The Agenda-Setting Power of Protest. Demonstrations, Media, Parliament, Government, and Legislation in Belgium,

Transcription

1 The Agenda-Setting Power of Protest. Demonstrations, Media, Parliament, Government, and Legislation in Belgium, Stefaan Walgrave Rens Vliegenthart Abstract This study assesses to what extent the number and size of demonstrations affects the political agenda in Belgium ( ). We conduct pooled time-series analyses for different political agenda s for twenty-five issues. We find that protest matters and that the parliamentary, the governmental and the legislative attention are significantly affected by preceding protest activities. Protest size matters more than protest frequency. Some political agendas are more affected by protest than others. The media act as an intermediary variable: media coverage reacts on protest and, in turn, affects the political agenda afterwards. External (movement) and internal (political elites) factors setting the political agenda interact: the effect of protest increases when the issue was already politically salient, as shown by the share of attention in the preceding party manifestos. Stefaan Walgrave is the corresponding author: Stefaan.walgrave@ua.ac.be Word count (without tables or figures): 11,166 words Stefaan Walgrave Department of Political Science Research group Media, Movements, and Politics (M²P) University of Antwerp Sint-Jacobsstraat Antwerp Belgium stefaan.walgrave@ua.ac.be Rens Vliegenthart Department of Communication Science & Amsterdam School of Communications Research (ASCoR) University of Amsterdam Kloveniersburgwal CX Amsterdam The Netherlands r.vliegenthart@uva.nl

2 The Agenda-Setting Power of Protest. Demonstrations, Media, Parliament, Government, and Legislation in Belgium, Abstract This study assesses to what extent the number and size of demonstrations affects the political agenda in Belgium ( ). We conduct pooled time-series analyses for different political agenda s for twenty-five issues. We find that protest matters and that the parliamentary, the governmental and the legislative attention are significantly affected by preceding protest activities. Protest size matters more than protest frequency. Some political agendas are more affected by protest than others. The media act as an intermediary variable: media coverage reacts on protest and, in turn, affects the political agenda afterwards. External (movement) and internal (political elites) factors setting the political agenda interact: the effect of protest increases when the issue was already politically salient, as shown by the share of attention in the preceding party manifestos. Introduction At first sight, it seems obvious that social movements can have sweeping effects on politics the peaceful revolutions in Central and Eastern Europe after 1989 are maybe the clearest examples of this. But our knowledge about the political influence of social movements and the protest events they stage remains limited. Social movement scholars and students of political protest have produced extensive knowledge about social movement emergence and about protest mobilization dynamics. Yet, probably the most crucial question and the question that interests non-scholarly observers of social movements most whether social movements and their protests matter politically has until very recently only received modest attention and has only been answered partially (Amenta, Caren, Chiarello, and Su 2009). 2

3 The reasons for this lacuna are threefold: theoretical, methodological and empirical. First, there is a longstanding theoretical disagreement about how to conceptualize social movements influence. The potential outcomes of social movements are very broad; contentious events do not only have potential impact on politics but also on the public, on the movements themselves, on counter movements, and their impact can be cultural and long-term as well. But even the more strictly delineated political effects of protest have been conceptualized in a dissimilar way. In technical terms: protest effect studies have not agreed on the dependent variable. This hindered the subfield to build a cumulative body of evidence and left it scattered and non-integrated. In this study, we conceptualize the impact of social movements as the effect protest events have on the political agenda, that is the issues political actors on different levels and in different institutional settings are dealing with. The agenda-setting approach has developed into one of the main policy theories in political science; it offers clear concepts and suggests testable hypotheses that can systematize our knowledge of when and why movements matter. Second, the subfield of movement effects has struggled with the methodological problem of assessing causality. Frequently the methods employed did not allow students to make convincing causal inferences. In line with most work on social movements, the case study approach has largely prevailed. Few studies have used statistical methods taking into account many cases and assessing causality through statistical modeling. The lack of comparison and the reliance on thick descriptions prevented generating systematic and especially generalizable knowledge. Also, a control measurement of alternative sources of the observed changes in policy was not included in many studies. In this study, we propose a straightforward and simple measurement of impact and we draw on many cases: do protest events on an 3

