Opportunity Structures for Selective Exposure

The transition from low-choice to high-choice media environments has raised new concerns about selective exposure. In this context, two types of selective media exposure are relevant. One is selectivity based on political ideological preferences, the other selectivity based on political interest. Evidence for both has been found primarily in an American context, while there is less research on European countries. This is problematic, as the opportunity structures for different forms of selectivity vary across media environments. Against this background, the purpose of this study is to investigate the two types of selective exposure in a country—Sweden—where the opportunity structures for selective exposure differ from the American context. This study investigates both types of selective exposure in relation to televised party-leader interviews. Based on panel survey data, the findings show that selective exposure based on political interest is substantially more important than selective exposure based on ideological preferences in explaining exposure to party-leader interviews. To substantiate this finding, the results are replicated with partisan learning as the dependent variable.

The transition from low-choice to high-choice media environments has fundamentally reshaped contemporary political information environments and drastically expanded the supply of all kinds of information. As a consequence, media consumers have more opportunities than ever to select and to avoid media content based on their own personal interests and preferences (Bennett and Iyengar 2008). These rich opportunity structures, in turn, may increase the importance of people's motivations and abilities when deciding what media and media content they expose themselves to (Luskin 1990;Prior 2007).
One key motivation to select media content is rooted in people's ideological leanings. Several scholars have argued that when people can choose among a variety of sources, they tend to select content that supports their political attitudes and beliefs while avoiding opinion-challenging information (Bennett and Iyengar 2008;Mutz and Martin 2001).
Another key motivation is interest in politics (Prior 2007;Strömbäck and Shehata 2010). Not only has the proliferation of media made it more easy to find attitudeconsistent or avoid attitude-discrepant information, but it has also made it easier to find or avoid political information altogether. The growing media supply has made political interest a more important predictor of news media use, and some studies suggest an increasing gap between news seekers and news avoiders in terms of their news media consumption (Ksiazek et al. 2010;. Both kinds of motivated selectivity present potential challenges to democracy, either by an increased polarization of political views and a lack of common ground for democratic talk, or by widening knowledge gaps. From a democratic point of view, it can be argued that it is essential that people are exposed to political information and opposing perspectives and viewpoints, as it tends to promote political tolerance, more careful information search, and political knowledge (Delli Carpini et al. 2004;Mutz 2006;Stroud 2010Stroud , 2011Sunstein 2007).
Although the evidence for selective exposure is not entirely consistent (Mutz and Young 2011), American studies have provided compelling evidence for selective exposure based on both ideological preferences and political interest (Arceneaux and Johnson 2013;Iyengar and Hahn 2009;Prior 2007). Thus far, there is, however, only limited research on selective exposure beyond the American context. This is problematic, not least as the opportunity structures for selective exposure vary across media environments. Thus, it cannot be assumed that evidence of selective exposure found in the United States can be generalized to other countries, for example, in Europe. This holds particularly true with respect to television, where in most European countries, there are strong public service broadcasting outlets, characterized by norms of impartiality and internal pluralism but no partisan channels equivalent to FOX News or MSNBC (Esser et al. 2012;Hallin and Mancini 2004).
Against this background and focusing on television, the purpose of this study is to investigate the presence of selective exposure based on ideological preferences and political interest in a country-Sweden-where the opportunity structures provide plenty of scope for selectivity based on political interest but a narrow scope for selectivity based on ideological preferences. More specifically, based on a panel study during the 2010 Swedish national election campaign, we analyze both forms of selectivity using exposure to televised party-leader interviews as well as partisan learning as outcome variables. By doing this, we also seek to highlight the theoretical importance of opportunity structures for selective exposure for a full understanding of selective exposure in contemporary media environments. The main reason to focus on television is that it remains one of the media sources where most people get their information about politics (Shehata and Strömbäck 2014;Strömbäck 2016).

Two Types of Selective Exposure and the Influence of Media System Characteristics
As noted above, the transformation from low-to high-choice media environments has prompted a concern for selective exposure to political information. In essence, growing media supply is said to increase the importance of personal motivations as predictors of what information people consume (Prior 2007). In Changing Minds or Changing Channels, Arceneaux and Johnson (2013: 52) define motivation "broadly as any goal-directed preference regarding what to watch on television." In other words, increasing media supply makes media consumers more likely to select content in line with their preferences (Bennett and Iyengar 2008;Iyengar and Hahn 2009;Mutz and Young 2011;Prior 2007).
