INTRODUCTION

In spite of the introduction of a range of reforms generally referred to as “new public management” (NPM; Hood, 1991; Osborne & Gaebler, 1992), the delivery of public services continues to pose problems at local, national, and EU levels in terms of cost, outcome, configuration, and governance. Europe as a whole and its individual member states grapple with these issues with mixed success (Dunsire et al., 1994; Mackintosh, 1998; Hood, 2000; Kay, 2002). It is worth noting that, after 15 years of “new” public management, OECD figures show that taxes and government outlays as a proportion of GDP in the EU were roughly the same in 2002 as they were in 1987, although net government debt soared to 49 percent from 37 percent over the same period.

Although NPM has yet to fulfill its promise to improve significantly on either the costs or effectiveness of the public sector, it has nevertheless significantly increased the overall complexity of public services. Public service systems now encompass a broad range of public, private, and nonprofit organizations that may work independently, in partnership, or even in conflict with one another to meet the needs of the consumers and/or citizens in their domain of operation. Emerging network theories of public policy and management (Rhodes, 1996, 1997; Kickert et al., 1997) have added to the complexity of public administration theory by introducing a new level of “actor” in public service systems—the network itself—as well as highlighting the importance of interactions among participants in contributing to the outcomes of the system. Finally, the concept of social capital, and how it is created and/or destroyed by institutions and their interactions (Fukuyama, 1995; Putnam, 1995; Adler & Kwon, 2002), has added yet another thread to the mosaic of public administration theory that confronts researchers and practitioners.

The structural changes resulting from NPM reforms, along with the emerging awareness of network and social capital effects, have created a need for new models of public service systems to integrate the various threads of theory and practice and improve our understanding of the public service domain. Several authors (Pierce, 2000; Blackman, 2001; Chapman, 2002) suggest that CAS (complex adaptive systems) theory has considerable potential as a framework for the development of the required models in public administration. Organizational complexity theorists (Anderson, 1999; Stacey, 2000) argue that CAS theory can be applied to develop useful analyses of organizational systems in general. However, critics of attempts to apply complexity and/or CAS models to organizations cite the overly general nature of CAS theory, the lack of empirical verification of claims, and the use of concepts not suited to an organizational context (Rosenhead, 1998; Arndt & Bigelow, 2000). Even supporters such as Anderson (1999) admit that there is much to be done before CAS theory can be effectively applied to organizational contexts. As yet there are no examples of how this framework might be operationalized in models of public service systems.

This article seeks to progress the discussion of how CAS may be applied to public service systems (PSS) by describing the core elements of a CAS framework and briefly describing how the application of this framework could address some of the outstanding issues in public administration theory and practice. Applying this framework, however, requires further clarity around the elements of CAS in an organizational context, and we propose more precise definitions of CAS elements for the purpose of modeling PSS, based on existing organizational and public administration theory. However, it is first necessary to establish the basic definitions of CAS and PSS that delineate the scope of our analysis.

DEFINITIONS OF CAS AND PSS

BASIC ELEMENTS OF A COMPLEX ADAPTIVE SYSTEM

While readers of Emergence are likely to be familiar with the elements of a CAS framework, the following brief description is provided as context for the subsequent application of this framework to public service systems. More detail and examples of working with CAS in an organizational context may be found in Organization Science, Special Issue on Applications of Complexity Theory to Organization Science, May-June 1999. In terms of the core CAS elements to be defined for modeling PSS, the following are proposed:

  1. A description of the agents that make up the system. In an organizational system the definition of the agent is still somewhat controversial, with researchers opting for individuals (Pavard & Dugdale, 2002), organizations (Stacey, 1996), value-chain activities (McKelvey, 1999), and/or individuals and organizations (Carley, 2002). In this article we propose two general types of agent: the individual and the organizing initiative. The case for the latter is described at some length later.

  2. A description of the nature of the schemata for given agent types. This is comprised of the inputs to the agents’ decision processes along with the filtering rules that the agent applies to decide what information to process. Holland (1998) describes the inputs as including a “set of generators” or the totality of all possible actions/decisions that agents are capable of taking1; “transition functions,” which are the set of all possible pairings of current and future states of the agents in the system; and “strategy,” which is a combination of the concept of desired outcome or “fitness function” (see below) and the agent’s understanding of the series of steps that will lead to the outcome desired. Gell-Mann (1994) describes the filtering process as one of “coarse-graining” of information and Holland (1995) contributes the concept of “tagging” of similar observations for the purpose of creating cognitive groupings. Finally, Stacey (2001) suggests that the process for the mutation and adaptation of schema must be defined as part of the elaboration of CAS schemata.

  3. A description of the fitness function(s) for agents in the system, which may be either exogenous or endogenous to the system, or possibly a combination of both (Gell-Mann, 1984; Holland, 1998). Fitness functions govern how the agent will choose among alternative actions and Anderson (1999) provides examples of some likely fitness functions for private firms (e.g., returns to exploitation, returns to exploration, returns to reputation, market position, etc.). However, public management literature would suggest that these are too limited a set of potential functions for PSS (Stewart & Ranson, 1994; Johnson & Scholes, 2001).

  4. The nature and level of connections among agents. Defining the nature of the connections among agents is an ongoing discussion among academics engaged in formulating these models. Connections can take the form of exchanging information relating to the “state” of the agents (cellular automata), the giving and receiving of instructions (neural networks), and/or the exchange of genetic material (genetic algorithms; Anderson, 1999). Any one or all of these could have relevance in the case of public service systems. In addition, the number of agents and the number of connections between agents are likely to be important factors in modeling the system (Kauffman, 1993), as will the cognitive and relational complexity of the information that is exchanged and processed (Boisot & Child, 1999).

