An Information Resource Model of Agenda-Setting

Transcription

1 An Information Resource Model of Agenda-Setting Rune Stubager 1 Department of Political Science University of Aarhus Universitetsparken 8000 Aarhus C Denmark stubager@ps.au.dk 1 Previous version of the paper have been presented at the Nordic Political Science Association s XV meeting in Tromsø, August 2008, at the annual meeting of the American Political Science Association in Boston, MA, August 2008, and in the Department of Political Science, University of Aarhus. In addition it has benefitted from the comments of Christoffer Green-Pedersen, Stefaan Walgrave, and Rune Slothuus as well as research assistance from Peter Thorgaard. The research was supported by the Danish Social Science Research Council. Errors or omissions naturally remain my responsibility. 1

2 Abstract The question of Who Sets the Agenda? the media, politicians, the entire public or different segments of it has strong democratic implications. Agenda control, thus, is an important aspect of power. Based on a novel information resource model the paper analyses the variation in agendasetting power across issues and citizens with varying levels of political sophistication. Using timeseries methods on a unique dataset the analyses find clear evidence for an interaction between issue type and citizen sophistication. Whereas all citizens influence the media and politicians on obtrusive issues, there are strongly unequal patterns of influence on non-obtrusive issues where the most sophisticated citizens more often influence and less often are influenced by media and politicians than less sophisticated citizens. These findings have relevance for both democratic theory and agenda-setting studies. 2

3 Recent research has seen renewed interest in Dahl s (1961) famous question of Who Governs? Attention has been directed at inequalities in the representation of the views and interests of voter groups differing in income, education, and political engagement (Soroka and Wlezien 2007; Bartels 2008; Adams and Ezrow 2009) as well as at the comparable absence of such inequalities when it comes to presidential evaluations and policy mood (Enns 2007; Enns and Kellstedt 2008). However as noted by Schattschneider (1960), a crucial component of the answer to the Who Governs?-question is found in the outcome of the conflict of conflicts, i.e. the conflict over who sets the agenda (cf. Green-Pedersen 2007). In this area extant research leaves us with fewer answers. Possible inequalities in agenda-setting have only been explored in the context of the relationship between the media and the public (e.g. Zhu and Boroson 1997) or between the public and policy making (Soroka and Wlezien 2007). Thus, the most advanced agenda-setting studies that include all three core agenda-setting actors the public, the media, and politicians as well as real world events in the analysis have not distinguished between different segments of the public (Kleinnijenhuis and Rietberg 1995; Soroka 2002a; 2002b). What has been documented, on the other hand, is the contingency of agenda-setting effects (Walgrave and Van Aelst 2006; Green-Pedersen and Stubager forthcoming). Most consistently and pervasively it has been shown that the dynamics of agenda-setting differ across different types of issues (Walgrave, Soroka, and Nuytemans 2008): While the media act as agenda-setters on some issues, contrary to the prevalent media malaise perspective (Newton 2006) politicians or the public set the agenda on other issues. In other words, agenda-setting power resides with the media, politicians, as well as with the public which, so it will be argued below, derives its power from politicians need for votes and the media s need for an audience. All three of these actors set the agenda of the others, but the distribution of agenda power differs across issues. The purpose of this paper is to analyze the Who Sets the Agenda?-question, thereby illuminating the underlying Who Governs?-question, with special emphasis on inequalities across 3

4 different segments of the public. The point of departure is the development in the next section of a theoretical new model of agenda-setting that can encompass the contingencies discovered in past research; in particular issue differences and differences within the public. The core of the model presented is information: Those who have it have the potential to set the agenda of others. Clearly, the locus of information varies across issues thereby providing a theoretical underpinning of the different effects that have been found. However, the information perspective also points to a probable cause for inequalities in the agenda-setting balance of power within the public: differences in political sophistication. Those who have the information resources and analytical capacity entailed in a high level of sophistication are much better equipped to hold on to their own agenda; to reject agenda-setting, in other words. Therefore, when formulating hypotheses, in section three, the basic dynamics of the information centred model have to be re-specified with regard for the impact of both issues and different sophistication groups and the interaction of these two factors. The hypotheses are subsequently tested on a unique dataset which tracks the agendas of the media, politicians and different segments of the public in the period and thus permits a thorough evaluation by means of time-series techniques. The analyses find the expected differences across issues and also generally support the expectations regarding differences across sophistication groups. Overall, thus, the analyses document the fruitfulness of the information centred agenda-setting model. The Basic Dynamics of Agenda-Setting In large parts of the existing agenda-setting literature the mechanisms underlying the influence of one agenda on another seems undertheorized. Particularly when the public agenda appears as the dependent variable in the analysis, the influence from other agendas is seen as almost automatic thereby disregarding the possible impact of the different constitution in terms of interests and resources of the actors involved. This state of affairs is not only theoretically dissatisfying it also prevents a proper specification and examination of the agenda-setting logic. To achieve this we 4

