Intellectual Appeal

The intellectual appeal of complexity science derives from its extension of the boundaries of science to include phenomena that previously were unknown or are resistant to scientific explanation. In the novel The Unbearable Lightness of Being, the character Sabina describes her paintings in this way: “On the surface, an intelligible lie; underneath, the unintelligible truth” (Kundera, 1999, p. 63). The concepts and tools of complexity science allow researchers to explore the “unintelligible truth”—the messy, murky, intimidating realities of organizational life (Begun, 1994). Complexity is to conventional scientific method as relativity and quantum physics are to Newtonian mechanics. Just as Newtonian mechanics can be considered a special case of relativity theory, linear models can be viewed as a special case of nonlinear models. Near equilibrium points, nonlinear systems can be effectively modeled with linear models. However, science needs the complex, nonlinear models and tools to study systems operating far from equilibrium.

The intellectual appeal of complexity science to many scientists and practitioners is enhanced by its roots and continuing development in the natural and physical sciences. Complexity science grew from physics and mathematics and spread to biology and the social sciences. Many practitioners and academics in the health sector have backgrounds in the natural and physical sciences, which makes receptivity to complexity science more likely. As a result, communities of learning in complexity science have emerged in many clinical health professions, particularly nursing (Lindberg, Nash, and Lindberg, 2008; Plexus Institute, 2011; Suchman, Sluyter, and Williamson, 2011).

Similarly, complexity science appeals cognitively to many individuals who embrace systems theory and systems thinking. While there are clear differences between traditional systems thinking and complexity science (Phelan, 2001), complexity science also can be viewed as part of the continuum of systems thinking approaches (Jackson, 2003), differentiated from but similar to hard systems thinking, systems dynamics (following Senge, 1990), and organizational cybernetics. This tight relationship of complexity science to systems thinking particularly creates affinity between complexity science and the field of public health, where systems thinking fuels the search for prevention through identification and alteration of root causes and is embedded in the educational competencies of the master's in public health degree (Association of Schools of Public Health, 2010). Systems thinking has direct application to intransigent public health problems like tobacco control and obesity (Best et al., 2007). As a macrolevel theory, it surpasses the limits of national boundaries (see De Savigny and Adam, 2009). The high level of abstraction ensures that the theory becomes less embedded in the context of specific national scientific cultures, and systems thinking provides a means for scientists and practitioners to communicate across cultural boundaries.

Practical Appeal

At a practical level, complexity science helps researchers and managers investigate and cope with challenges that are not amenable to traditional approaches. Many would argue that the most interesting and important issues for science and management to tackle in health care are the most intractable and complex issues—for example, lowering the cost and raising the quality of patient care, building cultures of safety and quality, improving the dissemination of clinical guidelines and evidence-based medicine and management, integrating services at the organizational level, stimulating disruptive innovation, and improving clinical care for patients with multiple chronic illnesses. These issues have been referred to in the literature as “wicked problems”: they defy complete definition, resist all the usual attempts to resolve them, have no final solution (since any resolution generates further issues), and have solutions that are not either true or false or either good or bad, but are the best that can be done at the time (Brown, Harris, and Russell, 2010).

Scholars have made a useful distinction among systems characterized as simple, complicated, or complex. Often referred to as the Stacey matrix or the certainty matrix, one way of categorizing issues, systems, or contexts is by the two dimensions of degree of certainty about cause-and-effect linkages and level of agreement within the system (e.g., the team or organization) about the issue (Stacey, 1996; see also the Cynefin model: Snowden and Boone, 2007). Simple issues are high on both dimensions and can be managed by technically rational decision making. Simple contexts invite “straightforward management and monitoring” (Snowden and Boone, 2007, p. 70), using assessment of fact and response based on established practice. Complicated issues have either low consensus or low certainty about cause-and-effect linkages, such that compromise, negotiation, ideological control, or logical incrementalism come into play to resolve complicated issues. Complicated contexts call for analysis and choice among several alternatives, some of which may be equally “good.”

Complex issues are low on both dimensions, with low certainty about cause-and-effect and low consensus. Complexities require that managers use dialectical inquiry, intuition, muddling through, agenda building, and other messier decision-making means. Complex contexts are characterized by many competing ideas, no clear right answer, and the need to learn as patterns emerge (Snowden and Boone, 2007).

