• Aucun résultat trouvé

Quality of deliberation in the global justice movement groups: A multilevel analysis

N/A
N/A
Protected

Academic year: 2022

Partager "Quality of deliberation in the global justice movement groups: A multilevel analysis"

Copied!
34
0
0

Texte intégral

(1)

Working Paper

Reference

Quality of deliberation in the global justice movement groups: A multilevel analysis

GIUGNI, Marco, NAI, Alessandro

GIUGNI, Marco, NAI, Alessandro. Quality of deliberation in the global justice movement groups: A multilevel analysis. Geneva Laboratory in Political Science, Blue Lab, 2010

Available at:

http://archive-ouverte.unige.ch/unige:103602

Disclaimer: layout of this document may differ from the published version.

(2)

Quality of Deliberation in the Global Justice Movement Groups:

A Multilevel Analysis

Marco Giugni and Alessandro Nai

Blue Lab N° 1 • 2010

Geneva Laboratory of Political Science

(3)

                                                     

(4)

                   

(5)

Afin de mieux se présenter, le Département de science politique de l'Université de Genève a lancé en 2009 deux nouvelles publications. Sous le nom de Geneva Laboratory of Political Science, ces publications aideront à diffuser les travaux qui se font au sein du Département et à nourrir les échanges avec l'extérieur. Le nom souligne la vocation de « laboratoire » de ces textes, c’est-à-dire de lieu de réflexion, d'expériences et de débat. La publication se distingue par deux couleurs.

La série Blue Lab a été créée afin de favoriser la diffusion de travaux en cours de la part des membres du Département. Il peut s’agir de communications présentées à des colloques ou d’autres textes théoriques ou de recherche.

La série Green Lab accueille des travaux plus achevés et généralement plus longs. Elle est destinée en particulier à la publication des meilleurs mémoires de Master des étudiantes et étudiants qui achèvent leurs études auprès du Département.

(6)

Quality of Deliberation in the Global Justice Movement Groups: A Multilevel Analysis

_________________

Marco Giugni and Alessandro Nai

University of Geneva

(7)

Introduction1

Theories of social movements have traditionally focused on the origin of movement emergence and protest activities. Similarly, comparative analyses have focused on explaining cross-national differences in the movements’ mobilization and action repertoires (e.g. Kriesi et al. 1995; Rucht 1994), sometimes in their outcomes (e.g. Giugni 2004). Such comparative studies have often followed the political process approach, stressing the impact of different political opportunity structures on the mobilization and forms of action of social movements.

Another research tradition has looked more closely at the movements’ internal dynamics, focusing on their organizational structures and processes. These other studies have often referred to resource mobilization theory and focused on single social movement organizations. Yet very few works have compared organizational dynamics across contexts, specifically across countries. When this has been done, the main focus was on explaining variations in the structural characteristics or patterns of development of the organizations (e.g.

Kriesi 1996; Rucht 1996). To our knowledge, there are virtually no studies of the ways in which social movement organizations take internal decisions following a comparative perspective and, more specifically, of the ways in which conflicts arising during internal discussions are handled by the group.

This paper deals with conflict handling occurring in the global justice movement. We look at controversies arising during internal discussions within groups that are part of this movement in different countries. More specifically, we aim to provide some answers to the following two questions: (1) When are controversies more likely to emerge during group discussions? (2) How are such controversies handled by the group and, in particular, what is their deliberative quality? To answer these two questions we consider explanations located at three different levels of analysis: the agenda item, the group, and the country. We believe that the determinants both of the presence and the deliberative quality of controversies have a hierarchical structure that cannot be underestimated.

The data come from on a series of participant observations conducted on twelve groups active in the global justice movements in six countries (France, Germany, Italy, Spain, Switzerland, and the UK). We studied two groups in each country (see the methodological

1 We thank Donatella della Porta and Dieter Rucht for their valuable comments on a previous draft.

(8)

section for more details). The data thus retrieved are hierarchically structured. In other words, the data concerning a given level is embedded in the data at a higher level: we have two groups in each country and, in turn, each group has engaged in a number of discussions on different agenda items (sometimes giving rise to controversies). This requires a method of analysis that takes such an embedded structure of data into account and allows us to model both direct effects and cross-level effects: multilevel analysis.

Studying when controversies arise in internal discussions within social movements and how they are dealt with by movement activists is important not only because it tells us something about the functioning of social movement organizations, but also because it is a crucial aspect of the movements’ internal practices of democracy. In particular, the quality of communication is crucial for implementing a deliberative and participative view of democracy aimed at being as equal and inclusive as possible (Habermas 1992, 1996, 1997).

The ways in which discussions within the group or organization unfold as well as the ways in which decisions are reached are good indicators of the success in practicing deliberation and consensus in social movements, especially when confronted with conflictive or at least divergent opinions. Although a great deal of studies address theoretical or philosophical aspects of deliberation, empirical works are far less developed (e.g. Steiner et al. 2004;

Spörndli 2003; Giugni and Nai 2007, 2009). Especially in the field of social movements, empirical studies of deliberation are very rare. We have dealt elsewhere with the structural and cultural determinants of internal decision making in the global justice movement and in particular of the adoption of a model of decision making that stresses deliberation and participation (Giugni and Nai 2009). That analysis bares some resemblance with the one that will be carried here, but it is focused on the factors that can explain the adoption of such model by the movement’s organizations. By looking at how conflicts are handled during discussions held by movement activists in small settings during meetings and what can favor the quality of deliberation during such discussions, we aim to contribute further to this body of literature.

Anyone who has at least some experience with collective deliberation, in broad or narrow terms, knows that the ideal deliberative situation rarely exists. When a group of people discusses about the group’s strategic view, about the attribution of tasks within the group, about some sensible issues, or simply about what to do after the meeting, chances are high that the discussion will take a confrontational stance. When researchers deal with the

(9)

discursive dynamics within social movement organizations, as in the present paper, such a chance is probably even higher. Furthermore, the ways in which controversial discussions are handled take on a different meaning in different situations. The answer to the question of what makes a simple discussion turn into a controversy (or, conversely, of why a highly sensible topic leads to a peaceful and relaxed discussion), however, remains unclear. In other terms, which factors help create an ideal speech situation (Goodin 2006; Habermas 1992)?

