Delft University of Technology

Faculty of Technology, Policy and Management

PO Box 5015, 2600 GA Delft, The Netherlands

Voice: +31 15 2788052 Fax: 31 15 2783422

E-mail: p.w.g.bots@tpm.tudelft.nl

Dynamic Actor Network Analysis (DANA) is a systematic modeling approach to capture the cognitive and political complexity of policymaking situations. Where Operational Research approaches traditionally aim at supporting decision making by capturing the ‘objective’ world in computable models, the philosophy behind DANA is that this ‘objective’ world is not the determinant for decision making. Central to DANA is the assumption that the situations by which actors are influenced and to which they adapt themselves do not automatically relate to the 'objective' world of the policy analyst. Instead, they stem from their own subjective perceptions, constructed and reconstructed in the networked context of a resource-based and rule-guided dynamic interaction process. In this paper, the philosophy, modeling concepts, and computerized support functionality of DANA are illustrated by their application to the policy making context of water management in the Scheldt estuary.

Dynamic Actor Network Analysis (DANA) is developed in the LWI ‘university project’ EPSILON, the acronym for Estuaria: Perceptions- and Stakeholder-based Inquiry to Leverage Open-ended Negotiation. The original aim of this project was to develop a method and analyst workbench for a ‘quick scan’ of the stakeholders involved in policy processes related to estuaries in order to facilitate participative problem formulation. The main characteristic of such policymaking contexts is the variety in actors, their particular interests, problem perceptions, and rationalities. In such contexts, without a principal actor that can impose his problem perception on others, authoritative decision making requires a certain degree of consensus on what is exactly the decision problem. If actors do not agree on the problem definition, the impact that in-depth problem analyses using decision support systems have on the policy outcome will be severely limited.

Since many of the LWI projects involve the development of decision support tools, the issue of diverging problem perceptions is relevant. It is also a difficult issue to resolve. The aim of the EPSILON project was therefore more modest: the definition of a conceptual framework and definition of a modeling tool that (1) supports the identification of actors that are relevant to some policy situation, and (2) facilitates structuring the interaction between these actors in early phases of a policy process. The main function of the modeling tool would be to capture knowledge about actors and their problem perceptions and positions in a policy situation. More specifically, the EPSILON project should yield the following products:

• a generic actor model (modeling concepts and conventions) that allows a structured representation of the objectives, instruments and inter-relationships of actors in an ‘actor base’;
• a ‘quick-scan’ method for identification of actors that take an interest in a particular issue, given an complete ‘actor base’ for a certain problem area, such as an estuary;
• a (graphical) representation or ‘language’ to construct conceptual models of problem perceptions in such a way that (automated) systematic comparison is possible;
• multi-user computer support for ‘joint modeling sessions’ in which two or more people (analysts and/or delegates of corporate actors) can work together on a problem statement.

The EPSILON project started in November 1997. Presently (fall 1999), most of the envisioned products have been realized as prototypes and are in the process of being tested. The most important product is the DANA workbench, which embodies the ‘actor base’ and the perception modeling language. In this paper, we present the philosophy behind actor network analysis, the basic modeling concepts, and the functionality of the tool. In section 2, we focus on the characteristics of policy contexts that make their ‘objective’ analysis so difficult. We turn to perception modeling in section 3, where we argue that policy making is about a confrontation of subjective views, and that assuming the existence of an ‘objective’ reality may hamper, rather than help an analyst to understand a policy process. Using the Scheldt estuary as an illustrative case, we outline the functionality of DANA in section 4. We conclude this paper with some reflections on the practical use of DANA in actor network management and policy process design.

There are people who believe that policy problems are an objective condition whose existence may be established simply by determining the ‘facts’ in a given case: What are the newest unemployment figures? How many people are homeless? What is the number of people killed in a car accident? This naive view of the nature of policy problems fails to recognize that the same facts – for example, government statistics which show that crime, pollution, and inflation are on the upswing – can and will often be interpreted in markedly different ways by the different actors involved in a policy making process. The same policy-relevant information can (and will most often) result in conflicting definitions and explanations of a ‘problem’. This is not so much because the facts of the matter are inconsistent (though often they are), but because policy makers, policy analysts, and other the other actors involved hold competing assumptions about problems and solutions, means and ends, cause and effect.