5 need an agenda-setting model based on a set of assumptions about the core actors and their relationships. A fruitful point of departure for the development of such an agenda-setting model can be found in Baum and Potter s (2008) metaphor of the foreign policy marketplace which seems readily extendable to policy areas other than foreign policy. As pointed out by the authors (ibid., 56) the central resource in the policy marketplace is information which is seen as the primary market commodity, that which enables actors to influence policy. The actors in the policy marketplace are the three familiar: the public, politicians, and the media. The two former are fundamentally interested in creating policy that suits their interests and preferences (ibid.). For members of the public this implies that they have to stay informed about developments i.e. they have a need for orientation (cf. McCombs 2004, 53-57). A need, that is, to become knowledgeable about developments in the world that surrounds them. In other words, they have to seek out information in order to be able to influence politicians to behave in accordance with their preferences. On the other hand, information and the capacity to understand, contextualise, and analyse it also enables the public to withstand agenda-setting attempts by politicians or the media. Also for politicians there is an extra element in play in addition to the need to stay informed. Thus, in order for politicians to be able to affect policy they need to remain politicians; i.e. they need to be re-elected (cf. e.g. Kriesi 2008). The media take up a role as information traders in the policy marketplace (Baum and Potter 2008). In this respect they are placed in between the two other actors on whom they are dependent either as providers of information (the politicians) or as consumers of information (the public and politicians) (cf. also Nannestad 2005). As is clear thus, the main driving force in the policy marketplace is the demand for information; that is, the actors need information in order to be able to achieve their interests. In this way information and the capacity to handle it provides the basis for the exercise of the power that is entailed in agenda-setting: an actor s direction of the attention of other actors towards those issues that the former regard as important, thereby causing the latter to also see these issues as important 5

6 (cf. Schattschneider 1960). Precisely because of this power element in agenda-setting we should not expect individuals who have the means to resist to willingly go along with any agenda that is presented to them. This point will be developed in detail for the public below; first, I will set out the overall dynamics of the model, however. The information need of the three actors has implications for their internal relationships. Beginning with the media, we should expect the media agenda to follow that of the politicians for (at least) two reasons. First, the media depend upon politicians for information simply because their activities (and sometimes even they) are newsworthy (van Noije, Kleinnijenhuis, and Oegema 2008, 457). Thus, politicians control information that the media would like to access in order to be able to publish it. Furthermore, politicians are increasingly aware of the potential impact of the media (cf. below); hence they devote considerable resources to influencing the media so that they relay their preferred information (Brandenburg 2002, 38-39; van Noije, Kleinnijenhuis, and Oegema 2008, 457). The outcome of these dynamics is the so-called indexing effect (Bennet 1990) implying that the media index developments at the political level. On the other hand, the media agenda can also be expected to follow the agenda of the public since it is the primary media consumer and, consequently, the customers that the media have to satisfy (Uscinski 2008). This might lead the media to give the public what it wants also in terms of the issues that are put on the agenda (Nannestad 2005; Baum and Potter 2008, 50-51). Turning to the politicians, we should expect their agenda to follow that of the public for exactly the reason mentioned above: They want to be re-elected and to be that they should behave in accordance with the wishes of the public (Stimson 2004; Hobolt and Klemmensen 2005; Kriesi 2008). Politicians may keep informed about the public agenda through their use of polls and focus groups or through direct calls, letters, or s from individual citizens, but in many instances their primary source of information about the public is likely to be the media. This is an important reason, thus, for the politicians agenda to follow that of the media: Politicians regard the latter as reflecting the public agenda to which they react for the reasons just given (Walgrave 2008). A 6

7 second element that may lead us to expect that the agenda of the politicians follows that of the media is the fact that the media also provide politicians with information about a range of issues outside of their immediate horizon (Edwards and Wood 1999, ; Walgrave and Van Aelst 2006, ; Kepplinger 2007). The public agenda, finally, can be expected to follow the media agenda due to individuals basic need for orientation (McCombs 2004, 54-57). Thus, for issues that are outside of their daily experience, members of the public may depend upon the media as providers of information about the importance of the issues. One type of information relayed by the media regards the activities of politicians. For the vast majority of citizens, the media constitute the only source of information about politicians they have no direct contact to them (particularly outside of election campaigns). Citizens can be expected to pay particular attention to media reports about the agenda of politicians, however, since this agenda can serve as a cue to citizens at least those who do not have sufficient resources to form their own impression (cf. below) about the issues that are of importance in a given situation (Miller and Krosnick 2000, ). To the extent, then, that the media convey the agenda of the politicians, we might expect the public to follow it, but the relationship is mediated: Politicians do not posses the means to communicate directly to voters; they have to go through the media. 1 In addition to the inter-agenda relationships we should expect the agendas of all three actors to be influenced by real-world events. Although Dearing and Rogers (1996, 29) state that a realworld indicator is neither a necessary nor a sufficient cause for an issue to climb the agenda it might nevertheless be expected that such events frequently influence one, two, or all three of the agendas of interest here. For this reason, it is necessary to take the influence of real-world events into account when examining the relationships between the three agendas (Behr and Iyengar 1985, 40). In the present context, real world events are seen as exogenous to the three actors a point which is further substantiated when discussing the analytical model below. The theoretically possible relationships among the actors are presented in Figure 1. 7