Not surprisingly, a danger in management is forcing complicated and complex processes into simpler domains—looking for “intelligible lies” instead of “unintelligible truths”—with the misguided hope of producing decisions and control. Among others, Hall and Johnson (2009) argue that many complex processes in organizations are being overstandardized, to the detriment of both customer service and organizational performance.

Although it can be argued whether a particular system is simple, complicated, or complex or whether it can be validly treated as such for management or research purposes, there is little doubt that researchers and managers need tools to differentiate, manage, and study all three types of systems.

Reasons for Modest Use of Complexity Science in Health Care

We just noted a relatively modest investment in complexity science in health organization practice and research over the past two decades. Why is this the case?

System Dynamics, Simulation, and Modeling

System dynamics (SD) is a modeling approach that allows organizational leaders to develop causal flow loops and computer models of their systems to evaluate challenging and complex problems. It was developed in the 1960s and 1970s by Jay Forrester and his operations research collaborators at MIT (Forrester, 1961). System dynamics belongs at the operations research end of the systems thinking spectrum, as opposed to the softer and postmodern branches of systems thinking (Jackson, 2003). That said, the method is a useful tool for exploring the structure and behavior of complex systems, especially because it emphasizes the importance and impact of feedback loops, a leading source of nonlinear behavior in complex systems.

System dynamics models are designed at the system level, not the level of the individual agents in the system. The models are constructed using one of several user-friendly software programs, including a freeware version, Vensim PLE. The modeling involves collecting known information about the components and behavioral dynamics of the system to be modeled. This process is often done as a group learning exercise so as to integrate the input and diverse perspectives of the work group into the model. The model itself is constructed from a set of “stocks,” “flows,” variables, and delays that describe the structure of the system. Stocks are countable entities like employees, cash, and pieces of equipment. Flows are the rates at which stocks change. Variables are used to represent other components of the model—such as reputation, expectations, and morale—that influence the magnitude of stocks and flows but are not the primary phenomena of interest. Delays allow for integration of phenomena with different time scales into the model.

Once the model has been constructed, it should be subjected to a variety of validation tests, such as confirming that it behaves as expected with extreme parameter values and that it has face validity with decision makers and other important stakeholders. After validation, the model can be used for explanation, prediction, and decision making, always with the understanding that it is a model, not a perfect replica of reality.

SD models have been used to model and address health-related problems at a variety of levels, including policy issues, organizational performance issues, and physiology. Brailsford (2007) has identified these as level III (system level), level II (organization level), and level I (disease) models. While SD has been used mostly in developing and evaluating health policies and public health initiatives (Jones et al., 2006; Levy, Bauer, and Lee, 2006; Sterman, 2006), it has also been used to explore health care organizational performance (Elf et al., 2007; Hovmand and Gillespie, 2010; Miller et al., 2011; Samuel et al., 2010).

In addition to SD, discrete event simulation (DES) and agent-based modeling (ABM) are two other simulation and modeling approaches used in health care management and research. Of the two, DES appears to be the more commonly used method for operational improvement. It studies a system's performance by building a model that represents the system as a set of entities, activities, and queues. Entities (such as people and resources) wait in queues until an activity (such as an operation) occurs. A DES model can be used to evaluate how changing the characteristics of the entities and activities in the system will alter its performance. Feedback loops play a less important role in DES models, while queuing theory plays a much more important role. Multiple articles have been published documenting the use of DES to address issues like improving throughput and efficiency (Hamrock et al., 2013; Parks et al., 2011; Reynolds et al., 2011; Rohleder et al., 2011) and cost efficiency (Getsios et al., 2010; Kobelt, Lindgren, and Geborek, 2009).

ABM is another approach to simulating systems based on the behavior of a large number of “agents” specified by the modeler. ABM is particularly well suited to studying the emergence of system-level behaviors arising from the simple rules controlling the behavior of the individual agents in the model. Evidence about the utility of ABM for health care management decision making is limited, however. It has been used to study the spread of methicillin-resistant Staphylococcus aureus infections (Lee et al., 2011), the effect of income inequalities and spatial segregation on diet quality (Auchincloss et al., 2011), and policy development for refugee communities (Anderson, Chaturvedi, and Cibulskis, 2007).

Simulation and modeling can add value for leaders and managers in several ways. Fully developed formal models are useful for exploring solutions to complex problems and making better decisions. These models may take months to develop, generally involve extensive input from multiple stakeholders in the organization, and require careful validation testing. No model can predict the future, but a well-developed simulation model improves decision making by increasing our ability to analyze and understand the complex behavior of health care systems and anticipate the outcomes of different courses of action.