In this context, there are basically two types of motivations that might lead to selective exposure. The first is rooted in people's political interest. Increasing supply enables people with low interest in politics to consume media while avoiding political information, whereas people with stronger interest can consume even more political information. In the United States, this gap in news media consumption between news avoiders and news seekers has been shown to increase the gap in political knowledge and turnout (Prior 2007). The other type of motivation is rooted in people's political or ideological preferences. Here, theory suggests that people prefer being exposed to arguments that are in line with their attitudes or beliefs, rather than to arguments running counter to them (Frey 1986;Lodge and Taber 2013). This leads to a tendency to select information or media sources that people expect will meet such demands (Knobloch-Westerwick and Meng 2009;Mutz 2006;Stroud 2008). This argument implies an ideological gap in exposure to political information.
The growing interest in ideological selective exposure is largely driven by the U.S. experience of increasing polarization in terms of both media and politics (Bennett and Iyengar 2008;Iyengar and Hahn 2009;Mutz and Young 2011;Stroud 2008Stroud , 2011. Apart from the fact that the Internet has introduced an almost infinite number of potential information sources for citizens in many Western democracies, a significant change in the American media environment is the political polarization of the broadcasting system, in particular, with the introduction of FOX News and MSNBC. In the United States, studies have also produced empirical evidence for ideological selective exposure (Iyengar and Hahn 2009; Pew Research Center for the People & the Press 2010). Stroud (2011), for example, finds substantial ideological selectivity of specific content on television (Democratic and Republican presidential nomination acceptance speeches) as well as of politically biased news outlets (newspapers, radio, and television networks).
While the debate on selective exposure is primarily based on the U.S. experience, important to note is that the opportunity structures for selective exposure vary across countries. By opportunity structures for selective exposure, we refer to the availability of different media, media formats, media genres, and media content, and the ease with which citizens can select media and media content based on their personal preferences. In some media environments, it is easier for citizens to find content that matches their preferences, for example, in terms of ideological leaning, while it is more difficult in other media environments. This holds true both on an aggregate level as well as with respect to different media types such as newspapers, television, and the Internet.
We believe the concept of opportunity structures for selective exposure is important, not least considering research indicating that media use is influenced by the media environment in which people live, alongside individual abilities and motivations (Aalberg et al. 2013;Althaus et al. 2009;Curran et al. 2009;Goldman and Mutz 2011;Prior 2007;Shehata and Strömbäck 2011). Similar to other behaviors, media use is shaped by opportunities as well as preferences and abilities (Luskin 1990). Thereby, differences in opportunities for selective exposure across media environments should have an impact on the presence and ubiquity of selective exposure in countries with different media environments.
Although the growing popularity of the Internet and the multiplication of cable channels are more or less similar across media systems, the situation is quite different with respect to terrestrial broadcasting. Here, the strong position of public service broadcasting in many European countries has (thus far) prevented a political polarization of the broadcasting system similar to what has emerged in the United States. In many countries, public broadcasters hold substantial market shares. They are also obliged to work in accordance with norms of impartiality, objectivity, and political neutrality (Esser et al. 2012;Hallin and Mancini 2004). Next to the public service stations, a number of international cable channels offer pure entertainment programming and provide people the chance to avoid news and political information altogether in favor of entertainment. Comparing media environments in the United States and in European countries with strong public service suggests that the opportunity structures for selectivity based on political interest are great across media environments, but more narrow when it comes to ideological selectivity in European countries. Therefore, to understand the role of selective exposure, it is essential to broaden research to include European countries with different opportunity structures for selective exposure than the United States. One such country is Sweden, a typical example of what Hallin and Mancini (2004) has identified as the democratic corporatist model of media and politics. Other countries belonging to this model are, for example, Austria, Germany, and the other Nordic countries.

Selective Exposure Opportunities in the Swedish Broadcasting System
In this study, we will focus on selective exposure with respect to television. As mentioned earlier, one key reason is that television remains one of the most important sources of information for most people. As a typical democratic corporatist country, the Swedish broadcasting system is dominated by two public service channels (SVT1 and SVT2) and one commercial broadcaster (TV4). TV4 is also restricted by some public service obligations. Although there is a commercial cable television market, in terms of news and current affairs, SVT1, SVT2, and TV4 dominate and attract broad segments of the population (Ohlsson 2015). Together, these three channels provide a rich opportunity structure for political information.