  5. A set of dimensions that describe the state of the system. Holland (1998) uses the game of checkers (draughts) as a simple source of examples for each of the elements of a complex adaptive system. The state of the system in a checkers game is the current position of all of the pieces remaining on the board. For a public service system, the definition of the elements that constitute the relevant state of the system is a rather more difficult exercise, and one that public administration researchers are currently working on in their efforts to define appropriate outcome measurements for public service (Boston, 2000; Paton, 2003).

Specifying the five elements described above for public service systems is part of the research program we have undertaken. This article reports on the preliminary findings from empirical research into the housing system in Dublin, Ireland and further theoretical development of CAS elements based on existing organizational and public administration literature.

DEFINITION OF A PUBLIC SERVICE SYSTEM

Currently, there is no generally accepted definition of a public service system in public management/administration theory that covers the broad nature of a phenomenon that we wish to explore. Furthermore, public administration theory is in a state of flux, with as many as six schools of thought as to the appropriate focus and identity of the discipline (Stillman, 2000). For the purpose of this article we propose a definition of PSS that is based on three related definitions drawn from public administration/public management theory.

Dimock et al. (in Stillman, 2000: 2) define the focus of public administrators as “the production of goods and services designed to meet the needs of citizen-consumers.” McKevitt & Lawton (1994: 2) contend that “public management is primarily concerned with the creation and realization of collective values.” Finally, Agranoff & Maguire (2001: 296) address the multi-agent aspect of public service systems when they define public management networks as “multi-organizational arrangements for solving problems that cannot be achieved, or achieved easily, by single organizations.” Drawing on these, we propose the following: Public service systems consists of multiple organizations engaged in the provision of a specific set of goods and services that are of value to the majority of consumer-citizens.

Note that this definition has two limiting cases: that of a monopoly (which may be either state owned or privately owned) and that of a service that is of value to less than a majority of consumer-citizens. In the first case, CAS theory is unlikely to have much more to contribute to the understanding of the dynamics of the system than traditional public management theory or economics. In the latter case, we suggest that, in an age of constrained resources, priority of attention should be given to those public service systems that serve the majority of citizens.

There are numerous service systems that would come under this definition that would not normally be considered public service systems. The automobile industry, the food industry, even the toy industry provide goods and services of value to the majority of citizens in most countries. The fact that these products and services are provided by private enterprises operating in the market (albeit often heavily subsidized and/or regulated) does not exclude them from being public service systems under the definition above.

That being said, we have confined our discussion to public services that Rose (1979) lists as having been core components of public service for decades. He described three classes of public service: defining activities, such as defense, internal policing, and establishing the legal and financial systems; mobilization of resources, including the activities of the departments of commerce, industry, agriculture, transport, communication, and public works; and social services, including healthcare, education, housing, and the provision of welfare benefits. Rose suggests that the third category is the most problematic for public policy makers, as the relations between cause and effect are unclear, there is often conflict over the goals to be achieved, and the services are frequently provided by a mix of public, private, and voluntary service providers. We concur with Chapman’s (2002) view that the understanding and delivery of services in this third category are likely to benefit significantly from the application of a CAS framework.

DEVELOPING THE CAS FRAMEWORK FOR PSS

Our proposed definitions of CAS-PSS elements are based on existing organizational and public administration theory, which we then further explore through a case study of the housing system in Dublin, Ireland. Three of the basic five elements of a CAS discussed in the previous section are defined below through a careful application of relevant organizational and public administration theory to the concepts proposed.

DEFINING AGENTS IN PSS

Organizational CAS theorists generally start with human beings as the key agents in a system and the organization as the system to be studied. Stacey (1996: 47) argued that complexity theory applied to “human systems, that is, individuals, groups, organizations, and societies.” However, more recently he has shifted his ground. He now questions whether it is always useful to think of organizations as a noun: “We are arguing for a move away from understanding ‘the organization’ as a system… we are interested in understanding the process of organizing as the ongoing joint action of communication… No one steps outside it, operates on it or uses it, for there is no simply objectified ‘it.’” (2000: 186-7). McKelvey (1999) suggests that neither an organization nor an individual is a fundamental agent in an organizational system, but that activities within the organization—which he groups into value chain competencies following Porter (1985)—are a more appropriate focus for modeling the behavior and performance of these systems.

Following Stacey and McKelvey, Rhodes and MacKechnie (2001) proposed that the basic unit of organizing is probably best conceived of as an initiative, which is defined as the cooperative “interlocked behaviors” (Weick, 1969) of individuals engaged in exploiting the potential of the division of labor through one of the organizing modes. Individual participants are also agents, as they can choose to participate or not in organizing initiatives. Initiatives are, in fact, processes engaged in by individuals that are essentially transient in nature, although a specific initiative may well continue relatively unchanged over a period of time. In addition, the individuals engaged in these interlocking behaviors may decide to establish a legal existence for the initiative, which further cements its existence over time.

It is clear that in public service systems there are a large number of well-established initiatives comprising recurrent and institutionalized interlocked behaviors (often buried in large formal organizations). There are also many reasonably persistent initiatives that are less stable and cannot rely on recurrent behaviors, but need to revise their interlocked behaviors frequently to cope with changing circumstances. A particular “market” is this type of initiative, as it is comprised of the many transactions (each of which is itself an initiative) among individual agents exchanging relevant goods and services. Firms in a “dynamic network” as described by Miles & Snow (1986) may also be considered to be both persistent and subject to rapid change as circumstances require. Finally, in any organizational system there are normally a large number of new initiatives, stimulated by new opportunities for successful exercises in division of labor, some of which may persist but might just as easily be abandoned if the value they create for their participants disappears. Public-private partnerships, government task forces, and many community projects are examples of this type of initiative in public service systems.