Furthermore, under which circumstances do the deliberative processes within the group fail to reach this goal?

Controversies and conflict handling in social movements

A controversy can be defined as a discourse situation in which two or more participants in the discussion display divergent, if not opposing, view on a given issue.2 Such divergence can bear on various aspects of the discussion. In our observation protocol we distinguished between seven main subjects: internal organization (questions which mainly regard the rather practical organization of the group itself or the organization of the meeting, of rather short term relevance); internal structure (questions which mainly regard the group itself, but not just organizational matters but more fundamental issues of the long term structure of the group);

external delegation (questions about the delegation of group members for a meeting of another group or subgroup); external group action (organization and tactics: debates about all actions that present the group to a larger public, be it within the movement or in the public space, media or vis-à-vis representatives of the political system); strategic decisions (questions of collective strategy of the group: decisions which are not primarily directed at a concrete collective activity, but which have a long term perspective regarding the group’s strategic orientation); metadiscourse (discourse about the discussion: style of speech, personal conflicts impeding a factual discussion, but also discussions about what the result of the preceding discussion is supposed to be); and value conflict (questions of principle addressing fundamental values of the group).

2 The following operational definition of a controversy was used in the codebook for the participant observation: “A controversy is an explicit and extended verbal disagreement in the group. It starts as soon as a dissenting voice is followed by a reaction (e.g. a justification) of at least three speech acts or of a reaction that lasts more than three minutes. If two or more controversies can be connected to one and the same issue, they are considered as a single controversy.”

(10)

Controversies can vary not only according to their subject, but also on a number of other counts. In particular, we can characterize them along several dimensions pointing to the ways in which internal democracy works within the group. For example, one can look at the degree of reciprocity and symmetry among participants in a controversy. The former indicates whether most speakers refer to others’ positions during the discussion implicitly or explicitly or whether they make statements regardless of what other speakers said before, the latter to what degree speakers treat each other as equal discussants. In this sense, a reciprocal and symmetric relationship is a precondition for creating a viable setting for deliberation and the search for consensus (della Porta 2005), a form of democratic decision-making that Attac as well as other global justice movement groups often emphasize. Similarly, deliberative democracy and the search for consensus also presuppose that participants make use of soft power rather than hard power during a controversy. The former is communicative power based on words and symbols, and the arguments are its key source, whereas the latter is non- communicative power ultimately based on material, physical, or similar kinds of sanctions (e.g. expressing a veto, threat of exit, or majority rule). Controversies can also vary according to the degree of cooperation, civility, and the general atmosphere (tense, relaxed, etc.) during discussions.

Here we propose to look at the extent to which controversies arising during group discussions are deliberative and oriented towards the search for consensus. The quality of deliberation presupposes a number of features of controversies favoring deliberation. We focus on six such features: soft power (speakers emphasize rational arguments, as opposed to stressing sanctions rather than persuasion), reciprocity (speakers refer to others’ positions, as opposed to making statements regardless of what others said before), symmetry (speakers treat each other as equal discussants, as opposed to considering the other side as inferior or less important), cooperation among participants in the discussion (as opposed to competition), a relaxed atmosphere (as opposed to a tense one), and the absence of incivility (as opposed to its presence).3

3 Our view of deliberative quality is broader than the definition used in previous works, where it is often limited to such aspects as the use of soft power and symmetry. By including aspects characterizing the atmosphere in which discussions take place, we aim to provide a more comprehensive understanding of the quality of deliberation.

(11)

Now, the question is, firstly, when do controversies arise during group discussions and, secondly, how can we account for differences in their deliberative quality? To our knowledge, no coherent theoretical framework exists for explaining variations in the ways social movements handle controversies. Here we would like to address these issues from a comparative perspective taking into account groups active in different contexts (countries). In addition, we suggest that the answer to this question require that we pay attention to at least three levels of analysis: at the micro-level of the form and content of the discussion, at the meso-level of the characteristics of the group, and at the macro-level of the cultural and institutional context in which the discussion takes place. From an analytical point of view, we can draw from three strands of literature: symbolic interactionism and work on group dynamics at the micro level; organizational sociology and resource mobilization theory focusing on the group itself; and structural accounts of social movements stressing differences in political opportunity structures and political culture.

The first and most proximate determinants of the presence and handling of controversies in social movements are to be found in the very characteristics of the discussions. From a micro-sociological perspective, works in the interactionist tradition in sociology (Blumer 1969; Goffman 1959) have stressed long ago how individual and collective behavior depends on the specific social settings in which encounters have place and in the interactive dynamics emerging from it. Recent studies in the social movement literature have walked along this path to show the impact of micro-dynamics on the movements’ democratic views and practices (Polletta 2002). Referring more specifically to the discussion situation, the number of participants and their characteristics (social, individual), for example, are likely to affect the ways in which controversies occur and are solved. More generally speaking, factors pertaining to the specific setting in which discussions are held are likely to influence the ways in which controversies are handled by the group. Here we will focus on seven aspects. Three of them relate to the agenda items being discussed: whether the discussion on the agenda item is oriented to taking a decision, whether it is unilateral or multilateral, and whether it involves many or few participants. We will look at these three aspects in analyzing the presence of controversies. Four additional aspects pertain to the controversies themselves and will be considered, together with the decision-orientation and multilateral character of the discussion on the agenda item, in analyzing the deliberative quality of the latter: whether the discussion on the controversy deals with one issue (focused) or many issues (unfocused), whether it is

(12)

short or long, whether it involves many or few participants, and whether it bears upon a core issue for the group (salient) or a more peripheral issue (not salient).

Although we refrain to advance formal hypotheses given the lack of previous work on these issues, we might tentatively expect controversies to be more likely to emerge when discussions are oriented to taking a decision, they are multilateral, and they involve a large number of participants. Controversies should have greater chances to emerge in decision- oriented discussions to the extent that, since a decision must be taken, everyone will tend to holds more strongly its position, which leads to more conflicting discussions. In addition, multilateral communication favors the presence of controversies as there are more arguments on the table that may give rise to disagreements and conflicts. Finally, we might expect crowded discussions to increase the likelihood that diverging opinions arise which eventually lead to controversies. This expectation is consistent with what has been shown elsewhere (Giugni and Nai 2009). In that analysis, we show that smaller movements have a higher likelihood to adopt a more deliberative decision-making model. Similarly, for the very same reasons, decision-oriented and multilateral discussions should also improve the deliberative quality of controversies. In addition, we also expect the latter to be favored when the discussion on the controversy is focused, short, when the number of participants is low, and when it is not salient. Focused discussions leaves less room to bring new arguments that might be contested by others. Short discussions are more straightforward and less likely to produce fatigues, which might be a source of conflict. Discussions restricted to few participants, again, make deliberation easier. Finally, salient controversies should more easily lead to disagreements and conflicts.