4 issue produce a subsequent measurable increase in political attention to that issue, while controlling for alternative explanations? Third, few studies have had good data to empirically assess whether movements matters or not. How political systems processes movement protests is difficult to gauge and gathering data on political elites responses to movement activities is often difficult. The empirical scope of extant studies has been narrow. They mostly dealt with one case, one movement or even one single decision, and as good as all studies were situated in the US context. The US political system, and its social movements, are arguably different from many other Western democracies; it is not obvious that US findings would simply apply to other contexts. Consequently, we hardly dispose of any comparative knowledge about what types of movements dealing with what kind of issues matter compared to other movements dealing with other issues. Therefore, in this study we tackle an entirely different political system: Belgium, a small consociational democracy with a strong protest tradition. We propose a systematic and comparative multi-issue design with a broad empirical scope in which all protests (demonstrations) on all types of issues are incorporated. More concrete, based on longitudinal data covering eight years ( ) of protest events in Belgium we assess through time-series analyses to what extent protest events affect the political agenda. Via protest event analysis we accumulate evidence about all protest demonstrations staged in Belgium during the research period; this is the key independent variable. The dependent variables consists of three series of data assessing the parliamentary agenda (questions), the governmental agenda (ministerial council), and the legislative agenda (passed legislation). We also incorporate an intermediary variable in our design, the media agenda, and test for interactions with party manifestos. Our operational research question is thus: to what extent do protest demonstrations affect the Belgian political agenda, being 4

5 parliament, government and legislation, and does mass media coverage play an intermediary role? The political agenda-setting impact of social movements Assessing movements impact is notoriously difficult. Until recently not so many had engaged in a systematic effort to do so (Giugni 1998; Soule, McAdam, McCarthy, and Su 1999; Tarrow 1999). Since a few years, though, the subfield of movement effect studies is quickly expanding (Amenta, Caren, Chiarello, and Su 2009). Movement outcomes can and have been defined very broadly. The volume by Giugni, McAdam and Tilly (1999) contains examples of movements alleged effects on very different things ranging from scientific institutions (Moore 1999), over activists life courses (McAdam 1999) to the discourse of protest policing (della Porta 1999). Most movement students agree that social movements can be defined as actors (or contentious events 1 ) making collective and political claims. Some consider social movements as being the primary engines of political change (McAdam, McCarthy, and Zald 1998). Hence, it is natural to consider especially the political outcomes of social movements as being among their most important consequences and this is what most students of movement outcomes have focused upon (Giugni 1999: xxi). Limiting the scope of social movements effect to politics, though, still leaves ample room for diverse conceptualizations of political impact; studies have used a host of different dependent variables (Amenta, Caren, Chiarello, and Su 2009; Amenta, Carruthers, and Zylan 1992). Gamson s (1990) typology of political movement outcomes based on two dimensions - the political recognition of the 1 Note that we, in this study, and in line with Tilly s (2004) seminal definition of social movements, do not distinguish between social movements and the contentious events they stage. 5

6 challenging groups and the gain of new advantages for a movement s constituency - is the best known. Kitschelt (1986), for example, distinguishes procedural, substantial and structural political consequences. We focus here on substantial effects, that is: effects on public policy and not on the cooptation of the movement in policy-making. Even within the public policy effect studies a host of different dependent variables have been employed. Extant studies have investigated whether movements were able to produce the intended changes in legislation (Banaszak 1996), whether they were able to increase legislative activity (Costain and Majstorovic 1994), whether they managed to increase public spending (Amenta, Carruthers, and Zylan 1992), whether they managed to stop or at least slow down a criticized policy (Kitschelt 1986), whether they affected parliamentary votes (Burstein and Freudenberg 1978; MacDougal, Minicucci, and Myers 1995), whether they contributed to state-level adoptions of constitutional amendments (Soule and Olzak 2002) etcetera. A decade ago Giugni criticized the political effects literature for being neither systematic nor comparative. Most studies do not systematically compare across countries the field is overwhelmingly dominated by US-studies and across movements (Giugni 1998). Consequently, one cannot develop robust generalizations about when movements matter and when they do not. Since by far most studies are case studies using a specific, non-standardized and non-comparative measurement of effect of the movement under study the field has not developed a strong cumulative body of evidence. Ten years later the same criticism was repeated by Amenta and colleagues in their overview of movement impact studies: Only rarely do studies even compare across a few movements or issue areas (Amenta, Caren, Chiarello, and Su 2009). Additionally, most impact studies do not control systematically for alternative sources of the supposed movement effects (Amenta, Caren, Chiarello, and Su 2009: 26-27). Even when some studies may have shown that movements matter 6