It is important to recognize that the behavior of a given initiative is not reducible to the sum of the behaviors of its component parts, as the interlocking behavior among participants generates its own patterns and logics that influence the behavior of the participants over time. “Emergence” is the term used in CAS theory to describe the phenomena of patterns at a higher level of abstraction that arise from interactions among lower-level agents. The basic dynamic underlying the emergence of initiatives is a process of “enactment” (Weick, 1969) by the participants, an exercise of imaginative intelligence, which employs feedback mechanisms but is not limited to simply a mechanical process of stimulus and response. All actions are based on interpretive processes (Burns & Stalker, 1961; Boisot & Child, 1999); that is, initiatives develop according to the way the organizers interpret the possibilities of creating some form of value through division of labor. This means that the behavior of initiatives will not be consistent nor easily predictable, since, almost invariably, interpretation will tend to vary among different organizers, even in very similar circumstances. Indeed, we suspect that these characteristics of organizational initiatives—interpretation, enactment, and emergence— are a principal source of nonlinearity in organizational systems.

This proposal raises the question of how the initiative relates to the more familiar concept of the organization. We suggest that, as initiatives become established, it becomes useful to define and stabilize the interlocked behaviors required to sustain the initiative. The most common way of doing so since the mid-nineteenth century has been to place the initiative in a framework of social understandings and legal entitlements. The result of this is the creation of what we know as “the organization.” However, we are arguing that it is crucial to recognize that the legal, social, and psychological trappings of “the organization” are there as a means of facilitating the survival and effective performance of the initiative; they are not the thing itself. Indeed, it is commonly noted that organizing often develops quite independently of organizational boundaries. For example, Karpik (1978) found that in technically advanced industries scientists from buyers and sellers worked closely together and had very little interest or contact with members of what was ostensibly their own organization. Similarly, Womack et al. (1990) found that in the automobile industry it seemed to make little difference whether assemblers acquired components from sources inside or outside their organizational boundaries.

ORGANIZATIONAL “MODES” AS A CORE ELEMENT OF AGENT SCHEMA

Organizational theorists have long emphasized that achieving difficult exercises in organizing is not a natural or straightforward task, but requires quite exacting patterns of relationships among humans, tasks, information, and resources (Chandler, 1962; Galbraith, 1977; Mintzberg, 1979). In this article we will use Galbraith’s (1977) phrase “organizing modes” to describe these patterns, as he was among the first to emphasize different modes of relations among individuals as the core characteristic of organizing. After decades of research and theorizing about alternative organizational modes in organizational contingency theory (Burns & Stalker, 1961; Mintzberg, 1979), institutional theory (Williamson, 1975; Powell, 1990), and sociology (Perrow, 1967; Thompson et al., 1991), it may be argued that markets, hierarchies, and networks represent a reasonable set of fundamental alternative organizing modes for human productive activity. Boisot & Child (1999) support this basic range of alternatives,2 arguing that these different patterns of relationships constitute ways of coping with informational complexity for adaptive agents in organizational CAS.

In brief, market relations are those in which transactions or exchanges between agents, principally but not exclusively short term, are the fundamental form of interaction. Hierarchical relations are those in which one agent has “power” over another agent to direct behavior. Network relations involve mutual dependency between agents and some level of trust around shared objectives. A more elaborate discussion of the difference between these three modes of organizing may be found in Powell (1990). Note that these modes are not the same as the alternative ownership structures often debated under NPM, in which public, private, or nonprofit forms of ownership are contrasted. Future research into the alternatives for organizing public service systems should clearly distinguish between organizing mode and ownership, as these are not equivalent organizational characteristics.

To summarize, a wide range of theorists across several disciplines spanning decades of organizational research and theory development would suggest that human agents in organizational systems may form initiatives via one of three modes: market, hierarchy, or network. The choice of mode is part of the cognitive framework that an agent will apply to interpret its environment, as well as having an impact on the performance of the agent under different environmental conditions. In CAS terms this translates into the organizational mode being both part of an agent’s schema as well as a component of its fitness function.

In addition to organizational mode, there are a number of other characteristics that management researchers have identified as key to understanding the behavior of a particular initiative. In the interest of brevity, we have identified the relevant references for each characteristic, rather than reviewing the arguments of the authors. Drawing from central themes in organization theory, we identify four additional elements:

  1. A set of techniques (technology), learned or devised by the participants, which is capable of achieving the benefits of division of labor (Mintzberg, 1979).

  2. The size and age of the initiative, which will influence the choice of organizational modes (Chandler, 1962).

  3. The level and type of resources controlled by the initiative (Pfeffer & Salancik, 1978).

  4. New knowledge and information to give guidance on how these elements can be combined effectively over time (Senge, 1990; Nonaka & Takeuchi, 1995).

These elements, along with the organizational mode and value proposition (see later), may well comprise the dimensions of the organizational demographics of agents sought by Anderson (1999). We would propose that these demographic elements are part of the initiative-agent’s schema, as they will affect both how the agent interprets the world around it and also how other agents interpret the behavior/state of the initiative itself.

INDIVIDUAL AND INITIATIVE-AGENT FITNESS FUNCTIONS

If, as we suggest above, the creation of an initiative by individuals is based on their perception of the potential for creating value, it raises the question: “What type of value?” This is central to the definition of the individual agent’s fitness function. One definition of “fitness” for the individual in an organizational system is the expected value to be generated from either joining/remaining in an existing initiative; creating a new initiative with one or more other individuals; or exiting from an existing initiative. Therefore, for individuals we propose a fitness function relating to the decision of the individual to participate in a given initiative. Initially, the perceived value of participation could be set up as a random variable, although, over time, individuals may come to prefer to participate in particular types of initiatives.3 A more precise specification of what constrains or encourages participation in particular initiatives will require further research.