There is no single correct view; problem definitions depend on the actors’ specific characteristics, loyalties, past experience, and even accidental circumstances of involvement. For an important part, policy problems are in the eye of the beholder (Dunn, 1981, p. 97). This applies to crime prevention or education, but also to health care, environmental planning, etcetera. Although there is a sense in which the policy problems we are dealing with in these sectors are objective, the same data with respect to these policy problems are typically framed from very different perspectives. The external conditions that give rise to a policy problem are selectively classified and evaluated. And what is more: depending on which explanation one chooses, the solution takes a different shape. Understanding the problem is synonymous with solving it. Every formulation of the policy problem corresponds to a statement of solution and vice versa. As a result the actors involved in a policy process may disagree on the definition and explanation of a policy problem, and even when there is consensus about this, they may yet disagree about its scope, severity, and importance.

To sum up, policy problems are products of subjective judgement. Policy problems exist only because the actors involved make judgements about the desirability of altering some problematic situation. Policy problems are therefore socially constructed, maintained and changed.

The inherent subjectivity and artificiality are not the only characteristics of the complex problems that policy analysts have to deal with in practice. In reality, policy problems are not independent and dynamic entities; they are parts of whole systems of problems, best described as messes (Ackoff, 1981). Problems are in constant flux; hence problems do not stay solved. Solutions to problems become obsolete even if the problems to which they are addressed do not. Moreover, policy problems are linked together. These connections sometimes circle back to form feedback loops. Opportunities for social change through public action in one area frequently affect the intensity and importance of policy problems in other areas as a side effect. So for instance, policies to reduce pollution may cause the unemployment figures to go up, because firms may have to close down. Policy problems may rarely be decomposed into independent, discrete, and mutually exclusive parts (Dunn, 1981, pp. 99-100). Every wicked problem can be considered as a symptom of another problem. At the same time there is no identifiable root cause; since curing symptoms does not cure problems, one is never sure the problem is being attacked at the proper level. Wicked problems have no definitive solution. It is important to realize here as well that the resolution of a policy problem may involve commitments that are very costly and, to a great degree, irreversible. There is not much room for trial and error. Once a solution is attempted, it is very difficult to undo what you have already done. In other words, most policy problems are wicked problems of organized complexity (Mason and Mitroff, 1981, p. 12).

Although policy making is often thought of as a process for solving problems, that is often not what happens. Problems are worked upon in the context of some choice, but choices are made only when the shifting combination of problems, solutions, and participating actors happen to make action possible. Quite commonly this is after some problems have left a given arena or before they have discovered it (Cohen, March and Olsen, 1972, p. 16). From this point of view, the policy process is a garbage can in which issues and feelings looking for decision situations in which they might be aired, solutions looking for issues to which they might be the answer, and decision makers and policy analysts are looking for work (Cohen, March and Olsen, 1972, p. 2).

In addition to all this we can say that the policy analyst has to work in an environment which also exhibits the following characteristics: the information available to the policy analyst is incomplete, the actors have multiple and conflicting objectives, there are conflicts or interest, and there is always more than one participant involved with power to influence the outcome (Radford, 1984).

In such messy contexts, how can a policy analyst do her job? We will argue that she should stay away from ‘hard’, solution-oriented models for the risk of a false fixation of the problem formulation. Instead, she should acquire knowledge by making a whole range of ‘soft’, perception-oriented models, trying to improve her understanding of how actors think. We propose DANA as an approach that facilitates this mode of policy analysis. By elaborating the actor network concept in depth, we try to convince the reader that it is rich enough to do justice to the essence of real-world policymaking, and that DANA may possibly support the policy analyst in crossing the intangible line between analysis and design/management.

From a naive perspective, the professional practice of policy analysis may seem a process of problem solving. Complex problems are solved through the selection, from available means, of the one best suited to established ends. But with this emphasis on problem solving, we ignore problem setting: the process by which we define the decision to be made, the ends to be achieved, the means which may be chosen. In real-world practice, problems do not present themselves to the practitioner as givens. They must be constructed from the materials of problematic situations which are puzzling, troubling, and uncertain (Schön, 1983, pp. 39-40).