Even less formal exercises, like developing a causal loop diagram rather than a complete SD simulation, can help clarify thinking, identify potential unintended consequences, and rule out options that might otherwise have received serious consideration. Causal loop diagrams map out the components, chains of possible causation or implication, and feedback loops in a complex system. They can be drawn by hand on the back of an envelope or developed using software tools such as SD programs or mind-mapping software. For clarity when constructing a causal loop diagram, it is best to use neutral or positive nouns to describe the components of the system. The links between the components are conventionally labeled with an “s” (or “+”) or “o” (or “-”), depending on whether a change in the level, size, or amount of the antecedent component leads to a change in the level, size, or amount of the subsequent component in the same or opposite direction, respectively.

Developing a simulation or model in a group can be a useful team learning exercise. Getting a group together to construct a simulation model or causal loop diagram taps into the wisdom of the crowd, mitigates the potential impact of individual blind spots, and provides an opportunity for all participants to develop a broader and more coherent understanding of the phenomenon in question. If the group is diverse and the group dynamics are constructive, the resulting model is likely to be superior to one developed by an individual.

In general, adoption of simulation and modeling in health care lags behind other industries (Brailsford, 2007; Young et al., 2009). Even when a model has been developed, leaders and decision makers might not adopt and implement the findings. Factors contributing to low utilization of simulation and modeling methods in health care organizations appear to include unfamiliarity of the methods, concerns about accuracy, unwillingness to “delegate decision making to a computer program,” and mismatch between the time required to build a rigorous model and the short time frames under which many operational leaders work. Attempts to address the latter problem include building “good enough” models or using model templates that can be quickly adapted for the situation at hand. Another challenge is that models developed in one setting may not be generalized to other organizations.

Fuzzy Set Qualitative Comparative Analysis

Fuzzy set qualitative comparative analysis (fsQCA) was developed in the social sciences to evaluate small sample data sets and semiquantitative or qualitative data and concepts. It is intended to fill an analytical need between the single case study and large sample statistical analyses (Ragin, 1987 1999, 2008). It complements more well-known analytical methods and is a useful tool for understanding and making decisions about complex organizational systems. fsQCA is based on fuzzy set theory and Boolean algebra. A comprehensive text on the method is now available in English (Schneider and Wagemann, 2012).

Conventional statistical analysis is based on the concepts of bivariate correlation and linear additive models. Multivariate regression analysis, one of the standard epistemological tools in health services and management research, is designed to develop a single causal model and partition the causality among the variables in the model, based on correlations, with the presumption of a necessary and sufficient relationship between the causal and outcome variables. fsQCA, in contrast, makes no assumption that there is only one causal configuration for an outcome, and it allows the possibility of equifinality, the condition where different configurations of conditions may lead to the same outcome. Moreover, because it is based on set theory, fsQCA explicitly recognizes necessary but not sufficient, sufficient but not necessary, and neither necessary nor sufficient associations between conditions and outcomes. In contrast, more traditional methods of data analysis focus on identifying correlations, which are by definition symmetric “both necessary and sufficient” relationships between variables. Necessary but not sufficient and sufficient but not necessary relationships often manifest themselves as heteroskedasticity in the data, and consequently they may more readily be overlooked in conventional variable-based statistical analysis (Ragin, 1987). fsQCA is well designed to explore situations like organizational performance where the number of cases, such as nursing units, is small, the potential for experiment is limited, and uncontrollable contextual factors may determine the effectiveness of interventions.

