarly, impersonal contact is coded as 1 if a party contacted a respondent by mail, e-mail, or telephone (including text messages) and as 0 otherwise.

Party Ambivalence

We follow Schmitt-Beck and Partheymüller (2012) to operationalize party ambivalence employing 11-point feeling thermometers (i.e., like-dislike scales) of the respective parties. This allows us to account simultaneously for the dissimilarity and the intensity of attitudes in a multiparty setting and to quantify the balance between agreement and disagreement of all parties (Schmitt-Beck & Partheymüller, 2012). High values indicate that voters like more than one party (= high ambivalence), and low values indicate the absence of party ambivalence (= low ambivalence).

We also control for relevant attitudinal indicators: Respondents’ political knowledge, their general attention paid toward the campaign, and votes’ evaluation of the economy, as well as age, gender, education, religiousness, and union membership. To account for the panel design, we also control for the number of days until Election Day. Finally, we include whether (=1) or not (=0) respondents reported a party preference in the pre-election wave capture differences between late deciders and voters whose voting preferences changed from pre- to post-election wave.

Analysis Strategy

As voters differ in the information channels they rely on and the extent to which they are exposed to political information, we need to operationalize their individual information environment. Each respondent needs to be represented multiple times in the analysis, once for each of the parties analyzed (e.g., Banducci et al., 2017; Barabas & Jerit, 2009; van der Eijk, Franklin, & Oppenhuis, 1996; van der Eijk, van der Brug, Kroh, & Franklin, 2006). We achieve this by stacking the data in a way that respondents are represented for the respective parties, in our case four times. By doing so, we are able to assess for each individual how different levels of media and party communication about the respective parties influence the likelihood of changing their voting preferences. Such a within-subject design allows us to compare for the same individual how variations in news information about all parties influence changes in party preferences (Banducci et al., 2017). The dependent variable equals 1 if voters have changed their voting preferences in favor of a particular party and 0 for all the other parties.

Due to the within-subject design and the fact that we incorporate both alternative-specific and individual-specific variables, we estimate an alternative-specific conditional logit model (ASCLM; e.g., Alvarez & Nagler, 1994; Karimi, Hotz, & Johansson, 2016; Lifschitz, Sauder, & Stevens, 2014; McFadden, 1974). The ASCLM allows the media indices and party contact variables (alternative-specific variables) to vary by party and enables us to estimate the impact of voters’ individual political information environment. At the same time, the ASCLM allows us to control for variables that are constant within the respondents, such as the sociodemographic controls (individual-specific variables). When presenting our results, we only focus on the impact of the alternative-specific media and party communication variables.

In total, we run seven models. Model 1 is our baseline model, which only includes the individual-specific control variables. We include media visibility measures by quality versus tabloid outlet and impersonal and interpersonal party contact in Model 2. Model 3 provides a similar test for the tone of news media reporting by outlet and impersonal and interpersonal party contact. Next, we reestimate our second model, focusing on media visibility and party contact by voters with low levels of party ambivalence (Model 4) and respondents displaying high levels of ambivalence (Model 5). We conduct a similar analysis focusing on the tone of news media reporting and party contact by non-ambivalent (Model 6) and ambivalent voters (Model 7).

Results

We begin by presenting the descriptive results. Overall, voters displayed volatile electoral behavior during the electoral campaign: A significant proportion of voters (approximately 31%) reported that they either developed or changed their party preferences during the electoral campaign. The SPÖ benefitted most from intra-campaign changes in voting preferences, attracting roughly 32% of voters, followed by the Greens (26%) and the ÖVP (23%). The FPÖ was least successful, with only 19% of voters changing their party preferences towards this party.

We also note differences in party canvassing. Table 3 summarizes the proportion of interpersonal and impersonal party contacts as reported by the respondents, suggesting that party contact in general is an important mean to reach voters. It also indicates that respondents have predominantly been contacted impersonally by phone, mail, or e-mail rather than interpersonally by a party member at the doorstep or at a rally. This makes sense, as parties tend to rely on mass campaigns nowadays. Most respondents have been contacted by the SPÖ (interpersonally: 8.4%, impersonally: 35.8%), followed by the ÖVP (interpersonally: 6.7%, impersonally: 28.6%), and the FPÖ (interpersonally: 5.0%, impersonally: 25.5%). The smallest proportion of voters reports contact with the Greens (interpersonally: 3.5%, impersonally: 12.0%).

