Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Selective Exposure, Information Utility, and the Decision to Watch Televised Debates

Selective Exposure, Information Utility, and the Decision to Watch Televised Debates Abstract Why do people watch televised election debates? This article considers the role of three factors highlighted in recent research on political news exposure: selective approach toward attitude-congruent information; selective avoidance of attitude-incongruent information; and information utility. I test the distinct effects of these three factors on viewing decisions by taking advantage of Austria’s unique organization of televised debates as pairwise match-ups between the main party leaders. I find that selective approach dominates selective avoidance and that information utility increases exposure propensity. Beyond explaining debate viewership, these findings have general implications for understanding exposure to political information, particularly during campaigns. Political information-seeking has long been explained using the theory of selective exposure (Lazarsfeld, Berelson, & Gaudet, 1944). The basic argument is simple: we seek out and engage with sources that provide attitude-congruent information, and we avoid sources that provide attitude-incongruent information. A key motivation for such behavior would be to reduce the potential for cognitive dissonance, that is, to reduce exposure to information that challenges existing opinions and is thus unsettling and uncomfortable (Festinger, 1957; Frey, 1986). The theory of selective exposure remains highly relevant today for at least a couple of reasons. For one, politics is increasingly polarized in many countries such as the United States, giving people greater incentives for selective exposure. Moreover, technological changes leading to the fragmentation of media consumption (Prior, 2007) may make it easier for individuals to act on these incentives (Iyengar, Hahn, Krosnick, & Walker, 2008). More recently, the theory of selective exposure has been revisited and amended. First, it is argued that positive incentives to seek out congruent information are more important than negative incentives to avoid incongruent information. In other words, selective approach is more important than selective avoidance (Garrett, 2009). The consequences of distinguishing between approach and avoidance are significant, as they imply that polarization and technological change may have a less severe impact than previously argued. Second, it has been pointed out that explaining exposure using affect is too limited. Instead, we also need to consider the extent to which information is useful to citizens, for instance in assessing threats and opportunities, formulating opinions, and determining electoral choices (Knobloch-Westerwick, 2014; Knobloch-Westerwick & Kleinman, 2012). This article examines the influence of selective approach and selective avoidance as well as of information utility on decisions to watch televised debates between candidates for high public office. Televised debates have become near-ubiquitous: in addition to the famous presidential debates in the United States, such events are now organized in countries such as the UK, France, Germany, Denmark, and the Netherlands. Individuals who tune in to televised debates are necessarily exposed to more than one viewpoint: in Lang and Lang’s (1977) words, debates create “double exposure”. Televised debates are thus similar to panel discussions, roundtables, and even newscasts in that there are multiple actors and diverse points of view. While still largely predictable in terms of the constellation of opinions, the information setting of televised debates is therefore more complex than, say, a newspaper comment piece expressing a single opinion. The focus of this article is on explaining why people tune in to some debates but not others; in doing so, it tests the distinct impacts of approach, avoidance, and information utility. In other words, the core concern of this article is not on why certain individuals are more likely than others to watch debates or why certain candidates generally lead to high viewing figures (on these questions, see, e.g., Baum & Kernell, 1999; Kenski & Jamieson, 2011; Kenski & Stroud, 2005). Instead, the aim of this article is to study what kinds of debates viewers will tune in to see, using as the key explanatory approach the candidate perceptions that create approach, avoidance, and utility. These explanations are difficult to assess if there is just one debate. However, the analysis in this article takes advantage of a unique observational setting that makes it possible to disentangle the different drivers of exposure behavior. In the run-up to the 2013 parliamentary election in Austria, the top candidates of the six major parties faced off in pairwise debates, leading to a total of 15 debates between the different candidates. Hence, individuals who support one candidate can watch all his or her debates—or they can selectively avoid debates against candidates they dislike. Moreover, an individual who sees two candidates as potential choices in the election can selectively choose to watch the debate where those two candidates square off. As a result, the set-up studied here leads to an intra-individual variety of affect and utility configurations that differs for each pair of candidates, making effective statistical tests possible. Moreover, this set-up allows us to control for individual-level and debate-level confounders that may be associated with affect toward candidates; these include political interest among individuals and political polarization between candidates. In sum, the case used in this article provides an important setting for understanding exposure to multiple-source settings and for testing the distinct impacts of selective approach, selective avoidance, and information utility. This article is structured as follows. First, I summarize the recent debate about selective approach and avoidance and consider information utility as an additional driver of exposure decisions. Then, I discuss how these perspectives can be applied to the decision to view televised debates. In terms of data, viewership of the debates is assessed using a multi-wave online access panel (Kritzinger et al., 2014a). The key results are replicated using a probability-based rolling cross-section survey carried out throughout the period in which debates were held (see Supplementary Appendix 7; Kritzinger et al., 2014b). After presenting key results from these analyses, I conclude by discussing the broader relevance of the findings, also for discussions concerning the impact of political polarization of citizens’ information-seeking behavior. Explaining Exposure to Political Information Recently, it has been argued that selective exposure results from two behavioral regularities, namely, selective approach and selective avoidance (Garrett, 2009, Garrett & Stroud, 2014). Selective approach refers to the tendency to actively seek out information that is likely to bolster preexisting attitudes, while selective avoidance refers to actions designed to prevent exposure to information that contradicts these beliefs. Hence, individuals can avoid cognitive dissonance by seeking information that reinforces their opinions and by avoiding information that challenges them (Garrett, 2009). Crucially, the positive association between affect or attitudes and information-seeking may result from either of both of these behaviors. The empirical occurrence of selective exposure can therefore be the consequence of the fact that individuals seek out reinforcing information, avoid challenging information, or both. Hence, in testing selective exposure, we need to separate out the distinct influence of approach and avoidance (Garrett & Stroud, 2014). Existing evidence indicates that selective approach determines our information-seeking behavior to a greater degree than selective avoidance (Chaffee, Saphir, Graf, Sandvig, & Hahn, 2001; Frey, 1986; Garrett, 2009; Garrett & Stroud, 2014). There are several reasons why individuals may be less likely to avoid attitude-incongruent information than to seek out attitude-congruent material. For instance, the desire to avoid cognitive dissonance does not require avoiding attitude-incongruent information altogether: when faced with such information, individuals can also attempt to seek out additional, attitude-reinforcing information or find arguments that contradict the challenging information (Garrett, 2009; Taber and Lodge, 2006).1 It may be that the strategies of avoiding exposure completely, of arguing against attitude-incongruent information, and of seeking out reinforcing information may all be similarly effective at decreasing dissonance. Moreover, it may be less difficult to engage in countervailing strategies rather than to try to avoid all attitude-incongruent information in the first place (Garrett, 2009). As any amount of exposure to political information is likely to lead to being faced with opinion-challenging information, individuals may be used to other ways of reducing dissonance rather than just simple avoidance. Selective avoidance may also be relatively infrequent because utility is derived from the very fact that information contradicts one’s opinion. As Holbert, Garrett, and Gleason (2010, p. 21) note, “there are times when it will be more valuable to understand an issue fully than to avoid discrepant information, such as when an individual needs to defend a position or to critique the opposition.” It is useful to know what people with an opposing opinion think and how they argue their position and against one’s own view (Garrett, 2009). Counter-attitudinal information may thus allow individuals to better argue their position or dismiss others. Partisans might also seek out information “to see what ‘the enemy’ is up to” (Chaffee, Saphir, Graf, Sandvig & Hahn, 2001, 262). So, information from the other side of the debate may on occasion be actively sought out by individuals instead of being subject to selective avoidance. Selective exposure may not be the only driver of exposure decisions. Another important explanation concerns whether individuals believe they will derive some utility from their exposure to information (Knobloch-Westerwick, 2014; Knobloch-Westerwick & Kleinman, 2012; Stroud, 2011, p. 25). While attitude reinforcement itself of course provides utility (Knobloch-Westerwick & Kleinman, 2012), there are various other ways in which information can be perceived as useful. Knobloch-Westerwick (2008, 2014) highlights how information may help to assess threats (surveillance function). Building on Atkin (1973), Knobloch-Westerwick and Kleinman (2012) also point out that information can aid us in determining “what to feel” (guidance function) and “what to do” (performance function). Hence, information may help us to know what to think about the parties and candidates competing and how we should vote in the election (Knobloch-Westerwick & Kleinman, 2012).2 In sum, selective exposure based on individual affect and attitudes remains an important approach to understanding how people select which political news and information to consume. However, within that paradigm, seeking out exposure to attitude-congruent information (selective approach) may be a stronger motivator than avoiding exposure to attitude-incongruent information (selective avoidance), as the latter type of information may in reality not be unattractive to individuals. Moreover, information utility may be a further important driver of exposure decisions. Selective Exposure to Televised Debates? These insights into selective exposure can be applied to the decision whether or not to watch televised debates between candidates. The key characteristic of debates that is important for how we study the effects of incentives for selective exposure is the fact that they involve “double exposure” (Lang & Lang, 1977, p. 277): by deciding whether or not to watch a debate, individuals have to choose between exposing themselves to the personality and arguments of both candidates or to those of neither candidate. Selective exposure to just one of the two debaters is not possible, so individuals who watch a debate may have to see a disliked debater and hear unwelcome arguments. To some extent, debates therefore cannot give full reign to individuals’ desire to pick and choose the information they see and hear. As argued below, the Austrian setting of multiple debate match-ups provided individuals with more freedom in this regard, making it possible to determine the drives of exposure to televised debates. In general, studying televised debates can give us an insight into the relative weight of the two affect-based incentives in determining how voters seek out information. Do voters who like one candidate and dislike the other still tune in, so does selective approach dominate selective avoidance? Or do such voters instead choose not to expose themselves to counter-attitudinal information? Televised debates can tell us how citizens deal with situations where multiple exposure is likely. Such situations are not unusual, as much political information—such as roundtables or newscasts—provides two or more sides to every argument and presents a somewhat balanced view of events and opinions. Despite the fear that the advent of new technologies would increase tendencies toward the creation of “echo chambers,” it is close to impossible to follow politics closely and still avoid attitude-incongruent information. Applied to televised debates, the theory of selective exposure generates two hypotheses. First, the likelihood of watching a debate should increase as affect toward the debaters increases, while low levels of affect should decrease viewing probabilities. If both debaters are strongly liked, then this should reinforce viewing probabilities. Hence, the first hypothesis is: H1: The more positive their affect toward the debaters, the more likely individuals are to watch the debate; the more negative their affect toward the debaters, the less likely individuals are to watch the debate. However, as debates create multiple exposure, viewers may be faced with a situation where they feel positive affect toward one debater and