7 for public policy, we do not have a clue about the relative contribution of social movements compared to that of other political actors because studies have rarely been multivariate (Amenta, Caren, Chiarello, and Su 2009; Baumgartner and Mahoney 2005: 1). Thus, although assessing movements effects is crucial for understanding the role of social movements in contemporary democracies, the subfield of movement effect studies as a whole remains quite diverse and scattered. Therefore, this study proposes to measure the effects of social movements by assessing whether the attention for a movement s issues increases or not after the movement staged (a) protest event(s). This simple agenda-setting operationalization of movement effect is far from unique. In the recent literature, we found nine empirical studies that explicitly or implicitly tap whether the attention within the political system is partially driven by social movements. Table 1 presents a systematic overview of these studies. <Table 1 about here> Schumaker (1975: 494) was probably the first to state that one of social movements main political effects is setting the political agenda, he called this agenda responsiveness. Other students have pursued the same track taking a measurement of movement size or of movement protest activity as the independent variable. Burstein and Freudenberg indirectly focused on agenda-setting in the US Senate regarding the Vietnam war when they found that, initially, antiwar protest (demonstrations) were effective in increasing the salience of the Vietnam war issue (while these same demonstrations later became counterproductive in affecting the actual antiwar Senate votes) (Burstein and Freudenberg 1978). Costain and Majstorovic (1994) assessed what determines legislative activity on women s issues in 7

8 the US Congress and concluded that social movement activities especially reinforce the link between the congressional legislative agenda and public opinion. They contend that the women s movement activities do not directly impact Congress but only via public opinion. Especially during the most recent years, work adopting the agenda-setting perspective has gained momentum. McAdam and Su (2002) examined the roll-call votes about the war in Vietnam in the US House and Senate; they found that the size of the antiwar protests effectively affected the number of votes on war-related issues; hence, they conclude that the size of protest events contributes to setting the political agenda. Baumgartner and Mahoney (2005), in one of the rare studies comparing across issues and movements, compared five issue areas in the US and tested whether social movements contributed to setting the political agenda. They found that the number of social movement organizations (or membership) on women, environmental, elderly, civil rights, and human rights issues affects the number of congressional hearings on these topics. King and colleagues started developing a more systematic theory of agendasetting by social movements (King, Cornwall, and Dahlin 2005; see also: Soule and King 2006). They depart from the idea that each succeeding stage in the legislative process imposes more constraints as more stringent rules are in place and as each stage becomes more consequential in terms of actual policy changes. They maintain that social movements influence especially plays at the start of the policy cycle, when the political agenda is set, rather than later on in the policy cycle when policies are actually decided upon and implemented. They corroborated these ideas with a case study about woman suffrage in US States and show that women s movements - mainly through their lobbying and organizational strength - were influential in bringing the issue to the agenda while they had less impact in the voting stage. Drawing upon these ideas, in a study about the political effects of the US 8

9 environmental movement, Johnson found that the size of the environmental movement is positively associated with the agenda (hearings) in Congress (Johnson 2008). Sarah Soule, with several co-authors in a series of recent studies, further contributed to the idea that social movements may affect the political agenda. In a first longitudinal study on US Congress hearings and roll call votes about women s issues the authors could not find any proof of the fact that women s protest, what they called outsider tactics, affected political attention in Congress (Soule, McAdam, McCarthy, and Su 1999). Rather, the authors suggest that protest incidence is a consequence of political attention (hearings in Congress) and not a cause. In a second study, Soule and King (2006) did find effects from social movements on the agendasetting of the Equal Rights Amendment (ERA) in the legislative branches of the US states controlling for alternative explanations (elite support and public opinion). The authors argue, and show, that the women s movement managed to impact the agenda-setting stage but that it was much less successful in determining the real ratification in the second House. A strong presence of the women s movement in a state (number of local chapters of organization) combined with the Democrats in power in the state, increases the chances that the ERA bill is tabled. In a third study, the same authors take protest as the independent variable when trying to model US Congressional attention to a range of rights issues while controlling for a whole range of alternative agenda-setters (King, Bentele, and Soule 2007). They confirm their previous finding that protest matters and that there are more congressional hearings when the number of protests goes up. This ground-breaking study is innovative as it compares different types of rights issues and does not stick to a single issue or conflict; the authors also explicitly address the issue or generalizability; yet, the 9

10 comparison remains limited to a series of very similar issues that are related to each other. The brief overview establishes that, during the most recent period, the agendasetting approach gradually attracted more interest from students of social movements. We believe this is the case because the agenda-setting approach and associated design has the potential to solve many of the problems associated with earlier movement effect studies. First, although setting the agenda clearly refers only to a part of the policy process, it arguably refers to a crucial element. Garnering attention for an issue is a necessary condition for and precursor of policy change on the issue. The same applies to social movements and their policy effects: they need to attract attention, otherwise they will not succeed (Johnson 2008; McAdam and Su 2002). In political science, and more specifically in public policy, agenda-setting has developed into one of the main theories (Baumgartner and Jones 1993; Jones and Baumgartner 2005). Second and closely connected to the first point, movements are not homogenous entities; different movement subgroups and factions may stand for different aims and, as a consequence, it is hard to say whether movements reached their goals or not since this depends on the subgroup at stake (Giugni 1999: xx). An increased issue attention level on the political agenda, however, is something the entire movement can probably agree upon. All types of policy consequences require increased attention and movements invariably demand more attention for their issues. Third, simply assessing whether movements manage to make politics spend more attention to their issues avoids the tricky debate about movements substantive success and failure (Giugni 1999: xxi). Assessing increases in attention does not imply an interpretation or preceding knowledge about the precise goals of a movement or about its different composing groups. It only tests whether attention goes up or not and does not claim this represents success or not. Fourth, by 10