Secondly, the characteristics of the groups are also likely to influence the presence of controversies and their handling. Resource mobilization theory provides us with analytical tools to grasp factors located at this level. This approach has emphasized the role of endogenous resources and organizational factors for the emergence and mobilization of social movements (see Clemens and Minkoff 2004; Edwards and McCarthy 2004). Again, we can borrow certain insights from this theoretical perspective to account for variations in conflict handling within the global justice movement. The latter may be influenced both by structural and cultural factors relating to the organizations involved in a controversy. On the structural side, we may for example expect larger and more formalized organizations to adopt different strategies for handling controversies than smaller and more informal groups, as they would

(13)

follow more established routines and procedures. On the cultural side, conflict handling is likely to differ according to the tradition of contention upon which rests the group. For example, groups and organizations that share the heritage of the new social movements tend to stress participatory and deliberative democracy in decision-making rather than follow the representative model (Giugni and Nai 2007, 2009). Concerning this level, here we focus on three aspects: the size of the group, its degree of homogeneity, and the type of communication usually adopted by the group. We might expect small and homogeneous groups to be less likely to encounter controversies in their internal discussions and, when this happens, to handle them in a more deliberative fashion. We also think that groups who tend to engage in multilateral discussions are more likely to face controversies and to deal with them in a less deliberative way.

Finally, at the macro-structural level, we consider the broader context in which discussions arise and, more generally, in which the groups engaged in them are active. The concept of political opportunity structures might prove helpful in this regard. Political opportunity theory certainly offers the most developed set of concepts and analytical tools to deal with variations in social movements today. This theory, or set of theories, stresses (cross- national) variations in social movements and protest activities stemming from their political context (see Kriesi 2004). Although most comparative works in this research tradition focus on movement emergence and mobilization, especially their action repertoires (e.g. Beissinger 2002; Kitschelt 1986; Kriesi et al. 1995; Rucht 1994), some scholars have looked at the impact of the movements’ political-institutional settings on their organizational forms and processes (Kriesi 1996; Rucht 1996). While these studies tend to focus on the structural characteristics of organizations such as their size and degree of professionalization, or on their development over time, for example towards a less grass-roots and more interest-group oriented functioning, we may as well apply this perspective to organizational communication processes and the ways in which internal democracy is practices, in particular when conflicts over a given issue arise among movement activists. Since, in the end, we are dealing with the quality of deliberation as a form of democracy, to capture this aspect in our analysis we refer to Lijphart’s (1999) well-known distinction of majoritarian and consensual democracies.

More specifically, we take into account both the executive-parties and the federal-unitary dimensions of his typology. In this regard, one might expect groups that are active in more consensual democracies in such a context to be less likely to engage in discussions in which controversies arise and to be more deliberative in handling them.

(14)

To be sure, our goal is not to test competing theories of conflict handling within social movements in general and within the global justice movement in particular. We have a more modest aim. However, our analysis will take some of the insights provided by these three research traditions in order to shed light on the ways in which the groups studied handle controversies emerging during internal discussions.

Data and methods

The empirical evidence on which our analysis is based has its origin in one part of the DEMOS project (see further Haug and Teune 2008). In this part we used the method of participant observation to study practices of deliberative democracy in the global justice movement. Two groups were analyzed in each of the six countries included in the project (including one at the transnational level). Here we exclude the transnational-level groups to focus on ones located at the (sub)national-level. The number of observations (meetings attended by the researcher) varies across the twelve groups, depending on the length of the period during which they were carried on each group and the frequency at which the groups met.4

The methodology for the retrieval of the data is based on the principle of participant observation; the protocol included a quantitative part in which we coded a number of features of controversies, such as their duration, the number of participants, the issue at stake, the mode of decision, and so forth. More precisely, we coded information on three distinct levels:

the session (the specific meeting of the group observed), the agenda item (the issue being discussed), and the controversy (a divergence of views occurring during a discussion).5 In this paper we use this quantitative information to describe and explain controversies in discussions by the selected groups of the global justice movement. Of course, not all agenda items discussed yield a controversy. In the first part of our analysis we look at the effect of a

4 See chapter 2 for the full list of groups studied and researchers who conducted the participant observation.

5 Unfortunately, the information concerning the characteristics of the session cannot be used for our purpose. We therefore do not use the session as a potential explanatory level. However, in addition to information about the controversy (first level of analysis), we include specific information about each group (second level of analysis) as well as certain characteristics of the country in which the group is located (third level of analysis).

(15)

number of variables located at three different levels (agenda item, group, and country) on the likelihood that controversies emerge in group discussions. This is our first dependent variable.

In the second part we shift from the presence of controversies in group discussions to the deliberative quality of those controversies that have emerged. Here the ways in which controversies are handled by the groups is our dependent variable. More specifically, we examine the effect of a number of variables located at three different levels (controversy, group, and country) on the degree of “deliberativeness” of controversies. Inspired by a similar approach by Steiner et al. (2004), we operationalize this concept through an additive index made of six components: the type of power used in the deliberation (hard, mixed, soft), the degree of reciprocity of the deliberation (low, medium, high), the degree of symmetry of the deliberation (asymmetric, mixed, symmetric), the competitiveness among participants in the deliberation (competitive, mixed, cooperative), the atmosphere or emotional charge of the deliberation (tense, mixed, relaxed), and the degree of incivility among participants in the deliberation (frequent, some, none). Each component can take three values (0, 1, 2) according to the characteristic at hand. Thus, a deliberation characterized by soft power, high reciprocity, symmetric, cooperative, relaxed atmosphere, and no incivility obtains the highest score in our Index of Deliberative Quality (IDQ), which we consider as a continuous variable ranging from a minimum of 0 to a maximum of 12.