negative affect toward the other. Here, the traditional perspective would predict that selective approach and avoidance would cancel each other out. Indeed, proponents of the traditional perspective might even expect viewership among this group to be low, as individuals might prefer to seek out other sources of information where they are not confronted with attitude-incongruent information. In contrast, proponents of the more recent perspective that emphasizes selective approach over avoidance (Garrett, 2009; Garrett & Stroud, 2014) would argue that affect congruence outweighs affect incongruence as an incentive to watch debates. Hence, if there is high affect toward one debater and low affect toward the other, viewership should still be high, indeed possibly as high as when both debaters are well-liked. Hence, the second hypothesis is: H2: The level of affect toward the less-liked debater has no effect on the decision to watch a televised debate. The third hypothesis concerns information utility. Here, we expect individuals to tune in to televised debates if this may help them decide whom to vote for. Hence, voters will seek out information that fulfills the guidance and performance functions of information (Atkin, 1973; Knobloch-Westerwick & Kleinman, 2012). Debates will be particularly useful to voters who are hesitating between the two leaders or their parties. For such voters, the debate may contain important information on whom they should choose. Hence, the third hypothesis is: H3: Individuals are more likely to watch debates if they see both candidates as potential electoral choices than if they see only one or no candidate as a potential choice. Data and Methods The focus in this article is on understanding why individuals watch some debates but not others. This question is studied using the case of Austria in 2013: before the parliamentary elections, there were 15 debates where leaders of each of the six parties represented in parliament debated with every other party leader. As a result, there were naturally varying configurations of affect and support for the two debaters among voters, allowing us to run models that include affect and information utility as separate predictors. For example, it is possible to distinguish between voters who feel positively about both candidates and those who would consider voting for both of them. In settings such as the United States, where opinions about the two candidates tend to be strongly and negatively correlated, it is difficult to parse out the distinct effects of candidate affect and information utility. Another advantage of the data is that the data allow us to use individual- and debate-level fixed effects to consider within-respondent variation in viewing decisions, thus providing a particularly stringent test of the hypotheses. The debates were broadcast each Tuesday and Thursday in the last four weeks of the campaign, with two debates shown each evening.3 The leaders of the six main parties took part in the debates: Chancellor Werner Faymann of the Social Democratic SPÖ, Vice-Chancellor Michael Spindelegger of the Christian Democratic ÖVP, Heinz-Christian Strache of the radical-right FPÖ, Eva Glawischnig of the Greens, Josef Bucher of the radical-right BZÖ (a splinter party of the FPÖ), and Frank Stronach of the Team Stronach. At the time, the Austrian party system was therefore made up of two mainstream parties (SPÖ/ÖVP), a Green party, one large (FPÖ) and one small (BZÖ) radical-right party, and one new, personalized party (Team Stronach). The governing coalition was made up of the SPÖ and the ÖVP; such a grand coalition between the two mainstream parties is typical of Austrian politics. However, the two parties were challenged at the polls mainly by the resurgent radical-right FPÖ and, to a lesser extent, by the new party founded by the billionaire Frank Stronach. Both leaders had significant personal appeal, at least to their followers, though Figure 1 shows that, on average, they were less popular than their rivals. In the end, however, the SPÖ and the ÖVP continued to hold enough seats to carry on in coalition despite both losing votes overall (Dolezal and Zeglovits, 2014). This outcome was foreshadowed in the polls during the election campaign. Figure 1 Open in new tabDownload slide Histograms of like–dislike scores for the six debaters. Note. Data from the probability-based rolling cross-section telephone survey (Kritzinger et al., 2014b); responses not weighted Figure 1 Open in new tabDownload slide Histograms of like–dislike scores for the six debaters. Note. Data from the probability-based rolling cross-section telephone survey (Kritzinger et al., 2014b); responses not weighted Of the 15 debates, the highest viewing figures were for the debate between the Chancellor Werner Faymann and the leader of the radical-right Freedom Party, Heinz-Christian Strache. This debate reached almost 1.1 million people out of a total Austrian population of 8.5 million. The least-watched debate, that between the Strache and his radical-right rival Josef Bucher, nevertheless reached over 650,000 people. A complete list of the debates and their respective viewing figures can be found in Supplementary Appendix 1. All in all, while the debates were thus a central part of the last month of the election campaign, there is also clear variation in the extent to which Austrians tuned in to each debate. Concurrently with the campaign, the Austrian National Election Study ran a four-wave panel study (Kritzinger et al., 2014a). This survey was carried out online using an access panel, from which TNS drew a sample of respondents that was representative of the population on key characteristics. For the analyses, I make use of data from the first and third waves; these were carried out on 16–28 August 2013 (Wave 1) and on 26–29 September 2013 (Wave 3). In all, 3,085 respondents participated in Wave 1 and 1,318 in Wave 3. The supplemental analyses in Supplementary Appendix 4 also make use of data from Wave 2 (9–12 September), which was completed by 2,044 respondents. One particular useful feature of this multiwave panel is that the independent variables can be measured before the debates occurred, thus eliminating the possibility that the debates themselves influenced the variables of interest (such as candidate sympathy). Given the obvious concerns that data from an online access panel are not a probability sample, the key results are replicated in Figure 3 and Supplementary Appendix 7 using data from a rolling cross-section telephone survey (Kritzinger et al., 2014b; on the design of rolling cross-section studies, see Johnston & Brady, 2002). This survey is not used for the main analyses because it asks about debate viewership in a more limited fashion and measures the key covariates after exposure to the debates. The dependent variable is whether respondents say they watched a particular debate or not. Respondents were presented with a list of all debates and were asked to select those debates they had seen (complete question wording in Supplementary Appendix 2). In total, 1,421 respondents answered this question; as I stack the data by debate, this results in a total of 21,315 responses overall. Responses are coded as “watched” for respondents who said they had seen the debate, with all other responses (including don’t knows and refusals) classed as “not watched”. Unsurprisingly, the proportion of respondents who said they viewed the debate is markedly higher than the proportion of actual viewers as recorded by the broadcasters. Partly, this will be because people caught up on the debates in other news programs or only watched part of the debate, but it might also be a result of misremembering and social desirability (Prior, 2012) as well as of biased sampling. To address this in part, Supplementary Appendix 4 presents additional analyses that use a stricter measurement of debate viewership, where I take advantage of the fact that the respondents were asked twice whether they had seen each debate. In these analyses, the respondents are only coded as having viewed a debate if they gave consistent responses in both Wave 2 and Wave 3. Measures of candidate sympathy are needed to assess the extent to which selective exposure affects the decision to watch a debate (H1 and H2). I use two variables here.4 First, I include one variable that measures the sympathy score for the more liked of the two debaters. This sympathy is measured using the 0–10 like–dislike scale for each of the six party leaders.5 The greater the sympathy for the more-liked candidate, the more an individual should want to watch a debate. Second, I include as a second predictor the sympathy score of the less-liked debater.6 To test H3, I need a measure of how likely the respondent is to vote for the party of the debater. I therefore include a measure the respondent’s self-declared probability to ever vote for the candidate’s party, provided on a 0–10 scale. This captures the extent to which the citizen would consider voting for that party; while related to affect ratings, this is nevertheless conceptually distinct. Again, this measure is included as two variables, one for the higher and one for the lower of the two values. I also present models that include several other variables that vary by individual and candidate. First, I include the extent to which the respondent finds each candidate charismatic, measured on a 1–4 scale. This captures the attractiveness of a candidate beyond positive or negative affect. Second, I include a dummy variable for whether the respondent identifies with the party of either candidate debating (1 = yes). This captures the extent to which group loyalties may drive viewership decisions. Finally, I include a variable measuring whether the citizen expects the candidates to do well in the debate. This could capture the extent to which a debate might be seen as entertaining. Obviously all these variables correlate with affect ratings and propensity-to-vote scores to a certain degree, creating a tough test for the hypotheses.7 I run a series of linear probability models, with debate viewing as the dependent variable. Because the data are stacked, with 15 cases per respondent, I use fixed effects for each respondent, as this captures individual-specific differences in the propensity to view debates. (This also means that it is not possible to include any additional predictors that do not vary by individual, such as political interest, age, education, or gender.) In the results section, I also present multilevel regression models with a set of key control variables; these models produce substantively similar results for key covariates as those in the fixed-effects models. It is the use of fixed effects that also requires the use of linear probability models rather than binary logistic regression: there are naturally a number of respondents who did not watch any debates and would thus be dropped in fixed-effects logistic regression analyses. All models also include fixed effects for each debate. This captures factors such as the effect of viability on interest in candidates (Utych & Kam, 2014), idiosyncratic effects linked to the two debaters, or simply other circumstantial aspects such as the weather or competing TV programming. Given that fixed effects for each debate are included, it is neither possible nor necessary to include further debate-level factors (such as the average popularity of debaters or their policy positions) in the models. In Supplementary Appendix 5, I present results disaggregated by debate; overall, the results are stable across debates, with moderate variation in the magnitude of effects. Results Table 1 presents the core models, all of which include fixed effects for all respondents and all debates. Model 1 presents an “empty” model without any predictors apart from the fixed effects. Model 2 then includes the four main predictors: the sympathy score for the more well-liked and that for the less-liked of the two debaters, and the probability to vote for the debaters’ parties. Model 3 includes interactions between the variables to capture differences in the effect of affect for one debater according to affect for the other debater and in the effect of voting probability for one party according to voting probability for the other debater. For instance, a three-unit sympathy score for the less-liked debater may have a different impact if the score for the more-liked debater is three or if it is ten. Models 4 and 5 repeat these Models, but include the series of controls described above.8 Table 1 OLS fixed-effects regression results Predictors . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Sympathy, more-liked candidate 0.00844*** (0.001) 0.00805*** (0.001) 0.00716*** (0.002) 0.00687*** (0.002) Sympathy, less-liked candidate 0.000327 (0.002) −0.00511 (0.005) −0.00110 (0.002) −0.00443 (0.005) Interaction: Sympathy scores 0.000735 (0.001) 0.000460 (0.001) Propensity to vote for party of candidate, higher value 0.00378*** (0.001) 0.00341** (0.001) 0.00278* (0.001) 0.00236 (0.001) Propensity to vote for party of candidate, lower value 0.00440** (0.002) −0.00420 (0.005) 0.00527** (0.002) −0.00540 (0.006) Interaction: Propensities to vote 0.00103 (0.001) 0.00128 (0.001) Party ID for one candidate 0.0223* (0.009) 0.0207* (0.009) Expected performance, higher value −0.00679 (0.008) −0.00509 (0.008) Expected performance, lower value −0.0273 (0.016) −0.0237 (0.016) Interaction: Expected performances 0.00448 (0.003) 0.00360 (0.004) Charisma of candidate, higher value −0.0182 (0.012) −0.0151 (0.012) Charisma of candidate, lower value −0.0266 (0.021) −0.0206 (0.021) Interaction: charisma scores 0.00777 (0.006) 0.00602 (0.006) Constant 0.335*** (0.008) 0.270*** (0.010) 0.276*** (0.011) 0.404*** (0.050) 0.394*** (0.052) Observations 21,315 19,965 19,965 17,402 17,402 R2 0.53 0.53 0.53 0.53 0.53 Adjusted R2 0.49 0.50 0.50 0.49 0.49 AIC 9,462 9,030 9,027 8,762 8,760 BIC 9,582 9,180 9,193 8,964 8,978 Predictors . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Sympathy, more-liked candidate 0.00844*** (0.001) 0.00805*** (0.001) 0.00716*** (0.002) 0.00687*** (0.002) Sympathy, less-liked candidate 0.000327 (0.002) −0.00511 (0.005) −0.00110 (0.002) −0.00443 (0.005) Interaction: Sympathy scores 0.000735 (0.001) 0.000460 (0.001) Propensity to vote for party of candidate, higher value 0.00378*** (0.001) 0.00341** (0.001) 0.00278* (0.001) 0.00236 (0.001) Propensity to vote for party of candidate, lower value 0.00440** (0.002) −0.00420 (0.005) 0.00527** (0.002) −0.00540 (0.006) Interaction: Propensities to vote 0.00103 (0.001) 0.00128 (0.001) Party ID for one candidate 0.0223* (0.009) 0.0207* (0.009) Expected performance, higher value −0.00679 (0.008) −0.00509 (0.008) Expected performance, lower value −0.0273 (0.016) −0.0237 (0.016) Interaction: Expected performances 0.00448 (0.003) 0.00360 (0.004) Charisma of candidate, higher value −0.0182 (0.012) −0.0151 (0.012) Charisma of candidate, lower value −0.0266 (0.021) −0.0206 (0.021) Interaction: charisma scores 0.00777 (0.006) 0.00602 (0.006) Constant 0.335*** (0.008) 0.270*** (0.010) 0.276*** (0.011) 0.404*** (0.050) 0.394*** (0.052) Observations 21,315 19,965 19,965 17,402 17,402 R2 0.53 0.53 0.53 0.53 0.53 Adjusted R2 0.49 0.50 0.50 0.49 0.49 AIC 9,462 9,030 9,027 8,762 8,760 BIC 9,582 9,180 9,193 8,964 8,978 Note. Standard errors in parentheses, fixed effects for debates and individuals included. * p < .05, **p < .01, ***p < .001. Table 1 OLS fixed-effects regression results Predictors . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Sympathy, more-liked candidate 0.00844*** (0.001) 0.00805*** (0.001) 0.00716*** (0.002) 0.00687*** (0.002) Sympathy, less-liked candidate 0.000327 (0.002) −0.00511 (0.005) −0.00110 (0.002) −0.00443 (0.005) Interaction: Sympathy scores 0.000735 (0.001) 0.000460 (0.001) Propensity to vote for party of candidate, higher value 0.00378*** (0.001) 0.00341** (0.001) 0.00278* (0.001) 0.00236 (0.001) Propensity to vote for party of candidate, lower value 0.00440** (0.002) −0.00420 (0.005) 0.00527** (0.002) −0.00540 (0.006) Interaction: Propensities to vote 0.00103 (0.001) 0.00128 (0.001) Party ID for one candidate 0.0223* (0.009) 0.0207* (0.009) Expected performance, higher value −0.00679 (0.008) −0.00509 (0.008) Expected performance, lower value −0.0273 (0.016) −0.0237 (0.016) Interaction: Expected performances 0.00448 (0.003) 0.00360 (0.004) Charisma of candidate, higher value −0.0182 (0.012) −0.0151 (0.012) Charisma of candidate, lower value −0.0266 (0.021) −0.0206 (0.021) Interaction: charisma scores 0.00777 (0.006) 0.00602 (0.006) Constant 0.335*** (0.008) 0.270*** (0.010) 0.276*** (0.011) 0.404*** (0.050) 0.394*** (0.052) Observations 21,315 19,965 19,965 17,402 17,402 R2 0.53 0.53 0.53 0.53 0.53 Adjusted R2 0.49 0.50 0.50 0.49 0.49 AIC 9,462 9,030 9,027 8,762 8,760 BIC 9,582 9,180 9,193 8,964 8,978 Predictors . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Sympathy, more-liked candidate 0.00844*** (0.001) 0.00805*** (0.001) 0.00716*** (0.002) 0.00687*** (0.002) Sympathy, less-liked candidate 0.000327 (0.002) −0.00511 (0.005) −0.00110 (0.002) −0.00443 (0.005) Interaction: Sympathy scores 0.000735 (0.001) 0.000460 (0.001) Propensity to vote for party of candidate, higher value 0.00378*** (0.001) 0.00341** (0.001) 0.00278* (0.001) 0.00236 (0.001) Propensity to vote for party of candidate, lower value 0.00440** (0.002) −0.00420 (0.005) 0.00527** (0.002) −0.00540 (0.006) Interaction: Propensities to vote 0.00103 (0.001) 0.00128 (0.001) Party ID for one candidate 0.0223* (0.009) 0.0207* (0.009) Expected performance, higher value −0.00679 (0.008) −0.00509 (0.008) Expected performance, lower value −0.0273 (0.016) −0.0237 (0.016) Interaction: Expected performances 0.00448 (0.003) 0.00360 (0.004) Charisma of candidate, higher value −0.0182 (0.012) −0.0151 (0.012) Charisma of candidate, lower value −0.0266 (0.021) −0.0206 (0.021) Interaction: charisma scores 0.00777 (0.006) 0.00602 (0.006) Constant 0.335*** (0.008) 0.270*** (0.010) 0.276*** (0.011) 0.404*** (0.050) 0.394*** (0.052) Observations 21,315 19,965 19,965 17,402 17,402 R2 0.53 0.53 0.53 0.53 0.53 Adjusted R2 0.49 0.50 0.50 0.49 0.49 AIC 9,462 9,030 9,027 8,762 8,760 BIC 9,582 9,180 9,193 8,964 8,978 Note. Standard errors in parentheses, fixed effects for debates and individuals included. * p < .05, **p < .01, ***p < .001. The first two hypotheses relate to the impact of candidate affect on watching a debate. In Model 2, only the sympathy score for the more liked-debater is a statistically significant predictor: the more this debater is liked, the more likely it is that a voter will tune in. This fits with the selective approach theory. Note that the sympathy score of the less-liked debater also has a much smaller effect on viewing decisions than the sympathy score of the more-liked debater. In Model 4, the effect of sympathy scores for the less-liked debater is indeed slightly negative (but not statistically significant). Moreover, the interaction terms in Models 3 and 5 are not statistically significant at the 0.05 level, providing a first indication that sympathy for the less-liked debater does not vary in impact depending on sympathy for the more-liked debater. To interpret the effects of sympathy toward debaters more precisely, I present the results graphically (as recommended by Brambor, Clark, & Golder, 2006). The left panel of Figure 2 shows the predicted probability of watching the debate for varying levels of sympathy for the two candidates; scores for the more-liked debater are on the x-axis, scores for the less-liked debater on the y-axis. This Figure is based on Model 5, as this most closely relates to the hypotheses to be tested. Each cell contains the predicted probability of watching the debate; higher probabilities have darker shades of gray. For instance, the predicted probability of watching the debate is 26% if the sympathy score for both debaters is 0, 28% if both scores are at 5, and 33% if both scores are at 10. Figure 2 Open in new tabDownload slide Predicted probability of watching a debate conditional on candidate affect. Note. Probabilities calculated based on Model 5, Table 1. In the left panel, the x-axis represents the sympathy scores for the more-liked debater, the y-axis sympathy scores for the less-liked debater; higher probabilities in darker shades of gray. In the right panel, the x-axis represents the vote intention (propensity to vote) scores for the more-preferred debater, the y-axis the vote intention (propensity to vote) for the less-preferred debater; higher probabilities in darker shades of gray Figure 2 Open in new tabDownload slide Predicted probability of watching a debate conditional on candidate affect. Note. Probabilities calculated based on Model 5, Table 1. In the left panel, the x-axis represents the sympathy scores for the more-liked debater, the y-axis sympathy scores for the less-liked debater; higher probabilities in darker shades of gray. In the right panel, the x-axis represents the vote intention (propensity to vote) scores for the more-preferred debater, the y-axis the vote intention (propensity to vote) for the less-preferred debater; higher probabilities in darker shades of gray By looking at the changes along the x-axis, we can see how sympathy for the more-liked debater increases viewing probabilities. If the less-liked debater has a sympathy score of 0 (lowest row), then shifting from 0 to 10 for the more liked-debater increases viewing probabilities by 7% (from 26 to 33%). Similar patterns hold for all rows: increasing affect to the more-liked debater increases viewing probabilities. In contrast, the sympathy scores for the less-liked debater matter less. We can see this by looking at the columns, so by moving up and down the y-axis. Here, we can see that predicted viewing probabilities are mostly stable. For instance, if one debater is well-liked (sympathy score of 9), it does not matter a great deal what individuals think of the other debater: in the columns on the right, all predictions are very similar, at around 32%. In other words, the sympathy individuals have for the less-liked debater has little impact on decisions whether or not to watch the debate; only the sympathy scores for the more-liked debater have an appreciable effect. Importantly, the predicted probability to watch a debate if both candidates are disliked is not zero, as many other factors—opinions about the candidates and individual-level characteristics—also drive this decision. Nevertheless, the fact that 26% who dislike both candidates are predicted to tune in does somewhat go against stronger version of the selective exposure theory. Interestingly, individuals who strongly favor one candidate and strongly dislike the other have the same viewing probability as those who favor both candidates: 33%. Overall, there is therefore strong evidence in favor of H2, so the results indicate that a theory based on selective approach applies better than a theory—reflected in the simpler H1—that places equal weight on selective avoidance. Turning to H3, the respondents’ self-reported likelihood of ever voting for a party yields interesting results. First, the regression models in Table 1 indicate that the scores for both debaters are statistically significant in Models 2 and 4. If an interaction effect is included, this is significant at the 0.1 level in Models 3 and 5. Hence, it appears that the propensity to vote for the debater’s party has an impact for both debaters. To understand this effect, graphical illustration is again useful. Figure 2 therefore presents the effects of vote intention on the probability of tuning in to the debates (right panel). Calculating predicted viewing probabilities indicates that the highest probability of viewing a debate is for candidates from parties that the respondent could both imagine voting for (two scores of ten: 37%). If the respondent can imagine voting for just one of the two parties, the probability of tuning in declines (one score of ten, one of zero: 30%). The probability is lower again if the respondent cannot imagine voting for either candidate’s party (two scores of zero: 27%). Unlike for the affect scores in the left panel, it matters whether individuals might vote for both parties or are only considering voting for one of them. Hence, there is evidence in favor of individuals choosing to watch debates that are particularly useful in terms of ultimately deciding whom to vote for. Information utility in terms of guidance and performance functions may therefore be an important driver of TV debate viewing, in addition to selective approach. Overall, the models explain about 50% in the variation in the decision to view the televised debates. While the reduction in the explained variance created by the additional variables is low, the AIC/BIC values do decline by large amounts, indicating a superior fit of the larger models to the data; this applies in particular to Models 2 and 4. However, note that the interaction terms between the sympathy scores and probability-to-vote scores add little to model fit in terms of AIC/BIC, providing an indication that simple patterns of association dominate. As a robustness check, Figure 3 presents the results for two alternative model specifications; the full models can be found in Supplementary Appendices 6 and 7. The first alternative specification uses random- rather than fixed-effects at the individual level, along with a suite of additional controls: age, education, general levels of TV viewing, turnout propensity, political interest, whether the respondent sees televised debates as entertaining, and whether the respondent has already decided whom to vote for. The key patterns from Figure 2 are replicated for both sympathy scores and vote intentions. Figure 3 Open in new tabDownload slide Predicted probability of watching a debate conditional on candidate affect and vote intention. Note. Probabilities calculated based on the models in Supplementary Table A6.1 and Supplementary Table A7.1. Results for the rolling cross-section survey based on a logistic regression with standard errors clustered by individual. See Supplementary Appendices 6 and 7 for full results Figure 3 Open in new tabDownload slide Predicted probability of watching a debate conditional on candidate affect and vote intention. Note. Probabilities calculated based on the models in Supplementary Table A6.1 and Supplementary Table A7.1. Results for the rolling cross-section survey based on a logistic regression with standard errors clustered by individual. See Supplementary Appendices 6 and 7 for full results The second key robustness check uses the telephone-based rolling cross-section survey mentioned above; this only asked respondents about the two most recent debates, so these data have the drawback that we cannot include individual-level fixed effects. The response is modeled using logistic regression with a small set of controls and standard errors clustered by respondent. As the rightmost panel in Figure 3 shows, the key patterns are again replicated: a strong increase in viewing probabilities in the rows, comparably stable probabilities in the columns. An interesting nuance in the rolling cross-section results is that there is a stronger reinforcing effect: viewing probabilities are much higher if individuals feel positively about both candidates rather than just one; however, strong positive affect toward one debater remains a key determinant of viewing decisions, so H2 is confirmed here as well. These data do not allow us to control for information utility. Because these additional models do not include individual-level fixed effects, we can also include a series of controls that show interesting results. We thus find that respondents who are interested in politics, who watch TV frequently, who find debates entertaining, and who are middle-aged are most likely to tune in to watch the debates. Interestingly, self-declared undecided voters are in fact less likely to watch the debates, perhaps because such voters are generally less interested in politics. For more detail on these findings, see Supplementary Appendices 6 and 7. Conclusions Two key findings emerge from this study, which used an advantageous setting to test the distinct influences of approach, avoidance, and information utility on watching televised debates. First, the decision to watch televised debates is determined more by selective approach than by selective avoidance. It matters more whether one candidate is well-liked by a potential viewer than whether the