8 FIGURE 1 ABOUT HERE At the present stage of model development the three actors are seen as unitary actors. This will allow for an overall test of the predictions of the model. However, as discussed above the precise manifestations and outcome of the processes described may vary across different kinds of individual actors. Therefore, such differences have to be taken into account in concrete analyses. It should be noted, though, that such contingencies are easily incorporated into the basic model. Expectations for different kinds of individual actors, that is, can be deduced by combining factors that are characteristic for a given individual actor with the basic interests and incentives of each of the three main actor types as discussed above. Agenda-Setting across Issues and Voter Groups As noted, the relationships among the three agendas have been found to vary. While other studies have addressed political contingencies (Green-Pedersen and Stubager forthcoming) and variations across different types of media (Walgrave, Soroka, and Nuytemans 2008) attention here is directed towards issue type and the political sophistication of the public, i.e. the extent to which an individual pays attention to politics and understands what he or she has encountered (Zaller 1992, 21). 2 In previous studies (cf. the discussion below) these two variables have to varying degrees been shown to interact with the basic agenda-setting mechanism. So far, however, these variations have not been integrated into a larger theoretical framework that can guide their interpretation and/or they have appeared inexplicable. This integration can be achieved by means of the policy marketplace model. The key is information: Put simply, the actor who possesses information about a given issue is expected to lead the others. In the following I spell out the implications of this basic logic. Obviously issues vary by their subject matter. This variation has implications for the information flows associated with different issues; i.e. for the availability of information to the three actors in the policy marketplace. A fundamental distinction pertains to the obtrusiveness of an issue 8

9 (Zucker 1978; cf. also Zhu and Boroson 1997; McCombs 2004, 60-62; Walgrave and Van Aelst 2006, 93). Obtrusive issues directly influence the everyday lives of people. This means that they are in possession of information about such issues and have clear interests in them since they have something at stake (in Soroka s (2002a, 19-22; cf. Soroka 2002b; Walgrave, Soroka, and Nuytemans 2008) terminology such issues are prominent ). Following the logic of the information resource model of agenda-setting, we should therefore expect the public agenda to lead those of the politicians and the media on such issues (cf. Newton 2006, 216). Among unobtrusive issues, Soroka (ibid.) differentiates between a sensational and a governmental type according to the concreteness and drama associated with a given issue. Governmental issues are either concrete or abstract but they are not associated with the kind of drama or sensation that makes an issue interesting to the media (cf. below). Rather, they are often technical in nature. Coupled with their unobtrusiveness, this implies that neither the media nor the public possess sufficient information about such issues to act as agenda-setters. Information, rather, rests with politicians who deal with this kind of issues as part of their law making and other activities. On such issues, therefore, politicians are able to lead the agendas of the public and the media (cf. Bartels 1996, 16-17; Walgrave and Van Aelst 2006, 94). Sensational issues, finally, are both concrete and characterised by drama and sensation. These features make such issues particularly suited for a profit- or audience-seeking media: They are easy to communicate and they can be expected to attract attention. Hence, the media can be expected to devote resources to developing such issues; i.e. to gather information about them and, subsequently, to publish this information, thereby leading the agendas of politicians and the public. These considerations give rise to the following three hypotheses: - H1: The agenda-setting power of the public is largest for obtrusive issues with which citizens have first hand experience. - H2: The agenda-setting power of the politicians is largest for governmental, i.e. technical, abstract, and undramatic, issues. 9