Table 3

Interpersonal and impersonal party contact (weighted results)

	Interpersonal Contact	Impersonal Contact
SPÖ	8.4%	35.8%
ÖVP	6.7%	28.6%
FPÖ	5.0%	25.5%
Greens	3.5%	12.0%
N	2,181	2,116

Note. Interpersonal contact includes direct face-to-face contact with a candidate, party member, or other party personnel such as a visit at the doorstep or contact at rallies. Impersonal contact includes all indirect non-face-to-face contact with a candidate, party member, or other party personnel by telephone, e-mail or mail contact.

Although the visibility of parties in each media outlet differed from day to day, the different media outlets attributed similar attention to the four parties across the whole time period of the analysis: looking at the average Relative Media Visibility (RMV) in Table 1, the news media most frequently attended to the SPÖ in all seven media outlets (RMV: .82), closely followed by the coalition partner ÖVP (RMV: .73). With the exception of the tabloid Kronen Zeitung, the FPÖ (RMV: .37) was more visible than the Greens (RMV: .32) in most outlets.

Looking at the average Relative Media Tone (RMT) as reported in Table 2, the Greens (RMT: –.07) is the party that was best evaluated by the news media, on average. By contrast, the ÖVP was most criticized (RMT: –.20), followed by the SPÖ (RMT: –.17) and the FPÖ (RMT: –.15). However, the media outlets largely differed tone of their party evaluations. Depending on the individuals’ news diets, respondents could be exposed to quite different evaluations: For example, frequent readers of the tabloids Kronen Zeitung and Heute would be exposed to rather negative evaluations of the Greens (RMT Kronen Zeitung: –.14; RMT Heute: –.19). In comparison, viewers of the main public newscast Zeit im Bild 1 would observe by far the most Greens-friendly coverage (RMT: .00), but also quite negative evaluations of the FPÖ (RMT: –.07). Readers of the leftist quality paper Der Standard would encounter the most negative evaluations of the ÖVP (RMT: –.29).

The main effects of the ASCLM are presented in Table 4. Model 1 is the baseline model. Model 2 reveals that party visibility has no impact on changes in voting preferences in the quality or tabloid news media. Solely, the direction of the coefficient for the quality media points in a positive direction, indicating that more frequent attention to a particular party coincides with voters’ changing their voting preferences in favor of this party, but it does not reach conventional levels of statistical significance. These results indicate that we may have to reject H1(a). We find positive and statistically significant effects for party canvassing: Both interpersonal and impersonal campaign efforts seem to convert respondents to vote for the respective party. Furthermore, the effect of interpersonal contact appears to be much stronger than the impact of impersonal canvassing. H2 is supported by the results.

Table 4

Alternative-specific conditional logit model predicting intra-campaign vote switching

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
				Low Ambivalence	High Ambivalence	Low Ambivalence	High Ambivalence
Main Effects
Visibility: Quality Media		0.21		0.44*	–0.07
		(0.11)		(0.18)	(0.15)
Visibility: Tabloid Media		0.00		0.01	0.12
		(0.12)		(0.24)	(0.17)
Inter-personal Contact		2.52***	2.54***	2.74*	2.54***	2.72*	2.50***
		(0.47)	(0.46)	(1.33)	(0.55)	(1.38)	(0.55)
Im-personal Contact		0.62**	0.62**	0.62	0.88**	0.66	0.87**
		(0.20)	(0.20)	(0.38)	(0.28)	(0.36)	(0.29)
Tone: Quality Media			0.19			–0.90	1.11*
			(0.39)			(0.74)	(0.52)
Tone: Tabloid Media			0.17			0.12	0.16
			(0.45)			(0.87)	(0.55)
N	1648	1648	1648	636	1012	636	1012
LL	–505.04	–465.74	–467.90	–143.19	–286.63	–146.29	–284.31
Chi2	72.98	130.22	130.74	88.33	90.48	83.59	96.59
AIC	1082.08	1011.49	1015.79	360.38	647.26	366.59	642.62

Notes. Weights applied. Standard errors in parentheses.

* p < .05. ** p < .01. *** p < .001.