other debater is liked or disliked. People seek out attitude-congruent information more than they shy away from attitude-incongruent material. This pattern was broadly constant across debates, though future research should consider how aspects of debates can influence the impact of selective approach on viewing decisions. Second, the decision to watch televised debates is related to information utility. Specifically, this article found that people are more likely to watch a debate between two candidates from parties they would consider voting for. Hence, debates can fulfill the guidance and performance functions of information (Knobloch-Westerwick & Kleinman, 2012): they can help individuals to determine what they feel about candidates and whom they should vote for. This finding provides further evidence that information utility is an important factor in understanding exposure decisions. There are of course limitations to this study. While televised debates, and in particular the paired match-ups in Austria, provide a useful setting to examine incentives toward selective exposure, it is important to restate that debates also have the rather unique characteristic of creating “double exposure” to political information. This feature of debates is similar to the tendency toward balance on panel shows and newscasts. (In contrast, the ideal-type single-exposure information settings may in reality be quite rare and atypical.) The results in this study should therefore be applicable to media events or televised panel debates that feature more than one partisan point of view in a stable and predictable manner. However, it is important for future research to examine less fixed types of media coverage and political information and how these differences affect the behavior of individuals. Future studies should also consider further drivers of exposure decisions. For one, information utility is a complex concept with at least four components (Knobloch-Westerwick & Kleinman, 2012), and only two—guidance and performance—were applied in this study. It would also be worthwhile to distinguish between the type of information utility empirically examined in this study and the information utility of attitude-incongruent information, which I have only discussed theoretically. Moreover, researchers should consider in more depth explanations related to political interest and entertainment, two factors this study has neglected because they vary at the individual and not the debate level. While this article focused on why people tune in to a specific debate (compared with other debates), future research could focus on factors that explain why individuals watch debates. Finally, these results have important broader implications for understanding the impact of political polarization on exposure to political information. This study finds that there is no effect of polarization per se on exposure decisions. Instead, only affect toward the more-liked candidate drives exposure: high sympathy scores encourage individuals to seek out and engage with information from that candidate, even if attitude-incongruent information is present. The results of this study therefore paint a positive picture of the influence of sympathy toward candidates. Moreover, people are more likely to watch debates if this will help them form electoral decisions, providing weight to the information utility approach. Finally, individuals do not completely tune out from debates among candidates they dislike. Overall, the extent of engagement with political information should therefore be related more with how much people support candidates and parties and less with how polarized their opinions are. As long as political information is generally not provided in one-sided settings, strong affect and also polarization should therefore lead to greater levels of political information-seeking. These insights are important for understanding how and when citizens use the media to find out and inform themselves about political events. Acknowledgement I would like to thank Hajo Boomgaarden and Sylvia Kritzinger for comments on an earlier draft of this article. Data collection was carried out by the Austrian National Election Study (AUTNES), a National Research Network (NFN) sponsored by the Austrian Science Fund (FWF) (S10902-G11). Supplementary Data Supplementary Data are available at IJPOR online. Markus Wagner is an Associate Professor in the Department of Government at the University of Vienna. His research focuses on the role of issues and ideology in party competition and vote choice. His research has been published in Comparative Political Studies, the British Journal of Political Science and the European Journal of Political Research. Footnotes 1 " Some authors also argue that the challenge of cognitive dissonance may be overstated. For example, highly opinionated individuals secure in their views may be faced with few costs in terms of potential cognitive dissonance (Holbert, Garrett, & Gleason, 2010). 2 " Selective exposure and information utility are not the only possible drivers of exposure. Some people may simply be interested in politics and public affairs, and all kinds of information may serve to satisfy that curiosity (Sears, 1968) and allow them to engage with politically involved peers (Chaffee, Saphir, Graf, Sandvig & Hahn, 2001, p. 262). Indeed, for some people, all kinds of information may have entertainment value, so people will read about politics because of the intrinsic benefits this creates (Knobloch-Westerwick & Kleinman, 2012). 3 " There were two exceptions: One set of debates was held on a Monday (9 September) instead of a Tuesday; and the debate between the party leaders of the two government parties was the only one that evening. 4 " Supplementary Appendix 9 presents an alternative way of coding candidate affect using sums and distances of the two leaders’ scores, which is simply an arithmetic transformation of the coefficients presented in Table 1. The key conclusions concerning approach and avoidance remain the same. 5 " Supplementary Appendix 8 presents results if we exclude respondents with only moderate levels of affect for both debaters (as recommended by Feldman, Stroud, Bimber, & Wojcieszak, 2013). The conclusions concerning approach and avoidance are the same. 6 " One concern may be that affect toward parties is more important than affect toward candidates. Unfortunately, the online panel survey does not contain measures of affect toward parties. The telephone-based rolling cross-section does, however. Here, the correlation between the two measures of affect is generally around 0.75, and substituting party affect for candidate affect leads to substantively similar results (see Supplementary Appendix 7). 7 " The various variables in the model are related assessments of candidates. The highest bivariate correlation between these is 0.8, the lowest 0.26, with most in the range of 0.5 and 0.65. The highest bivariate correlations are between affect toward candidates and their perceived charisma (average correlation: 0.65), with the average correlation between propensity-to-vote scores and candidate affect at 0.64. In the presence of a large amount of interrelated predictors, multicollinearity may be a concern. However, collinearity diagnostics for the predictors in Model 4 (Table 1) indicate that no variable has a higher variance-inflation factor than 3. 8 " As there may be nonlinear effects for sympathy scores, Supplementary Appendix 3 presents results that include a quadratic term for the sympathy score for the more well-liked of the two debaters. Specifically, there may be particularly strong effects for those with high levels of affect, who will be highly motivated to seek out attitude-congruent information (Taber & Lodge, 2006). The results do not alter the conclusions in this article. References Atkin C. K. ( 1973 ). Instrumental utilities and information seeking. In Clark P. (Ed.), New models of communication research (pp. 205 – 242 ). Newbury Park, CA : SAGE . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Baum M. A. , Kernell S. ( 1999 ). Has cable ended the golden age of presidential television? American Political Science Review , 93 , 99 – 114 . Google Scholar Crossref Search ADS WorldCat Brambor T. , Clark W. R., Golder M. ( 2006 ). Understanding interaction models: Improving empirical analyses . Political Analysis , 14 , 63 – 82 . Google Scholar Crossref Search ADS WorldCat Chaffee S. , Saphir M. N., Graf J., Sandvig C., Hahn K. S. ( 2001 ). Attention to counter-attitudinal messages in a state election campaign . Political Communication , 18 , 247 – 272 . Google Scholar Crossref Search ADS WorldCat Dolezal M. , Zeglovits E. ( 2014 ). Almost an earthquake: The Austrian parliamentary election of 2013 . West European Politics , 37 , 644 – 652 . Google Scholar Crossref Search ADS PubMed WorldCat Feldman L. , Stroud N. J., Bimber B., Wojcieszak M. ( 2013 ). Assessing selective exposure in experiments: The implications of different methodological choices . Communication Methods and Measures , 7 , 172 – 194 . Google Scholar Crossref Search ADS WorldCat Festinger L. ( 1957 ). A theory of cognitive dissonance . Stanford, CA : Stanford University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Frey D. ( 1986 ). Recent research on selective exposure to information. In Berkowitz L. (Ed.), Advances in Experimental Social Psychology, 19, 41–80. Garrett R. K. ( 2009 ). Politically motivated reinforcement seeking: Reframing the selective exposure debate . Journal of Communication , 59 , 676 – 699 . Google Scholar Crossref Search ADS WorldCat Garrett R. K. , Stroud N. J. ( 2014 ). Partisan paths to exposure diversity: Differences in pro‐ and counterattitudinal news consumption . Journal of Communication , 64 , 680 – 701 . Google Scholar Crossref Search ADS WorldCat Holbert L. , Garrett K., Gleason L. ( 2010 ). A new era of minimal effects? A response to Bennett and Iyengar . Journal of Communication , 60 , 15 – 34 . Google Scholar Crossref Search ADS WorldCat Iyengar S. , Hahn K. S., Krosnick J. A., Walker J. ( 2008 ). Selective exposure to campaign communication: The role of anticipated agreement and issue public membership . Journal of Politics , 70 , 186 – 200 . Google Scholar Crossref Search ADS WorldCat Johnston R. , Brady H. E. ( 2002 ). The rolling cross-section design . Electoral Studies , 21 , 283 – 295 . Google Scholar Crossref Search ADS WorldCat Kenski K. , Jamieson K. H. ( 2011 ). Presidential and vice presidential debates in 2008: A profile of audience composition . American Behavioral Scientist , 55 , 307 – 324 . Google Scholar Crossref Search ADS WorldCat Kenski K. , Stroud N. J. ( 2005 ). Who watches presidential debates? A comparative analysis of presidential debate viewing in 2000 and 2004 . American Behavioral Scientist , 49 , 213 – 228 . Google Scholar Crossref Search ADS WorldCat Knobloch-Westerwick S. ( 2008 ). Informational utility. In Donsbach W. (Ed.), International encyclopedia of communication (pp. 2273 – 2276 ). Oxford : Basil Blackwell . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Knobloch-Westerwick S. ( 2014 ). Choice and preference in media use: Advances in selective exposure theory and research . Abingdon : Routledge . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Knobloch-Westerwick S. , Kleinman S. B. ( 2012 ) Preelection selective exposure: Confirmation bias versus information utility . Communication Research , 39 , 170 – 193 . Google Scholar Crossref Search ADS WorldCat Kritzinger S. , Johann D., Glantschnigg C., Aichholzer J., Glinitzer K., Thomas K., Wagner M., Zeglovits E. ( 2014a ): AUTNES TV Debates Panel Study 2013. GESIS Data Archive, Cologne. ZA5858 Data file Version 1.0.0. doi:10.4232/1.11951 Kritzinger S. , Johann D., Aichholzer J., Glinitzer K., Glantschnigg C., Thomas K., Wagner M., Zeglovits E. ( 2014b ): AUTNES Rolling-Cross-Section Panel Study 2013. GESIS Data Archive, Cologne. ZA5857 Data file Version 1.0.0. doi:10.4232/1.11950 Lang K. , Lang G. E. ( 1977 ). Reactions from viewers. In Kraus S. (Ed.), The great debates: Kennedy vs. Nixon, 1960, a reissue (pp. 313 – 330) . Bloomington : Indiana University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Lazarsfeld P. F. , Berelson B., Gaudet H. ( 1944 ). The people’s choice . New York, NY : Columbia University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Prior M. ( 2007 ). Post-broadcast democracy: How media choice increases inequality in political involvement and polarizes elections . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Prior M. ( 2012 ). Who watches presidential debates? Measurement problems in campaign effects research . Public Opinion Quarterly , 76 , 350 – 363 . Google Scholar Crossref Search ADS WorldCat Sears D. O. ( 1968 ). The paradox of de facto selective exposure without preference for supportive information. In Abelson R., Aronson E., McGuire W. J., Newcomb T. M., Rosenberg M. J., Tannenbaum H. (Eds.), Theories of cognitive consistency: A sourcebook (pp. 777 – 787 ). Chicago, IL : Rand McNally . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Stroud N. J. ( 2011 ). Niche news: The politics of news choice . Oxford : Oxford University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Taber C. S. , Lodge M. ( 2006 ). Motivated skepticism in the evaluation of political beliefs . American Journal of Political Science , 50 , 755 – 769 . Google Scholar Crossref Search ADS WorldCat Utych S. M., , Kam C. ( 2014 ). Viability, information seeking, and vote choice . Journal of Politics , 76 , 152 – 166 . Google Scholar Crossref Search ADS WorldCat © The Author 2016. Published by Oxford University Press on behalf of The World Association for Public Opinion Research. All rights reserved. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png International Journal of Public Opinion Research Oxford University Press

Selective Exposure, Information Utility, and the Decision to Watch Televised Debates

Loading next page...
 
/lp/oxford-university-press/selective-exposure-information-utility-and-the-decision-to-watch-Drw8IcY1vQ

References (28)

Publisher
Oxford University Press
Copyright
© The Author 2016. Published by Oxford University Press on behalf of The World Association for Public Opinion Research. All rights reserved.