11 potentially including the agenda consequences of many types of movements and protests, even the non-successful ones, the agenda-setting approach deals with the problem that cases are sometimes selected on the dependent variable (Amenta, Caren, Chiarello, and Su 2009: 14; Giugni 1999: xxiv). Fifth, and perhaps most importantly, relying on issue attention as an indicator of a policy effect can solve some of the thorny causality problems that show up when drawing on case study designs (Giugni 1998: 373). Agenda-setting studies rely on a longitudinal time series analysis and can use Granger causality leading to robust statistical causality inferences. Agenda-setting, better than some other designs, also offers plenty of opportunities to control for the agendas of alternative actors within and outside the political system. This permits to isolate the impact of the protest and to analyze the movement s influence in conjunction with other potential influences (Amenta, Caren, Chiarello, and Su 2009: 5-6). Sixth, since there does not exist such a thing as the political agenda but a whole range of different political agendas, students may use agenda-setting to devise and test more subtle theories about movements diverging impact on different political institutions and their agenda. In other words: affects may be disaggregated and may be assessed separately for different political institutions. Finally, the agenda-setting approach and its related design allow for comparison of the effects of different protest events on different issues and of movements in different countries. As measures of attention are fairly easy to formalize and to standardize, the approach is well-suited for broad comparisons spanning several movements and countries. That the agenda-setting approach and related design has all these potential advantages for the study of movement effects does not mean that the extant studies have effectively made full use of the ample analytical possibilities agenda-setting offers. Most agenda-setting studies reviewed above are analytically strongly 11

12 developed and, drawing on robust statistical methods, offer much better understandings and measurements of causality than previous studies. However, many of the potential analytical strengths of the agenda-setting approach of social movements have not been employed fully. Many studies only take one political agenda into account as dependent variable and discard the possibility of assessing diverging effects on different actors and agendas (for exceptions see: Johnson 2008; King, Cornwall, and Dahlin 2005; McAdam and Su 2002). Few studies have drawn upon a tight time series design with predefined lags between independent and dependent variables; most studies have just relied on yearly (or three-monthly) observations and have analyzed synchronous yearly changes in the dependent and independent variables which considerably weakens the causal conclusions one can draw (again the exception is the study of McAdam and Su 2002; and also Soule, McAdam, McCarthy, and Su 1999). Additionally, one can raise theoretical doubts whether causes and effects that take such a long time to materialize (at least three months) really indicate causality. As McAdam and Su (McAdam and Su 2002: 700) rightfully state, working with much closer temporal connections (months, weeks, days) reinforces the chance that we are really dealing with true causal effects. Other work in political agenda-setting focusing on the impact of mass media coverage on political agendas has shown that, for example, parliament and government tend to react quite immediately to media cues. It is a matter of days before political actors adjust their attention and adopt mediatized issues (Vliegenthart and Walgrave 2009; Walgrave, Soroka, and Nuytemans 2008; Walgrave and Van Aelst 2006). We believe these short term effects are also likely when it comes to reacting to specific protest events staged by social movements. Moreover, the reviewed studies do not take into account the problem of autocorrelation and control the impact of the independent (movement size, protest etc) on the dependent variable (political attention) for the 12

13 dependent variable s own past (earlier political attention) (the two exceptions being King, Bentele, and Soule 2007; Soule, McAdam, McCarthy, and Su 1999). This also undermines the strength of the causal inferences one can draw as it does not respect the rules of Granger causality: two correlated trends may not point to causality but to a spurious relationship. Most studies did not control the movements impact for other, alternative explanations of the rising or diminishing political attention for issues; other political actors (e.g. political parties) may at the same time as movements be trying to raise attention independent of what movements are doing (but see McAdam and Su 2002; Soule and King 2006). Finally and most importantly, previous work on agenda-setting and social movements has hardly profited from the large comparative opportunities that the standardizing of the measurement of political attention offers. Only two studies compared different issues and different movements (Baumgartner and Mahoney 2005; King, Bentele, and Soule 2007). Doubtlessly, this lack of comparative work is the single weakest point in the movement effects literature so far. In the same vein, all extant work on agendasetting and social movements has dealt with the US. Remarkably enough, none of the reviewed studies considers this to be a problem and raises concerns about the generalizability of its findings. The implicit assumption seems to be that what applies to the US applies to the rest of the world as well. This American generalism is all the more remarkable as some studies recognize explicitly the importance of the specific institutional setting and the peculiarities of the political system in which the agendasetting process is studied (see for example: Soule and King 2006). 13