The tone of news media reporting by outlet is included in Model 3. While the impact of interpersonal and impersonal contact remains stable and statistically significant, we are unable to identify a statistically significant influence of the tone of news media reporting, neither for quality nor tabloid news media. Once again, the coefficients point in the expected direction. These results indicate that we may have to reject H1(b).

Assuming that the effect of media and party communication differs across voters with different levels of party ambivalence, Model 4 separates voters with low levels of party ambivalence. Interestingly, the results suggest that media visibility in quality media positively relates to the likelihood of lowly ambivalent voters changing their voting preferences in favor of the respective party supporting H1(c). We do not find a similar effect of visibility in the tabloid press. Furthermore, the impact of interpersonal contact remains stable: Lowly ambivalent respondents who had interpersonal contact with a party representative are more likely to change their voting preference in favor of this party during the campaign compared to their counterparts who had no interpersonal contact with a party representative. However, the influence of impersonal campaigns vanishes. We do not identify a statistically significant influence among voters displaying low levels of ambivalence, lending support for H3.

Looking at highly ambivalent voters in Model 5, the results indicate that media visibility has no longer an impact on changes in voting preferences, whereas interpersonal and impersonal party canvassing seems to drive changes in voting preferences of ambivalent voters. Once again, the impact of interpersonal contact is stronger than that of impersonal canvassing.

We distinguish the impact of the tone of news media reporting and party contact among voters with low levels of ambivalence and respondents displaying high levels of party ambivalence in Models 6 and 7. Turning to voters with low party ambivalence first, we find no influence of the tone of news media reporting for quality or tabloid outlets. However, interpersonal contact appears to have a rather strong impact on changes in voting preferences in favor of the respective party. While the coefficient for impersonal party contact points in a positive direction, it does not reach conventional levels of statistical significance. Among voters with high levels of party ambivalence, a positive tone of news media reporting in quality media seems to convert respondents to vote for the respective party. We find a positive and statistically significant coefficient supporting H3 for the quality media. We are unable to isolate a similar effect for the tonality in the tabloid press. In addition, interpersonal and personal party contacts appear to be effective campaign tools to persuade highly ambivalent voters. Both coefficients are positive and statistically significant. The influence of interpersonal means yet seems to be stronger than that of impersonal canvassing.

In sum, the results lend empirical support to our hypotheses, but reveal more diverse patterns than initially thought (see Table 5). We find some support that the visibility of a party in the news media matters for short-term changes in party preferences (H1[a]), even though this applies only to the influence of visibility in quality news outlets among voters with low levels of party ambivalence (H1[c] and H3). Furthermore, the results seem to support our media tone hypothesis (H1[b]). Accordingly, the tone of news media reporting only matters for highly ambivalent voters relying on quality news media (H1[b], H1[c], and H3). Most strikingly, our analysis provides empirical evidence for our assumptions about party canvassing. We find robust results for the influence of interpersonal contact in all models, suggesting that respondents who have been personally contacted by a party are more likely to change their voting preferences in favor of this party. The coefficients are similar in strength and significance. Moreover, we also identify that interpersonal contact seems to be a stronger predictor than impersonal contact (H2). Impersonal party contact seems to be effective among highly ambivalent but not among lowly ambivalent voters (H3).

Table 5

Evaluation of hypotheses against core results

Hypothesis	Core Result
H1[a]: Media Visibility	Applicable to quality news media reporting among lowly ambivalent voters
H1[b]: Media Tone	Applicable to quality news media reporting among highly ambivalent voters
H1[c]: Quality Media	Applicable to the visibility of news media reporting in quality outlets among lowly ambivalent voters;
	also applicable to the tone of news media reporting in quality outlets among highly ambivalent voters
H2: Party Canvassing	Applicable to both impersonal and interpersonal party contact, but stronger impact for interpersonal canvassing
H3: Ambivalence	Applicable to the tone of news media reporting in quality outlets;
	also applicable to impersonal party contact among highly ambivalent voters

Discussion and conclusion

This article investigated the impact of media and party communication on intra-campaign changes in voting preferences following an integrative data approach by linking public opinion data to the media content. We were interested as to whether and to what extent voters’ party preferences are volatile and how they are affected by citizens’ individual information environments in electoral campaigns.