ISSN
0954-2892
eISSN
1471-6909
DOI
10.1093/ijpor/edw016
Publisher site
See Article on Publisher Site

Abstract

Abstract Why do people watch televised election debates? This article considers the role of three factors highlighted in recent research on political news exposure: selective approach toward attitude-congruent information; selective avoidance of attitude-incongruent information; and information utility. I test the distinct effects of these three factors on viewing decisions by taking advantage of Austria’s unique organization of televised debates as pairwise match-ups between the main party leaders. I find that selective approach dominates selective avoidance and that information utility increases exposure propensity. Beyond explaining debate viewership, these findings have general implications for understanding exposure to political information, particularly during campaigns. Political information-seeking has long been explained using the theory of selective exposure (Lazarsfeld, Berelson, & Gaudet, 1944). The basic argument is simple: we seek out and engage with sources that provide attitude-congruent information, and we avoid sources that provide attitude-incongruent information. A key motivation for such behavior would be to reduce the potential for cognitive dissonance, that is, to reduce exposure to information that challenges existing opinions and is thus unsettling and uncomfortable (Festinger, 1957; Frey, 1986). The theory of selective exposure remains highly relevant today for at least a couple of reasons. For one, politics is increasingly polarized in many countries such as the United States, giving people greater incentives for selective exposure. Moreover, technological changes leading to the fragmentation of media consumption (Prior, 2007) may make it easier for individuals to act on these incentives (Iyengar, Hahn, Krosnick, & Walker, 2008). More recently, the theory of selective exposure has been revisited and amended. First, it is argued that positive incentives to seek out congruent information are more important than negative incentives to avoid incongruent information. In other words, selective approach is more important than selective avoidance (Garrett, 2009). The consequences of distinguishing between approach and avoidance are significant, as they imply that polarization and technological change may have a less severe impact than previously argued. Second, it has been pointed out that explaining exposure using affect is too limited. Instead, we also need to consider the extent to which information is useful to citizens, for instance in assessing threats and opportunities, formulating opinions, and determining electoral choices (Knobloch-Westerwick, 2014; Knobloch-Westerwick & Kleinman, 2012). This article examines the influence of selective approach and selective avoidance as well as of information utility on decisions to watch televised debates between candidates for high public office. Televised debates have become near-ubiquitous: in addition to the famous presidential debates in the United States, such events are now organized in countries such as the UK, France, Germany, Denmark, and the Netherlands. Individuals who tune in to televised debates are necessarily exposed to more than one viewpoint: in Lang and Lang’s (1977) words, debates create “double exposure”. Televised debates are thus similar to panel discussions, roundtables, and even newscasts in that there are multiple actors and diverse points of view. While still largely predictable in terms of the constellation of opinions, the information setting of televised debates is therefore more complex than, say, a newspaper comment piece expressing a single opinion. The focus of this article is on explaining why people tune in to some debates but not others; in doing so, it tests the distinct impacts of approach, avoidance, and information utility. In other words, the core concern of this article is not on why certain individuals are more likely than others to watch debates or why certain candidates generally lead to high viewing figures (on these questions, see, e.g., Baum & Kernell, 1999; Kenski & Jamieson, 2011; Kenski & Stroud, 2005). Instead, the aim of this article is to study what kinds of debates viewers will tune in to see, using as the key explanatory approach the candidate perceptions that create approach, avoidance, and utility. These explanations are difficult to assess if there is just one debate. However, the analysis in this article takes advantage of a unique observational setting that makes it possible to disentangle the different drivers of exposure behavior. In the run-up to the 2013 parliamentary election in Austria, the top candidates of the six major parties faced off in pairwise debates, leading to a total of 15 debates between the different candidates. Hence, individuals who support one candidate can watch all his or her debates—or they can selectively avoid debates against candidates they dislike. Moreover, an individual who sees two candidates as potential choices in the election can selectively choose to watch the debate where those two candidates square off. As a result, the set-up studied here leads to an intra-individual variety of affect and utility configurations that differs for each pair of candidates, making effective statistical tests possible. Moreover, this set-up allows us to control for individual-level and debate-level confounders that may be associated with affect toward candidates; these include political interest among individuals and political polarization between candidates. In sum, the case used in this article provides an important setting for understanding exposure to multiple-source settings and for testing the distinct impacts of selective approach, selective avoidance, and information utility. This article is structured as follows. First, I summarize the recent debate about selective approach and avoidance and consider information utility as an additional driver of exposure decisions. Then, I discuss how these perspectives can be applied to the decision to view televised debates. In terms of data, viewership of the debates is assessed using a multi-wave online access panel (Kritzinger et al., 2014a). The key results are replicated using a probability-based rolling cross-section survey carried out throughout the period in which debates were held (see Supplementary Appendix 7; Kritzinger et al., 2014b). After presenting key results from these analyses, I conclude by discussing the broader relevance of the findings, also for discussions concerning the impact of political polarization of citizens’ information-seeking behavior. Explaining Exposure to Political Information Recently, it has been argued that selective exposure results from two behavioral regularities, namely, selective approach and selective avoidance (Garrett, 2009, Garrett & Stroud, 2014). Selective approach refers to the tendency to actively seek out information that is likely to bolster preexisting attitudes, while selective avoidance refers to actions designed to prevent exposure to information that contradicts these beliefs. Hence, individuals can avoid cognitive dissonance by seeking information that reinforces their opinions and by avoiding information that challenges them (Garrett, 2009). Crucially, the positive association between affect or attitudes and information-seeking may result from either of both of these behaviors. The empirical occurrence of selective exposure can therefore be the consequence of the fact that individuals seek out reinforcing information, avoid challenging information, or both. Hence, in testing selective exposure, we need to separate out the distinct influence of approach and avoidance (Garrett & Stroud, 2014). Existing evidence indicates that selective approach determines our information-seeking behavior to a greater degree than selective avoidance (Chaffee, Saphir, Graf, Sandvig, & Hahn, 2001; Frey, 1986; Garrett, 2009; Garrett & Stroud, 2014). There are several reasons why individuals may be less likely to avoid attitude-incongruent information than to seek out attitude-congruent material. For instance, the desire to avoid cognitive dissonance does not require avoiding attitude-incongruent information altogether: when faced with such information, individuals can also attempt to seek out additional, attitude-reinforcing information or find arguments that contradict the challenging information (Garrett, 2009; Taber and Lodge, 2006).1 It may be that the strategies of avoiding exposure completely, of arguing against attitude-incongruent information, and of seeking out reinforcing information may all be similarly effective at decreasing dissonance. Moreover, it may be less difficult to engage in countervailing strategies rather than to try to avoid all attitude-incongruent information in the first place (Garrett, 2009). As any amount of exposure to political information is likely to lead to being faced with opinion-challenging information, individuals may be used to other ways of reducing dissonance rather than just simple avoidance. Selective avoidance may also be relatively infrequent because utility is derived from the very fact that information contradicts one’s opinion. As Holbert, Garrett, and Gleason (2010, p. 21) note, “there are times when it will be more valuable to understand an issue fully than to avoid discrepant information, such as when an individual needs to defend a position or to critique the opposition.” It is useful to know what people with an opposing opinion think and how they argue their position and against one’s own view (Garrett, 2009). Counter-attitudinal information may thus allow individuals to better argue their position or dismiss others. Partisans might also seek out information “to see what ‘the enemy’ is up to” (Chaffee, Saphir, Graf, Sandvig & Hahn, 2001, 262). So, information from the other side of the debate may on occasion be actively sought out by individuals instead of being subject to selective avoidance. Selective exposure may not be the only driver of exposure decisions. Another important explanation concerns whether individuals believe they will derive some utility from their exposure to information (Knobloch-Westerwick, 2014; Knobloch-Westerwick & Kleinman, 2012; Stroud, 2011, p. 25). While attitude reinforcement itself of course provides utility (Knobloch-Westerwick & Kleinman, 2012), there are various other ways in which information can be perceived as useful. Knobloch-Westerwick (2008, 2014) highlights how information may help to assess threats (surveillance function). Building on Atkin (1973), Knobloch-Westerwick and Kleinman (2012) also point out that information can aid us in determining “what to feel” (guidance function) and “what to do” (performance function). Hence, information may help us to know what to think about the parties and candidates competing and how we should vote in the election (Knobloch-Westerwick & Kleinman, 2012).2 In sum, selective exposure based on individual affect and attitudes remains an important approach to understanding how people select which political news and information to consume. However, within that paradigm, seeking out exposure to attitude-congruent information (selective approach) may be a stronger motivator than avoiding exposure to attitude-incongruent information (selective avoidance), as the latter type of information may in reality not be unattractive to individuals. Moreover, information utility may be a further important driver of exposure decisions. Selective Exposure to Televised Debates? These insights into selective exposure can be applied to the decision whether or not to watch televised debates between candidates. The key characteristic of debates that is important for how we study the effects of incentives for selective exposure is the fact that they involve “double exposure” (Lang & Lang, 1977, p. 277): by deciding whether or not to watch a debate, individuals have to choose between exposing themselves to the personality and arguments of both candidates or to those of neither candidate. Selective exposure to just one of the two debaters is not possible, so individuals who watch a debate may have to see a disliked debater and hear unwelcome arguments. To some extent, debates therefore cannot give full reign to individuals’ desire to pick and choose the information they see and hear. As argued below, the Austrian setting of multiple debate match-ups provided individuals with more freedom in this regard, making it possible to determine the drives of exposure to televised debates. In general, studying televised debates can give us an insight into the relative weight of the two affect-based incentives in determining how voters seek out information. Do voters who like one candidate and dislike the other still tune in, so does selective approach dominate selective avoidance? Or do such voters instead choose not to expose themselves to counter-attitudinal information? Televised debates can tell us how citizens deal with situations where multiple exposure is likely. Such situations are not unusual, as much political information—such as roundtables or newscasts—provides two or more sides to every argument and presents a somewhat balanced view of events and opinions. Despite the fear that the advent of new technologies would increase tendencies toward the creation of “echo chambers,” it is close to impossible to follow politics closely and still avoid attitude-incongruent information. Applied to televised debates, the theory of selective exposure generates two hypotheses. First, the likelihood of watching a debate should increase as affect toward the debaters increases, while low levels of affect should decrease viewing probabilities. If both debaters are strongly liked, then this should reinforce viewing probabilities. Hence, the first hypothesis is: H1: The more positive their affect toward the debaters, the more likely individuals are to watch the debate; the more negative their affect toward the debaters, the less likely individuals are to watch the debate. However, as debates create multiple exposure, viewers may be faced with a situation where they feel positive affect toward one debater