In relation to the initiative itself, however, Rhodes & MacKechnie (2001) argued that it would rarely be the case that participants in a given initiative would all perceive the same value being derived from the initiative, and that it was therefore impossible to establish an aggregate view of the value proposition of an initiative based on the value perceptions of each participant. However, public administration theory considers the value orientation of agencies (and their staff members) as critical to their performance (Aucoin, 1995; Cohen, 2001). Furthermore, the concept of “core values” in organizations is well established in both public management literature (Stewart & Ranson, 1994) and organizational culture literature (Hofstede, 1980; Kotter & Heskett, 1992) as the basis of difference among organizations. There is little consistency among authors around what constitutes the content of “values,” although there is general agreement that public-sector managers have a more difficult job than private-sector managers in responding to competing value priorities. Given the above, it appears that establishing the value proposition relevant to a given initiative is crucial to the definition of its fitness function.

CAS organizational theorists generally agree that the notion of fitness function is difficult to pin down. Some consider the definition of fitness to be an emergent property of agents (Gell-Mann, 1994; Stacey, 2001), and therefore specifying the content of fitness functions is secondary to specifying the agents’ processes of defining fitness and then acting based on that definition. This dynamic process is one that we propose should be incorporated into a CAS model of public service systems. However, this does not imply that there are innumerable alternatives for defining the value propositions for initiative-agents in these systems. Interactions among agents and the learning that arises therefrom (Senge, 1990), processes of structuration (Giddens, 1984) and of isomorphism (Dimaggio & Powell, 1983) are likely to lead to persistent patterns of value propositions among initiatives. We therefore undertook to identify patterns of value proposition for agents in the housing case study in order to determine if this were a feasible line of inquiry. The findings from this research are discussed later.

SUMMARY: DEFINING AGENTS, SCHEMA, AND FITNESS FUNCTIONS IN PSS

In this section we have argued the following in relation to three of the CAS elements of PSS.

The key agents in public service systems are organizing initiatives that may be as short-lived as a transaction between consumers and service providers or as long-lived as a government agency. Initiatives are formed at many levels. The lowest level is formed by individuals choosing to engage in organizing behavior, and higher levels are formed by organising interactions among initiatives. Individual participants are also agents within an organizational system, but only so far as they participate in initiatives. There are at least six key characteristics of an initiative that existing organizational theory would suggest are fundamental to defining its performance over time: organizational mode; value proposition; technology employed; size and age; resources; and knowledge level and generation processes.

The schema of an initiative includes the organizational mode and the range of values that may be considered for inclusion in a value proposition. Information from the environment and connections to other agents will be interpreted through the particular mode and value orientation of the initiative. Three alternative organizational modes were identified in organizational and institutional theory, while the possible range of values was left for further research (see later for findings from our case study).

With respect to the individual participants in initiatives, organizational mode is also part of their schema, but value is deemed to be unique to the individual and irrelevant to the modeling of a PSS. What is important from a modeling perspective is whether or not the individual perceives there to be value in participating (or not) in an initiative—not what the content of that value is. The range of actions that an individual can engage in was limited to three: joining an initiative; remaining in an initiative; and leaving an initiative.

Fitness functions of initiatives will be largely reliant on the value proposition that they choose. Although some CAS theorists suggest that defining fitness functions a priori is of little use, we pointed out that organizational and public administration theories suggest that values are highly relevant to performance and are therefore a key component of fitness functions. We also identified theoretical support for the likely convergence of value propositions for agents in a given organizational system around a limited set of possible combinations.

While we did not address the definition of connections among agents in a PSS, it must be acknowledged that a complication in the definition of the connections between agents is the confusion between the nature of the connection and the behavior that results from the connection. “Relationship” is the term often used to encompass both the connection and the resulting behavior, which is a definitional weakness. In our definition of initiatives we could be accused of this very weakness, as the organizing mode of an initiative is, in fact, a connection among individuals in an initiative, as well as a fundamental component of the resulting initiative’s schema.

However, this connection among individual agents is for the purpose of forming an initiative and we therefore propose that, once engaged in, these connections are more a characteristic of the resulting initiative rather than a characteristic relating to each constituent participator in the initiative. This still leaves us with other types of connections to be defined, which could include the exchange of information between initiatives, individuals leaving one initiative and joining another (similar to the exchange of genetic material), and instructions passed from national agencies to local agencies. We have not yet begun the analysis of the research data to establish the nature of connections among agents, other than the special connection that is organizational mode. Our research into this CAS element will focus initially on the exchange of “state” information among agents, for example information relating to the characteristics of initiatives and that relating to the state of the system.

Finally, system state is also left undefined in this article. Consistent with our views on the subjective and dynamic nature of agent value propositions, we will rely on the views of the participant agents in the system under study to inform the definition of what is important system state information. In addition, we must consider that there may be some particular set of system or environmental dimensions that is critical for describing the “initial state” of the system, which may or may not be the same set of dimensions that pertain to a description of the current state. Further interrogation of the data in our case study is required to propose a definition of PSS system state elements.

CASE STUDY

In order to explore further the relevance of CAS concepts to PSS, we undertook to apply the concepts developed above to a case study of the housing system in Dublin, Ireland. Broadly following the case study methodology of Eisenhardt (1989), the case study data collected to date has included interviews with over 60 participants in the system representing a cross-section of agent types over six years (1998-2003); review of documents and literature relating to the Irish housing system (Blackwell, 1988; Murphy, 1995; Clark & Dieleman, 1996; Drudy & Punch, 1999; Norris, 2003); and the collection of quantitative measures identified by the participants as having relevance to their decisionmaking processes for the last 30 years. The specific focus of the research has been on the decision-making processes of agents, applying the processual methodologies of Pettigrew (1997) and Poole et al. (2000). More detail on the case study approach we took may be found in Rhodes (2002).