Traditional demands, made through the literature and implicitly through the separation of solving from implementation, lead to a tendency to believe that time devoted to the activities of problem construction and making sense of the problem is time wasted (Eden, Jones and Sims, 1983). This tendency is illustrated by the requirement for actors in a given situation to express problems in terms of a solution; otherwise their contributions are not seen as ‘constructive’. The thinking that has gone into arriving at the suggested solution is camouflaged. What most of us do, however, is to ‘think around’ a problem; we redefine it a few times, we mentally simulate some of the possible outcomes from possible courses of action, we try to make sense of the situation (Eden, Jones and Sims, 1983).

The process of problem definition, which is an inherent part of the policy analyst’s role, does not essentially involve modeling an objective reality, but does entail identifying the actors whose realities (perceptions) are to be attended to and mapping these. The situations by which actors are influenced and to which they adapt themselves is not based on the ‘objective’ world of the policy analyst, but on their own subjectively perceived world. Finding out how actors perceive a situation or – to be more realistic – reflecting on the consequences if actor perceptions were such-and-so, may well be the primary task of a policy analyst.

DANA is a conceptual modeling approach which intends to portray the perceptions of actors and their relationship to one to another in a form which is amenable to study, analysis and (re-)design. The DANA workbench we are in the process of developing is based on the assumption that the situations by which actors are influenced and to which they adapt themselves do not stem from the ‘objective’ world of the policy analyst, but from their own subjectively perceived world. The representation of an actors’ perception in DANA is not supposed to be an objective reality in the way some influence diagrams are, but rather a representation of a part of the world as a particular person sees it. As a result, it can never be shown to be right or wrong, in an ‘objective’ sense (Eden, Jones and Sims, 1983).

This rejection of an ‘objective’ truth does not mean that modeling with DANA is by definition ‘invalid’ or ‘unscientific’. Through seeing as and thinking as an actor, a policy analyst using DANA produces knowledge that is objective, in the sense that it is refutable (in the scientific sense). She can discover that it cannot satisfactory forecast change or that the perceptions of one or more actors are not well understood. But the knowledge she produces is also subjective, bounded by the perceptions of the actors involved and restricted by the analyst’s perspective on the situation as represented in DANA, not the objective reality that exists beyond her personal angle and scope of observation.

It is of some importance here to realize that, using the words of Schön (1983, p. 163) the analyst functions as an agent/experient in the reflective conversation she has with a situation to construct her own perspective on this situation. Through her transaction with the situation, she shapes it and makes herself a part of it. Hence, the sense she makes of the situation must include her own contribution to it. Through the unintended effects of action, the situation talks back. The analyst, reflecting on this back-talk, may find new meanings in the situation which lead her to a new reframing. Thus, she judges a problem-setting by the quality and direction of the reflective conversation to which it leads. This judgement rests, at least in part, on her perspective of the potentials for coherence and congruence which she can realize through her further inquiry (Schön, 1983, p. 135).

The aim of the EPSILON project is to construct a workbench to support policy analysts in their representation and analysis of information on the relevant actors in a certain policy situation. Our DANA prototype offers a set of methods to collect and analyze information on the position of actors in a network. The design of the DANA workbench is largely determined by the underlying method of dynamic actor network analysis, which in turn is based on our interpretation of the literature this research area. This method leads the analyst to think in terms of actors who all have their own perception on the situation at hand. Since a full account of the structure of the DANA modeling language is beyond the scope of this paper, we merely outline its conceptual base and refer to Bots et al. (1999, 2000) for more details.