In contrast with the United States, there are no partisan television channels in Sweden. Thus, it makes little sense to study ideological divides in the audience of different television channels. The lack of partisan channels does not, however, prevent programming that is partisan in nature. One prime example is the institutionalized party-leader interviews that are televised prior to each national election in Sweden. In this study, and inspired by Stroud's (2011) study on Democratic and Republican presidential nomination acceptance speeches, we thus focus on exposure to these partyleader interviews. In arguing why she studies exposure to presidential nomination acceptance speeches, Stroud (2011, 42) stresses the popularity and the blatant partisanship of the speeches as reasons to expect ideological selective exposure. In comparative terms, the televised party-leader interviews in Sweden are equivalent to the acceptance speeches in terms of being central to Swedish national election campaigns and reaching wide audiences (Esaiasson and Håkansson 2013;Petersson et al. 2006;. Moreover, they are equivalent in the sense that a television viewer can easily anticipate each interview to be either attitude-consistent or attitude-discrepant.
The party-leader interviews are broadcasted live during the final weeks of the Swedish election campaign. A pair of experienced professional journalists interviews each leader of the parties having seats in parliament for a full hour. The interviews are broadcasted live on prime time by one of the public service television channels, thereby providing a unique opportunity for parties to get their message across unedited and without having to face political opponents. Furthermore, as people know which party leader is being interviewed, they have every chance to select or avoid these interviews based on their political interest or ideological preferences.
With respect to the Swedish political system, it is essential to know that Sweden had seven parties in parliament at the time, but also that these parties formed two blocs, one Center-Left and one Center-Right. In the election, the parties in the Center-Right bloc-who had been in government since 2006-won about 49 percent of the votes, while the Center-Left bloc won about 44 percent of the votes. An eighth partythe Sweden Democrats-also entered parliament after having received 5.7 percent of the votes.

Hypotheses: The Role of General Political Interest and Ideological Preferences
Turning to our hypotheses, and based on the notion of different opportunity structures for selective exposure, our basic argument is that the Swedish broadcasting system provides more choice opportunities at the genre level than at the ideological level. As a consequence, we expect that political interest will be more important than ideological leaning in explaining exposure to party-leader interviews. Viewing decisions at the genre level refer to choices between watching news and current affairs, entertainment, sports, movies, sitcoms, and so on, while choices at the ideological level refer to the political leaning of a television program or channel (liberal, conservative, right wing or left wing, etc.; Webster 2014).
With respect to party-leader interviews, people with divergent political beliefs have the chance to either actively change channel or turn the television off to avoid these interviews, while citizens who share the political views of a particular party can tune in. Following Stroud's findings concerning ideological selective exposure to presidential acceptance speeches, we should expect some ideologically driven selective exposure to the party-leader interviews in Sweden as well. However, previous research suggests that television viewing is driven at least as much by habits and routines as by personal preferences (Diddi and LaRose 2006;LaRose 2010;Rosenstein and Grant 1997;Webster 2014;Wonneberger et al. 2011). Following Mutz and Young (2011), ideological selective exposure should thus be affected by habitual television viewing or channel loyalty. In a polarized broadcasting system such as the American, the opportunity structure enables citizens to develop viewing habits based on ideological preferences, resulting in what Mutz and Young term passive selective exposure. In contrast, the substantial audience share for public service broadcasting in Sweden means that a significant part of the habitual television viewing will take place on the main public service channels (Aalberg and Curran 2012;Shehata et al. 2015). Thus, the lack of opportunities for ideological selective exposure at the channel level combined with the importance of viewing habits can be expected to reduce ideological selective exposure to specific political content.
Despite this, we still expect to find some ideological selective exposure to the party-leader interviews. The main gap in exposure to Swedish party-leader interviews, we argue, is however not between citizens with different ideological leanings but between people with an ideological preference-either left-leaning or right-leaningand those without. This argument is based on two propositions. First, on the individual level, previous studies indicate that people with ideological preferences are generally more politically active, interested, and knowledgeable than people who are neither left-nor right-leaning (Curran et al. 2012;Oscarsson and Holmberg 2008). Therefore, they are also more likely to watch the party-leader interviews. Second, while the Swedish broadcasting system provides limited opportunities to select television channels based on ideological orientations, there are a large number of national and international cable channels that broadcast entertainment, movies, or television shows rather than politics and current affairs. This makes it significantly easier to select media content based on general interest in politics than on ideological preferences. Thus, while there are ample opportunities to seek out nonpolitical television programs for those uninterested in politics, people who are equally interested in politics but divided ideologically are dependent on the same channels for getting televised campaign information.