AGENTS AND ORGANIZATIONAL MODES IN THE HOUSING SYSTEM IN DUBLIN, IRELAND

The types of initiative-agents that we identified in the housing system in Ireland may be grouped into five categories. The first category contains initiatives that produce housing; that is, developers, builders, local authorities, nonprofit housing organizations; and self-build initiatives. We call this category of initiative the primary production initiatives. Generally, these initiatives adopted a hierarchical organizational mode, although the last two types were often organized as networks. The second category, resource transfer initiatives, includes all sales, rentals, and allocations of land and/or homes between initiatives and individual agents or between two (or more) individual agents. Although the market mode is dominant in this category, a hierarchical relationship may also be found in social housing allocations.

Our third category, supplier initiatives, contains those initiatives that play a supporting role through the production of goods and services that are key elements of the production process for housing. These include architectural firms, banks and building societies, the Housing Finance Agency (finances local authority and nonprofit housing), building material supply firms, and so on. We noted that, while it is possible for primary production initiatives (PPIs) to produce these goods and services internally through the addition of particular individuals and/or necessary resources, there appears to be a process of institutionalization of supplier structure in the housing system. Over time this process leads to particular resources grouping together to form supplier initiatives (SIs), which then (generally) engage in market mode relationships with PPIs. Note that this type of interaction between SIs and PPIs is itself a type of initiative, which we are currently grouping together with the resource transfer initiatives (RTIs) described above. It may be necessary in the future to split the RTI category into two separate categories due to the different types of resource being transferred.

A fourth category of initiative is formed through the decision by individuals and/or initiatives to join together for the purpose of increasing their influence over the environment in which they operate. Labor unions, the Royal Institute of Architects in Ireland (RIAI), the “Homeless Initiative,” and the Construction Industry Federation (CIF) are examples of associative initiatives. These we consider special-purpose initiatives because they are not primarily engaged in division of labor to increase productivity; rather, their primary purpose is to increase the power that one group of individuals or initiatives has in relation to others in the system. While these appear to begin in the network mode, they often adopt the hierarchical mode over time. The Homeless Initiative in Dublin provided a fascinating case study of this process over the course of four years.

The final category of initiative encompasses centers established to influence or develop policy in universities, government task forces and departments, private thinktanks, and so on. We propose that the actual writing of the laws is external to the PSS, following the classical distinction in public administration between the law makers and those charged with carrying out those laws. However, the inclusion of policy-influencing initiatives in the scope of the PSS recognizes the close connection between policy formulation and implementation activities (Pressman & Wildavsky, 1973), as well as facilitating modeling of the workings of policy networks (Rhodes, 1997) within the system.

In the course of collecting information from our interviewees regarding the key elements of their strategic decision-making process(es), we identified an additional type of agent in the system: the information agent. While we are aware of organizational CAS models that incorporate this type of agent (Carley, 2002), we originally considered information to be a property of one of the other agents and/or of the connections between agents. However, it became clear over the course of analyzing the data collected to date that information agents had an independent existence both physically and in the perceptions of human agents, and that they played a key role in the housing system. Over the past 40 years, building regulations and planning regulations have been introduced, resisted, modified, ignored, used to further other agendas, and, ultimately, accepted as part of the system. The characteristics of these types of agents and the nature of the connections between these and other agents will need to be further explored.

AGENT SCHEMATA: MODES AND VALUE DIMENSIONS

In the discussion above we used organizational modes to describe the differences (and similarities) among the different agents. While there were some surprises in the mode(s) that particular types of agents might employ, there were no fundamentally different modes from those described in existing organizational theory; that is, market, hierarchy, and network modes. One of the interesting findings relating to modes was that the nature of ownership did not appear to be correlated with the mode adopted by a particular agent. Rather, the size of the initiative (in terms of the number of individuals participating) was more indicative of a particular mode, with the smallest tending toward market mode, network mode in medium-sized initiatives, and hierarchy in large initiatives.

In terms of the way that mode influences the decisions made within the initiative, preliminary research indicates that the value orientation may be related to the mode, in that different value “biases” are exhibited by initiatives depending on the mode of the initiative. More formal classification of the data and additional field research are required in order to flesh out our understanding of this relationship.

As noted in the previous section, values are likely to play a significant role in influencing the decisions made by individuals participating in initiatives. While our data is not fine-grained enough to distinguish between values espoused by an individual and those shared by individuals in a particular initiative, the interview protocol was designed to facilitate the identification of values affecting decisions relating to the initiative. Whether or not these are personal values of the interviewee or values that are held by some or all of the individuals in the initiative is not critical to the analysis. In order to identify the values underpinning agent decision making in the housing system in Dublin, we asked participants in initiatives to describe key strategic decisions4 along with the considerations that led them to choose one action over another. In addition, interviewees were asked to comment on the “environment,” choosing whatever dimensions they deemed relevant to their own organization. These questions elicited responses that we analyzed to determine if there were consistent dimensions of value across the various initiatives and/or if these were combined in any recognizable patterns. We found that there were four dimensions that regularly appeared in the responses, and the relative importance of these dimensions varied across initiative and over time. The four dimensions relate to economic, social, political, and innovative pressures on/in the system.

In addition a fifth dimension, which appeared relevant to privately owned companies only, was a “family” dimension, which often influenced major decisions having to do with the company’s strategic direction. The family dimension incorporates concerns about the level and/or distribution of income to owners within a family and/or the effect of the firm’s activities on the personal reputation(s) of individuals associated with the firm.

HOW THE VALUE DIMENSIONS AFFECT FITNESS FUNCTIONS

Having established the range of potential value dimensions, we then looked at how initiative-agents weight each of these value dimensions in their decision processes. Understanding the weights that agents assign to different value dimensions should facilitate the development of fitness functions for agents, which may or may not be similar across agents and have the potential for changing over time as agents shift their priorities. Based on the interviews of decision makers in 60 initiative-agents, we found the following.