Our results revealed important findings. The core message we take from the empirical analysis is that party campaigns matter. Especially citizens who were personally approached by a political party appear to be more likely to change their voting preferences in favor of this party. While impersonal contact through mass campaigns also seems relevant, it appears to be more effective among voters who are rather ambivalent about their party preferences. Our findings confirm claims about the effectiveness of party campaigning made by previous research (e.g., Fisher et al., 2015; Górecki & Marsh, 2012; Green & Gerber, 2015), suggesting that personal campaign efforts pay off. It almost seems as if interpersonal party contact “shuts down” people’s defense mechanisms with regard to information selection and processing, and thus makes them more susceptible to persuasion efforts (Arcuri et al., 2008).

These results have important implications for future research as well as party campaign strategies: In times of cost-efficient mass campaigns, impersonal canvassing appears to be sufficient in terms of persuading ambivalent voters, but less effective to reach voters with fairly clear party attachments. By contrast, interpersonal campaigns appear to be more effective as they reach all kinds of voters by allowing them to develop a personal rapport with the candidates or their parties. This emphasizes the need for parties to carefully define their campaign aims and to precisely consider their campaign strategies (who to reach and by what means?), if they want to more successfully canvas voters in times of increasing short-term electoral volatility.

Media reporting seems to matter as well. However, media effects appear to be less uniform and to depend on the individual’s usage of news outlets (quality versus tabloid media) and level of party ambivalence. While media visibility in quality outlets seems to affect voters who are already fairly settled regarding their party preferences, the tone of news media reporting in quality media appears to have the ability to convert ambivalent voters. Furthermore, it appears that the tabloid media have no impact, even though they are frequently ascribed increased power to change public opinion. We were unable to identify an effect of the visibility or tone of party evaluations in the tabloid media on changes in voting preferences whatsoever, which previous studies also suggest (e.g., Banducci et al., 2017). Although our research design did not allow us to explicitly test for different levels of cognitive processing, our findings are in line with dual processing theories, such as the elaboration likelihood model (Petty & Cacioppo, 1986). The results of our large-N population study thus further support dual processing theory initially developed on the basis of experimental research: We were able to show that highly ambivalent voters are more likely to be more motivated to process information in (quality) media outlets thoroughly and are thus more susceptible to the tone of party evaluations when changing their party preferences (= systematic or central processing). By contrast, lowly ambivalent voters are less likely to be motivated to process the information in the news, but rather rely on peripheral cues, such as the simple frequency of evaluations in their decision process (= peripheral or heuristic processing).

Our research has important theoretical, but also methodological, implications for the field of political communication. Regarding theory, our research confirms initial assumptions and weaker tests of party communication effects, suggesting that different channels of party communication are perceived differently by different voters. Inherently, they also have a diverse impact on their electoral behavior. The results also provide further evidence on the applicability of long-standing processing theories in media communication effects confirmed by a methodologically neat design using a representative population sample. Finally, the study design contributes a better suitable model for a not-so-new data structure. While studies linking media content to survey data have been conducted before, the alternative-specific conditional logit model is superior to analyze the resulting and more complex stacked data structure when studying campaign effects. To our knowledge, this model has not been applied when studying political communication effects.

We yet see leeway for future research. To fully understand the impact of electoral campaigns on voters, it may be useful to incorporate additional information on party communication. For instance, it is important to understand the context in which campaigning takes place as well as particular campaign strategies by the political parties, their budgets, and legal regulations (see Fisher et al., 2015; Green & Gerber, 2015). Furthermore, it would also be interesting to include important issues that were relevant in the respective campaign and which may have the ability to persuade voters, too.

Overall, we conclude that with an increasing number of volatile voters and late deciders, political communication efforts during electoral campaigns become more important. We provided convincing evidence that the information environment is able to shift electoral choices.