In this context, it is important to note that selecting information in line with one's attitudes does not necessarily imply avoidance of attitude-discrepant information. In fact, people have several reasons to expose themselves to attitude-discrepant information: It can be emotionally rewarding to argue against such information and reject it; such information can help people prepare a defense for their own position; or it could be useful to learn about attitude-discrepant information to reexamine one's own position (Garrett et al. 2013). However, it requires a certain degree of political interest to engage in such cognitive-demanding activities. Consequentially, those politically interested are more prone to embrace both attitude-discrepant as well as attitude-consistent information, and this general motivation is likely to be stronger among people with an ideological leaning than among those without.
Based on these arguments, we expect to find some ideological selective exposure to the party-leader interviews in Sweden, but also that political interest is more important than ideological leanings in explaining why people watch interviews. Our hypotheses, thus, are as follows: Hypothesis 1a: People with a right-leaning ideology are more likely to watch party-leader interviews with right-wing parties, while people with a left-leaning ideology are more likely to watch party-leader interviews with left-wing parties. Hypothesis 1b: People with an ideological leaning are more likely than those who are neither left-nor right-leaning to watch party-leader interviews, independently of the direction of their ideological leanings. Hypothesis 1c: General political interest is more important than ideological preferences for explaining why people watch party-leader interviews.
To extend our analysis and assess the robustness of our findings, we will also test our hypotheses in relation to partisan learning. If increased media supply results in less cross-cutting media exposure, it could lead to low awareness of attitude-inconsistent information, compared with the awareness of attitude-consistent information. Thus, by partisan learning, we refer to one-sided campaign learning, that is, learning about the proposals of parties with whom voters agree ideologically. Partisan learning should thereby be contrasted with general campaign learning, which reflects learning about parties from all political camps.
Intuitively, it would make sense that people remember information with which they agree better than information with which they disagree. However, this seems not to be the case (Eagly et al. 1999). The tendency to counterargue attitude-discrepant messages is part of the mechanism that makes counterattitudinal messages as memorable as pro-attitudinal messages (Eagly et al. 2000). A similar point is made by Taber and Lodge (2006), who show that people engage more in denigrating arguments they disagree with than bolster arguments they agree with. Doing so requires a certain level of motivation and political interest, however. Therefore, we expect to find a similar pattern for partisan learning as for exposure to party-leader interviews.

Hypothesis 2a:
People with a right-leaning ideology are more likely to learn about the proposals of right-wing parties, while people with a left-leaning ideology are more likely to learn about the proposals of left-wing parties, during the election campaign. Hypothesis 2b: People with an ideological leaning are more likely than those who are neither left-nor right-leaning to learn about the proposals of all political parties, independently of the direction of their ideological leaning. Hypothesis 2c: General political interest is more important than ideological preferences for campaign learning.

Data and Method
To investigate the hypotheses above, we will rely on a four-wave panel study conducted during the 2010 Swedish national election campaign. Compared with crosssectional surveys, using panel data provides the opportunity to study how ideological preferences and motivations formed already prior to the election campaign influence information selectivity during the campaign, that is, whether citizens with certain ideological orientations measured months before Election Day are more or less likely to subsequently expose themselves to attitude-consistent and inconsistent media content. Thus, by analyzing these dynamics over time, the panel design provides substantially better opportunities to capture the causal effect of ideological preferences on selectivity and retention (Finkel 1995). The panel survey was conducted by the Centre for Political Communication Research at Mid Sweden University in cooperation with the polling institute Synovate in Sweden.
The sample was drawn using stratified probability sampling from a database of approximately twenty-eight thousand citizens from Synovate's pool of Web survey participants. Those included in this pool are recruited continuously using both random digit dialing and mail surveys based on random probability samples. Approximately 5 percent of those who are initially contacted and invited agree to be part of this pool of respondents. The pool of Web survey participants covers different segments of the population in terms of, for example, residence, age, education, and occupation.