14 Hypotheses: protest and political agenda-setting One of the central debates in the literature is whether disruption rather than moderation leads to social movement effects (Giugni 1999; McAdam and Su 2002). This discussion deals with the tactics of the movements and the type of action they adopt. Some claim that disruptive and even violent action leads to impact (McAdam 1983; Tarrow 1998) while others contend that rather moderate actions forms are associated with movement gains. This study explores to what extent demonstrations (their number and size) affect the political agenda and is entirely situated within the moderate tactics. Since Barnes and Kaase s Political Action demonstrations are considered as belonging to the moderate actions forms (Barnes and Kaase 1979). The routinized character of most demonstrations have lead to a normalization of this action type (Van Aelst and Walgrave 2001). Consequently, the vast bulk of the demonstrations in our sample are legal, authorized, non-violent and orderly assemblies. So, in this study we control for the moderate-disruptive dimension and assess whether moderate action forms affect politics or not. The idea that mobilization in itself may lead to policy change may seem natural and is the basic hypothesis of many studies of movements political outcomes (Johnson 2008: 975) but not all scholars tend to underscore it. Amenta et al. (2009), for example, state that what helps challengers mobilize may sometimes thwart a movement s impact. Some scholars have suggested that agenda-setting can be done by protesting while really weighing on the actual decision requires more organizational strength (King, Cornwall, and Dahlin 2005) while others have suggested just the opposite (Soule and King 2006: 1898). In this study, we will make use of evidence relating to the number and the size of demonstrations (see below) and, thus, our first hypothesis simply 14

15 states: the number and size of protest demonstrations on an issue increase attention for that issue on the political agenda (H1). If not the protest strategy and the tactics a movement employs, what accounts for movement outcomes? One of the main research tracks has been to test whether a movement s organizational strength is related to the realization of its political goals (Giugni 1998: ). Following the resource mobilization approach and especially the seminal work by Gamson (1990), the idea is that strongly organized and not internally-divided movements are better positioned to generate political impact than loosely organized movements (see also: Amenta, Carruthers, and Zylan 1992). As we will elaborate below, our two key independent variables in this paper are (1) the number of organized demonstrations regarding a certain issue in a given time frame and (2) the number of demonstrators showing up at these demonstrations. We suppose that strongly centralized and homogenous movements are especially able to organize big events with a large attendance rather than staging several smaller events. An accumulation of smaller events, we contend, rather points to a disorganized and disintegrated movement. Many protest events are set up by a temporary campaign or coalition; if the composing parts do not agree on a common theme or claim they decide to each organize their own (smaller) event resulting in a multiplication of events. We do not expect that the number of demonstrations does not matter at all but rather that the size of demonstrations is a more important factor. Additionally, also Lohmann (1993) has argued that the number of protesters is closely associated with the strength of the signal that is sent to policy makers who seek to make decisions that are advantageous for a majority of the population. This argument has been relayed by Burstein and Linton (2002) when they claim that the size of the protest gives political elites cues about public opinion and, thus, affects what they undertake. All this leads to our second hypotheses: the size (turnout) of 15

16 demonstrations rather than the number of demonstrations has an impact on the political agenda (H2). Another core debate in the movement effects literature deals with the question whether movement success is mainly the consequence of the own strength of the movement or rather of a pre-existing willingness in the political system to accommodate a movement s wishes (Giugni 1998). The political effects movements produce can in fact be attributed to internal movement factors as was hypothesized by H2 or to external political context which comes close to the basic tenet of the political opportunity structure approach (Kriesi, Koopmans, Duyvendak, and Giugni 1995; Tarrow 1998). Do movements have the power to force the political system to react or are they dependent on internal support? Is it strategy or structure that determines movement success? We will test these different propositions by controlling the demonstrations impact for the preceding attention for the demonstration issue manifestos of political parties. Political parties can be considered to be the most important political actors in Belgium (De Winter, della Porta, and Deschouwer 1996). We expect that demonstrations addressing issues that have previously received ample attention from political parties that can thus count on a favorable political context will result in larger increases in political attention than demonstrations dealing with issues that have a much lower standing on the party agenda. In other words, in line with the most recent literature, we expect internal and external factors to be complementary and interacting. In their recent overview Amenta c.s. state that a good deal of the recent studies dealing with movement effects found an interaction between internal movement indicators and external political context factors (Amenta, Caren, Chiarello, and Su 2009: 15). Earlier, Amenta and colleagues have coined their Political Mediation Model that holds that movement mobilization only leads to success when it is situated in a favorable 16