Table A1

Alternative-specific conditional logit model predicting intra-campaign vote switching

				Model 4	Model 5	Model 6	Model 7
	Model 1	Model 2	Model 3	Low ambivalence	High ambivalence	Low ambivalence	High ambivalence
Main Effects
Visibility: Quality Media		0.21		0.44*	–0.07
		(0.11)		(0.18)	(0.15)
Visibility: Tabloid Media		0.00		0.01	0.12
		(0.12)		(0.24)	(0.17)
Interpersonal Contact		2.52***	2.54***	2.74*	2.54***	2.72*	2.50***
		(0.47)	(0.46)	(1.33)	(0.55)	(1.38)	(0.55)
Impersonal Contact		0.62**	0.62**	0.62	0.88**	0.66	0.87**
		(0.20)	(0.20)	(0.38)	(0.28)	(0.36)	(0.29)
Tone: Quality Media			0.19			−0.90	1.11*
			(0.39)			(0.74)	(0.52)
Tone: Tabloid Media			0.17			0.12	0.16
			(0.45)			(0.87)	(0.55)
ÖVP
Campaign Attention	0.49	0.50	0.48	0.19	0.94*	0.13	0.89*
	(0.31)	(0.32)	(0.33)	(0.53)	(0.44)	(0.52)	(0.44)
Political Knowledge	–0.09	–0.18	–0.20	0.94	–1.01*	0.97	–0.99*
	(0.31)	(0.32)	(0.33)	(0.54)	(0.46)	(0.56)	(0.46)
Union Membership	0.15	0.37	0.37	0.77	0.27	0.80	0.32
	(0.48)	(0.47)	(0.47)	(0.71)	(0.59)	(0.81)	(0.60)
Religiosity	–0.03	–0.14	–0.15	0.09	–0.42	0.16	–0.47
	(0.33)	(0.34)	(0.34)	(0.55)	(0.45)	(0.55)	(0.45)
Gender: Females	–0.19	−0.35	−0.36	−0.20	−0.56	−0.09	−0.64
	(0.32)	(0.33)	(0.33)	(0.57)	(0.43)	(0.57)	(0.42)
Age	0.02*	0.03**	0.03*	0.03	0.03	0.03	0.03*
	(0.01)	(0.01)	(0.01)	(0.02)	(0.01)	(0.02)	(0.01)
Qualification for Higher	0.61	0.70*	0.70*	1.11*	0.76	1.14	0.73
Education	(0.33)	(0.34)	(0.34)	(0.55)	(0.44)	(0.58)	(0.44)
Attitudinal	0.05	0.04	0.04
Ambivalence	(0.07)	(0.07)	(0.07)
Economic Evaluation	0.64	0.40	0.42	–1.47	1.51	–1.11	1.40
	(0.66)	(0.67)	(0.68)	(1.08)	(0.98)	(1.06)	(0.99)
Days to Election Day	–0.03*	–0.03*	–0.03*	–0.03	–0.03	–0.03	–0.03
	(0.01)	(0.01)	(0.01)	(0.02)	(0.02)	(0.02)	(0.02)
Party Preference	–0.42	–0.56	–0.59	0.15	–1.28*	0.24	–1.32*
(pre-election)	(0.39)	(0.41)	(0.41)	(0.68)	(0.56)	(0.71)	(0.55)
Constant	–1.26	–0.97	–1.09	–1.12	–0.84	–1.78	–0.81
	(0.70)	(0.75)	(0.74)	(1.13)	(1.02)	(1.09)	(1.01)
FPÖ
Campaign Attention	0.12	0.25	0.19	–0.33	0.86	–0.40	0.74
	(0.37)	(0.39)	(0.38)	(0.75)	(0.47)	(0.73)	(0.47)
Political Knowledge	–0.30	–0.43	–0.41	–1.47	–0.72	–1.36	–0.64
	(0.40)	(0.41)	(0.41)	(0.86)	(0.51)	(0.81)	(0.51)
Union Membership	0.30	0.60	0.56	1.42	0.70	1.19	0.69
	(0.57)	(0.57)	(0.57)	(0.99)	(0.69)	(0.94)	(0.72)
Religiosity	–0.36	–0.47	–0.50	–0.16	–1.03*	–0.13	–1.11*
	(0.40)	(0.41)	(0.41)	(0.90)	(0.46)	(0.85)	(0.46)
Gender: Females	–0.40	–0.57	–0.57	–1.81*	–0.04	–1.73*	–0.11
	(0.38)	(0.40)	(0.40)	(0.82)	(0.46)	(0.77)	(0.46)