The probability sample of 4,760 respondents aged eighteen to seventy-four from this pool was stratified by gender, age, county size, political interest, and Internet use, so as to be as representative of the Swedish population aged eighteen to seventy-four years as possible. Among these, we base our analyses on those 4,010 respondents who were invited to participate in all waves of the panel. These respondents were asked to complete a Web-based survey at four times during a period of five months leading up to the election. Wave 1 of the panel took place in May (May 3−May 20), wave 2 in mid-June (June 14−June 23), wave 3 in mid-August (August 16−23), and, finally, wave 4 immediately after Election Day (September 20−September 27). The total cooperation rates were 63 percent in wave 1, 48 percent in wave 2, 43 percent in wave 3, and 39 percent in wave 4. However, not everyone participated in all waves, but a total number of 1,413 respondents did so-amounting to a 35 percent of all respondents initially contacted. While the overall response rates are relatively high given common problems of panel attrition, the sampling procedure and final cooperation rates certainly raise issues concerning external validity. Compared with available population statistics, the final sample is broadly representative with regard to sex, age, and geography (county size), but weaker in terms of education (47 percent with higher education in the sample, compared with 29 percent among the population). Compared with cross-sectional surveys based on traditional national probability samples, the panel respondents are also more interested in politics (63 percent compared with 52 percent) and more frequent users of the Internet (92 percent compared with 78 percent). Although the primary strength with panel data is the opportunity to analyze changes in communication and opinions over time-and, more specifically, to use measures of personal motivations (interest and ideological leaning) collected prior to the campaign as predictors of viewing behaviors during the campaign-we will base descriptive statistics on weighted data.

Measures
Our key variables in the present study are ideological orientation, ideological selective exposure, and partisan learning. While ideological orientation was measured in the first panel wave, both selective exposure and learning were tapped in the following waves-waves 2, 3, and 4. In addition, we use several control variables to identify the unique effect of ideological orientation on ideological exposure and partisan learning.
Ideological orientation was measured prior to the election campaign based on a standard Left-Right ideological scale. Respondents were asked to identify their orientation on an 11-point scale from 0 (clearly to the Left) to 10 (clearly to the Right). A three-level categorical variable was created by distinguishing respondents with a leftleaning orientation (0-4 on the original 11-point scale), from those lacking a clear ideological orientation (5 on the original 11-point scale), as well as respondents with a right-leaning orientation (6-10 on the original 11-point scale)-creating three groups of left-leaning (n = 1,000), center (n = 566), and right-leaning (n = 1,135) citizens.
As discussed above, the presence of selective exposure will be analyzed in relation to televised party-leader interviews. Exposure to party-leader interviews broadcasted on the main public service channels in the final weeks of the election campaign was measured in the fourth panel wave, based on a battery of survey items asking whether respondents watched each of the seven one-hour-long party-leader interviews. For each party-leader interview, the response categories ranged from 0 (no) to 1 (yes, partly) and 2 (yes, the entire interview). Apart from analyzing each of these items separately, we also computed two additive scales: (1) exposure to left-wing partyleader interviews (range = 0-6, M = 1.95, SD = 1.98), and (2) exposure to right-wing party-leader interviews (range = 0-8, M = 2.34, SD = 2.53).
To measure learning during the campaign, the panel survey included several items tapping campaign knowledge, that is, awareness of events taking place and policy proposals presented during the election campaign-measured in waves 2, 3, and 4. To capture potential partisan learning, we focus here on knowledge about specific policy proposals presented by the parties during the campaign. For instance, respondents were asked questions such as, "Which of the following proposals were presented by the left-wing parties in their joint party manifesto?" Five response categories, including don't know, were given for each knowledge question to minimize the chance of randomly guessing the correct answer. A time limit of 20 seconds for answering each question was used to avoid Web searches for the correct answers. For each knowledge question, respondents who gave a correct answer were given the value 1, while incorrect and don't know answers were coded 0. We identified nine items focusing on knowledge about the left-wing parties, and six items about the right-wing parties-creating two separate learning scales ranging from 0 to 9 (left-wing party learning) and 0 to 5 (right-wing party learning). Based on these items, we constructed a (1) left-wing party knowledge index (range = 0-9, M = 3.47, SD = 2.11), as well as a (2) right-wing party knowledge index (range = 0-5, M = 2.55, SD = 1.28).