10 - H3: The agenda-setting power of the media is largest for sensational, i.e. concrete and dramatic, issues with which the other actors have no direct experience. The information centred model thus provides a theoretical framework for the distinctions between different types of issues that have appeared in previous agenda-setting research. In itself this is an important contribution. But the real strength of the model lies in its capacity to go beyond this level. Thus, the framework can be extended to encompass differences in the level of political sophistication across different segments of the public. In doing so, however, it should be acknowledged that the issue variation just identified may result in different agenda-setting effects for different segments of the public i.e. that there might be an interaction between the two factors. Previous research has revealed a set of seemingly incoherent (cf. Miller and Krosnick 2000, 303) results regarding the impact of political sophistication (often proxied by education, cf. below) on agenda-setting effects: McCombs (1997) refers to Shaw and Martin (1992) who show a high degree of agenda similarity across different demographic. These findings are backed by the longitudinal analyses of Zhu and Boroson (1997). However, other studies (McKuen 1984; Hill 1985) have found that agenda-setting effects are strongest among the most sophisticated segment of the public and yet others (Iyengar and Kinder 1987) have found the opposite result. A probable cause for these mixed findings is the insufficient attention paid to variations in the information context across issues (Zaller 1992, ; cf. also Takeshita 2005; Enns 2007; Enns and Kellstedt 2008). 3 That is, depending upon the difficulty of the issues i.e. how hard it is to obtain information about them and relate to them we should expect different effects of sophistication on the degree to which the public adopts new information. The key to conceptualising these differences can again be found in the information centred approach. Thus, in studies of communication effects it is common to distinguish between the processes of reception (or exposure) and acceptance (cf. e.g. Miller and Krosnick 1996). The former pertains to whether an individual receives a given piece of information while the latter pertains to whether a piece of information is accepted once received. The processes are both correlated with 10

11 sophistication although in opposite directions. Ceteris paribus, while sophistication increases reception those who are more sophisticated receive more information 4 it decreases acceptance: due to their larger stock of prior knowledge the more sophisticated are less easily persuaded by new information. In relation to attitude change, authors like Zaller (1992) have shown how both processes have to be taken into account and the same may be the case for agenda-setting. Thus political sophistication and the critical sense that is part of it may make individuals political views less malleable in general and therefore less susceptible to agenda-setting in particular (Iyengar and Kinder 1987, 55). Likewise, Zaller (1992, 121) notes that, among the more sophisticated, new information about a given issue may engender resistance because sophisticated individuals already posses large stocks of knowledge. Sophisticated individuals are not easily moved in other words; they often already know how much importance to attach to a given issue before receiving the information, entailed in e.g. the media agenda, that others find the issue important. For the less sophisticated the situation is the exact opposite. They are less attentive, have a smaller (/no) stock of prior knowledge on which to base their judgements, and posses less intellectual capacity with which to understand and relate new information to the wider picture. Therefore, they more easily fall prey to agenda-setting by media and politicians. However, the discussion above has shown that the effects may vary across different issues as these are associated with different information environments a point that has also been noted in the literature on priming (cf. e.g. McGraw and Ling 2003; Togeby 2007). Thus, when it is easy to become informed about an issue such as is the case on obtrusive issues that impact directly on people s lives i.e. the information is unmediated, the public itself is the first to know (cf. above) all members of the public can be expected to obtain this information and to accept it. Hence, everyone is potentially influenced by the development and therefore we should expect effects for all levels of sophistication. This can be formulated as hypothesis four: - H4: On obtrusive issues all sophistication levels are affected by the real world developments. 11

12 On non-obtrusive issues citizens are not directly influenced by real world developments. This means that it is not immediately clear whether an issue should be placed on the agenda or not. In such situations we should expect sophistication to play a moderating role on agenda-setting effects so that the least sophisticated individuals are most affected by the agenda of the media while the most sophisticated are affected the least. Thus when the importance of an issue is not immediately clear, the least sophisticated will have to rely on the media and politicians to make their judgements. This should cause such individuals to adopt the agenda of these other actors, i.e. to be subject to their agenda-setting. The most sophisticated, on the other hand, can rely on their knowledge and analytical capacity to form their own importance judgement. They should, hence, be least affected by the media agenda. The medium sophistication group is found in between the extremes. This gives rise to the following hypothesis: - H5: On non-obtrusive issues the agenda of citizens with lower levels of sophistication is affected more by the agenda of the media than is the agenda of those with higher levels of sophistication. Just as the different segments of the public react differently to information from politicians and the media, we might expect the latter two to react differently to different segments of the public. Unlike (all members of) the public, however, these two actors are highly sophisticated; i.e. we should expect them to receive any new information available and to be able to understand it. Given their interest in re-election, however, the politicians might not follow the agendas of all voter groups equally close. Again, the effects can be expected to depend on the information context. Thus, on issues where information is easy to obtain and relate to, politicians should follow the agenda of all voter groups since they are all informed about and respond to the situation (cf. above). This is the content of hypothesis six: - H6: On obtrusive issues politicians follow the agendas of all sophistication groups. In more difficult information contexts, on the other hand, politicians should pay less attention to the less sophisticated who are relatively less interesting from a political point of view. On the one hand, individuals with low levels of sophistication tend to vote at a lower rate than those with higher 12