17 political context. In other words, they claim that internal and external factors reinforce each other (Amenta, Carruthers, and Zylan 1992; for a similar argument see: Soule and King 2006; Soule and Olzak 2002). Hence, we expect to see the largest increases in political attention when the number and size of demonstrations is big and when the main political parties have proven to be sensitive for the issue. The influence of demonstration size and demonstrator numbers on the political agenda depends on the attention political parties devote to the issue at stake: the more attention for the issue in party manifestoes, the larger the effects of demonstration size and demonstrator numbers (H3). If social movements and the protest they stage matter, we may expect them to matter to a dissimilar extent for different political agendas. Indeed, the political agenda does not exist. Rather, politics consists of a range of different political agendas. Some of these agendas are probably more affected by protest than others. According to Jones and colleagues agendas can be categorized according to whether they are situated at the beginning (e.g. congressional hearings) or at the end (e.g. budget) of the policy cycle (Jones, Sulkin, and Larsen 2003). Like for lobbying groups (Milbrath 1963), we expect social movements to have most impact on the early agendas rather than on the more institutional and constrained agendas. This idea more or less but not entirely fits King and colleagues (2005; 2006) argument that movement impact diminishes as one moves through the stages of the legislative process (see also: Johnson 2008). In this study, we measure three different political agendas in Belgium: questions and interpellations in parliament, decisions by the ministerial council (government), and passed legislation. Note that these three agendas are not necessarily different stages of the same policy process as King c.s. have examined; interpellations and questions, for example, are not a precursor of legislation. Rather, these are really distinct agendas each with their own finality and 17

18 logic. Either way, we expect the impact of protest to differ and, consequently, our fourth hypotheses states: the protest impact diminishes from questions and interpellations in parliament, over government decisions to passed legislation (H4). When assessing the impact of social movements on public policy, some scholars have taken into account the intervening role of public opinion (McAdam and Su 2002) or of the media (Giugni 1998: 380). Costain and Majstorovic (1994), for example, claim that media coverage worked together with protest (and public opinion) to generate women s rights legislation in the US. Starting with Gitlin s The Whole World is Watching (1980), many movement scholars have investigated the interaction between social movements and the mass media. In this study, we consider the media as an intermediary variable. Many studies in communications and political science established that mass media coverage affects the political agenda (for an overview, see: Walgrave and Van Aelst 2006). Here we test the hypothesis that mass media coverage reacts to protest events (Gamson and Wolfsfeld 1993) and, then, affects the political agenda. So, we suppose that protest not only has a direct but also an indirect influence on the political agenda. Our fifth hypotheses states: mass media act as an intermediary variable, they are affected by movement protest and, in turn, affect the political agenda (H5). Data and methods The present study focuses on Belgium, a small consociational democracy (Lijphart 1999). Belgium is a polity with very strong political parties (De Winter, della Porta, and Deschouwer 1996). It is a parliamentary democracy that is typically governed by a coalition government just like most other West-European democracies. We take these key features of the Belgian system into account by incorporating 18

19 partisan issue preferences in our analyses. According to the surveys of the World Values Study the Belgian polity is characterized by a large incidence of demonstration activism (Norris, Walgrave, and Van Aelst 2005). We would expect this to diminish the impact of protest demonstrations on the political agenda as demonstrations are a routine and normalized feature of political action. Added to that comes the typical decentralized character of the Belgian state-with a French-speaking and a Dutchspeaking part. This probably mitigates the impact of protest on politics. Indeed, when protest is organized by a movement that only mobilizes at one side of the language border chances are smaller that this signal would be picked up by the federal political institutions. Although the specifics of our analysis and the differential strengths of the effects we may find certainly cannot be generalized to other political systems, we believe our results can be considered as a first step to cautiously develop some generalizations about when protest matters for politics. In any case, the study generates badly needed evidence from outside of the US. The main analytical innovation our study proposes is that we, in contrast to earlier work, do neither focus on one case of decision-making, nor on one issue or even on a series of neighboring issues. We simply take all issues and entire agendas into account. As we will explain below, based on a detailed codebook, we issue-coded all parliamentary activities, all governmental decisions, all passed legislation, all party manifestos during an eight-year time frame. We also accumulated evidence on all (recorded) protest demonstrations and coded them according to the same issue coding scheme. Then, we pool all these data in one large dataset that has sufficiently analytical strength to test the hypotheses. We tackle the questions raised above relying on a design with two key independent variables (number of protest demonstrations and number of demonstrators), three dependent variables assessing different types of political 19