In addition to these focal variables, the panel survey also included a number of key control variables such as age, gender, education, and income. Most important, however, given our focus on general political motivations and resources, the analyses will account for individual differences in political interest and general political knowledge. Political interest was measured in the first panel wave (t − 4) based on two four-level items focusing on respondents' interest in (1) politics, as well as in (2) the election campaign (Pearson's r = .73). The two items were summed to form a political interest index ranging from 0 (no interest) to 6 (strong interest). General political knowledge-representing the stock of political knowledge that citizens brought with them at the beginning of the election campaign-was also measured in wave 1 (t − 4). This is an additive index based on eight political knowledge questions focused on personalities (three items), political processes (three items), and issue positions (two items)-with acceptable reliability levels (Kuder-Richardson = .72). Table 1 presents some basic descriptive statistics covering initial differences between left-leaning, center, and right-leaning voters at the beginning of the election campaign. As can be seen, there are no major differences with respect to age between these groups, indicating that fundamental political orientations do not follow an age pattern. This is not the case for political motivation and resource variables, however. While right-leaning citizens score slightly higher than left-leaners on both news attention and general political knowledge, their interest in politics is fairly equal. The major gap, however, is found between people who are neither right-nor left-leaning and those who are either right-or left-leaning. Citizens lacking an ideological leaning are substantively less interested in politics, pay less attention to political news in traditional media, and score significantly lower on political knowledge, than do both left-wing and right-wing citizens.

Ideological Selective Exposure
The research question at the heart of this study concerns selective exposure, that is, the extent to which Left-Right orientations as well as general political interest influence (1) what media content citizens turn to, as well as (2) what political information they acquire during the election campaign. We start confronting the exposure dimension in Figure 1, which displays the relationship between Left-Right orientations and exposure to televised party-leader interviews. As can be seen, there is a consistent relationship between ideological orientations expressed prior to the campaign, and exposure to televised party-leader interviews in the final weeks of the election campaign. Citizens considering themselves as left-leaning are more likely to watch interviews  with party leaders representing the left-wing parties-the Left Party, the Social Democrats, and the Green Party-than are right-leaning citizens. The gap amounts to approximately 6 to 12 percentage points between the two ideological camps. A similar pattern is found for exposure to right-wing party-leader interviews. These are primarily watched by citizens with a right-leaning orientation, and this time, the Left-Right gap amounts to approximately 6 to 15 percentage points. Furthermore, there is another striking pattern in Figure 1: Those who watch partyleader interviews to the least extent are citizens lacking a clear political Left-Right orientation, suggesting again that the primary gap in political information exposure is not due to ideological but to more general motivation factors. It is not the ideological divide that matters, but rather whether citizens possess ideological convictions or not, which is likely to be related to differences in interest, attention, and general knowledge, as documented above.
So far, the results presented have been descriptive-revealing a consistent pattern of ideological selective exposure to televised interviews. However, even though the findings indicate that ideological orientations are related to what partisan content citizens are exposed to-and that citizens seem to prefer attitude-consistent information-additional individuallevel analyses suggest that these types of media use are far from ideologically divided. The bivariate correlation between watching party-leader interviews of the left-wing parties on one hand, and of the right-wing parties on the other hand, is very strong (Pearson's r = .80). This again suggests that general political motivations-or viewing habits-rather than ideological motivations drive television exposure during the election campaign.
In Table 2, we present a more critical test of the ideological selective exposure hypothesis by analyzing whether there are independent effects of Left-Right orientations on television exposure, controlling for a host of socioeconomic background, political resource (knowledge), and motivation (interest) variables. Here, we use leftleaning respondents as the reference group when estimating the effect of ideological orientation. Furthermore, we introduce general political interest and knowledge variables sequentially in the analysis. 1 The patterns unraveled in Table 2 are revealing. First, model 1 includes ideological orientation and background variables only, and the results confirm previous bivariate findings indicating the presence of ideological selectivity. They also indicate that voters who are neither left-nor right-leaning are the least exposed to televised interviews irrespective of who is interviewed. Second, however, this "curvilinear" pattern disappears when political interest and knowledge are added to the regression in model 2. Once these variables are added, ideological orientation has a more linear impact on exposure to party-leader interviews: both center (b = −.28, p < .05) and right-leaning (b = −.42, p < .001) voters are significantly less likely to watch interviews with leftwing party leaders, while center (b = .55, p < .01) and right-leaning (b = .99, p < .001) voters are more likely to watch interviews with right-wing leaders-compared with citizens with a left-leaning ideological orientation. It is also evident that political interest has a much stronger and consistent effect on watching party-leader interviews than general political knowledge. Together, these two variables increase the amount of explained variance by approximately 16 to 17 percentage points (adjusted R 2 increases from 10 percent to 27 percent and from 12 percent to 28 percent when these variables are added).