fsQCA is a case-oriented analytical method, not a variable-oriented method. Focused on cases, not abstracted variables, it is intrinsically more holistic and less reductionist. Consequently, it is well suited to evaluating complex systems in which the emergent behavior is not amenable to reductionist analytical approaches. The qualitative comparative method in general is based on the assumption that causality is dependent on context (the surrounding environment in which any complex system is embedded) and that organizational performance is a consequence of the interaction of multiple causal factors, which may themselves be interactions rather than discrete characteristics or interventions (Byrne, 2011). Health care services and organizational management research is strongly influenced by the methodology of the randomized controlled trial, the paradigmatic analytical method in biomedical research. This and related analytical methods isolate the independent and dependent variables of interest and control for variation in all other contextual factors. This method works relatively well for simple interventions like a new drug treatment regimen; however, it is not at all well suited to more complex interventions in social systems. Here the outcome of an intervention is critically dependent on context, and achieving success often requires tailoring to address contextual factors. Evaluation in such a setting requires that contextual factors be included as components in the analysis, not handled like control variables (Pawson and Tilley, 1997). Conventional meta-analytical techniques also encourage the attempt to control for contextual factors and thereby achieve a summary assessment of whether an intervention is effective across a broad range of social settings. Meta-analysis and randomized controlled trials are designed to answer questions like, “Is there an effect?” In the case of social and organizational interventions, the more valuable question is often in what settings and under what circumstances there is an effect. fsQCA prompts us to frame both research and management questions in this way and provide a method that supports the subsequent analysis.

Conventional multivariate regression looks for relationships among the variables and treats all cases the same, without distinction. It neither requires nor encourages deep knowledge of individual cases. In contrast, fsQCA specifically calls for investigators to do in-depth analyses of any situation where the relationship between a given configuration and the outcome is poorly understood. Such exploration can yield further insights that allow investigators to enrich the explanatory model. Thus, qualitative comparative analysis studies are often iterative and exploratory in nature. They may appear to grow or evolve organically, a process that is anathema in conventional analysis and evaluation because of the potential for the introduction of investigator bias. Investigators using the fsQCA method must guard against this risk by being transparent about the process and the evolution of their thinking, using good technique, and treating the causal recipes identified by fsQCA as falsifiable hypotheses to be tested with additional data sets and alternative analytical methods.

fsQCA is gaining traction in the health care services and management literature. Thygeson and colleagues (2012) used it to explore the relationship between quality outcomes and the degree to which primary care clinics have characteristics of the medical home. Of note, contextual factors like the sociodemographic characteristics of the patient population were found to have substantial impact on the quality outcomes. A review of the method and its application to implementation of the patient-centered medical home is available online (Thygeson, Peikes, and Zutshi, 2013). Another recent paper demonstrated the use of qualitative comparative analysis (QCA), the “crisp set” version of fsQCA, for studying complex organizational relationships, outcomes, and causal relationships in small data sets (Rizova, 2011). In crisp set QCA, units are classified by membership either in or out of a given set, while in fuzzy set QCA, units can be assigned to partial membership in the set. For example, measuring physician affiliation to a hospital, physicians could be classified as in or out of the set of affiliated physicians (crisp set analysis) or affiliated in some proportion from 0 to 1 (fuzzy set analysis). QCA or fsQCA has also been used to evaluate weight loss program effectiveness (Kahwati et al., 2011), the use of critical pathways and guidelines (Dy et al., 2005), the impact of organizational change on sickness absence (Baltzer et al., 2011), patient and customer segmentation (Woodside and Zhang, 2012), performance of primary care trusts in the National Health Service (Byrne, 2011), and reasons for differential progress on addressing health care disparities (Blackman and Dunstan, 2010). QCA methods can also be combined with statistical methods; the two are not mutually exclusive (Dixon-Woods et al., 2005).

Research by Fiss (2011) using configurational theory and fsQCA methods illustrates the concordance of fsQCA with complexity theory. A configurational approach views organizations as holistic entities comprising interconnected and interacting elements (Bedford and Malmi, 2010; Short, Payne, and Ketchen, 2008). This approach “allows researchers to express complicated and interrelated relationships among many variables without resorting to artificial oversimplication of the phenomenon of interest” (Dess, Newport, and Rasheed, 1993, p. 776). Fiss (2011) uses fsQCA methods to relate configurations of organizational structure, strategy, and environment to organizational performance. A similar approach is promising for studies of integrated health networks and systems, medical homes, accountable care organizations, and other complex arrangements of health care delivery organizations, where elements of environment, structure, strategy, and culture are interconnected and difficult to separate.

Social Network Analysis

Social network analysis (SNA) is a formal method for analyzing relational data (see chapter 10 for a comprehensive presentation of SNA). It has been a standard method of analysis in sociology and other social sciences since the mid–twentieth century (Wasserman and Faust, 1994; Scott, 2000), but has only recently begun to be used in health care management and research (Anderson and Talsma, 2011; Blanchet and James, 2012; Cunningham et al., 2012; Hossain and Kit Guan, 2012; O'Malley and Marsden, 2008; Willis et al., 2012). Given that health care organizational performance issues like patient safety, quality of care, adoption of new practices, and innovation are clearly related to the nature of the relationships and the quality of communication between individual health care workers and teams (Colon-Emeric et al., 2006; Gittell et al., 2000; Gurtner et al., 2007; Piven et al., 2006), having a standardized method for measuring and analyzing these phenomena should be highly useful in health care services research and management.