and negative affect toward the other. Here, the traditional perspective would predict that selective approach and avoidance would cancel each other out. Indeed, proponents of the traditional perspective might even expect viewership among this group to be low, as individuals might prefer to seek out other sources of information where they are not confronted with attitude-incongruent information. In contrast, proponents of the more recent perspective that emphasizes selective approach over avoidance (Garrett, 2009; Garrett & Stroud, 2014) would argue that affect congruence outweighs affect incongruence as an incentive to watch debates. Hence, if there is high affect toward one debater and low affect toward the other, viewership should still be high, indeed possibly as high as when both debaters are well-liked. Hence, the second hypothesis is: H2: The level of affect toward the less-liked debater has no effect on the decision to watch a televised debate. The third hypothesis concerns information utility. Here, we expect individuals to tune in to televised debates if this may help them decide whom to vote for. Hence, voters will seek out information that fulfills the guidance and performance functions of information (Atkin, 1973; Knobloch-Westerwick & Kleinman, 2012). Debates will be particularly useful to voters who are hesitating between the two leaders or their parties. For such voters, the debate may contain important information on whom they should choose. Hence, the third hypothesis is: H3: Individuals are more likely to watch debates if they see both candidates as potential electoral choices than if they see only one or no candidate as a potential choice. Data and Methods The focus in this article is on understanding why individuals watch some debates but not others. This question is studied using the case of Austria in 2013: before the parliamentary elections, there were 15 debates where leaders of each of the six parties represented in parliament debated with every other party leader. As a result, there were naturally varying configurations of affect and support for the two debaters among voters, allowing us to run models that include affect and information utility as separate predictors. For example, it is possible to distinguish between voters who feel positively about both candidates and those who would consider voting for both of them. In settings such as the United States, where opinions about the two candidates tend to be strongly and negatively correlated, it is difficult to parse out the distinct effects of candidate affect and information utility. Another advantage of the data is that the data allow us to use individual- and debate-level fixed effects to consider within-respondent variation in viewing decisions, thus providing a particularly stringent test of the hypotheses. The debates were broadcast each Tuesday and Thursday in the last four weeks of the campaign, with two debates shown each evening.3 The leaders of the six main parties took part in the debates: Chancellor Werner Faymann of the Social Democratic SPÖ, Vice-Chancellor Michael Spindelegger of the Christian Democratic ÖVP, Heinz-Christian Strache of the radical-right FPÖ, Eva Glawischnig of the Greens, Josef Bucher of the radical-right BZÖ (a splinter party of the FPÖ), and Frank Stronach of the Team Stronach. At the time, the Austrian party system was therefore made up of two mainstream parties (SPÖ/ÖVP), a Green party, one large (FPÖ) and one small (BZÖ) radical-right party, and one new, personalized party (Team Stronach). The governing coalition was made up of the SPÖ and the ÖVP; such a grand coalition between the two mainstream parties is typical of Austrian politics. However, the two parties were challenged at the polls mainly by the resurgent radical-right FPÖ and, to a lesser extent, by the new party founded by the billionaire Frank Stronach. Both leaders had significant personal appeal, at least to their followers, though Figure 1 shows that, on average, they were less popular than their rivals. In the end, however, the SPÖ and the ÖVP continued to hold enough seats to carry on in coalition despite both losing votes overall (Dolezal and Zeglovits, 2014). This outcome was foreshadowed in the polls during the election campaign. Figure 1 Open in new tabDownload slide Histograms of like–dislike scores for the six debaters. Note. Data from the probability-based rolling cross-section telephone survey (Kritzinger et al., 2014b); responses not weighted Figure 1 Open in new tabDownload slide Histograms of like–dislike scores for the six debaters. Note. Data from the probability-based rolling cross-section telephone survey (Kritzinger et al., 2014b); responses not weighted Of the 15 debates, the highest viewing figures were for the debate between the Chancellor Werner Faymann and the leader of the radical-right Freedom Party, Heinz-Christian Strache. This debate reached almost 1.1 million people out of a total Austrian population of 8.5 million. The least-watched debate, that between the Strache and his radical-right rival Josef Bucher, nevertheless reached over 650,000 people. A complete list of the debates and their respective viewing figures can be found in Supplementary Appendix 1. All in all, while the debates were thus a central part of the last month of the election campaign, there is also clear variation in the extent to which Austrians tuned in to each debate. Concurrently with the campaign, the Austrian National Election Study ran a four-wave panel study (Kritzinger et al., 2014a). This survey was carried out online using an access panel, from which TNS drew a sample of respondents that was representative of the population on key characteristics. For the analyses, I make use of data from the first and third waves; these were carried out on 16–28 August 2013 (Wave 1) and on 26–29 September 2013 (Wave 3). In all, 3,085 respondents participated in Wave 1 and 1,318 in Wave 3. The supplemental analyses in Supplementary Appendix 4 also make use of data from Wave 2 (9–12 September), which was completed by 2,044 respondents. One particular useful feature of this multiwave panel is that the independent variables can be measured before the debates occurred, thus eliminating the possibility that the debates themselves influenced the variables of interest (such as candidate sympathy). Given the obvious concerns that data from an online access panel are not a probability sample, the key results are replicated in Figure 3 and Supplementary Appendix 7 using data from a rolling cross-section telephone survey (Kritzinger et al., 2014b; on the design of rolling cross-section studies, see Johnston & Brady, 2002). This survey is not used for the main analyses because it asks about debate viewership in a more limited fashion and measures the key covariates after exposure to the debates. The dependent variable is whether respondents say they watched a particular debate or not. Respondents were presented with a list of all debates and were asked to select those debates they had seen (complete question wording in Supplementary Appendix 2). In total, 1,421 respondents answered this question; as I stack the data by debate, this results in a total of 21,315 responses overall. Responses are coded as “watched” for respondents who said they had seen the debate, with all other responses (including don’t knows and refusals) classed as “not watched”. Unsurprisingly, the proportion of respondents who said they viewed the debate is markedly higher than the proportion of actual viewers as recorded by the broadcasters. Partly, this will be because people caught up on the debates in other news programs or only watched part of the debate, but it might also be a result of misremembering and social desirability (Prior, 2012) as well as of biased sampling. To address this in part, Supplementary Appendix 4 presents additional analyses that use a stricter measurement of debate viewership, where I take advantage of the fact that the respondents were asked twice whether they had seen each debate. In these analyses, the respondents are only coded as having viewed a debate if they gave consistent responses in both Wave 2 and Wave 3. Measures of candidate sympathy are needed to assess the extent to which selective exposure affects the decision to watch a debate (H1 and H2). I use two variables here.4 First, I include one variable that measures the sympathy score for the more liked of the two debaters. This sympathy is measured using the 0–10 like–dislike scale for each of the six party leaders.5 The greater the sympathy for the more-liked candidate, the more an individual should want to watch a debate. Second, I include as a second predictor the sympathy score of the less-liked debater.6 To test H3, I need a measure of how likely the respondent is to vote for the party of the debater. I therefore include a measure the respondent’s self-declared probability to ever vote for the candidate’s party, provided on a 0–10 scale. This captures the extent to which the citizen would consider voting for that party; while related to affect ratings, this is nevertheless conceptually distinct. Again, this measure is included as two variables, one for the higher and one for the lower of the two values. I also present models that include several other variables that vary by individual and candidate. First, I include the extent to which the respondent finds each candidate charismatic, measured on a 1–4 scale. This captures the attractiveness of a candidate beyond positive or negative affect. Second, I include a dummy variable for whether the respondent identifies with the party of either candidate debating (1 = yes). This captures the extent to which group loyalties may drive viewership decisions. Finally, I include a variable measuring whether the citizen expects the candidates to do well in the debate. This could capture the extent to which a debate might be seen as entertaining. Obviously all these variables correlate with affect ratings and propensity-to-vote scores to a certain degree, creating a tough test for the hypotheses.7 I run a series of linear probability models, with debate viewing as the dependent variable. Because the data are stacked, with 15 cases per respondent, I use fixed effects for each respondent, as this captures individual-specific differences in the propensity to view debates. (This also means that it is not possible to include any additional predictors that do not vary by individual, such as political interest, age, education, or gender.) In the results section, I also present multilevel regression models with a set of key control variables; these models produce substantively similar results for key covariates as those in the fixed-effects models. It is the use of fixed effects that also requires the use of linear probability models rather than binary logistic regression: there are naturally a number of respondents who did not watch any debates and would thus be dropped in fixed-effects logistic regression analyses. All models also include fixed effects for each debate. This captures factors such as the effect of viability on interest in candidates (Utych & Kam, 2014), idiosyncratic effects linked to the two debaters, or simply other circumstantial aspects such as the weather or competing TV programming. Given that fixed effects for each debate are included, it is neither possible nor necessary to include further debate-level factors (such as the average popularity of debaters or their policy positions) in the models. In Supplementary Appendix 5, I present results disaggregated by debate; overall, the results are stable across debates, with moderate variation in the magnitude of effects. Results Table 1 presents the core models, all of which include fixed effects for all respondents and all debates. Model 1 presents an “empty” model without any predictors apart from the fixed effects. Model 2 then includes the four main predictors: the sympathy score for the more well-liked and that for the less-liked of the two debaters, and the probability to vote for the debaters’ parties. Model 3 includes interactions between the variables to capture differences in the effect of affect for one debater according to affect for the other debater and in the effect of voting probability for one party according to voting probability for the other debater. For instance, a three-unit sympathy score for the less-liked debater may have a different impact if the score for the more-liked debater is three or if it is ten. Models 4 and 5 repeat these Models, but include the series of controls described above.8 Table 1 OLS fixed-effects regression results Predictors . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Sympathy, more-liked candidate 0.00844*** (0.001) 0.00805*** (0.001) 0.00716*** (0.002) 0.00687*** (0.002) Sympathy, less-liked candidate 0.000327 (0.002) −0.00511 (0.005) −0.00110 (0.002) −0.00443 (0.005) Interaction: Sympathy scores 0.000735 (0.001) 0.000460 (0.001) Propensity to vote for party of candidate, higher value 0.00378*** (0.001) 0.00341** (0.001) 0.00278* (0.001) 0.00236 (0.001) Propensity to vote for party of candidate, lower value 0.00440** (0.002) −0.00420 (0.005) 0.00527** (0.002) −0.00540 (0.006) Interaction: Propensities to vote 0.00103 (0.001) 0.00128 (0.001) Party ID for one candidate 0.0223* (0.009) 0.0207* (0.009) Expected performance, higher value −0.00679 (0.008) −0.00509 (0.008) Expected performance, lower value −0.0273 (0.016) −0.0237 (0.016) Interaction: Expected performances 0.00448 (0.003) 0.00360 (0.004) Charisma of candidate, higher value −0.0182 (0.012) −0.0151 (0.012) Charisma of candidate, lower value −0.0266 (0.021) −0.0206 (0.021) Interaction: charisma scores 0.00777 (0.006) 0.00602 (0.006) Constant 0.335*** (0.008) 0.270*** (0.010) 0.276*** (0.011) 0.404*** (0.050) 0.394*** (0.052) Observations 21,315 19,965 19,965 17,402 17,402 R2 0.53 0.53 0.53 0.53 0.53 Adjusted R2 0.49 0.50 0.50 0.49 0.49 AIC 9,462 9,030 9,027 8,762 8,760 BIC 9,582 9,180 9,193 8,964 8,978 Predictors . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Sympathy, more-liked candidate 0.00844*** (0.001) 0.00805*** (0.001) 0.00716*** (0.002) 0.00687*** (0.002) Sympathy, less-liked candidate 0.000327 (0.002) −0.00511 (0.005) −0.00110 (0.002) −0.00443 (0.005) Interaction: Sympathy scores 0.000735 (0.001) 0.000460 (0.001) Propensity to vote for party of candidate, higher value 0.00378*** (0.001) 0.00341** (0.001) 0.00278* (0.001) 0.00236 (0.001) Propensity to vote for party of candidate, lower value 0.00440** (0.002) −0.00420 (0.005) 0.00527** (0.002) −0.00540 (0.006) Interaction: Propensities to vote 0.00103 (0.001) 0.00128 (0.001) Party ID for one candidate 0.0223* (0.009) 0.0207* (0.009) Expected performance, higher value −0.00679 (0.008) −0.00509 (0.008) Expected performance, lower value −0.0273 (0.016) −0.0237 (0.016) Interaction: Expected performances 0.00448 (0.003) 0.00360 (0.004) Charisma of candidate, higher value −0.0182 (0.012) −0.0151 (0.012) Charisma of candidate, lower value −0.0266 (0.021) −0.0206 (0.021) Interaction: charisma scores 0.00777 (0.006) 0.00602 (0.006) Constant 0.335*** (0.008) 0.270*** (0.010) 0.276*** (0.011) 0.404*** (0.050) 0.394*** (0.052) Observations 21,315 19,965 19,965 17,402 17,402 R2 0.53 0.53 0.53 0.53 0.53 Adjusted R2 0.49 0.50 0.50 0.49 0.49 AIC 9,462 9,030 9,027 8,762 8,760 BIC 9,582 9,180 9,193 8,964 8,978 Note. Standard errors in parentheses, fixed effects for debates and individuals included. * p < .05, **p < .01, ***p < .001. Table 1 OLS fixed-effects regression results Predictors . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Sympathy, more-liked candidate 0.00844*** (0.001) 0.00805*** (0.001) 0.00716*** (0.002) 0.00687*** (0.002) Sympathy, less-liked candidate 0.000327 (0.002) −0.00511 (0.005) −0.00110 (0.002) −0.00443 (0.005) Interaction: Sympathy scores 0.000735 (0.001) 0.000460 (0.001) Propensity to vote for party of candidate, higher value 0.00378*** (0.001) 0.00341** (0.001) 0.00278* (0.001) 0.00236 (0.001) Propensity to vote for party of candidate, lower value 0.00440** (0.002) −0.00420 (0.005) 0.00527** (0.002) −0.00540 (0.006) Interaction: Propensities to vote 0.00103 (0.001) 0.00128 (0.001) Party ID for one candidate 0.0223* (0.009) 0.0207* (0.009) Expected performance, higher value −0.00679 (0.008) −0.00509 (0.008) Expected performance, lower value −0.0273 (0.016) −0.0237 (0.016) Interaction: Expected performances 0.00448 (0.003) 0.00360 (0.004) Charisma of candidate, higher value −0.0182 (0.012) −0.0151 (0.012) Charisma of candidate, lower value −0.0266 (0.021) −0.0206 (0.021) Interaction: charisma scores 0.00777 (0.006) 0.00602 (0.006) Constant 0.335*** (0.008) 0.270*** (0.010) 0.276*** (0.011) 0.404*** (0.050) 0.394*** (0.052) Observations 21,315 19,965 19,965 17,402 17,402 R2 0.53 0.53 0.53 0.53 0.53 Adjusted R2 0.49 0.50 0.50 0.49 0.49 AIC 9,462 9,030 9,027 8,762 8,760 BIC 9,582 9,180 9,193 8,964 8,978 Predictors . Model 1 . Model 2 . Model 3 . Model 4 . Model 5 . Sympathy, more-liked candidate 0.00844*** (0.001) 0.00805*** (0.001) 0.00716*** (0.002) 0.00687*** (0.002) Sympathy, less-liked candidate 0.000327 (0.002) −0.00511 (0.005) −0.00110 (0.002) −0.00443 (0.005) Interaction: Sympathy scores 0.000735 (0.001) 0.000460 (0.001) Propensity to vote for party of candidate, higher value 0.00378*** (0.001) 0.00341** (0.001) 0.00278* (0.001) 0.00236 (0.001) Propensity to vote for party of candidate, lower value 0.00440** (0.002) −0.00420 (0.005) 0.00527** (0.002) −0.00540 (0.006) Interaction: Propensities to vote 0.00103 (0.001) 0.00128 (0.001) Party ID for one candidate 0.0223* (0.009) 0.0207* (0.009) Expected performance, higher value −0.00679 (0.008) −0.00509 (0.008) Expected performance, lower value −0.0273 (0.016) −0.0237 (0.016) Interaction: Expected performances 0.00448 (0.003) 0.00360 (0.004) Charisma of candidate, higher value −0.0182 (0.012) −0.0151 (0.012) Charisma of candidate, lower value −0.0266 (0.021) −0.0206 (0.021) Interaction: charisma scores 0.00777 (0.006) 0.00602 (0.006) Constant 0.335*** (0.008) 0.270*** (0.010) 0.276*** (0.011) 0.404*** (0.050) 0.394*** (0.052) Observations 21,315 19,965 19,965 17,402 17,402 R2 0.53 0.53 0.53 0.53 0.53 Adjusted R2 0.49 0.50 0.50 0.49 0.49 AIC 9,462 9,030 9,027 8,762 8,760 BIC 9,582 9,180 9,193 8,964 8,978 Note. Standard errors in parentheses, fixed effects for debates and individuals included. * p < .05, **p < .01, ***p < .001. The first two hypotheses relate to the impact of candidate affect on watching a debate. In Model 2, only the sympathy score for the more liked-debater is a statistically significant predictor: the more this debater is liked, the more likely it is that a voter will tune in. This fits with the selective approach theory. Note that the sympathy score of the less-liked debater also has a much smaller effect on viewing decisions than the sympathy score of the more-liked debater. In Model 4, the effect of sympathy scores for the less-liked debater is indeed slightly negative (but not statistically significant). Moreover, the interaction terms in Models 3 and 5 are not statistically significant at the 0.05 level, providing a first indication that sympathy for the less-liked debater does not vary in impact depending on sympathy for the more-liked debater. To interpret the effects of sympathy toward debaters more precisely, I present the results graphically (as recommended by Brambor, Clark, & Golder, 2006). The left panel of Figure 2 shows the predicted probability of watching the debate for varying levels of sympathy for the two candidates; scores for the more-liked debater are on the x-axis, scores for the less-liked debater on the y-axis. This Figure is based on Model 5, as this most closely relates to the hypotheses to be tested. Each cell contains the predicted probability of watching the debate; higher probabilities have darker shades of gray. For instance, the predicted probability of watching the debate is 26% if the sympathy score for both debaters is 0, 28% if both scores are at 5, and 33% if both scores are at 10. Figure 2 Open in new tabDownload slide Predicted probability of watching a debate conditional on candidate affect. Note. Probabilities calculated based on Model 5, Table 1. In the left panel, the x-axis represents the sympathy scores for the more-liked debater, the y-axis sympathy scores for the less-liked debater; higher probabilities in darker shades of gray. In the right panel, the x-axis represents the vote intention (propensity to vote) scores for the more-preferred debater, the y-axis the vote intention (propensity to vote) for the less-preferred debater; higher probabilities in darker shades of gray Figure 2 Open in new tabDownload slide Predicted probability of watching a debate conditional on candidate affect. Note. Probabilities calculated based on Model 5, Table 1. In the left panel, the x-axis represents the sympathy scores for the more-liked debater, the y-axis sympathy scores for the less-liked debater; higher probabilities in darker shades of gray. In the right panel, the x-axis represents the vote intention (propensity to vote) scores for the more-preferred debater, the y-axis the vote intention (propensity to vote) for the less-preferred debater; higher probabilities in darker shades of gray By looking at the changes along the x-axis, we can see how sympathy for the more-liked debater increases viewing probabilities. If the less-liked debater has a sympathy score of 0 (lowest row), then shifting from 0 to 10 for the more liked-debater increases viewing probabilities by 7% (from 26 to 33%). Similar patterns hold for all rows: increasing affect to the more-liked debater increases viewing probabilities. In contrast, the sympathy scores for the less-liked debater matter less. We can see this by looking at the columns, so by moving up and down the y-axis. Here, we can see that predicted viewing probabilities are mostly stable. For instance, if one debater is well-liked (sympathy score of 9), it does not matter a great deal what individuals think of the other debater: in the columns on the right, all predictions are very similar, at around 32%. In other words, the sympathy individuals have for the less-liked debater has little impact on decisions whether or not to watch the debate; only the sympathy scores for the more-liked debater have an appreciable effect. Importantly, the predicted probability to watch a debate if both candidates are disliked is not zero, as many other factors—opinions about the candidates and individual-level characteristics—also drive this decision. Nevertheless, the fact that 26% who dislike both candidates are predicted to tune in does somewhat go against stronger version of the selective exposure theory. Interestingly, individuals who strongly favor one candidate and strongly dislike the other have the same viewing probability as those who favor both candidates: 33%. Overall, there is therefore strong evidence in favor of H2, so the results indicate that a theory based on selective approach applies better than a theory—reflected in the simpler H1—that places equal weight on selective avoidance. Turning to H3, the respondents’ self-reported likelihood of ever voting for a party yields interesting results. First, the regression models in Table 1 indicate that the scores for both debaters are statistically significant in Models 2 and 4. If an interaction effect is included, this is significant at the 0.1 level in Models 3 and 5. Hence, it appears that the propensity to vote for the debater’s party has an impact for both debaters. To understand this effect, graphical illustration is again useful. Figure 2 therefore presents the effects of vote intention on the probability of tuning in to the debates (right panel). Calculating predicted viewing probabilities indicates that the highest probability of viewing a debate is for candidates from parties that the respondent could both imagine voting for (two scores of ten: 37%). If the respondent can imagine voting for just one of the two parties, the probability of tuning in declines (one score of ten, one of zero: 30%). The probability is lower again if the respondent cannot imagine voting for either candidate’s party (two scores of zero: 27%). Unlike for the affect scores in the left panel, it matters whether individuals might vote for both parties or are only considering voting for one of them. Hence, there is evidence in favor of individuals choosing to watch debates that are particularly useful in terms of ultimately deciding whom to vote for. Information utility in terms of guidance and performance functions may therefore be an important driver of TV debate viewing, in addition to selective approach. Overall, the models explain about 50% in the variation in the decision to view the televised debates. While the reduction in the explained variance created by the additional variables is low, the AIC/BIC values do decline by large amounts, indicating a superior fit of the larger models to the data; this applies in particular to Models 2 and 4. However, note that the interaction terms between the sympathy scores and probability-to-vote scores add little to model fit in terms of AIC/BIC, providing an indication that simple patterns of association dominate. As a robustness check, Figure 3 presents the results for two alternative model specifications; the full models can be found in Supplementary Appendices 6 and 7. The first alternative specification uses random- rather than fixed-effects at the individual level, along with a suite of additional controls: age, education, general levels of TV viewing, turnout propensity, political interest, whether the respondent sees televised debates as entertaining, and whether the respondent has already decided whom to vote for. The key patterns from Figure 2 are replicated for both sympathy scores and vote intentions. Figure 3 Open in new tabDownload slide Predicted probability of watching a debate conditional on candidate affect and vote intention. Note. Probabilities calculated based on the models in Supplementary Table A6.1 and Supplementary Table A7.1. Results for the rolling cross-section survey based on a logistic regression with standard errors clustered by individual. See Supplementary Appendices 6 and 7 for full results Figure 3 Open in new tabDownload slide Predicted probability of watching a debate conditional on candidate affect and vote intention. Note. Probabilities calculated based on the models in Supplementary Table A6.1 and Supplementary Table A7.1. Results for the rolling cross-section survey based on a logistic regression with standard errors clustered by individual. See Supplementary Appendices 6 and 7 for full results The second key robustness check uses the telephone-based rolling cross-section survey mentioned above; this only asked respondents about the two most recent debates, so these data have the drawback that we cannot include individual-level fixed effects. The response is modeled using logistic regression with a small set of controls and standard errors clustered by respondent. As the rightmost panel in Figure 3 shows, the key patterns are again replicated: a strong increase in viewing probabilities in the rows, comparably stable probabilities in the columns. An interesting nuance in the rolling cross-section results is that there is a stronger reinforcing effect: viewing probabilities are much higher if individuals feel positively about both candidates rather than just one; however, strong positive affect toward one debater remains a key determinant of viewing decisions, so H2 is confirmed here as well. These data do not allow us to control for information utility. Because these additional models do not include individual-level fixed effects, we can also include a series of controls that show interesting results. We thus find that respondents who are interested in politics, who watch TV frequently, who find debates entertaining, and who are middle-aged are most likely to tune in to watch the debates. Interestingly, self-declared undecided voters are in fact less likely to watch the debates, perhaps because such voters are generally less interested in politics. For more detail on these findings, see Supplementary Appendices 6 and 7. Conclusions Two key findings emerge from this study, which used an advantageous setting to test the distinct influences of approach, avoidance, and information utility