The importance of political interest as a predictor of exposure to party-leader interviews compared with ideological orientation can be further tested in two ways. First, what is the relative importance of each variable in terms of overall contribution to the model? By comparing the change in R 2 when both variables are excluded from the full model (model 2), one at a time, their individual contribution can be assessed. Dropping ideological orientation (the two dummy variables) from model 2 yields a decrease in adjusted R 2 of approximately 1-3 percentage points. Excluding political interest, however, results in a drop in adjusted R 2 of approximately 12 to 13 percentage points. Second, what is the substantive effect on exposure to party-leader interviews resulting from a change in ideological orientation as well as in political interest? One way to assess this is to look at the maximum possible effects. For instance, citizens with a right-leaning ideological orientation score, on average, 0.42 points lower on the exposure to left-wing party-leader interviews scale compared with left-wing voters, while the maximum effect of political interest is 3.42 on the exposure scale (6 × 0.57). Similarly, while having a right-wing leaning increases exposure to right-wing partyleader interviews by 0.99, the maximum effect of interest is 4.32 on the exposure scale (6 × 0.72). Although such comparisons are not straightforward, they illustrate the relative importance of general political interest as a motivation behind party-leader interviews. 2

Partisan Learning
As discussed above, one of the key concerns frequently raised is that growing opportunities for media choice will not only elevate the impact of personal motivations and preferences on what media content citizens are exposed to, but also what information they acquire. In terms of ideological selectivity, growing supply is assumed to result in less cross-cutting media exposure and, thereby, decreasing awareness of attitudeinconsistent information. Above, we found that while ideological selective exposure does occur, citizens who are exposed to partisan media content about parties with their preferred ideological leaning are also likely to be exposed to information about the nonpreferred parties. Thus, in this regard, a more general political motivation seems to be what distinguishes people who in terms of ideology are neither right-nor leftleaning and those who are either right-or left-leaning. The question is whether the same pattern can be found with respect to partisan learning, that is, the extent to which citizens primarily learn about the policies and proposals of their own favored parties during the campaign. Figure 2 gives a first glimpse at this issue by illustrating partisan learning among citizens based on their Left-Right orientations. The findings reveal very little evidence of extensive partisan learning. While right-leaning citizens are slightly more likely to learn about the activities of the right-wing parties, they are also more likely to learn about the left-wing parties than left-leaning voters are. Again, however, the most striking learning gap is not found between partisans of different ideological camps, but between people with no clear ideological leaning and those who are either right-or left-leaning. Citizens lacking a clear left-right orientation score lowest on both left-wing party knowledge (21 percent correct answers) and right-wing party knowledge (36 percent correct), compared with the left-leaning (31 and 47 percent) and right-leaning (38 and 51 percent correct) voters, respectively.
In Table 3, we conduct a more critical test of the partisan learning hypothesis based on a series of regression models predicting both left-wing and right-wing partisan learning. For both types of learning, two models are estimated to see how general political knowledge and interest influence the effect of ideological orientations on learning. Focusing on left-wing partisan learning, model 1 confirms the findings displayed in Figure 2, even when controlling for several background characteristics: Compared with citizens who are neither right-nor left-leaning (the reference category), both left-wing and right-wing citizens learn significantly more about the activities of the left-wing parties during the campaign. Once we include political interest and general political knowledge, however (model 2), this effect of ideological orientation is substantially reduced, while the important role of general motivations and knowledge is reflected both in their highly significant coefficients as well as the dramatic increase in R 2 when these variables are included (from .19 to .46). A very similar pattern is found for rightwing party knowledge. People who are neither left-nor right-leaning learn significantly less than people with an ideological leaning, irrespective of direction. But once interest and general political knowledge are included, these effects disappear.

Conclusion and Discussion
While most research on selective exposure has focused on the United States, in this study we extended research to a country that in many respects differs significantly from the United States in terms of the opportunity structures for selective exposure. This, we argue, is important as there is ample evidence that media use is influenced by media environments as well as by individual-level variables (Aalberg et al. 2013;Althaus et al. 2009;Curran et al. 2009;Goldman and Mutz 2011;Shehata and Strömbäck 2011). Because the Swedish broadcasting system provides better opportunity structures at the genre level than at the ideological level, our expectation was that political interest would be more important than ideological leanings in explaining exposure to televised party interviews, without eradicating the importance of ideological leanings.