Most data collected in health care management and research are attribute data. Age, gender, disease category, and preferences, for example, are attributes that are assigned at the level of the individual agent or subject. Attribute data at the unit or organizational level (e.g., culture, climate, employee engagement, and quality measures) are also collected.

In contrast, relational data apply to relationships between agents. These data are intrinsically dyadic, and the relationship is the object of interest. Social network analysis is a method for collecting and analyzing relational data. In addition to existence (i.e., a relationship exists), relations can have magnitude, such as the frequency of contact, and direction, for example, A seeks advice from B. Relational data are often multifold. That is, agents (egos) A and B may have multiple different relations, such as “A and B are friends,” “A and B work together,” and “B seeks advice from A.”

SNA allows documentation and evaluation of the social structure of an organization. The relevance of this to complexity science is immediately evident in that all complex systems can be thought of as having an underlying social structure that determines the processes and emergent behaviors of the system. SNA is thus a method well designed to support one of the key tools of leadership in a complex world: studying and manipulating the social structures in an organization (see the discussion of adaptive leadership that follows).

Social networks are ways of describing relationships. They can be represented by sociograms—graphs with nodes representing individual “egos” or “agents” (people, teams, or organizations, depending on the organizational level being modeled). The lines that connect the nodes represent the relationships that exist. Attributes of the individual egos can be represented by the color and shape of the nodes. Relationship information can be encoded in the directionality and width or magnitude of the links among nodes. Social maps constructed in this way are typically nonlinear structures with multiple cross-links, branching paths, and fractal characteristics. Again, the connection with complexity science is evident.

Social network analysis has a set of metrics specific to the approach. These metrics are of two types: metrics at the level of the individual node or ego and metrics at the network level. For instance, “degree” is a node-level measure. A node's degree is the number of links connecting it to other nodes. “Density” is a network-level measure. A network's density is the number of existing internodal links divided by the maximum possible number of links. Low-density, sparse networks have relatively few links between the nodes in the population of interest; in high-density networks, most of the nodes are connected with most of the other nodes. Other commonly used SNA measures include centrality (node-level measure of how central a node is in the network), centralization (network-level measure of how centralized the network is), reciprocity (network-level measure of how reciprocal directional relations are), path length (network-level measure of distance between nodes in the network), and clustering coefficient (network-level measure of the degree to which the network is aggregated into clusters). Many of these concepts map directly to organizational phenomena. For instance, one would expect information to flow more rapidly through an organization with a shorter average path length. Hierarchical, bureaucratic organizations would be expected to be highly clustered.

Social network analysis has been useful in management in a number of ways. It can be used to explore performance and cultural differences between different health care units or organizations. For instance, Effken and colleagues (2011) used SNA to study the relationship between social network structures and quality and patient safety metrics on seven nursing units. They found associations between social network metrics and quality measures. However, different quality metrics were associated with different social structures, such that interventions to change the social structure of the nursing units might improve some quality metrics but cause others to deteriorate (Effken et al., 2011). Scott and colleagues (2005) used SNA to identify significantly different communication and interaction patterns related to decision making in two primary care practices—one much more hierarchical and centralized, with fewer collaborative groups, than the other.

SNA has also been used to explore influence and diffusion of innovation in organizations. Keating and colleagues (2007) used SNA to identify factors associated with being viewed as an influential colleague with respect to women's health issues in an academic practice. Being viewed as an expert on women's health, having a lot of women patients, and practicing in the same clinic were all associated with being influential. Lurie, Fogg, and Dozier (2009) used SNA to study social structures in three settings in an academic organization: team interactions in the intensive care unit, the composition of advisory committees for career development awardees, and relationships between directors of a clinical translational sciences institute. The interaction patterns in the two ICU teams were different, possibly related to the different patients each team cared for. SNA revealed important differences in the interdisciplinary nature of the various advisory committees and highlighted potential relationship challenges between departments engaged in the clinical translational sciences institute. SNA provided a “useful and standardized set of tools for measuring important aspects of team function, interdisciplinarity, and organizational culture that may otherwise be difficult to measure in an objective way” (Lurie et al., 2009, p. 1029).