The dimension that appeared most often was the economic dimension, which included concerns about house prices, supply prices (most often land), supply of housing, and so on. Owner-occupiers dominate the Irish housing market and the government has had a policy of encouraging home ownership since the founding of the state. With this in mind, it is not surprising that concerns about a mismatch between the supply and demand of housing assumes such importance in this system.

The next dimension that appeared in our interviews was the political dimension. In nearly all of the interviews with individuals in the primary production initiatives (PPIs), the importance of having good relations with local and/or national government representatives was stressed and examples were given about different “eras” in housing policy that had significantly different impacts on the production capabilities and decisions of the PPIs. Individuals in policy-formulation initiatives (PFIs) put the political dimension high on their list of decision factors and were generally polarized either for or against the current government’s housing policies. Interviewees from the other three types of initiatives had various views as to the relevance of the political dimension.

In several cases, interviewees discussed concerns about the effect of the housing system on the social fabric of the city. This was particularly germane to the local authority and nonprofit housing PPIs and several of the PFIs. Builders and developers also commented on this issue, often voicing criticism of local authorities for exacerbating the problem through ill-considered development and allocation policies. In addition, architects had strong opinions on this issue and saw addressing it as an important dimension of their role in the housing system. Most often the issue identified by participants related to concerns about increasing marginalization and segregation resulting from access to particular types of housing or particular locations. We refer to this as the social dimension of the value proposition.

Finally, in discussing key changes in the housing system over the past 30 years, interviewees were likely to mention significant innovations in the housing system. These included the introduction of timber-frame housing, prefabricated structural components, “lifetime” housing, EU legislation, and planning regulations. Some interviewees were very sensitive to innovation in the system, reacting quickly to either adopt or resist the change. Others were slow to recognize the existence of the potential for change and continued to make decisions as if the innovation did not exist.

These four dimensions—economic, political, social, and innovative— are the principal dimensions of the value proposition for the majority of initiatives that we studied. Individuals in the initiatives might consider the relative importance of these dimensions differently, but few had other dimensions that were significant contributors to decisions that were made. As noted, the family dimension applied only in privately owned PPI and SI initiatives.

Not only does the relative importance of these dimensions vary across different types of initiatives, but it varies across time for specific initiatives. Our interviewees recalled decisions in the past that were influenced by a dimension that was now considered quite unimportant, but at the time of the decision was seen as critical. Our CAS model must therefore include fitness functions that incorporate these dimensions with varying importance across initiatives, as well as a process for adapting the relative weightings over time.

SUMMARY OF CASE STUDY FINDINGS

Our case study data has provided further detail as to the characteristics of agents, schemata, and fitness functions that are consistent with the concepts we proposed in our theoretical melding of CAS, organizational, and public administration theory. The six types of agents, three organizational modes, and five value dimensions that were identified in the housing system in Dublin will form the basis for future CAS model specifications. Further exploration of the case data to flesh out the nature of connections among agents, system state variables, and the decision process of agents is required to realize the potential for modeling public service systems as CAS.

HOW PUBLIC ADMINISTRATION THEORY CAN GAIN FROM A CAS APPROACH

So far we have explored definitions for three of the five elements required for modeling PSS using a CAS theoretical framework, based on existing organizational and public administration theory and findings from our case study. We believe that definitions of the remaining two elements are within reach by continuing to interrogate the case study data and comparing existing literature to the proposed definitions and relationships that are identified. However, the question remains, “Why bother?” What will a CAS model of PSS enable us to explore? What value does a CAS approach bring to either practitioners or theoreticians in public administration?

We suggest that there are at least three areas in which a CAS approach will contribute to an improved understanding of public service systems. These are to facilitate the integration of existing strands of public administration theory into a coherent model of PSS; to provide theoretical support for the incorporation of adaptability measures into the evaluation of policy interventions; and to generate novel hypotheses through the application of existing CAS theory to PSS.

FACILITATE THE INTEGRATION OF EXISTING STRANDS OF PUBLIC ADMINISTRATION THEORY INTO A COHERENT MODEL OF PSS

We noted at the beginning of this article that public administration theory is currently in a state of flux, in which there are several competing perspectives on the nature of the phenomenon to be studied and the approach to be taken. New public management (NPM), network theory, and social capital all focus on different elements of public service systems and prescribe different interventions, not so much in conflict with one another, but more as if the adherents of the various approaches are choosing to ignore one another. A CAS framework, as we have defined it here, has the capacity to incorporate the core elements from each of these perspectives, thereby facilitating research that seeks to explore how NPM reforms affect social capital—or how network effects might offset or enhance the impact of NPM. In support of this claim, we offer the following redefinition of the central elements of each of the three strands of public administration theory in terms of the CAS framework proposed.

According to Hood (1991), NPM includes three central elements: privatization, decentralization, and managerialism. Theoretically, the argument for privatization of public services is based on public choice theory (Arrow, 1963) and institutional economics (Friedman, 1964; Williamson, 1975). In practice, privatization has involved either the transfer of ownership of public-sector organizations to private sector or the outsourcing of some functions of the public sector to the private sector (“contracting”). However, there is a growing body of literature that suggests that private firms do not always perform more efficiently/effectively than public or nonprofit organizations in the delivery of public services (Dunsire et al., 1994; Peters & Savoie, 1998; Cohen, 2001; Kay, 2002). Under the CAS framework proposed, privatization of activities that were previously performed by government agencies may be modeled as either changes in the ownership of a particular initiative5 or changes in its organizational mode. The effect of this change on the performance of the target initiative and/or system as a whole could then be studied through computer simulations based on the CAS framework.