The IDQ has both theoretical validity and empirical reliability. First of all, it is built based on a large literature on deliberation. We follow recent empirical works in combining certain dimensions of deliberation (Steiner et al. 2004; Spörndli 2003) which are also discussed in theoretical works on deliberation as discursive practice (Habermas 1992, 1996, 1997; Dryzek 1990, 2001; Benhabib 1996). Furthermore, our combination of six dimensions has high empirical reliability. Any other combination of these dimensions obtains lower reliability scores.

In both cases we assess the relative weight of explanatory factors pertaining to different levels of analysis in accounting for variations in the ways in which controversies are handled by the groups. The multilevel and hierarchical structure of our data (discussions are embedded in groups which in turn are embedded in countries) requires a technique of analysis allowing us to best exploit such data. The multivariate analysis spells out a number of interesting findings quite straightforwardly. However, we must stress a major limitation. As we

(16)

mentioned earlier, our data have a hierarchical structure, with explanatory factors distributed among three levels: the controversy or the agenda item (the same level as our dependent variable), the group, and the country. In standard regression models all the independent factors are entered simultaneously, thus disaggregating the values for the highest level variables into the number of observations at the lowest level (the characteristics of the highest level are assigned to the observations at the lower levels). Even if quite common in the literature, this entails two major problems. First, it underestimates the hierarchical structure of the data, overlooking the fact that the context influences the behavior of the entities that compose it (Books and Prysby 1991). Second and even more dramatically, disaggregating the values leads to “the miraculous multiplication of the number of units” (Snijders and Bosker 1999: 15) produced by the fact that the values of the highest level’s variables are assigned to each observation at the lowest level. The problem is straightforward: the statistical tests are biased, as they are based on a number of observations much higher than the actual one. The null hypothesis is therefore more easily rejected, and this leads to an overestimation of the significance level of the relationships (Hox 2002: 3). Briefly put, the higher the level of the variable entered in the model, the lower the robustness of the results found in our non- hierarchical model. A reasonable solution to this problem consists in building empirical models capable of taking into account the hierarchical structure of the data (i.e. the fact that observations are nested into higher levels factors). Hierarchical models (or, more simply, multilevel models) are designed to do so (Jones, Johnson and Pattie 1992; Snijders and Bosker 1999).

Our analyses are conducted through two different sets of hierarchical models. The first one concerns the factors that may cause the emergence of a controversy within a specific agenda item. As mentioned earlier, we believe that determinants located at three hierarchical levels can explain the presence of controversies in group deliberations. We suggest that the characteristics of the agenda item itself, those of the group deliberating, and those of the country in which the group is located play all an important role in the explanation of the presence of controversies.

On the first level (agenda item) we expect the presence of a controversy to be facilitated by a decision-oriented discussion, a multilateral discussion, and a high participation in the discussion. The decision-orientation of the discussion is a dummy variable based on a specific question in the observant participation questionnaire asking the coder to describe the specific

(17)

decisional procedure followed by the group. If the discussion did not end up with a decision, the coder was told to indicate that the question is non-applicable. The multilateral (vs.

unilateral) quality of the discussion is a dummy variable based on a question about the type of communication. We recoded the raw data on the type of communication as follows:

“discussion,” brainstorming,” and “go-round” are considered as multilateral communication, while “input or proposal,” “separate contributions,” “output productions,” and “other” (which turned out to be always interventions such as providing information or reporting to the group) were coded as unilateral communication. Participation in the discussion is a dummy variable obtained by recoding the absolute number of participants, starting from the median of the distribution of the overall variable. A number of participants between 1 and 4 are considered as low participation, while high participation occurs when at least 5 people actively participate in the discussion.

On the second level (group) we identify three potential determinants, as we expect controversies to be more likely to arise when the group is large, heterogeneous, and with an inclination for multilateral discussions. The size of the group is a dummy variable that distinguishes between small and large groups based on the median of the average participation in the distribution of meetings. The threshold used is an average of 11.3 participants in the meetings for a group. The homogeneity (vs. heterogeneity) of the group is a dummy variable built on our knowledge of each group and distinguishing between political or ideologically homogeneous groups and heterogeneous ones. The type of communication used by the group (multilateral vs. unilateral) is a dummy variable created by aggregating the types of communication of each agenda item discussed during the group's meetings.

On the third level (country) we look at whether given democratic and institutional settings is more likely to another to induce the emergence of controversies than others. Here we use Lijphart (1999) distinction between majoritarian and consensual democracies and consider Britain and France as majoritarian countries, Germany and Switzerland as consensual countries, and Italy and Spain as mixed models. Two dummy variables (majoritarian vs. others, consensual vs. others) will be entered in our models.6

6 As often happens in this kind of models, the number of observations in the upper structure is quite low (e.g., our data include six observations on level 3). However, this should not be problematic, even if there is a risk that significant relationships are found less easily. Hierarchical models are built based on the idea that relationships among independent factors on the lower level and the outcome (which is necessarily at the lowest level as well) may vary across different upper-level observations. This is not very different

(18)

While our first set of models deals with the factors that could explain the emergence of controversies, our second set of hierarchical models looks at the quality of the deliberation (IDQ) that occurs when a controversy arises. In other words, here we analyze what occurs once a controversy has emerged and, above all, under which conditions. Only discussions that are struck with controversies are therefore included in the second set of models.

The factors pertaining to level 2 and level 3 which we hypothesize as having an effect on the deliberative quality of controversies are the same as for the first set of models: the size of the group, its homogeneity, and the type of communication for level 2; the patterns of democracy (following Lijphart) for level 3. As determinants of the deliberative quality for level 1 we considered, like in the first set of models, the decision-orientation of the agenda item, the type of discussion (multilateral vs. unilateral), and the participation (high or low, but based here on the absolute number of participants during the discussion on the controversy and not on the agenda item as before).7 However, we also include three controversy-related factors: the focusedness of the discussion (based on a specific question the coder was asked to answer explicitly in the questionnaire), the length of the discussion (1-5 minutes, 6-10, 11-20, and 20+ minutes), and the saliency of the issue. The latter is a dummy variable obtained through a recoding of the variable classifying the subject of the controversy, assuming that controversies dealing with strategic decisions or principle values have higher saliency, while those dealing with internal organization or structure, external delegation or group action, and metadiscourse have lower saliency.