SNA also is useful for designing organizational change or performance improvement interventions (Sales, Estabrooks, and Valente, 2010). Quantitative measurement of an organization's social structure provides useful information that can be used to select influential members of the network to participate in change initiatives or facilitate transmission of new information or practices. Social network analysis and mapping can also be used less formally to provide feedback to teams engaged in quality improvement work. Visualization of the social network of agents engaged in a quality improvement project may provide real-time insight to the team about gaps in participation—or it may unexpectedly identify influential or knowledgeable individuals whose role should be enhanced (Buscell, 2008; Lindberg and Clancy, 2010).

Positive Deviance

Positive deviance (PD) is a social and organizational change method that has been used in a variety of settings to facilitate lasting, sustainable improvement in pervasive, persistent problems without requiring additional resources. It was first used to address childhood undernutrition in Vietnam, with remarkable success. Positive deviance has since been applied with similar results in a variety of settings around the world, including in health care (Pascale et al., 2010).

Positive deviance enacts fundamental principles from complexity science (Lindberg and Clancy, 2010). The process is a framework that engenders the emergence of group learning and adaptation that can result in radical improvements in system performance. The process is consistent with the conceptualization of complex adaptive systems as aggregates of multiple, heterogeneous agents that are dynamic, massively entangled, emergent, and robust. Both PD and complexity science emphasize the importance of promoting a multiplicity of relationships between a variety of agents with diverse points of view in a constructive process of information sharing and collaborative exploration of solutions. Also consistent with complexity science is the role of leadership in the PD change process; leadership occurs through facilitation of self-organization and self-discovery by agents in the system (Lindberg and Schneider, 2012). The alignment between PD and complexity science theory may well explain why PD has been so successful at achieving sustainable change in complex social systems.

Positive deviance works by identifying sustainable solutions, “hidden in plain sight” and spreading them throughout a community using minimal additional resources. It complements or surpasses Lean and Six Sigma for certain kinds of problems. Although there is some overlap, Lean and Six Sigma are process improvement approaches primarily focused on eliminating waste and improving the reliability of existing processes. Positive deviance is designed to identify and spread processes—previously existing or new—that deliver qualitatively better solutions for a persistent, pervasive problem.

Positive deviance is a change management technique that has promise to close the knowing-doing gap that arises when we know what to do but not how to do it. For instance, it is one thing to know that hand-washing is important for preventing hospital-acquired infections. It is something else to know how to get everyone to do it.

Hospital-acquired methicillin-resistant Staphylococcus aureus (MRSA) infection is one of the first health care delivery system problems that was addressed with the PD method. The impressive results of are described in a number of publications (Buscell, 2008; Lindberg and Clancy, 2010; Singhal, Buscell, and Lindberg, 2010). Positive deviance has also been used to improve hand-washing in transitional care units (Marra et al., 2011), treatment of iron deficiency anemia in rural Africa (Ndiaye et al., 2009), and the care of myocardial infarction (Bradley et al., 2009). Positive deviance pilots are underway to address inpatient pain management (at Allina Hospitals and Clinics in Minnesota) and increasing understanding and use of palliative care (at the Billings Clinic in Montana). Potential other areas of application are such persistent problems as avoidable readmissions, other hospital-acquired infections, surgical safety, health care provider burnout, and falls.

The process of PD starts with identifying or defining an issue or problem. It is important that the people who will be doing the PD work take an active role in defining what the problem is. One best practice is to invite the community to meet together and explore the issue. They may reframe the issue in ways that are more meaningful to them. Community reframing is an example of the bottom-up nature of the PD process. In PD, “leaders” are facilitators, not experts. Their job is to make sure that the team doing the PD work has the resources it needs to do the work. The members of the community such as nurses and other staff working on MRSA abatement commit to doing the work and are accountable for the results. Having outside experts or nominal leaders “solve problems” would deprive the community of the essential trial and error required to really learn how to do something, not just what to do. Ownership of the problem and solutions by the frontline team is essential for adoption and spread of a successful practice.