13 levels of sophistication (Franklin 2004, 154) a fact that, in general, makes the group less relevant to vote-seeking politicians. Highly sophisticated individuals, on the other hand, should be of particular interest to politicians since this group is most likely to contain individuals that act as opinion leaders (cf. Stimson 2004, 20-21; Adams and Ezrow 2009). To the extent, therefore, that politicians adapt to the agenda these citizens they might hope to score a bonus, so to speak, since the other groups may follow the lead of the more sophisticated. Further, the more sophisticated members of the public are also those most likely to make their voices heard and to act in accordance with their convictions (ibid.) both of which are elements that should increase their leverage over vote-seeking politicians. Thus, hypothesis seven is formulated as: - H7: On sensational issues the political agenda is affected more by the agenda of high sophistication citizens than by those of citizens with lower levels of sophistication. As for the media, on obtrusive issues they are in the same situation as the politicians: all segments of the public receive information about new developments and we should therefore expect the media to follow all sophistication groups on such issues: - H8: On obtrusive issues the media follow the agendas of all sophistication groups. On sensational issues the main expectation (cf. above) is that the media or the politicians lead the public agenda. However, to the extent that effects flow in the opposite direction we might expect the media to cater more to the interests of the more sophisticated segments of the public. Thus, not only are the less sophisticated less likely to take part in politics, they are also less likely to be large consumers of news media (cf. Bennet and Iyengar 2008) a feature that makes the group less interesting to profit seeking media outlets. For this reason and because the media, like politicians, might also be expected to pay particular attention to opinion leaders, hypothesis nine may be formulated as: - H9: On sensational issues the media agenda is affected more by the agendas of more sophisticated citizens than by that of citizens with less sophistication. 13

14 Data and Method Proper tests of the expectations set fairly high requirements in terms of data and analytical design. The design has to incorporate measures pertaining to all three actors as well as real world developments and to permit the flow of effects in all directions depicted in Figure 1. The design should also cover periods of both high and low/no attention to a given issue. Finally, the design has to allow for variation across issues and levels of sophistication. These requirements also in terms of data are met in the context of the Danish agenda-setting project ( where suitable quarterly data series on the agendas of the media, politicians, and the public and on a set of real world indicators are available for the period 1985Q2-2003Q3 for the issues of unemployment, inflation, violence and crime, immigrants and refugees, pollution, and the level of taxation. In relation to the expectations formed above (and in accordance with Soroka 2002a, 22-31) the issues of inflation and unemployment are seen as obtrusive. These are both issues that have a very direct impact on the lives of many citizens and thus, information about these issues is readily available. This implies the expectation that the public agenda is leading the others on these issues. Taxation (including duties etc.) for its part, is seen as a governmental issue characterised as it is by technical details concerning the organization of the revenue system with its multitude of rules and regulations (cf. Page and Shapiro 1992, 166). In such matters, politicians are the only actors among those in focus here that posses an overview sufficient enough to set the agenda. Immigrants and refugees, crime and violence, and pollution are seen as unobtrusive but concrete and to some extent associated with dramatic events. Not many people, that is, are directly affected by crime or visible pollution nor is it possible for most people to gauge from their daily lives how much of a problem immigrants and refugees may constitute for society, but the drama that is often associated with these topics makes them well suited for media coverage. Therefore, the media agenda should lead the two others on these issues. The data series are constructed as follows: 14

15 The public: In most agenda-setting analyses some version of the well-known Most Important Problem (MIP) question is used. The mechanisms underlying this question have been debated, however (see Dearing and Rogers 1996, 45-49; Takeshita 2005, ; Wlezien 2005), and since the MIP-question is not available in a sufficient number of Danish surveys the analyses below rely on a different type of measure. This measure stems from the private Danish polling company IFKA which, on a quarterly basis, has asked samples (13 of which comprise 1100 respondents; 3 comprise 500, 2 comprise 300, and the rest comprise 600) of Danes aged 15 and above how much they are concerned about certain issues. Clearly, this measure is not the same as the MIP-question. There is, however, sufficient overlap between the two to warrant using the concern question in the analyses (cf. also Trumbo 1995, 9-14). Thus, when a respondent indicates that a given issue is the most important problem it seems fair to see this as an indication that she is concerned about the issue otherwise it would not be a problem. While the closed format used here might increase the absolute levels of concern for some issues as compared to the MIP-format, this is less of a problem since focus here is on the dynamics and not the absolute levels. Following Trumbo (1995) and Nannestad (1999; 2005) the measure employed indicates the percentage of respondents in the IFKA polls who say that they worry A lot (as opposed to Some, Only little, or Not at all ) about the specific issues presented to them. 5 The overall percentage of respondents who worry a lot is used in the first set of analyses concerning differences across issues. Subsequently, the overall measure is disaggregated into series for the high, medium and low sophistication groups. In accordance with common practice in the field (cf. e.g. the extensive discussion in Enns and Kellstedt 2008; Zaller 1992), the respondents were split into three roughly equally sized groups based on their level of school education (-9 years: low awareness; 10 years: medium awareness; high school: high awareness 6 ). For the lack of better indicators education is chosen as a reliable proxy for sophistication. Although some conceptual slippage is unavoidable it should be noted that education is not only correlated with attention and knowledge, but also 15