To briefly summarize, the results show support for several of our hypotheses. First, people with a right-leaning ideology were more likely to watch party-leader interviews with right-wing parties, while people with a left-leaning ideology were more likely to watch party-leader interviews with left-wing parties (H1a). These ideological selectivity effects-showing that citizens are more likely to seek out attitude-consistent than attitude-discrepant information-held up even when controlling for a range of background, political motivation, and resource factors. Second, however, the results indicate that basic political interest is substantially more important than ideological leaning in explaining why people watch party-leader interviews (H1c). Thus, rather than being driven by ideological preferences, exposure to party-leader interviews is primarily dependent on other types of motivations that differentiate people who are neither rightnor left-leaning from those who are either right-or left-leaning (H1b). According to the data presented here and elsewhere (Oscarsson and Holmberg 2008), people lacking an ideological orientation are less interested in politics, pay less attention to politics in traditional news media, and are less knowledgeable about politics. Finally, these patterns were also replicated using partisan learning instead of exposure as the dependent variable. That is, we found very little evidence of partisan learning occurring at all (H2a). Instead, gaps in learning emerged between people who are neither right-nor leftleaning and those who are either right-or left-leaning, and these gaps disappeared when accounting for political interest and general political knowledge (H2b and H2c).
Beyond these empirical findings, the main theoretical contribution of this study relates to the notion of opportunity structures for selective exposure. Defined as the availability of different media, media formats, media genres, and media content and the ease to which citizens can select media and media content based on their personal preferences, the concept of opportunity structures for selective exposure has several implications for theory and research on selective exposure. First, it is a reminder that empirical findings of selective exposure from any particular media environment and country, with its specific opportunity structure for selective exposure, cannot be transferred to media environments and countries with other opportunity structures for selective exposure. Second, it highlights and offers a framework for further research on how opportunitiesa macro-level variable-influence and moderate the importance of individual preferences and abilities-micro-level variables. As part of this, it highlights the importance of distinguishing between different forms of selectivity in terms of what the opportunity structures look like and in terms of its presence and ubiquity. Third and related, it offers a framework for studies investigating differences in opportunity structures for selective exposure across different countries, media environments, media types, and genres. With increasing media supply follows increasing selectivity, and media environments across democracies have all transformed from low-to high-choice media environments. However, the antecedents as well as the prevalence and consequences of selectivity might vary depending on the opportunity structures for different forms of selectivity. Fourth and related to the findings of this study, it suggests that media policy and policies aimed at strengthening public service broadcasting might offer a means toward counteracting selective exposure based on political or ideological preferences. This, of course, assumes that limited opportunity structures for selectivity with respect to broadcasting does not create incentives for those with strong preferences to migrate to other media where the opportunity structures are more conducive to selectivity based on political or ideological preferences or on political interest. This highlights the importance of understanding how different opportunity structures for selectivity with respect to different media are linked to each other and people's media use.
As this study is a single-country study, a key question, though, is how far the results can be generalized. While ultimately an empirical question, based on our reasoning above, our best estimate is that the findings can be generalized to countries with similar opportunity structures for selective exposure. What matters is not the country per se, but the opportunity structures for different types of selective exposure. Concerning more everyday coverage of political affairs, this also means that it is likely the case that ideological selective exposure to television is even more limited than suggested by the results in this study, as there is usually less partisan programming on television than during election campaigns. This should, however, not limit selective exposure based on political interest. Again, the key is the opportunity structures for different forms of selective exposure.
In essence, if we want to understand selective exposure across contemporary media environments and countries, we must understand the different opportunity structures for different types of selective exposure. We cannot think of or study selective exposure as a matter of individual preferences and traits only.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was partly funded by the research project "Changing Media Environments, Changing Democracies," supported by the Axel and Margaret Ax:son Johnson Foundation.

Notes
1. Given the character of our dependent variables, we also estimated all models using ordered logit regression. The substantive results were, however, very similar in terms of effects and statistical significance.
2. For instance, while our political interest measure is a continuous scale capturing the strength of this motivation, ideological orientation is a categorical variable that does not take ideological strength into account. Thus, going from a minimum to a maximum value on these two measures has different implications substantively.