Decentralization under NPM is the disaggregation of decisionmaking authority and the empowering of local government and/or communities to create and deliver services (Osborne & Gaebler, 1992). The argument for the shift of authority is found in classic organizational theory (Chandler, 1962) and is also consistent with Pressman and Wildavsky’s (1973) analysis of the disconnect between central government policy intention and local implementation. In essence, the argument is that organizations at a lower level6 are better able to perceive and adapt to changes in the environment than are organizations at higher levels, which must operate with more aggregated information. Under CAS, decentralization could be modeled as a change in the connection between initiatives (and possibly the creation of new initiatives), along with an increased ability in the lower-level initiatives to respond to the environment faster. The impact of this increase in agent adaptability on the performance of a given agent, and/or on the system as a whole, could then be analyzed via CAS-PSS computer simulations.

Managerialism in NPM covers a range of management practices initially proposed by Simon (1947) and developed over time by business management writers and consultants. Under NPM these practices were repackaged to focus on goal setting, performance measurement, and the application of best practice from the private sector in areas such as human resource management, operations, financial management, and strategy. Public-sector managers are encouraged to engage in a systematic approach to setting realizable goals, establishing good management practices in support of those goals, and then measuring performance and making adjustments as required. Public administration literature would suggest that the first and third steps continue to pose significant challenges for both practitioners and theorists (Behn, 1995; Boston, 2000). In the CAS model proposed, public service goals would appear as the fitness functions for initiatives, as well as being candidates for measures of the system state. Furthermore, public service goals are likely to be defined in terms of the value proposition for initiatives. In the housing system in Dublin, we found that the value proposition for the range of initiatives we studied included measures of economic, social, political, and innovative impact—likely starting points for the selection of appropriate public service system goals.

While NPM has targeted the performance of individual initiativeagents in order to improve the overall performance of PSS, public management network theory (Kickert et al., 1997; Rhodes, 1997; Agranoff & McGuire, 2001) has focused on the interactions among agents and the identification of “meta-agents,” referred to variously as policy networks, public service networks, and/or inter-organizational networks. Kickert et al. (1997) group the various perspectives on public management networks into three “ideal types.” The first is the instrumental perspective that considers network management as an exercise in “creating conditions under which goal-oriented processes can take place” (de Bruijn & ten Heuvelhof, 1997), with government agencies in particular responsible for creating these conditions. This perspective is similar to the NPM managerialist approach in its emphasis on the setting of objectives and then steering the system toward their accomplishment. Performance evaluations at various “levels” of a public service network are a feature of this view of the role of public administration (Provan & Milward, 2001). In the CAS framework we described, there are two ways of modeling the establishment of network-level goals and intervening to create conditions that facilitate the accomplishment of these goals. The first is to change the information exchange between a government-“owned” initiative and other initiatives to communicate these goals, and the second is to change the “rules of the game” that a set of initiatives incorporates into its schema. In both cases, the change in system state dimensions as a result of these changes is the PSS element of interest.

In contrast to the instrumental perspective, the institutional and interactive perspectives in public management network theory incorporate neither central actors nor goal-attainment assessments. The institutional view is that the network consists of the interactions among participants, which create norms and organizational frameworks that, in turn, affect the behavior of the participants. The processes and structures of institutionalisation are of greatest interest here. Rhodes (1997) describes the phenomenon of “governing without government,” which incorporates the concept of public service provision through autonomous, self-organizing networks of public, private, and nonprofit organizations that interact to establish and achieve shared—or even conflicting—objectives. This perspective is quite consistent with a CAS approach to modeling PSS, with the norms and frameworks referred to in the institutional network view being emergent properties of the system and affecting the behavior of agents through modifications to schemata and/or fitness functions.

Finally, the interactive perspective focuses on the processes of interaction between participants, in particular those processes relating to collective decision making, implementation, and coevolution toward or away from a “common purpose” (Kickert et al., 1997). The management focus for practitioners applying this perspective is on how to minimize the effect of conflicts of interest that block collective performance. The CAS elements most relevant for exploring this perspective are the content and number of connections between initiatives, the schema that agents apply to interpret the meaning of these connections, and the impact that changes in these elements have on the adaptability of agents and their fitness functions.

For social capital theorists, the CAS framework offers a different way of capturing the quantity and quality of one type of trust relationship among individuals and initiatives through research aimed at establishing the occurrence of the network mode of organizing. In addition, the concept of shared norms of behavior among individuals and/or initiatives could be incorporated into a CAS framework through the modeling of schemata and the convergence (or not) of schemata across agents over time. Fukuyama (1995) makes the case for social capital having an impact on the economic performance of societies—the dynamics of which could be explored through the use of a CAS model.

PROVIDE THEORETICAL SUPPORT FOR THE INCORPORATION OF ADAPTABILITY MEASURES INTO THE EVALUATION OF POLICY INTERVENTIONS

In both organizational literature and CAS theory, the ability of agents to adapt to changing environmental circumstances has been shown to be crucial to the performance of the specific agent and of a system involving many agents. Furthermore, an underlying assumption of the decentralization strand of NPM is that the adaptability of lower-level public service organizations is at the heart of overall systems improvement. However, adaptability is rarely included in any proposed measures or dynamics of agent or system performance in public administration literature.

Our proposed CAS framework would facilitate the incorporation of adaptability measures and processes into proposed evaluations of policy interventions. For example, attention to the effect of policies of decentralization and contracting between agents on the individual agents’ ability to adapt may shed some light on why these reforms were either not fully implemented (Larbi, 1998; Hood, 2000) or resulted in unanticipated, negative outcomes (Mackintosh, 1998). In addition, measures of adaptability may be a productive avenue in the search for appropriate performance measures for both public service organizations and public service systems.

ESTABLISHING A FIRM BASIS FOR TESTING THE APPLICABILITY OF EXISTING CAS THEORY TO PSS

While the above might prove to be of assistance in clarifying existing concepts and resolving some outstanding issues in public administration theory, the real promise of a CAS approach to public service systems is in the potential for establishing the existence of relationships among elements of public service systems that we have not yet understood. There are at least four relationships that are well defined in CAS theory that may have significant ramifications for public administration theory as it relates to public service systems.