16 provides an indication of individuals levels of analytical capability (Enns and Kellstedt 2008, 435; Pallas 2000, 502) all of which are resources that contribute to an individual s level of sophistication. The media: The contents of the two main daily radio news broadcasts on the publicly owned Danish Broadcasting Corporation (DR) have been coded according to an adapted version of the coding scheme of the American Policy Agendas Project ( the codebook used is available on the Danish website). In the period analysed, DR enjoyed a virtual monopoly on radio news and these constituted a vital link in the daily news flow, which makes the broadcasts an ideal source for measuring the media agenda. Thus, studies of the Danish media system (Lund 2002), indicate that many stories originate in the major national newspapers, but that the radio news is the most important filter for stories raised in the newspapers to make it into the TV news in the evening. Radio news thus link newspapers and TV, thereby constituting the best single source for measuring the agenda of the mass media in general. The database was developed using individual radio features as the unit of analysis. Each feature was thus coded in terms of issue content, length, etc. (cf. also Green-Pedersen and Stubager 2007; Green-Pedersen and Stubager forthcoming). Politicians: For theoretical as well as analytical reasons the agenda of the politicians has been measured by means of oral and written Questions to the minister asked by the individual parliamentarians in the Danish Folketing. There are no limits to the number of questions that can be asked and previous analyses (Green-Pedersen forthcoming) have shown that the questions cover the entire span of issues on the agenda of the parliament. Most questions are asked by members of the opposition parties and the agenda measured by means of the questions, therefore, is the agenda of the opposition. In analytical terms this agenda is well suited for analyses of the present kind since a sufficient number of questions are asked each quarter to prevent the instabilities associated with e.g. data based on parliamentary debates. 7 Further, the questions-agenda has the theoretical advantage of being flexible. Asking questions does not require much preparation and therefore questions provide an opportunity for politicians to take up issues that they are alerted to by the media or the public. Thus whereas opposition parties can react with words, the government, primarily, has to 16

17 react with policy which usually takes considerable time to implement thereby obfuscating the relationships (Walgrave and Van Aelst 2006; Green-Pedersen and Stubager forthcoming). For this reason the questions-agenda is the most likely to be impacted by the media and public agendas as well as to impact these agendas if such impact occurs. However, to the extent that governments do react we should expect them to do so in accordance with the dynamics set out above. Therefore, the opposition agenda can (with one minor exception discussed below) be used as a sensible proxy for the entire political agenda in the analyses. The questions were content coded in accordance with a coding scheme that is compatible with the one used for the media agenda (see the Danish website for the codebook). Real world indicators: For each of the six issues except pollution for which it was impossible to find indicators measured on a quarterly basis, real world indicators have been collected from the website and publications of Statistics Denmark: Unemployment: measured in per cent; inflation: measured in per cent over the past 12 months; immigrants and refugees: the raw number of spontaneous asylum seekers (i.e. applications handed in after or upon arrival to Denmark); crime and violence: the raw number of police reports about violent crimes; taxation: the burden of taxation as calculated by Statistics Denmark. 8 In accordance with the thermostatic logic developed by Wlezien (1995; cf. also Soroka 2002a, 78) all real world series have been included in the models in first differences under the assumption that the three actors react to changes in the levels of the indicators rather than to their levels as such. As noted, the real world indicators are seen as exogenous to the other variables in the model. This reflects theoretical interest as well as the operationalizations applied. Firstly, main interest here is devoted to the relationships between the three actor agendas rather than the possibilities for each of these to influence real world developments; analytically, therefore, the three agendas are in focus. Secondly, the choice of the opposition agenda to operationalize the political agenda reduces the possibility for the latter to influence real world developments. We should expect, that is, such influence to require policy changes rather than agenda changes a variable that is not available in this context. 9 17

18 The combined dataset, thus, consists of 41 different time series: For unemployment, inflation, immigrants and refugees, violence and crime, and taxation there are seven quarterly series: one for the share of the media agenda devoted to the issue; one for the share of the political agenda devoted to the issue (these two agendas were aggregated over quarters); one for the share of the entire public worrying A lot about the issue; one for the share of each of the three groups of low, medium, and high sophistication individuals that worry A lot ; and one for the real world indicator. For pollution only the first six series were available. For simplicity, I shall refer to the first three series as, respectively, the media agenda, the political agenda, and the public agenda while the series for the three sophistication levels are referred to as the agendas of the respective sophistication groups. The time series range from 43 (taxation) over 65 (immigrants and foreigners) to 74 (all other issues) observations. The raw series (excluding, for the sake of simplicity, the separate series for the three sophistication levels) can be seen in Figure 2. FIGURE 2 ABOUT HERE The use of quarterly series (which is dictated by the availability of data for the public) clearly has implications for the kind of relationships that can be found. Thus, the relatively high level of aggregation means that short-run media hypes and the like are unlikely to show up in the data. Rather, what will be captured is the general sweep of long-run developments. More than the day to day information flow, the analyses will focus on these more enduring trends of social development. Obviously, the possibility exists that the causal arrows run in the opposite direction from what has been hypothesised. This possibility, which has not been addressed in much of extant agenda-setting research, has to be incorporated into the analytical design. In this situation Vector Auto Regression (VAR) models would seem an ideal tool (cf. Bartels 1996; Edwards and Wood 1999). Such models allow for the simultaneous estimation of equations in which all of the endogenous variables (the three agendas in this case) appear as dependent variables with lags of themselves and the other endogenous variables as well as the exogenous variables (the real world 18