DANA is not an established theory, but rather an integration and formalization of the kind of approaches that can be found in the policy science literature. In general, there is no lack of theoretical notions of networks and actors within the policy science. There is, however, a lack of practical aid for empirical research on basis of the actor model and the network approach. A simple and accessible (ICT-supported) policy technology for research into networks and actors is in fact still absent (VanTwist et al., 1998). The development of DANA and its supporting workbench should therefore be seen as an experiment to test the added value (by enhancing reflection, producing original insights, serving as an external memory, or otherwise) of actor network analysis. The present tool design includes support for actor network representation, graphical rendering of perceptions, and a range of analytic queries. DANA provides a domain of knowledge that presupposes a continuous switching of perspective, from the outsider/spectator to insider/participant.

To illustrate our approach to actor network analysis, we present the concepts with which actor networks are modeled as a sequence of topics, some of which can be illustrated with a DANA screen. As the common factor of the papers in this volume is the Scheldt estuary, the case we use as an illustration here hardly needs introduction. For an elaborate analysis of Scheldt-related policymaking, we refer to Meijerink (1998).

The key to ensuring a more complete understanding of the policy problem at hand lies in finding out which actors are involved. An actor could be defined as an acting unit. This unit can be an individual person or a collective, like an organization or an institute. Actors can be public or private, or semi public. One reason to gather information on actors is to get an indication of whose objectives and interests are at stake in a certain situation. Information on the actors involved is also needed to determine if and how they will act in a certain situation.

Like so many transboundary water systems, the management of the Scheldt estuary is characterized by conflict an coorperation between a variety of actors. First and foremost, the set of actors includes public authorities: the Belgian and Dutch governments, the riparian regional authorities such as the province of Zeeland, and municipalities such as Antwerp, Flushing, Ghent and Terneuzen. At a higher level, the European Commission and the United Nations-Commission for Europe (UN-ECE) can be placed in this category. Then there are the private enterprises that have an interest in the region, such as the shipping companies in Antwerp, the fishing, manufacturing, and tourist industries in the area. A third category encompasses the special interest groups, such as Greenpeace. In our example, we limit ourselves to actors from the first category only.

It is assumed here that actors behave in networks. A network consists of actors and represents the relation between them. The network indicates the position and the influence of an actor in relation to the other actors involved. The term network is meant to describe the factual relation between the actors involved in a certain complex policy problem and the way in which they interact to exchange resources (for example money, authority, information, expertise) to achieve their objectives, to maximize their influence on what is going on in the network, and to avoid becoming overly dependent from other players in the game.

Figure 1. Scheldt actors and a networked representation in DANA

The distinction between an actor and a network is of an analytical nature and can be defined empirically in a number of ways, depending on the situation at hand, see also Marsh (1998). What we consider an actor can also be defined as a network and, conversely, what is defined here as a network may be regarded as an actor. The distinction, then, is only intended as an aid to the analysis. When we use the term ‘actors’, here, it is understood that any further interaction on their part is not the subject of our research. The term ‘network’ indicates the opposite: here, it is precisely the interaction between actors and the impact which this interaction has on the outcomes of a policy-making process for example which we wish to highlight.

That fact that we highlight in DANA the conceptual importance of networks and actors as vehicles for policy analysis does not, by no means, intend to say that formal organizational boundaries disappear in our view, or become inoperative an irrelevant. By breaking down the interaction within and between organizations as something that involves actors and networks, we gain analytical access to the dynamic interaction that takes places at the organizational and at the intra-organizational and inter-organizational levels.

As might be clear from our argument so far, actors have their own perception of the world that surrounds them. A perception is an image through which the complex, ambiguous world that surrounds an actor can be made sense of and acted upon. It guides the stimuli that the actor experiences and helps shaping the responses. In the literature, several concepts can be found that all more or less refer to the same phenomenon, although the meaning is never exactly the same: frames of reference (Rein and Schön, 1986), belief systems (March and Olson, 1989; Sabatier, 1988), theories in action (Argyris and Schön, 1978), causal maps (Weick, 1979), etcetera.

Each actor has subjective perceptions of the relevant factors with respect to the problem. Each actor also has perceptions of the action space of the problem, the linkages to other problems, the characteristics of the environment of the problem, the constraints on courses of action, and the possibility of occurrence of future natural and quasi-natural events. In addition, each has perceptions of the perceptions of the other actors concerning these factors (Radford, 1984).