For each dependent variable we first look at the direct effect of the variables described above in each level (full model) and then at the combined effect between level 1 and level 2 factors (full model with cross-level effects). Specific interactions among variables pertaining to different levels can help us better understanding both the existence of controversies and

from the logic that compares similar regression models within different contexts and draws conclusions about the importance of the context on the differences found. Yet, multilevel modeling allows testing the effect of the upper structure in a straightforward manner. Some researchers argue that the number of observations in the higher level should be around 25-30 (see Luke 2004), but this is hardly consensual in the literature (see Gelman and Hill 2007).

7 A controversy exists only within a specific agenda item. We can therefore consider that the characteristics of each agenda item such as the type of discussion or the decision-orientation remain the same also during the controversy. The first level in the second set of hierarchical models therefore includes both the characteristics of the controversy itself and the characteristics of the agenda item within which the controversy arose.

(19)

their handling. Their estimation is a further advantage of multilevel models. The first set of hierarchical models is based on 510 observations on level 1 (agenda items), 12 on level 2 (groups), and 6 on level 3 (countries). For the second set of hierarchical models the samples remain the same for both levels 2 and 3, but are limited to 143 observations on level 1 (i.e.

143 agenda items in which a controversy arose). This lower number will probably have an effect on the robustness of our significance tests, as we shall discuss in more detail below.8

The presence of controversies during group discussions

When do controversies emerge during group discussions? We assume that the factors explaining the presence of controversies have a hierarchical structure and that therefore we need to take into account the characteristics of the discussion, those of the group, and the larger context in which discussions take place. Before we address this issue, however, table 1 provides a brief descriptive look at how the twelve groups studied score with regard to the content of discussions.

Table 1: Number of agenda items and controversies by group

Group Number of

agenda items

Number of controversies

Percentage of controversies

Ecologistas en Acciòn Cordòba (SPA) 24 24 100.0

Cordòba Solidaria (SPA) 9 7 77.8

Attac Berlin Financial Markets Group (GER) 22 14 63.6

Attac France (FRA) 42 25 59.5

Attac Geneva (SUI) 76 26 34.2

Forum Social Lémanique (SUI) 16 5 31.3

National Campaign on Water (ITA) 18 5 27.8

No-Vox Network (FRA) 43 9 20.9

Thanet Friends of the Earth (GBR) 40 6 15.0

Attac Florence (ITA) 66 7 10.6

Berlin Social Forum (GER) 67 7 10.4

Conscious Consumers (GBR) 87 8 9.2

Total 510 143 28.0

8 Since the dependent variable on the first set of models is dichotomous (presence or absence of controversies in the agenda item), we applied the Bernoulli algorithm. The dependent variable in the second set of models (deliberative quality of the controversies) is continuous. Therefore no particular transformation needs to be applied. Our models have been run with HLM 6.06 (Raudenbush et al. 2004). The first set of models has been run using Full PQL estimations, the second one using Full Maximum Likelihood (FML) estimations.

(20)

A group session (the unit observed) is divided analytically into agenda items understood as separate segments within the session. An agenda item typically comprises a discussion or a cluster of contributions related to one issue. The number of agenda items handled by each group varies considerably, both across countries and among the two groups studied in each country. For example, while the Spanish Cordòba Solidaria has dealt with 9 agenda items, for the British Conscious Consumers this number goes up to 87. At the same time, while the number of items handled by the two French groups is more or less the same, in the other countries the two groups vary very much in this respect. These differences, however, might in part be an artifact as the number of observations has also varied from one group to the other.

Discussions on agenda items sometimes give rise to controversies (discourse situations in which two or more participants have divergent or even opposing views and express them overtly). The twelve groups also differ in this regard, as the number of controversies vary from 5 (Forum Social Lémanique and National Campaign on Water) up to 26 (Attac Geneva).

Again, we observe important differences among groups within the same country. What is most interesting, however, is the percentage of controversies, that is, the how many agenda items have turned to divergent or opposing views. This gives us an indication of the degree of

“contentiousness” of discussions within the groups. In this regard, differences are as striking, if not more. The proportion of agenda items that have turned to controversies ranges from little more that 9 percent in the case of Conscious Consumers to a plain 100 percent for Ecologistas en Acciòn Cordòba (the overall average equals 28 percent). Here we find a somewhat more homogeneous distribution as groups in the same country tend to show relatively similar scores. This holds for Spain, Switzerland, less for Britain and Italy, and not really for France and Germany. The two Spanish groups are particularly “contentious” in this respect. In the case of the local group Ecologistas en Acciòn Cordòba, all agenda items have given rise to controversies, and the other Spanish group is not very far from this score. The two British groups, in contrast, both rank quite low concerning the percentage of controversies raised. Perhaps there is a country effect here. Given the low number of cases (two in each country), however, we refrain from drawing any conclusion about the impact of the context.

Table 2 shows the results of the multilevel analysis for explaining the presence of controversies in group discussions. At the most general level, our models clearly confirm that the data have a hierarchical structure. As we can see in the table's lower section, the interclass

(21)

correlation scores are quite high for both level 2 (group) and level 3 (country). The interclass correlation can be interpreted as the part of the variance that can be attributed to the higher level or as the correlation that exists between two level 1 units randomly drawn from a randomly selected level 2 unit (or, for level 3 interclass correlation, among two level 2 units randomly drawn from a randomly selected level 3 unit). The higher the score, the higher the correlation among units that are nested in the same upper-level unit (i.e. the stronger the hierarchical structure of the data).9 This can be seen as confirming the hierarchical structure of the data (i.e. the fact that the determinants are not all at the same level) and that such a structure has an important effect on the presence of controversies, although this does not necessarily mean that the variables we included in the models actually explain it. Multilevel models are therefore the most appropriate tool for analyze these data.