Positive deviance generally proceeds as a group process, aided by trained PD facilitators who motivate action by the community members rather than by performing tasks themselves. Facilitators help the community establish baseline conditions, including the extent of the problem, and develop an assessment plan to identify current common behaviors with respect to the issue. The facilitator then helps the community identify positive deviants—the people, units, and teams that are succeeding despite current constraints—and how the positive deviants do what they do. The community then designs a learning exercise that spreads positive deviant practices to other community members. Positive deviance emphasizes experiential learning, also known in the PD community as “acting your way into a new way of thinking.”

In addition to defining the problem and designing the solution, the community develops its own process for tracking outcomes and results. This often includes existing measurement processes, but it also involves developing new ways of measuring and tracking the phenomenon. One reason this is particularly important is that it makes the invisible visible. In the case of MRSA transmission, spread of the bacteria from person to some inanimate object or substance or from person to person is invisible. An important component of identifying new practices that reduce spread is making this process more visible to the staff so they are mindful of it. For inpatient pain management, patient pain is often invisible to staff. Finding ways to make this more visible so staff can engage with it has been an important component of this project.

Positive deviance has a number of benefits. First, it has been remarkably effective in addressing problems that had been refractory to other social change efforts (Bradley et al., 2009; Pascale et al., 2010; Singhal et al., 2010). In addition, it does not require substantial new resources. The solutions it finds can be implemented with the currently available resources and constraints. Indeed, positive deviants have by definition found solutions despite dealing with the same limitations that constrain the other members of the community.

Positive deviance also has a beneficial impact on the culture of an organization. It is a bottom-up change management strategy that releases the potential energy of employees and puts it to work addressing important, intractable problems. One of us (M.T.) has seen firsthand how PD engages and inspires hospital staff in ways that more conventional top-down approaches fail to do. It can transform the organizational climate from resignation to innovation, from alienation to engagement. In addition to culture and climate change and increased team reflexivity, PD may generate a halo effect in that teams trained on PD may start to apply these methods to other improvement opportunities in their work.

Leaders starting a PD project in a health care organization face some challenges. For one, it is hard for them to step aside and not be the source of the solutions and answers, to let others do the work (see the discussion of adaptive leadership in the next section). Nevertheless, the success of PD projects is highly dependent on strong support from the very top of the organization. In the absence of commitment from top leadership, middle managers will often resist it because it threatens the usual power hierarchy, and their lack of control over the process may create anxiety regarding their ability to achieve their management objectives.

The PD approach is contrary to another common health care cultural norm: it is very process oriented and appears at first to be slow, messy, and disruptive. Again, this quality creates discomfort for managers and staff who are used to more controlled and orderly, albeit often ineffective, approaches to change management. In PD parlance, you have to go slow to go fast.

Positive deviance is a particular example of a general approach to engaging complexity in organizations that uses emergent design principles and a positive solution-oriented approach to change relational patterns. The role of leadership in engaging complexity is addressed next.

Adaptive Leadership

Leadership theories flowing from complexity science generally focus on the role of leadership in setting the conditions for organizational participants to jointly address adaptive challenges, that is, complex and often mysterious problems that require learning and organizational behavior change and cannot be solved by technical solutions and experts (Goldstein, Hazy, and Lichtenstein, 2010; Marion and Uhl-Bien, 2001 2011; Uhl-Bien, Marion, and McKelvey, 2007). Leadership is viewed as a process rather than a role, in subtle contrast to the transformational leadership approach that is a popular prescription for addressing complex problems (see Nembhard et al., 2009). The process of leadership involves interaction among dynamic agents in complex feedback networks, from which leadership of the system emerges.

Even accepting the view that leadership is a process, the question of how they can best lead still engages managers and administrators. Adaptive leadership (AL) is one approach to individual organizational leadership based on the understanding that organizations are complex adaptive systems that face adaptive challenges. The term adaptive leadership emerged from a variety of sources—for example, the application of chaos theory and complexity science to organizational leadership theory in the 1990s (Hickman, 2010). Heifetz and associates, early proponents of the concept, have developed practical guidelines for implementing AL, making its adoption more feasible (Heifetz, 1994; Heifetz, Grashow, and Linsky, 2009a 2009b).

As developed by Heifetz and colleagues, AL is a set of specific behaviors designed to facilitate organizational learning and behavior change. Its practice starts with the recognition that the challenges organizations face can be roughly grouped into two categories: adaptive and technical. Technical challenges are simple or complicated problems that are amenable to technical solutions like new software or machinery or technical experts. The experts do the work, and the problem is solved. Buying and installing new imaging equipment or upgrading the billing system are examples of technical work undertaken to solve technical challenges.