Working from DANA, the perceptions of the actors in a network, in terms of relevant factors and actor-specific instruments and goals, should define the basis for evaluating relevant aspects in a specific situation. By making the perceptions explicit in a qualitative, conceptual language, the analyst can sharpen her insight by performing different types of comparative analysis. We define the perception of an actor to include all assumptions (regardless whether they are empirically true or false) that an actor makes about a situation. An actor acts on his perception in a rational way, i.e., his actions are consistent with his assumptions. In DANA, every actor perception is modeled in terms of factual, causal and teleological assumptions.

• Factual assumptions make a statement about the state of a particular factor, e.g., "water quality is adequate" or "sediment pollution < 50 mg Cd / kg".

• Causal assumptions represent the logic that an actor (says he) attributes to chains of events, e.g., "if strictness of water policies increases then freedom of navigation on the Scheldt will decrease", "if freedom of navigation decreases then economic position of Antwerp will probably decline", and so on. Formally, a link has a cause part and an effect part, where cause and effect are two different changes. Links may be uncertain, which is expressed by a hedge like ‘possibly’, ‘probably’ or ‘definitely’, typically associated with some value between 0 and 1. A link should be read as "if cause then certainty effect".
• Teleological assumptions represent an actor’s true (or stated) objectives. The actor’s insistence on a goal expresses how strongly he desires a factor to change, typically on some ordinal scale, e.g., from ‘preferably’ to ‘definitely’.

All three types of assumptions are the actor’s interpretations of factors. Change is the basic building block for models of actor perceptions: changes that an actor expects to occur anyhow, changes that he would (not) like to occur, and assumptions on how one change may lead to an other.

Figure 2. Specifying causal assumptions in DANA

The graphical language of DANA is a causal mapping convention similar to cognitive mapping as elaborated by Eden and Ackermann (1998), and supported by the Decision Explorer tool (see http://www.banxia.com). The main differences are:

• The nodes in the map are of two kinds: factors (modeled by noun phrases) and actions (modeled by verb phrases).
• A standard factor coloring convention: blue indicates a desired decrease, orange a desired increase. The color intensity reflects the actor’s insistence on a change. For constraints, the fill style of the oval is vertical bars, rather than solid fill.
• Factual assumptions are highlighted by showing the factor name in bold face.
• Causal links are not labeled with + or – ; instead, a solid arrow head indicates an increase, an open arrow head a decrease. Furthermore, the width of an arrow reflects the ratio of effect over cause and as the uncertainty in the causal assumption increases, a link takes on a lighter shade of purple.

The DANA dictionary provides the fundament for further analysis. It enforces ‘strong typing’ in the sense that all concepts considered relevant by the analyst fall into one of the basic categories of DANA: either arena, actor, factor (with its subcategory instrument), relation, or rule. Furthermore, it enforces name uniqueness within categories, effectively preventing the use of homonyms. Due to the formal description of perceptions in a database that maintains semantic integrity, a wide range of queries can be executed. For instance, questions like "Which factors are considered relevant by both actor A and actor B?" and "Which actors have conflicting goals on factor X?" could be brought to our special attention. In another paper (Bots et al. 2000) we explore a number of properties of actor networks that can be automatically derived from a DANA model, and discuss what insight they may provide to an analyst.

New concepts may be added either directly into the dictionary, or by means of DANA’s idea generation functionality: a brainstorming and organizing tool similar to e.g. the Group Outliner of Ventana’s GroupSystems, extended with DANA’s strong typing. Especially when several analysts are involved in the analysis, the use of this feature to generate and discuss concepts before entering them into the dictionary will not only mitigate the risk of synonyms, but also enhance their mutual understanding.

Actors who act strategically will, if it gives them an advantage, distinguish between their genuine beliefs and desires (their ‘private thoughts’) and that what they wish others to believe (their ‘public voice’). It may be strategic for an actor to make appear as though they aim for one thing, while the actually aim for something else. In such situations, the actor is said to have a hidden agenda. DANA can effectively represent such situation. One example – rather exaggerated for the purpose of illustration – is depicted in figure 3. The Dutch government may take the formal standpoint that their objections to dredging the Scheldt river bed and dumping dredged material elsewhere in the estuary are justified for ecological reasons. However, their hidden agenda might be that they are concerned for the position of the port of Rotterdam. If the maritime access to Antwerp is restricted, part of the transfer of goods through Antwerp will probably shift to Rotterdam. Thus, if the main objective of the Dutch government were indeed economic development, ratifying the UN-ECE convention would be the better tactic.