Table 2: Estimates of effects of selected independent variables on the presence of controversies (odds ratios from multilevel analysis with fixed effects)

Full model

Full model with cross-level interactions

Intercept .51 (.6) .49 (.6)

Level 1: agenda item (discussion)

Decision-oriented discussion on agenda item 2.62 (.2)*** 3.85 (.3)***

Multilateral discussion on agenda item 1.36 (.2) †† 1.29 (.3) High participation in discussion on agenda item 2.44 (.2)*** 2.53 (.3) **

Level 2: group

Large group 1.19 (.5) 1.29 (.5)

Homogeneous group 2.55 (.4)* 3.43 (.4) *

Multilateral communication 2.71 (.6) †† 2.67 (.7) ††

Level 3: country

Majoritarian democracy .2 (1.4) .2 (1.4)

Consensual democracy .44 (1.4) .45 (1.4)

Cross-level interactions (levels 1 and 2)

Decision-orientation * large group - .67 (.6)

Decision-orientation * homogeneous group - .25 (.7) * Decision-orientation * multilateral communication - 2.29 (.6) ††

Multilateral discussion * large group - 1.02 (.5) Multilateral discussion * homogeneous group - 1.4 (.5) Multilateral discussion * multilateral communication - .58 (.5)

High participation * large group - 1.2 (.5)

High participation * homogeneous group - .85 (.5) High participation * multilateral communication - 1.8 (.5) Interclass correlation (group) ρgroup = .41

Interclass correlation (country) ρcountry = .20

9 The interclass correlation for multilevel logistic models (as in our case, since the dependent variable is dichotomous) is calculated through the following approximation (Snijders and Bosker 1999: 224):

ρ = (σ2u0) / (σ2u0 + π2/3), where the variance of level-2 residuals (σ2u0) is divided by the total variance (i.e.

σ2u0 plus the variance of the logistic distribution for level-1 residuals π2/3=3.29).

(22)

N (level-1) 510

N (level-2) 12

N (level-3) 6

Dependent variable: presence of controversy. Distribution at level-1: Bernoulli.

Estimations computed through Full PQL interactive procedures. Se between parentheses.

††p<.2, p<.1, *p<.05, **p<.01, ***p<.001,

Let us now turn to the effect of the independent variables (full model). Coefficients are logistic odds ratios and should be interpreted as in non-hierarchical logistic regression models.

Our findings seem to confirm our expectations, at least for level 1 and level 2. The likelihood of a controversy within an agenda item is much higher when the discussion is decision- oriented and when the number of participants in the meeting is high. As expected, therefore, more crowded discussions or the knowing that a decision has to be taken at the end of the deliberation pushes toward a more confrontational discussion, which eventually ends up in an overt controversy between two or more participants.

A multilateral agenda item (i.e. a communication with interactive features such as discussions, brainstorming, or go-rounds) also increases the likelihood that a controversy arises. This is again in line with our expectations. The coefficients, however, are lower in this case, and the significance level is sufficient only if we accept a 0.2 threshold. This is obviously more inclusive than the threshold usually accepted in social science, but in this case it is justified by the relatively small number of observations (especially in the second set of analyses).

Our analyses also show some significant effects of level-2 variables. As hypothesized, a prevalence of multilateral style of communication within the group increases the chance that a controversy will emerge during the discussion on the agenda item. In addition, a homogeneous group is more likely to produce controversies. This could suggest that there were no interpersonal differences raising conflicts during discussions and that having similar ideological or political predispositions does not necessarily lead to the absence of confrontation in a debate. Finally and contrary to our expectations, we found no statistically significant effect for the size of the group.

Similarly, the institutional patterns of the countries do not seem to matter for explaining the presence of controversies in group discussions. As the table shows, neither the majoritarian democratic setting nor the consensual one have a significant effect on the

(23)

presence of controversies. Given the good results obtained in these models for the interclass correlation coefficients (which confirms that the upper levels of the data structure play a key role in explaining the presence of controversies), the fact that our level-3 variables do not have any effect suggests that we were unable to grasp the dimensions of the country level that affect our dependent variable. As the interclass correlation for the country level shows, about 20 percent of the variance can be attributed to level-3 variables. Unfortunately, our data do not allow us to determine which specific dimensions of the country level contribute to that.

The interactive effects between level-1 and level-2 factors do not add much to the picture. All independent variables remain statistically significant in this model, with the exception of multilateral discussions (level-1 variable). As before, however, controversies are more likely to appear when discussions are decision-oriented, when more people participate in the discussions about the agenda item, when the group is homogeneous, and when its prevailing type of communication is multilateral.

The right-hand column in table 2 (full model with cross-level interactions) also shows the shared effect of determinants at different levels. Among the full range of combinations, only two are statistically significant, but their interpretation is somewhat complex. Firstly, a decision-oriented agenda item and a homogeneous group seem to be partial substitutes of each other for the emergence of controversies within the agenda item. The interactive term between these two cross-level factors is statistically significant and much lower than 1 (which indicates the substitutive relationship between the two variables). Secondly, our findings also show a positive (although significant only at the 0.2 threshold) interactive effect between a decision- oriented agenda item and a multilateral style of communication within the group. These two conditions, however, work jointly and reinforce each other.

It comes as no surprise that the only two relevant cross-level effects concern an interaction between level-2 factors and the presence of a decision-oriented discussion. As we have seen earlier, both in the full model and in the model with cross-level interactions, this level-1 variable has proved to have the strongest effect on the presence of controversies within agenda items. Overall, however, we can conclude that the structure of the data has an interesting hierarchy among explanatory factors. If we aim to understand the presence of controversies, empirical models that “flatten” the explanatory factors in one same level either face major empirical problems or underestimate the importance of the upper structure of the

(24)

data, which is worse. That said, the higher level in our data has less effect than expected on the presence of controversies and the various levels are globally not strongly interconnected.

The deliberative quality of controversies

What explains the way in which a controversy is handled within a given group? Again, before we address this question in the next section, table 3 shows to what extent the twelve groups differ with regard to the deliberative quality of controversies. If we look at the distribution of the six variables that compose the IDQ across the twelve groups studied, we can see that they display some important variations across groups. Most of the scores are equal or above the average, therefore witnessing a fairly high deliberative quality, except in five cases: type of power for Attac Geneva, reciprocity for Thanet Friends of the Earth as well as cooperation for Attac Berlin Financial Markets Group, Attac France, and the Berlin Social Forum.