19 indicators in this case) as independent variables. However, due to the theoretical restriction imposed on the system that the political agenda is not allowed to impact the public agenda directly, hence precluding one of the relationships modelled in the VAR-model, it is more appropriate to use the Seemingly Unrelated Regression (SUR) technique (Soroka 2002a, 127) SUR also estimates all equations in the system simultaneously, thereby accounting for the possible correlations between the error terms across the equations, but in addition SUR takes into account the restrictions entailed in suppressing one of the relationships that exist in the VAR-model (ibid.). 10 Granger (1969) causality tests are subsequently used to examine whether past developments in the various series influence each other in accordance with the hypotheses posited above. The tests were conducted in two stages: At the first stage, the public agenda was represented by the combined series while at the second stage each of the three sophistication levels were represented in the models in turn. All models were run four times, thus. To ensure that all yearly and seasonal components as well as any auto-correlation in the data is captured in the models they were, in accordance with common practice, estimated with up to six lags of all variables and LR-tests were used to determine the appropriate lag length of each model (Brandt and Williams 2007, 24-26, 31). Due to the differing nature of the issues (Eyal, Winter, and DeGeorge 1981; Dearing and Rogers 1996, 68) the resulting lag lengths vary across issues; the lag lengths of the models are presented in Table A1 in the online appendix. Initial (Phillips-Perron) tests for stationarity showed no problems except for several of the unemployment series which were I(1). However, a (Johansen trace statistic-based) co-integration test showed that these series are co-integrated thus permitting the use of the SUR model despite their individual non-stationarity (Nannestad 2005). The models estimated, thus, have the following form: Me Po Pu it it it = α = α = α l= 1 6 l= 1 6 l= 1 β β β 1( l ) 5( l ) 9( l ) Me Po Pu it l it l it l l= 1 6 l= 1 6 l= 1 β β β 2( l ) 6( l ) 10( l ) Po Me Me it l it l it l l= 1 6 l= 1 6 l= 1 β β 3( l ) 7( l ) β 11( l ) Pu Pu it l it l Rw + + it l 6 l= 1 6 l= 1 + ε β β 3it 4( l ) 8( l ) Rw Rw it l it l + ε + ε 1it 2it 19

20 where i indexes issues and t time, while Me, Po, and Pu respectively are the media, political and public agendas. Rw is the real world indicator series. Results Overall, the results of the Granger causality tests for the models at the first stage (i.e. without distinguishing between different sophistication levels) presented in Table 1 provide support for hypotheses 1-3 thereby corroborating Soroka s (2002a; 2002b) results for Canada and those of Walgrave, Soroka, and Nuytemans (2008) for Belgium: First, the results clearly seem to support H1. Thus as can be seen from the inclusion of an arrows pointing from Rw to Pu and from Pu to Me and Po in the first column of the table, the public picks up on the real world development in unemployment and uses the information to influence the media and the politicians. This is precisely what was expected on the obtrusive prominent issues which are easy to comprehend and react to for the general public. On inflation, the media pick up on the real world development but are also influenced by the public which, again, acts as agenda-setter. H2 also meets with support. As expected, the only issue out of the six analysed here on which the politicians come out at the beginning of an agenda-setting process, hence, is the technical, governmental tax issue (the politicians also appear to influence the media agenda on unemployment but here they are influenced by the public). We may note, further, how the media succeed in passing on the information obtained from the politicians to the public thereby extending the agenda-setting power of the politicians. TABLE 1 ABOUT HERE Support can also be found for H3. Thus, the media are able to influence the politicians or the public on the concrete and dramatic sensational issues of pollution, immigrants and foreigners, as well as crime and violence. This is entirely as expected: For such issues the media have a clear incentive to seek out the often dramatic and thus highly newsworthy information that is subsequently relayed to the public and the politicians. We may also note, however, some unexpected effects on these issues. On immigrants and foreigners, first, the public is directly 20