Figure 3. Distinguishing between an actor’s private thoughts and public voice in DANA

One important reason for the interaction that underlies the communication between actors is the interdependence that exists between them. Actors are dependent on each other because they need each other’s resources to achieve their goals. Interdependencies imply that there is something to be gained by the interaction between the actors involved. Interdependencies cause interaction between actors, which create and sustain relation patterns (Benson, 1978; Aldrich and Whetten, 1991). The actor who needs resources from another actor in order to achieve his objectives is dependent on that other actor (or on others who may take his place). This means actors have to exchange their go alone strategies for contingent strategies: courses of action tailored to the behavior of others (Ostrom, 1990).

In DANA, these concepts are made explicit, not only by representing different relations that (according to the analyst!) exist between the identified actors as arrows like in figure 1, but also because DANA facilitates analysis of the implications of the actor perceptions for their inter-dependencies. As is explained in more detail in (Bots et al., 2000), DANA can infer which actions or instruments are favored by which actors. If some actor favors an action that not he, but only some other actor can take, this particular resource dependence can be detected by DANA.

The interactions between the actors in a network have a game-like character, regulated by rules of the game negotiated and agreed upon between the network participants. The fact that actors are part of a network places specific restrictions on an actor. These restrictions take the form of values and norms, customs and rules of interaction which together form the arena in which the game is played. The rules give guidance to what behavior is correct in the interaction between actors and what is not, what is acceptable and what is not, what is appropriate and what is not. This includes aspects such as: With whom should you (not) speak? Are personal attacks allowed? Are discussions mainly concentrated in negotiation and/or conferences? Is publicity permissible?

Typical rules that DANA should help articulate are entry rules, exit rules, and rewarding rules. Entry and exit of actors in a network can be made formally verifiable by, for example, introducing an institutionalized membership as a condition of participation in interaction (Hayward, 1986, p. 16, as cited by Jordan, 1990, p. 327). Examples of networks in which this occurs are for example religious communities and associations of professionals. Illustrative of the closed nature of professional associations is the strong aversion traditionally shown by the (Dutch) medical profession to admitting ‘alternative healers’ to their ranks. Rules do not always have to be formally confirmed, by the way. Frequently, informal rules of behavior are developed within a network which in fact regulate the inclusion of actors in and their exclusion from the interaction within the network, without this being explicitly indicated in formal rules.

Initially, the conclusion of our argument about the arena and its rules of the game would seem to be that actors simply have to put up with this network-context. If they fail to comply with the rules, if they do not conduct themselves appropriately, the network closes itself off from them: they cannot communicate with it and they no longer form a part of it. Such a conclusion, however, would be too simplistic. Actors in a network are not weak-willed creatures who automatically go along with the rules of the game, if only because they have their own interpretive framework, their own frame of reference. They are able to react in different ways to the network culture, whether it involves the codes of conduct required of them, or the condition that they speak comprehensibly and intelligibly. Thus we can ascertain that an interaction exists between actors and networks. The behavior of actors gives structure to networks. At the same time, the structure of the network conditions the behavior of actors.

In this respect, networks are both medium and outcome of the reproduction of the actors interacting with each other. Networks are the precondition for action: without the rules and resources no meaningful interlinked interaction would take place. At the same time the rules and the distribution of resources may well evolve during the interaction between the actors. The rules and resources that constitute the network enable the interaction, but are also interpreted and changed during the interaction. This duality of structure is of course well known (Giddens, 1979).

Neither actor perceptions nor actor networks are static entities; they change in time. DANA will have to accommodate such dynamics. For this reason, the tool we are in the process of developing should be able to support the policy analyst by exploring a variety of scenarios. This will, however, require more sophisticated inference mechanisms than we have conceived for DANA so far.