on watching televised debates. First, the decision to watch televised debates is determined more by selective approach than by selective avoidance. It matters more whether one candidate is well-liked by a potential viewer than whether the other debater is liked or disliked. People seek out attitude-congruent information more than they shy away from attitude-incongruent material. This pattern was broadly constant across debates, though future research should consider how aspects of debates can influence the impact of selective approach on viewing decisions. Second, the decision to watch televised debates is related to information utility. Specifically, this article found that people are more likely to watch a debate between two candidates from parties they would consider voting for. Hence, debates can fulfill the guidance and performance functions of information (Knobloch-Westerwick & Kleinman, 2012): they can help individuals to determine what they feel about candidates and whom they should vote for. This finding provides further evidence that information utility is an important factor in understanding exposure decisions. There are of course limitations to this study. While televised debates, and in particular the paired match-ups in Austria, provide a useful setting to examine incentives toward selective exposure, it is important to restate that debates also have the rather unique characteristic of creating “double exposure” to political information. This feature of debates is similar to the tendency toward balance on panel shows and newscasts. (In contrast, the ideal-type single-exposure information settings may in reality be quite rare and atypical.) The results in this study should therefore be applicable to media events or televised panel debates that feature more than one partisan point of view in a stable and predictable manner. However, it is important for future research to examine less fixed types of media coverage and political information and how these differences affect the behavior of individuals. Future studies should also consider further drivers of exposure decisions. For one, information utility is a complex concept with at least four components (Knobloch-Westerwick & Kleinman, 2012), and only two—guidance and performance—were applied in this study. It would also be worthwhile to distinguish between the type of information utility empirically examined in this study and the information utility of attitude-incongruent information, which I have only discussed theoretically. Moreover, researchers should consider in more depth explanations related to political interest and entertainment, two factors this study has neglected because they vary at the individual and not the debate level. While this article focused on why people tune in to a specific debate (compared with other debates), future research could focus on factors that explain why individuals watch debates. Finally, these results have important broader implications for understanding the impact of political polarization on exposure to political information. This study finds that there is no effect of polarization per se on exposure decisions. Instead, only affect toward the more-liked candidate drives exposure: high sympathy scores encourage individuals to seek out and engage with information from that candidate, even if attitude-incongruent information is present. The results of this study therefore paint a positive picture of the influence of sympathy toward candidates. Moreover, people are more likely to watch debates if this will help them form electoral decisions, providing weight to the information utility approach. Finally, individuals do not completely tune out from debates among candidates they dislike. Overall, the extent of engagement with political information should therefore be related more with how much people support candidates and parties and less with how polarized their opinions are. As long as political information is generally not provided in one-sided settings, strong affect and also polarization should therefore lead to greater levels of political information-seeking. These insights are important for understanding how and when citizens use the media to find out and inform themselves about political events. Acknowledgement I would like to thank Hajo Boomgaarden and Sylvia Kritzinger for comments on an earlier draft of this article. Data collection was carried out by the Austrian National Election Study (AUTNES), a National Research Network (NFN) sponsored by the Austrian Science Fund (FWF) (S10902-G11). Supplementary Data Supplementary Data are available at IJPOR online. Markus Wagner is an Associate Professor in the Department of Government at the University of Vienna. His research focuses on the role of issues and ideology in party competition and vote choice. His research has been published in Comparative Political Studies, the British Journal of Political Science and the European Journal of Political Research. Footnotes 1 " Some authors also argue that the challenge of cognitive dissonance may be overstated. For example, highly opinionated individuals secure in their views may be faced with few costs in terms of potential cognitive dissonance (Holbert, Garrett, & Gleason, 2010). 2 " Selective exposure and information utility are not the only possible drivers of exposure. Some people may simply be interested in politics and public affairs, and all kinds of information may serve to satisfy that curiosity (Sears, 1968) and allow them to engage with politically involved peers (Chaffee, Saphir, Graf, Sandvig & Hahn, 2001, p. 262). Indeed, for some people, all kinds of information may have entertainment value, so people will read about politics because of the intrinsic benefits this creates (Knobloch-Westerwick & Kleinman, 2012). 3 " There were two exceptions: One set of debates was held on a Monday (9 September) instead of a Tuesday; and the debate between the party leaders of the two government parties was the only one that evening. 4 " Supplementary Appendix 9 presents an alternative way of coding candidate affect using sums and distances of the two leaders’ scores, which is simply an arithmetic transformation of the coefficients presented in Table 1. The key conclusions concerning approach and avoidance remain the same. 5 " Supplementary Appendix 8 presents results if we exclude respondents with only moderate levels of affect for both debaters (as recommended by Feldman, Stroud, Bimber, & Wojcieszak, 2013). The conclusions concerning approach and avoidance are the same. 6 " One concern may be that affect toward parties is more important than affect toward candidates. Unfortunately, the online panel survey does not contain measures of affect toward parties. The telephone-based rolling cross-section does, however. Here, the correlation between the two measures of affect is generally around 0.75, and substituting party affect for candidate affect leads to substantively similar results (see Supplementary Appendix 7). 7 " The various variables in the model are related assessments of candidates. The highest bivariate correlation between these is 0.8, the lowest 0.26, with most in the range of 0.5 and 0.65. The highest bivariate correlations are between affect toward candidates and their perceived charisma (average correlation: 0.65), with the average correlation between propensity-to-vote scores and candidate affect at 0.64. In the presence of a large amount of interrelated predictors, multicollinearity may be a concern. However, collinearity diagnostics for the predictors in Model 4 (Table 1) indicate that no variable has a higher variance-inflation factor than 3. 8 " As there may be nonlinear effects for sympathy scores, Supplementary Appendix 3 presents results that include a quadratic term for the sympathy score for the more well-liked of the two debaters. Specifically, there may be particularly strong effects for those with high levels of affect, who will be highly motivated to seek out attitude-congruent information (Taber & Lodge, 2006). The results do not alter the conclusions in this article. References Atkin C. K. ( 1973 ). Instrumental utilities and information seeking. In Clark P. (Ed.), New models of communication research (pp. 205 – 242 ). Newbury Park, CA : SAGE . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Baum M. A. , Kernell S. ( 1999 ). Has cable ended the golden age of presidential television? American Political Science Review , 93 , 99 – 114 . Google Scholar Crossref Search ADS WorldCat Brambor T. , Clark W. R., Golder M. ( 2006 ). Understanding interaction models: Improving empirical analyses . Political Analysis , 14 , 63 – 82 . Google Scholar Crossref Search ADS WorldCat Chaffee S. , Saphir M. N., Graf J., Sandvig C., Hahn K. S. ( 2001 ). Attention to counter-attitudinal messages in a state election campaign . Political Communication , 18 , 247 – 272 . Google Scholar Crossref Search ADS WorldCat Dolezal M. , Zeglovits E. ( 2014 ). Almost an earthquake: The Austrian parliamentary election of 2013 . West European Politics , 37 , 644 – 652 . Google Scholar Crossref Search ADS PubMed WorldCat Feldman L. , Stroud N. J., Bimber B., Wojcieszak M. ( 2013 ). Assessing selective exposure in experiments: The implications of different methodological choices . Communication Methods and Measures , 7 , 172 – 194 . Google Scholar Crossref Search ADS WorldCat Festinger L. ( 1957 ). A theory of cognitive dissonance . Stanford, CA : Stanford University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Frey D. ( 1986 ). Recent research on selective exposure to information. In Berkowitz L. (Ed.), Advances in Experimental Social Psychology, 19, 41–80. Garrett R. K. ( 2009 ). Politically motivated reinforcement seeking: Reframing the selective exposure debate . Journal of Communication , 59 , 676 – 699 . Google Scholar Crossref Search ADS WorldCat Garrett R. K. , Stroud N. J. ( 2014 ). Partisan paths to exposure diversity: Differences in pro‐ and counterattitudinal news consumption . Journal of Communication , 64 , 680 – 701 . Google Scholar Crossref Search ADS WorldCat Holbert L. , Garrett K., Gleason L. ( 2010 ). A new era of minimal effects? A response to Bennett and Iyengar . Journal of Communication , 60 , 15 – 34 . Google Scholar Crossref Search ADS WorldCat Iyengar S. , Hahn K. S., Krosnick J. A., Walker J. ( 2008 ). Selective exposure to campaign communication: The role of anticipated agreement and issue public membership . Journal of Politics , 70 , 186 – 200 . Google Scholar Crossref Search ADS WorldCat Johnston R. , Brady H. E. ( 2002 ). The rolling cross-section design . Electoral Studies , 21 , 283 – 295 . Google Scholar Crossref Search ADS WorldCat Kenski K. , Jamieson K. H. ( 2011 ). Presidential and vice presidential debates in 2008: A profile of audience composition . American Behavioral Scientist , 55 , 307 – 324 . Google Scholar Crossref Search ADS WorldCat Kenski K. , Stroud N. J. ( 2005 ). Who watches presidential debates? A comparative analysis of presidential debate viewing in 2000 and 2004 . American Behavioral Scientist , 49 , 213 – 228 . Google Scholar Crossref Search ADS WorldCat Knobloch-Westerwick S. ( 2008 ). Informational utility. In Donsbach W. (Ed.), International encyclopedia of communication (pp. 2273 – 2276 ). Oxford : Basil Blackwell . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Knobloch-Westerwick S. ( 2014 ). Choice and preference in media use: Advances in selective exposure theory and research . Abingdon : Routledge . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Knobloch-Westerwick S. , Kleinman S. B. ( 2012 ) Preelection selective exposure: Confirmation bias versus information utility . Communication Research , 39 , 170 – 193 . Google Scholar Crossref Search ADS WorldCat Kritzinger S. , Johann D., Glantschnigg C., Aichholzer J., Glinitzer K., Thomas K., Wagner M., Zeglovits E. ( 2014a ): AUTNES TV Debates Panel Study 2013. GESIS Data Archive, Cologne. ZA5858 Data file Version 1.0.0. doi:10.4232/1.11951 Kritzinger S. , Johann D., Aichholzer J., Glinitzer K., Glantschnigg C., Thomas K., Wagner M., Zeglovits E. ( 2014b ): AUTNES Rolling-Cross-Section Panel Study 2013. GESIS Data Archive, Cologne. ZA5857 Data file Version 1.0.0. doi:10.4232/1.11950 Lang K. , Lang G. E. ( 1977 ). Reactions from viewers. In Kraus S. (Ed.), The great debates: Kennedy vs. Nixon, 1960, a reissue (pp. 313 – 330) . Bloomington : Indiana University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Lazarsfeld P. F. , Berelson B., Gaudet H. ( 1944 ). The people’s choice . New York, NY : Columbia University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Prior M. ( 2007 ). Post-broadcast democracy: How media choice increases inequality in political involvement and polarizes elections . Cambridge : Cambridge University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Prior M. ( 2012 ). Who watches presidential debates? Measurement problems in campaign effects research . Public Opinion Quarterly , 76 , 350 – 363 . Google Scholar Crossref Search ADS WorldCat Sears D. O. ( 1968 ). The paradox of de facto selective exposure without preference for supportive information. In Abelson R., Aronson E., McGuire W. J., Newcomb T. M., Rosenberg M. J., Tannenbaum H. (Eds.), Theories of cognitive consistency: A sourcebook (pp. 777 – 787 ). Chicago, IL : Rand McNally . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Stroud N. J. ( 2011 ). Niche news: The politics of news choice . Oxford : Oxford University Press . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Taber C. S. , Lodge M. ( 2006 ). Motivated skepticism in the evaluation of political beliefs . American Journal of Political Science , 50 , 755 – 769 . Google Scholar Crossref Search ADS WorldCat Utych S. M., , Kam C. ( 2014 ). Viability, information seeking, and vote choice . Journal of Politics , 76 , 152 – 166 . Google Scholar Crossref Search ADS WorldCat © The Author 2016. Published by Oxford University Press on behalf of The World Association for Public Opinion Research. All rights reserved.

Journal

International Journal of Public Opinion ResearchOxford University Press

Published: Dec 1, 2017

There are no references for this article.