20 agendas (parliament, government, and legislation), one intermediary variable (media), and one interaction variable (party manifestos). After explaining how we coded the raw data, we introduce these data series one by one. Issue codebook All codes for all data series are based on the internationally employed hierarchical EUROVOC thesaurus, designed for coding all EU-documents ( This thesaurus contains 6,075 different hierarchicallystructured descriptors. Mainly relying on aggregate categories but sometimes adopting more detailed EUROVOC categories to grasp typical media issues (e.g. different kinds of crime), we reduced the total number of codes to 110. Using all 110 issue categories means that many categories are very small, and equal to zero much of the time. Therefore, the 110 issues are further combined, and our analyses consequently rely on a collapsed form of the dataset, where the 110 issues are collapsed into twenty-five major issue categories and thus constituting twenty-five issue panels. All these panels are pooled in aggregate analyses. Dependent variables Media - We have a large media database to our disposal, comprising the main evening news of the four major TV-stations, two Dutch-speaking (TV1 and VTM) and two French-speaking (RTBF and RTL), and five major newspapers (Dutch-speaking: De Standaard, De Morgen and Het Laatste Nieuws; French-speaking: La Libre Belgique and Le Soir). We coded all front-page newspaper stories, with exception of the newspapers that appeared on Tuesdays and Thursdays, on a daily basis. The prime time TV news (7.00 p.m.) is coded in its entirety on a daily basis. Taken together, the Flemish and French-speaking media database contains 180,265 news 20

21 items (113,658 TV items and 66,607 newspaper items). Belgian newspapers are not issued on Sundays. Therefore, for the various television channels a mean score for each issue for Saturdays and Sundays is calculated to substitute the original Saturday score, while the Sunday score is deleted in order to keep data comparable with the newspaper data. Furthermore, newspaper data for Tuesdays and Thursdays, that are typically not coded, are estimated based on previous and subsequent values. Correlations between issue-attention scores in the various outlets are above.65 for the daily level and even higher if we aggregate data to weekly or monthly levels. Elsewhere, it is shown that the media studied here influence each other considerably (Vliegenthart and Walgrave 2008). When assessing the influence of protest on media coverage, we use daily-level analysis, since this is the typical news-cycle for traditional media such as newspapers and television. For some analyses, we treat newspapers and television broadcasts separately. Since all television news is broadcasted in the evening, we use the same-day protest events to predict this coverage. Since all newspapers are issued in the mornings, we use previous-day protest events to predict this coverage. In analyses where media coverage is an independent variable, we lump newspaper and television coverage together. Parliament Our first dependent variable is the weekly number of interpellations and questions made in Belgian Parliament by each of the twelve political parties that were represented during our research period on any of twenty-five issues. Questions and interpellations are the most important non-legislative activities of most Parliaments. To obtain those data, we coded all parliamentary records for the period , which contained 10,556 interpellations and parliamentary questions. We use a weekly aggregation level since it encompasses what one can call the shortest political cycle ; Question time is organized once a week. However, Belgian parliament 21

22 does not meet every week and those weeks in which no parliamentary activity took place are excluded, leaving 231 weeks for the eight years that are used in our analyses. Government To test the hypothesis that protest has a different impact of different political agendas we not only code parliamentary action but two other political agendas. Government s priorities are tapped via the communication about the weekly ministerial meetings in Belgium, taking place on Fridays. These briefings are coded in a similar way as the parliamentary interpellations and questions. In total we encoded 5,088 government decisions mentioned in the weekly publication Facts of the Belgian Information service. Again, we include 231 weeks in which the Ministers met. Legislation All passed legislation in the Belgian parliament during the research period has been attributed with an issue code, again relying on the same codebook. In total, the number of passed laws during the eight years of our research amounts to 1,198 laws passed in total. Independent variables Number of demonstrations Protest event analysis relying on mass media coverage is a widely used method in the sociology of social movements. There is some debate about the reliability of newspaper data (McAdam and Su 2002). We therefore relied not only on the coverage in two broadsheets, De Standaard and Le Soir, but also went to the national police archives to supplement our newspaper data with direct police information. For the last three years in the research period ( ) we got direct information from the national police. For more information about these data we refer to other publications (Van Aelst and Walgrave 2001). In total we recorded 3,839 demonstrations in the research period. Number of demonstrators To be able to test the hypothesis that the number of demonstrators rather than the amount of discrete demonstration events matters we 22