Age	0.02	0.03*	0.03*	0.08**	0.01	0.08**	0.01
	(0.01)	(0.01)	(0.01)	(0.03)	(0.02)	(0.03)	(0.02)
Qualification for Higher	–0.19	–0.39	–0.38	–0.07	–0.26	–0.10	–0.19
Education	(0.42)	(0.43)	(0.42)	(0.86)	(0.50)	(0.84)	(0.50)
Attitudinal	0.12	0.09	0.11
Ambivalence	(0.09)	(0.10)	(0.10)
Economic Evaluation	0.49	0.12	0.16	–3.17*	1.91	–2.77	1.67
	(0.96)	(1.05)	(1.06)	(1.59)	(1.20)	(1.57)	(1.26)
Days to Election Day	–0.02	–0.02	–0.02	0.00	–0.03	0.01	–0.04
	(0.02)	(0.02)	(0.02)	(0.03)	(0.02)	(0.03)	(0.02)
Party Preference	0.69	0.60	0.60	3.58***	–1.05	3.63***	–0.98
(pre-election)	(0.43)	(0.47)	(0.47)	(1.05)	(0.61)	(1.00)	(0.62)
Constant	–1.26	–0.91	–1.07	–2.63	–0.52	–3.52*	–0.42
	(0.80)	(0.85)	(0.82)	(1.90)	(1.16)	(1.70)	(1.15)
Greens
Campaign Attention	0.46	0.39	0.32	–0.02	0.69	–0.02	0.55
	(0.33)	(0.34)	(0.35)	(0.54)	(0.44)	(0.53)	(0.44)
Political Knowledge	–0.11	–0.20	–0.21	–0.30	–0.28	–0.22	–0.24
	(0.31)	(0.33)	(0.33)	(0.51)	(0.44)	(0.51)	(0.44)
Union Membership	0.59	0.93	0.89	1.10	0.74	1.00	0.76
	(0.47)	(0.48)	(0.48)	(0.81)	(0.58)	(0.88)	(0.60)
Religiosity	0.01	0.04	–0.00	0.27	–0.23	0.38	–0.44
	(0.33)	(0.34)	(0.35)	(0.54)	(0.45)	(0.55)	(0.46)
Gender: Females	0.02	–0.19	–0.18	0.52	–0.79	0.62	–0.81
	(0.31)	(0.33)	(0.33)	(0.56)	(0.43)	(0.52)	(0.43)
Age	–0.03*	–0.02	–0.02	–0.02	–0.02	–0.02	–0.02
	(0.01)	(0.01)	(0.01)	(0.02)	(0.02)	(0.02)	(0.02)
Qualification for Higher	0.67*	0.69	0.70*	1.56*	0.41	1.54**	0.42
Education	(0.34)	(0.36)	(0.35)	(0.64)	(0.44)	(0.60)	(0.44)
Attitudinal	–0.09	–0.10	–0.07
Ambivalence	(0.06)	(0.07)	(0.07)
Economic Evaluation	0.70	0.45	0.57	–0.67	0.72	–0.13	0.77
	(0.68)	(0.71)	(0.71)	(1.32)	(0.96)	(1.22)	(0.97)
Days to Election Day	–0.01	–0.02	–0.02	–0.04*	0.00	–0.04*	0.01
	(0.01)	(0.01)	(0.01)	(0.02)	(0.02)	(0.02)	(0.02)
Party Preference	–0.62	–0.49	–0.48	–0.11	–0.72	–0.10	–0.69
(pre-election)	(0.40)	(0.45)	(0.45)	(0.93)	(0.48)	(0.91)	(0.48)
Constant	0.66	1.22	0.93	1.72	1.01	0.81	0.65
	(0.70)	(0.81)	(0.75)	(1.31)	(1.08)	(1.12)	(1.02)
N	1648	1648	1648	636	1012	636	1012
LL	–505.04	–465.74	–467.90	–143.19	–286.63	–146.29	–284.31
Chi2	72.98	130.22	130.74	88.33	90.48	83.59	96.59
AIC	1082.08	1011.49	1015.79	360.38	647.26	366.59	642.62

Notes. The effects of the main alternative-specific independent variables are displayed at the top of the table, the individual-specific control variables at the bottom by party. The SPÖ serves as the baseline category. Weights applied. Standard errors in parentheses.

* p < .05. ** p < .01. *** p < .001.