Adaptive challenges are complex and require that the people facing the challenge learn and adopt new beliefs, attitudes, and behaviors. They themselves, not experts or outside consultants, have to do the work to overcome it. Changing organizational culture, developing a new business strategy, and improving safety and quality are all adaptive challenges that require adaptive work.

Adaptive work means letting go of old attitudes, beliefs, and behaviors, often experienced as a loss. The threat of this loss typically engenders fear, and fear leads to resistance. Thus, organizations resist doing the adaptive work necessary to overcome their adaptive challenges, and failure to deal with organizational adaptive challenges generally leads to underperformance and existential risk.

Given the resistance to doing adaptive work, perhaps the most common error leaders make is to try to address adaptive challenges by substituting technical interventions for adaptive work. This always results in failure. Not only is the problem unsolved by doing this, but also it makes the problem worse. Adaptive leadership is the process of helping organizations do the work to overcome their adaptive challenges. It is not dependent on authority and therefore can be practiced by anyone in an organization. However, AL can be hard to practice because of organizational resistance to adaptive work, especially for people with low authority.

It might seem at first glance that AL presumes a deterministic model of system change such that the adaptive leader can function like an expert, diagnose the problem, and prescribe a set of AL interventions to solve it with some reasonable expectation of success. This interpretation, however, is incorrect and represents a reversion to type—the persistence in seeing the adaptive leader as a technical expert in AL and organizational change rather than someone who is participating in the adaptive work himself or herself. The solution to an adaptive challenge will typically emerge from the work done by members of the organization to address the challenge. Adaptive leaders do not need to know the solution to the challenge ahead of time, and it is likely that if they think they know the solution, they will be wrong. This echoes the PD experience. Adaptive leadership facilitates a bottom-up process that engages the organization in addressing adaptive challenges and doing the adaptive work to overcome them.

Adaptive leadership has application beyond organizational leadership. It has been used as a framework for reforming the clinical practice of medicine by recasting the clinician's role as an adaptive leader making judicious use of technical interventions while facilitating the patient's adaptive work in order to overcome health challenges that generally have strong adaptive components (Thygeson, Morrissey, and Ulstad, 2010).

There is little empirical literature about AL. Anecdotes regarding the effectiveness of the approach are available in the business (Heifetz, Grashow, and Linsky, 2009b; Useem, 2010) and academic (Randall and Coakley, 2007; Eubank et al., 2012) literature. However, the scope of the concept remains incompletely specified, and there is much preempirical work to be done to identify the constituent components and behaviors of adaptive leadership with sufficient rigor that it can be studied empirically. This represents an opportunity for research in management and leadership.

Adaptive leadership is related to complexity science in a number of ways. Ontologically, the concept arose along with the emerging understanding of organizations as complex adaptive systems. The distinction between technical challenges and work and adaptive challenges and work is reminiscent of the Stacey matrix zones “simple” and “complex” described previously in this chapter. Where there exists a high level of agreement about what needs to be done and how to do it, technical interventions can be very effective. But where the solution is more mysterious (low consensus) and the possible actions are definitely countercultural or controversial (low certainty), technical interventions are typically counterproductive, and a more adaptive approach to leadership is warranted. The injunction not to try and solve adaptive challenges with technical interventions is also an application of Ashby's law of requisite variety: a system (intervention) must be as complex as the challenge it addresses (Ashby, 1958). Adaptive leadership is a set of techniques to change the normative conversational patterns and complex responsive processes of the act of relating in an organization. The intent is to influence the simple rules so as to facilitate addressing and resolving an adaptive challenge (Stacey, 2001).

Much AL explores how to facilitate the creation of new social structures such as relationships, committees, and funding streams. This includes conversations that allow people in an organization to address blind spots (sense making) and work together to address the adaptive challenges they face. These are the core activities of leadership in complex adaptive systems (Begun and White, 2008; Drath, 2001 2004a, 2004b). In the words of Marion and Uhl-Bien (2001), AL is a process not for “controlling the future” in an organization, but rather “fostering interactive conditions that enable a productive future” (p. 394).

An additional virtue of AL is that it largely avoids the jargon of complexity science. Thus, it may be less obscure and more acceptable to people with a variety of cognitive and social styles or positions in the organization. It may even avoid drawing the blank or hostile stares that seem so often to be the response to talking about complexity science.