Until recently the idea of networks was mainly used to explain why things go wrong and good policies fail. As a result the network-concept had a negative connotation associated with it. It was seen as an explanation for policy failure: the network as a non-transparent and impenetrable form of organization and interest representation, preventing policy innovations and threatening the effectiveness, efficiency and legitimacy of the policy process. A change in perspective is possible here, however (Kickert et al., 1997, p. 2). From this changing perspective, networks are seen as a fundamental and, moreover, inevitable characteristic of the policy process in modern societies. It is the task for the policy analyst to find ways to cope with these networks and even explore the potentials for public policy and modern governance. Used in this way, the concept of networks provides an alternative to the more conventional way of dealing with inter-organizational relations, which is bracketing or even ignoring them.

With a few exceptions (Hanf and Scharpf, 1978; Rogers and Whetten, 1982; Gage and Mandell, 1990), the potentials of networks for problem resolution and making better policies have received little attention (Kickert et al., 1997, p. 10). One thing is sure: managing networks is quite different from the management of organizations. Network management is in essence an inter-organizational activity. In a network situation, there is no single central authority; a formal hierarchy and a clear set of goals do not exist. None of the actors has the capacity to completely dominate the others. A top-down or even a holistic perspective is counterproductive in this situation. Hierarchical, top-down management does not work in a context where no top exists. Actors with the ambition to manage the network have to handle complex interaction settings and work out strategies to deal with different perceptions, preferences and strategies of the actors involved. Although there may be an officially mandated or self-selected overall-manager, the management in networks is different: it involves dealing with the relationships between interdependent but also more or less autonomous actors, having divergent goals and interests. In empirical terms the degree of problem solving success varies depending on the extent and the involvement of targeted individuals and organizations in ‘co-producing’ the cooperative effort. The kinds of circumstances in which no single actor can solve a problem nor compel others seek effective solutions are precisely the sort which indicate the need for some form of network management. Conventional management approaches are of little practical help in understanding and dealing with situations where the challenge is one of managing across the boundaries of single organizations. New perspectives are needed to address the problem solving requirements of such policy settings (O’Toole et al., 1997, p. 137).

As an approach to management, the network concept that is central to DANA underlines the highly interactive nature of policy processes while at the same time highlighting the institutional context in which these processes take place; an institutional context characterized by relatively stable relations between actors, sustained by ongoing resource flows between these actors. It is not our ambition here to give a exposé on the way DANA could be used as an instrument, not only for the analysis van actor-networks but also for the (re-)design and management of these. However, we would like to bring under attention here the fact that actors who try to manage the network have several strategies available to them. Without going into detail we will just present three examples here (Klijn et al., 1995 p. 439; O’Toole et al., 1997, p. 150):

• Bargaining and Compromise: One mode to stimulate agreement – regarding problem solving in a network context, or indeed anything else – is to encourage bargaining leading to compromises in which all parties get some but not all of their concerns met in the outcome. Network managers can assist these processes by identifying bargaining points and potential agreements, as well as by brokering bargaining processes and monitoring to encourage the maintaining of commitments over time.

• Changing Perspectives: The perceptions and preferences of actors are linked in networked patterns of interdependence; they are not constants but variables. Part of the challenge of network management is to recognize the circumstances in which apparent deadlock or unproductive compromise can be moved towards a more favorable outcome. Network managers can seek to shift the perceptions of the actors.

• Managing the Network-Context: Not only by encouraging changes towards compromise or by inducing changes in the perception of the networked actors, but also by changing the context of the interaction actors with management ambitions can try to alter the game. Sometimes they can re-structure the interdependence so as to create the possibility of cooperative outcomes where these did not exist before. This is certainly the case when the actor can at least formally fall back on certain unique resources (the law, for example). The trick here is to encourage movement towards new patterns of interaction that are more productive.

We believe that DANA may contribute to articulate and diagnose particular situations (trade-offs and win-wins, deadlocks and windows of opportunity, actor configurations and hidden agendas) and conceive and elaborate appropriate strategies.