23 not only recorded the number of demonstrations but also the number of people that took to the streets at these events. In total, 4,034,789 people were recorded to have participated in one of the almost 4,000 demonstrations mentioned above (on average 1,051 people participated in a demonstration). We do not perform calculations on these figures but simply use the mean number of participants per demonstration per issue/time period divided. Scores are divided by 1,000 for reasons of interpretation. Interaction variable Party manifestos We coded party manifestoes drawn up before the elections of 1991, 1995 and 1999 for each party that gained parliamentary seats using the same coding scheme and to test the same hypothesis: is protest more effective when it covers issues that have previously been found by the parties? Party manifestoes are issued in the months preceding a national election and we coded the party manifestoes from the eight parties that took part in all three elections that took place during the research period. Manifestoes were carefully encoded per (quasi)sentence following the classic methodology devised by the Manifesto Research Group (Budge, Klingemann, Volkens, Bara, and Tanenbaum 2001). In total and aggregated, 31,783 sentences of party manifestos were issue-coded. We calculated the share of attention devoted to each issue. This series thus indicates the aggregated importance all parties together attributed to each of the twenty-five issues. We use those values until the publication of the next party manifestos. The potential interaction effect of the political context is captured by multiplying the party manifestoes scores with the number of demonstrations and mean number of demonstrators. 23

24 Model We conduct various analyses, each time we consider another dependent variable; we perform analyses with parliament, government, legislation, newspapers and television news as dependent variables. It is important to note that we use different aggregation levels for each analysis. In fact, the temporal structure of the different dataseries varies extensively. Mass media coverage, for example, is recorded on a daily basis as there is new news every day. The analysis with mass media as dependent variable is based on truly daily (media, demonstrations) or intra-polated daily (parliament, government, legislation, government agreement, party manifestos) data. Lags are one day at a maximum; we expect media to react the same or very next day. For parliament and government the time unit is one week, with one week lags. Question Time in parliament is once a week and it makes sense that MPs would react the week immediately after a protest event. The same applies to the Council of Ministers. The models with legislation as the dependent variable draw on monthly data; we aggregate all legislation that has been passed in an entire month (as legislation is much rarer). The independent variables in the legislation models are lagged by one month but consist of the sum of their values during the three preceding months. This makes sense as legislation takes more time to be passed. Only very exceptionally does a law get passed in less than a month; the incubation and processing time of legislation is considerable. So, in short, we use a fairly complicated multi-issue and multi-lag design; the design of the study is graphically depicted in Figure 1. <Figure 1 about here> Our data have certain distinct characteristics that require a careful 24

25 consideration of the appropriate analysis technique: (1) We use a pooled data structure, including multiple scores for the twentyfive issues. For the various analyses we have 62,575 (2,503 days*25 issues), 5,775 (231 weeks*25 issues) and 2,232 (93 months*25 issues) observations. (2) A second characteristic of our data is apparent from the description presented above: it is taking the form of time series, with daily values as the units of analysis. As said, this offers opportunities in terms of more convincingly demonstrating causality and requires adequately modeling the dynamic structure of the series. (3) A third relevant element concerns the structure of our variables. The variables measure the occurrences of certain issues on certain agenda s and are therefore counts resembling a Poisson distribution. Furthermore, for most analyses data are overdispersed: the variance is usually larger than the mean. This latter characteristic makes the use of a negative binomial regression instead of a Poisson model appropriate. Combining these features of the evidence results in the choice for a longitudinal, pooled negative binomial model. However, within this type of analysis, again various options are available. The following considerations are of importance here: (1) The first question that needs to be addressed is whether the series are stationary, i.e. whether the mean of each series is unaffected by a change of time origin. We use the Fisher test that is based on the results of augmented Dickey-Fuller tests for each individual issue series. Results indicate that for all our dependent variables the null hypothesis of non-stationarity can be rejected. Consequently, the series do not have to be differenced. 25

26 (2) To establish the preferred type of analysis, it is generally recommended to first check for heterogeneity. Heterogeneity indicates the presence of panelspecific (in our case issue-specific) differences in the dependent variable that are not captured by the independent variables in the model. From a substantial point of view, it is highly likely that heterogeneity is present in our data: there are substantial differences across issues with some issues receiving way more attention by media and politicians than other issues. Fixed-effect analyses confirm heterogeneity for all our media. (3) Taking into account this heterogeneity, we have to choose between a fixedeffects or a random-effects analysis. The first resembles an analysis in which dummies for each of the issues are included as independent variables. The latter resembles an analysis in which for each issue a random deviation from a mean intercept is allowed. Depending on sample sizes (number of panels and time points) one or the other is more efficient. However, with a large number of time points the difference between the two ultimately disappears. For all our analyses, we conducted both a fixed effects and a random effects variant and compared the parameters for each of the independent variables using a Hausman test. This Hausman test indicated no or hardly any differences in the parameters obtained with both types of analyses. In general, the random effect models are slightly more efficient and therefore, we report the results from those models. (4) The last question that needs to be addressed is how to deal with the temporal structure of the data. Diagnostical statistics indicate that all dependent variables, except for the legislation, exhibit autocorrelation, indicating that the current value is correlated with the previous value(s). To control for this autocorrelation, we include the lagged value of the dependent 26