The first of these is the tendency toward “order” in complex adaptive systems. Order is defined here as a stable pattern of relationships among elements of the system. Stuart Kauffman (1993, 1995) demonstrates how, in the physical world, order can and does emerge from the interactions of independent elements and autocatalytic processes. CAS theorists often refer to this as “self-organizing” behavior. Kauffman calls this “order for free” and argues for the existence of a physical law relating to these types of systems, namely that they will evolve toward a more “ordered” state. In a similar vein, Herbert Simon, addressing the American Philosophical Society in 1962, observed that “complexity frequently takes the form of hierarchy,”7 and that “hierarchic systems will evolve far more quickly than nonhierarchic systems of comparable size” (Simon in Midgley, 2003: 387). If public service systems conform to the behavior rules proposed by Kauffman and Simon, then PSS researchers should look for the ways in which public service systems are creating “order,” what form/mode this order takes, and how this affects the overall performance of the system.

Secondly, it is well established that complex systems often do not reach a fixed-point or cyclical equilibrium, but may revolve around “strange attractors,” or state spaces that constrain the behavior of the system (Bar-Yam, 1997). Much of the current public administration theory and practice is still based on neoclassical economics in which the existence of equilibrium states is assumed and policy tools are developed accordingly. A CAS approach to public service systems would focus on the identification of strange attractors and their implications for public policy formulation.

A third relationship demonstrated by Arthur (1989) and Prigogine (1997) is the dependency of complex systems on initial conditions, particularly when the agents in the system are subject to positive feedback loops. The specification of feedback in economic and urban systems has long been a feature of models supporting public policy formation. However, these models tended to incorporate assumptions of stable equilibrium points and were designed to predict future system states based on static relationships among agents. CAS theory suggests that future predictions based on a set of static or even dynamic relationships among agents is not sufficient. Some understanding of the “history” of the system is required in order to specify possible future states.

Note, however, that few complexity theorists suggest that complex systems lend themselves to the sorts of predictions that economic and/or systems dynamics models produce. Among the causes of this lack of predictability are nonlinear relationships among agents, sensitivity to initial conditions, the fluctuations that result from systems in far-from-equilibrium states, and the inability to know for certain which “choice” a complex system may make at a critical point (bifurcation point) in its evolution. Stacey (2001) contends that a further difficulty in predicting outcomes in human complex systems is the generation of novelty that can arise from the interaction between individuals. There is clearly some difficulty with causality in complex adaptive systems that should caution against pinning any serious expectations of finding simple levers to improve our influence over public services systems via a CAS framework, however much we might be able to improve our understanding of how the present state of a given system was achieved.

Finally, a key characteristic of complex adaptive systems is that complex behavior can arise from the actions of and interactions among agents following relatively simple rules (Holland, 1995, 1998). This characteristic offers the possibility that the complex behavior of PSS may be reducible to a manageable set of elements and behavior rules without losing the ability to represent the complexity inherent in the system as a whole. If it is the case that observable patterns of system behavior could be modeled using relatively simple rules and agent demographics, then exciting opportunities open up for exploring the behavioral possibilities of PSS under different scenarios. Through CAS models and empirical verification, the black box of public service systems may become more transparent, thereby progressing our understanding of the link between agent decisions, agent interactions, and system outcomes.

SUMMARY: HOW PUBLIC ADMINISTRATION AND PSS MANAGEMENT CAN BENEFIT FROM A CAS APPROACH

Boston (2000) states that research into public management at the “macro or system-wide level” is crucial to advance public administration theory, and that researchers must address the effect of “interdependencies of various kinds, interorganizational dynamics, coordination issues, etc.” on the performance of the public management system as a whole. The CAS approach proposed herein provides an integrated framework for engaging in this kind of research, allowing the researcher to identify various levels of agent behavior (including a system-wide level) and to address interdependencies among the various levels. Applying CAS theory, simple rules, and a limited range of different agents may, in fact, be capable of generating aspects of the observable and currently inexplicable behavior of existing public service systems.

While there have been some efforts at linking agent behavior in a PSS to system-wide outcomes (Provan & Milward, 2001), in general public administration theory and practice deal with the system either as a simple aggregate of individual organizational outcomes, or as a black box, into which go resources and out of which flow desired (or unanticipated) outcomes. A more explicit connection between system outcomes and agent behavior is likely to be one result of a CAS approach. In the best case, this approach would lead to an improved understanding of the types of levers that might be applied at agent level to move the system toward desirable outcomes. More modestly, the effort would provide a framework for integrating the various perspectives of current public administration theory and allow for competing hypotheses regarding policy instruments, agent behavior, and system outcomes to be tested through CAS simulations. With this potential in mind, careful definition of the five CAS elements of a public service system based on relevant theory and empirical investigation, along with the elaboration of CAS models of the behavior of these systems, appears to be well worth the effort.

NOTES

  1. Holland also refers to these as “the rules of the game.”

  2. Although they split the network alternative into “fiefs” and “clans.”

  3. For example, individuals may have preferences relating to one or more of the initiative characteristics defined in this article.

  4. We defined a key strategic decision as one that has significant implications for the structure, direction, or purpose of the initiative.

  5. Note that ownership is considered to be neither a key characteristic of initiatives nor a fundamental driver of initiative performance in our elaboration of a CAS-PSS model. This suggests that changes in ownership are unlikely to have much effect on the performance of the system as a whole—a hypothesis that is consistent with the lack of impact that NPM privatization reforms have had on either cost or service quality to date.

  6. With a narrower span of control in organizational management terms. 7 In the sense that complex systems are frequently made up of many subsystems.