Table 3: Average scores on the Index of Deliberative Quality of controversies and its components by group

Group

Type of power Reciprocity Symmetry Cooperation Atmosphere Incivility IDQ

Attac Florence (ITA) 2.0

0 2.0 0

2.0 0

1.7 1

2.0 0

2.0

0 11.71

No-Vox Network (FRA) 1.0

0 1.8 9 2.0

0 1.7 8 1.7

8 2.0

0 10.44 Ecologistas en Acciòn Cordòba (SPA) 1.7

1 1.7 5

1.5 4

1.5 0

1.6 3

1.8

8 10.00 National Campaign on Water (ITA) 1.6

0 1.4 0 1.8

0 1.4 0 1.4

0 1.8

0 9.40

Cordòba Solidaria (SPA) 1.5

7 1.0 0

1.0 0

1.4 3

2.0 0

2.0

0 9.00 Attac Berlin Financial Markets Group (GER) 1.7

9 1.5 7 1.3

6 .50 1.1 4 2.0

0 8.36

Conscious Consumers (GBR) 1.6

3 1.0 0

1.2 5

1.2 5

1.5 0

1.5

0 8.13

Forum Social Lémanique (SUI) 1.4

0 1.0 0 1.2

0 1.2 0 1.4

0 1.6

0 7.80 Thanet Friends of the Earth (GBR) 1.5

0 .67 1.0 0

1.3 3

1.5 0

1.5

0 7.50

Attac Geneva (SUI) .85 1.2

7 1.1 9 1.0

0 1.1 5 1.6

9 7.15

Attac France (FRA) 1.3

6 1.8 0

1.2 8

.52 1.0 4

1.0

4 7.04

Berlin Social Forum (GER) 1.2 1.1 1.2 .67 .43 2.0 6.71

(25)

The last column in the table shows the average score of the IDQ resulting from averaging on the six indicators that compose it. The deliberative quality of controversies varies from a lower 6.71 to a higher 11.71 (out of a maximum possible score of 12). Attac Florence is the most deliberative group, the Berlin Social forum the least deliberative. Scores vary both across countries and among the groups in the same country. Interestingly, some of the groups that have a high percentage of controversies also display a high IDQ. This is the case, for example, of Ecologistas en Acciòn Cordoba. At the same time, however, Attac Florence is the most deliberative, but ranks low with regard to the percentage of controversies. Thus, the main trend emerging from these descriptive analyses is that there is no homogeneity in the ways in which conflicts are handled among the groups we observed.

Next we discuss the multilevel models explaining the different degrees in the deliberative quality of the controversies that have emerged during group discussions. As for the explanation of the presence of controversies, our empirical models rest on the assumption that the factors accounting for differences in the handling of controversies are to be found not only in the characteristics of the discussion itself, but also in some higher levels. As we shall see, this is both correct and wrong.

9 4 9 0

Total 1.4 1 1.4

4 1.3 8 1.0

9 1.3 5 1.6

9 8.22 Notes: IDQ is a 12-point scale, (0) low deliberative quality, (12) high deliberative quality.

Type of power: (0) hard power prevailing, (1) mixed situation, (2) soft power prevailing.

Reciprocity: (0) low, (1) medium, (2) high.

Symmetry: (0) asymmetric, (1) mixed, (2) symmetric.

Cooperation: (0) competitive, (1) mixed, (2) cooperative.

Atmosphere: (0) tense, (1) mixed, (2) relaxed.

Incivility: (0) frequent, (1) rare or some, (2) none.

(26)

Table 4 shows the results of these analyses. Like before, a look at the interclass correlation coefficients suggests that the upper levels of the data structure have an interesting effect on the quality of the deliberation during the controversy. About 20 percent of the variance can be attributed to the second level (group) and a little more than 10 percent to the third level (country).

This means, once again, that level-1units (agenda items and controversies) are somewhat correlated to each other within level-2 units and that level-2 units are somewhat correlated to each other within level-3 units. In other words, this means that a hierarchy of explanatory factors has to be taken into account in the explanation of the IDQ and that multilevel models are an appropriate tool for that.

Unlike what we found earlier for the emergence of controversies, here we find few statistically significant effects of the independent variables (full model). The coefficients should be read as traditional linear regression standardized coefficients. Only the decision- orientation of the issue has a significant effect on the quality of the deliberation.

Table 4: Estimates of effects of selected independent variables on the quality of deliberation (standardized regression coefficients from multilevel analysis)

Full model

Full model with cross-level interactions

Intercept 8.53 (.3) *** 8.66 (.3) ***

Level 1: agenda item and controversy (discussion)

Decision-oriented discussion on agenda item 1.39 (.5) ** .5 (.7) Multilateral discussion on agenda item -.1 (.4) -.17 (.5) Focused discussion about controversy .21 (.3) .17 (.4) Long discussion about controversy -.08 (.2) -.18 (.2) High participation in discussion on controversy -.11 (.5) .14 (.6) High saliency of controversy -.29 (.5) -.4 (.7) Level 2: group

Large group -1.13 (.9) -.83 (1.1)

Homogeneous group -.19 (.6) -.02 (.5)

Multilateral communication -.6 (.9) -.73 (1.3)

Level 3: country

Majoritarian democracy -1.58 (.8) †† -1.98 (.6)

Consensual democracy - 2.81 (.9) -2.75 (.6) *

Cross-level interactions (levels 1 and 2)

Decision-orientation * large group - .33 (1.3)

Decision-orientation * homogeneous group - 1.78 (1.5) Decision-orientation * multilateral communication - .1 (1.6) Multilateral discussion * large group - -1.8 (1) Multilateral discussion * homogeneous group - -.91 (1) Multilateral discussion * multilateral communication - -1.97 (1.2)

(27)

Focussed discussion * large group - -.21 (.8) Focussed discussion * homogeneous group - .06 (.7) Focussed discussion * multilateral communication - .27 (1)

Long discussion * large group - -.61 (.9)

Long discussion * homogeneous group - -.66 (.5) ††

Long discussion * multilateral communication - -.32 (.9)

High participation * large group - 1.56 (1) ††

High participation * homogeneous group - -2.2 (1.1) High participation * multilateral communication - -3.03 (1.6)

High saliency * large group - -1.14 (1.2)

High saliency * homogeneous group - -1.51 (1.3)

High saliency * multilateral communication - -1.26 (1.4) Interclass correlation (group) ρgroup = .21

Interclass correlation (country) ρcountry = .12

N (level-1) 143

N (level-2) 12

N (level-3) 6

Dependent variable: deliberative quality of controversy (DQI).