21 influenced by the (change in the) number of asylum-seekers. This seems to indicate that information is more widely available on this issue than expected. On the issue of crime and violence the public influences the media agenda along with the development in the number of police reports about violent crimes. While the impact of the real world indicator is entirely in accordance with the expectations, the influence of the public agenda on the media agenda is unexpected on this issue. One possible explanation for the effect might be that the issue is characterised by a particularly high level of drama. This makes it ideal for the media and hence they may be anxiously following the development among the public on the issue so as to be able to provide the public s desired level of coverage even if unrelated to the actual development. Finally, the relationships that appear for pollution resemble the pattern found on the obtrusive issues where the public plays the central role in the information flow. However, due to the unavailability of real world indicators of a sufficient quality and frequency on the pollution issue, the estimates have not been controlled for the possible impact of pollution levels. Therefore it is impossible to rule out a spurious element in these findings. Before inspecting the results of the tests of hypotheses 4-9 it is worth noting that the series for the three sophistication levels show some measure of divergence. Thus although none of the correlations between the three series presented in Table A2 in the appendix fall below 0.7 it is remarkable that a number of them actually come out at that level at all. It might also be noted that the correlations, on average, seem lower for the non-obtrusive issues. These findings clearly indicate that it is worth examining whether the relationships between the three agendas differ as hypothesised. The results of the tests appear in Table 2 and in condensed form in Table 3. TABLE 2 ABOUT HERE TABLE 3 ABOUT HERE As is clear from the tables, the prediction of H4 that all sophistication levels are impacted by real world developments on obtrusive issues meets with support: Only one (for the low sophistication group on inflation) out of the six possible relationships fails to manifest itself thus 21

22 indicating the unmediated nature of information about such issues. Also H5 gets support. Thus, the agenda of the high sophistication group is influenced on one issue (out of the four possible nonobtrusive), that of the medium sophistication groups is influenced on two issues, and the figure for the low sophistication group is three. Just as expected, hence, on non-obtrusive issues sophistication enables citizens to reject the agenda-setting of the media most likely because high sophistication individuals are able to determine the importance of issues on their own. This finding provides strong support for the information resource centred policy marketplace model. For H6 the results are also supportive, at least as regards unemployment on which all sophistication levels influence the political agenda as was expected for this obtrusive issue where the public is directly impacted. On inflation, however, no relationship can be detected between the public and the politicians. The latter even appear to be completely disconnected from the other actors on this issue a finding which may reflect that inflation with an average share below 1% has occupied much less of the political agenda than the other issues. H7 similarly meets with support. Thus as expected on the three sensational issues, the political agenda is influenced twice by that of the high sophistication group while neither the low nor the medium sophistication group appears to influence the politicians. This pattern is entirely as expected and probably reflects the attraction of the most sophisticated as possible opinion leaders since this group is clearly demarcated from the two others. For the media-focused hypotheses H8 and H9 the results parallel those for H6 and H7. On the obtrusive issues in focus for H8, thus, the expectations are fulfilled for unemployment where all sophistication groups influence the media agenda. On inflation, however, only the high sophistication group influences the media agenda. This result may reflect that inflation also constitutes the smallest issue on the media agenda. For H9, finally, the medium sophistication group turns out to have the most pervasive influence on the media agenda which it affects on all three sensational issues thereby surpassing the high sophistication group which, with influence on two issues, is still ahead of the low sophistication group, however. This result may stem from the choice 22

23 of media indicator. Thus, it might be speculated that the editorial line of this media outlet (which has regularly had an audience of about 1/5 of the entire population) targets some sort of average citizen who, presumably, would be found in the medium sophistication category. When taken together the results for hypotheses 5, 7, and 9 reveal a clear pattern in the balance of net agenda-power on non-obtrusive issues across the sophistication groups. Thus, the most sophisticated citizens are only influence by the media on one issue while they influence the politicians and media on two issues each. This yields a net balance of three in favour of the high sophistication group. The medium sophistication group is influence by the media on two issues and influence the politicians on zero and the media on three issues. For this group, hence, the net balance is one. The least sophisticated are influenced by the media on three issues and influence the politicians on zero issues and the media on one. The group s net agenda-power balance, therefore, is negative two. These simple calculations thus clearly demonstrate the inequality in agenda-power across the three sophistication levels. Discussion The analyses reported extend our knowledge in several ways. At the empirical level it is clear that the answer to the Who Sets the Agenda?-question and thereby to an important part of the Who Governs?-question varies in accordance with the availability of information resources: those who have them can influence others. While on prominent issues we see a bottom-up pattern of agendasetting dominated by the public and with all members of the public equally implicated we see examples of a top-down politician-controlled process on governmental issues and a mediacracy model (cf. Kleinnijenhuis and Rietberg 1995) on sensational issues. For the two latter issue types where information is scarcer, further, we clearly observe an unequal involvement of the public in the processes in the sense that the more resourceful members of the public i.e. those with higher levels of political sophistication exert more influence on and are less influenced by politicians and the media than those less resourceful. 23