By no means do we wish to conceal that DANA is work in progress, conceptually, practically, as well as technically. The modeling language will evolve as we perform more detailed case studies – no doubt, and so will the analyst workbench. The development of DANA is as much a scientific inquiry as it is an engineering job. But we have good hopes that, in due course, DANA will indeed provide effective decision support in multi-actor situations, if only by constantly reminding the analyst of the fluidity of the ‘problem’ concept in policy making.

This research was in part funded by the Dutch Land Water Milieu Informatietechnologie fund.

Ackoff, R.L. (1981). The Art and Science of Mess Management, Interfaces, Vol. 11, p. 22.

Aldrich, H. & D.A. Whetten (1991). Organization-sets, action-sets and networks: making the most out of simplicity. In P.C. Nustrom and W.H. Starbuck (eds.). Handbook of Organizational Design (pp. 3-27). New York: Oxford University Press.

Argyris, C. & D.A. Schön (1974). Theory in Practice: increasing professional effectiveness. London: Jossey-Bass.

Bots, P.W.G., M.J.W. VanTwist & R. VanDuin (1999). Designing a Power Tool for Policy Analysts: Dynamic Actor Network Analysis. In R.H. Sprague & J.F. Nunamaker (eds.) Proceedings HICSS-32, Los Alamitos: IEEE Press.

Bots, P.W.G., M.J.W. VanTwist & R. VanDuin (2000). Automated Pattern Detection in Stakeholder Networks. In R.H. Sprague & J.F. Nunamaker (eds.) Proceedings HICSS-33, Los Alamitos: IEEE Press.

Cohen, M.D., J.G. March & J.P. Olson (1972). A Garbage Can Model of Organizational Choice. Administrative Science Quarterly, Vol. 17, No. 1, pp. 1-25.

Dunn W. N. (1981) Public Policy Analysis: An Introduction, Englewood Cliffs: Prentice-Hall.

Eden, C. & F. Ackermann (1998). Making Strategy: the Journey of Strategic Management. London: Sage.

Eden, C, S. Jones & D. Sims (1983). Messing About in Problems, Oxford: Pergamon Press.

Giddens, A. (1979). Central Problems in Social theory. London: University of California Press.

Hanf, K. & F. Scharpf (eds.) (1978). Interorganizational Policy-making, Limits to Central Coordination and Control. London: Sage.

Kickert W.J.M. et al., eds. (1997), Managing Complex Networks, Strategies for the Public Sector, London: Sage.

March, J.G. and J.P. Olson (1989). Rediscovering institutions; the organizational basis of politics. New York/London.

Marsh, D. (ed.) (1998). Comparing Policy Networks. Buckingham, Philadelphia: Open University Press.

Mason, R. O. & Mitroff, I. I. (1981). Challenging strategic planning assumptions. New York: Wiley.

Meijerink, S.V. (1998). Conflict and Cooperation on the Scheldt River Basin. Doctoral dissertation, Delft University of Technology.

Ostrom, E. (1990). Governing the Commons: The Evolution of Institutions for Collective Action. Cambridge: Cambridge University Press.

Radford, K.J. (1984). Simulating Involvement in Complex Situations. Omega, Vol. 12.

Rein, M. & D.A. Schön (1986). Frame-reflective policy discourse. Beleidsanalyse, Vol. 4, pp. 4-19.

Rogers, D.L. & D.A. Whetten (eds.) (1982). Interorganizational coordination: theory, research, and implementation. Iowa: Iowa State University Press.

Schön, D.A. (1983). The Reflective Practitioner: How Professionals Think in Action. New York: Basic Books.

Sabatier, P.A. (1988). An Advocacy Coalition Framework of Policy Change and the Role of Policy-Oriented Learning Therein. Policy Sciences, Vol. 21, No. 2, pp. 129-168.

VanTwist, M.J.W., P.W.G. Bots, and R. VanDuin (1998). Dynamic Actor Network Analysis (DANA) – On the application of policy technology to institutional design. Proceedings NIG conference, April 1998, Oisterwijk, The Netherlands.

Weick, K.E. (1979). The Social Psychology of Organizing, Reading, Addison-Wesley.