Estimations computed through Full Maximum Likelihood (FML) interactive procedures. Se

between parentheses.

††p<.2, p<.1, *p<.05, **p<.01, ***p<.001

Contrary to our expectations, however, the pressure coming from the awareness that a decision must be reached after the deliberation enhances the quality of the latter, although the relationship is not very strong. All other level-1 variables (qualities of the agenda item or the controversy itself) yield disappointing results.10 Neither the qualities of the discussion in terms of multilateralism, length, focusing, or participation, nor the saliency of the subject at stake have any effect on the way controversies are handled. When a controversy arises (which, as we saw earlier, can be explained by certain factors), the discursive and deliberative dynamics among participants are less influenced by the factors that produced it.

A similar reasoning applies to the group-level factors. Among the three level-2 variables we included in our models, none seems to affect how controversies are handled. This confirms what we found for level-1 variables: that which affects the presence of controversies does not necessarily have an impact on the way they are handled, and vice-versa. This holds even more once we specify the model by including cross-level interactions. Now none of the level-1 or level-2 variables display a significant effect, even considering the lower. The full model with cross-level effects does not really add anything relevant other than reinforcing our feeling that lower levels do not matter for how controversies are handled. None of the

10 We tried to add some interactive effects between level-1 variables, but this did not produce any relevant results.

(28)

interactive effects is significant at the 0.1 threshold, and their interpretation is rather difficult, but this is not surprising since no effect is found for level-1 and level-2 variables.

Our analysis shows, however, that the upper structure of the data (country) has an interesting effect on the quality of deliberation. The two variables measuring the cultural and institutional setting of the countries in which the groups studied are located are both significantly linked to the IDQ. This is especially the case for the full model with cross-level interactions. Yet the interpretation of the coefficients is not straightforward. On the one hand, groups located in majoritarian democracies, which are characterized by a strong concentration of power both within and among institutions, are less likely to handle controversies in a deliberative way. On the other hand, the chances to have a high deliberative quality of controversies also diminish in consensual democracies, where power is shared and more diffuse. While this makes sense in the former case, the fact that we observe a lower deliberative quality of deliberations in countries where there is a specific institutional culture of consensus is more difficult to understand. Furthermore, the negative coefficient for consensual countries is stronger than the one for majoritarian countries. How can we make sense of this seemingly contradictory result?

One way to do so is to recall that both variables are built by contrasting a given type of democracy (e.g. the majoritarian one) against other two situations: the opposite type (e.g. the consensual one) and the mixed type. Given that the effects concerning the two “pure”

democratic types are both negative and have more or less the same magnitude, we can but conclude that what really matters for the qualitative quality of controversies is the fact to be in a country characterized by a mixed democratic type. In other words, controversies are more likely to be handled in a deliberative way when groups are located in a type of democracy that combines a strong concentration of power within institutions and a decentralized institutional setting (i.e. a sharing of power among institutions) or vice-versa.

Our data are probably not robust enough to push this interpretation further. In our models for the explanation of the way controversies are handled, level-3 factors clearly seem to have the strongest effect. This can be seen both at the level of the data structure (interclass correlations) and on the effect of the independent variables pertaining to this level. More generally speaking, we aimed to test the hypothesis that the factors accounting for differences in the qualitative quality of controversies are to be found not only in the characteristics of the

(29)

discussion itself, but also in some higher levels. This hypothesis is confirmed only in part. On the one hand, the lower levels of the data structure (agenda items with controversies and group) do not have a strong effect on the IDQ. On the other hand, the upper level (country) has the strongest effect. All in all, however, we are not in a position to acknowledge what really enhances deliberation in the upper levels of the data structure, as we found signs that something is going on, but we only scratched the surface of the problem.

Conclusion

This paper has dealt with conflict handling occurring in the global justice movement.

Specifically, we tested a number of hypotheses about the presence of controversies during internal discussions within groups that are part of this movement in six European countries as well as the deliberative quality of such controversies and, therefore, of the discussions as a whole. To test our hypotheses we have applied multilevel analysis to a hierarchically structured dataset that includes indicators of the characteristics of the discussions within groups (agenda items and controversies), the groups themselves, and the country in which the groups are located.

Our findings suggest that controversies are more likely to arise when the discussion on the agenda item is oriented towards taking a decision, when it involves a large number of participants, and when the group is homogeneous. Two other variables are found to have an impact, but only at a lower threshold: a multilateral discussion on the agenda item and a multilateral style of communication within the group. Regarding the deliberative quality of controversies, we found a lower number of factors displaying an effect. The IDQ is only influenced by decision-oriented discussions on the agenda item and by the two country- related variables (majoritarian, respectively consensual democracy). The size of the group also has an impact, but only at a lower threshold. However, once we include the cross-level interactions in the model, only the two country-related variables hold. Furthermore, their effect is negative, meaning that the deliberative quality of controversies in both majoritarian and consensual democracies is likely to be lower than in mixed contexts.

Yet the most interesting result of our analysis lies perhaps in the fact that the two aspects that we have addressed are not explained by the same factors. Level 1 and level 2

Références

Documents relatifs

any doubly transitive permutation group G of degree greater than 2, and let .g be the stabilizer of one letter), the class of normal subgroups is properly

element in the compact semigroup SB we obtain the following propo- sition : a summable sequence (xn) in X is uniformly distributed if for every summable sequence (Yn)

They also agreed to develop, strengthen and implement effectively their national food safety programmes; to continue to monitor and evaluate their food safety activities; and

In the paper, we consider that governments act as Stackelberg leaders (stage 1), before trade unions set wages (stage 2), capital owners choose the location of their capital

To test whether queen homozygosity affects founding success, we compared the homozygosity of queens collected just after mating flight (i.e. before colony foundation) with that of

The semantics of a multilevel ontological dimension defines a legal instance of a syntactically correct dimension as a sequence of legal instances of the class models in the

Motivated by applications in mathematical imaging, asymptotic expansions of the heat generated by single, or well separated, small particles are derived in [5, 7] based on

2014 Health Behaviour in School-aged Children study (n ¼ 131,101), multilevel regression models with cross-level interactions examined whether country-level meritocratic