SEEING THE REST OF THE COMMUNITY: USING COMPLEX SYSTEMS TO REVEAL THE STRUCTURE AND FUNCTION OF INTERDEPENDENCE By Kyle R. Metta A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Community Sustainability Ð Doctor of Philosophy 2020 ABSTRACT SEEING THE REST OF THE COMMUNITY: USING COMPLEX SYSTEMS TO REVEAL THE STRUCTURE AND FUNCTION OF INTERDEPENDENCE By Kyle R. Metta Interdependence occurs when autonomous elements of a syst em interact, enabling the emergence of an overall system and its behavior. The humbling act of studying interdependence requires a shift from a reductionist world view that understands reality through its components. Instead, it holds that we can only unde rstand reality by accounting for the whole and appreciating its components' mutual interaction. Studying the patterns that underlie in ter dependence can yield insights into causal dynamics revealing how structure can lead to innovation, novel system states, and problems resistant to intervention. ...... In this dissertation, I present three studies to broaden the literature on how complex systems and systems thinking can unmask the structure and function of interdependence in place -based sustainability problems . In my first study, I examine the field of participatory modeling and use document citation and network analysis to reveal communities of practice in this field. By understanding the connections and communities of scholars in this work, I show the emergen ce of separate but related research fronts and how they diverge in their approach to participation and modeling. My next two studies are situated in Flint, MI, a community still responding to the water crisis's social -ecological disaster. These studies exa mine how a community can use systems thinking to elevate and target their positive change efforts. The second study explores how interdependent connections in the network governing the food system can explain the community's capacity to foster social learn ing, innovation, and !adaptation. It uses the small -world network model to assess social -ecological resilience as a function of a network's clustering and density. My third study deals with system archetypes or system structures that produce characteristic patterns of problematic behavior due to the interdependence of components. Though system archetypes are a well -documented tool for communicating the structure and behavior of systems and have been applied across various contexts, their identification is of ten difficult. This study demonstrates an explicit process for identifying system archetypes. It uses a qualitative coding scheme adapted from Wolstenholme's (2003) definition of isometric archetypes to elicit structure and behavior from purposive text dat a generated from a community visioning process. This process increases the narrative's connectedness to the model, which can enhance the modeling process and give specific insights into systems thinking pedagogy and practice. !iv! For C.M.D. and M.E .M. Thank you for being both my compass and guide s. !v!ACKNOWLEDGEMENT S This dissertation would not have been possible without the support and guidance from faculty and colleagues at Michigan State University. I would especially like to acknowledge the men torship of Dr. Laura Schmitt Olabisi as her commitment to participatory research inspired and made this work possible. I would also like to give special acknowledgement to Dr. Miles McNall who se mentorship in community engagement, critical reflection, and evaluation was vital to this dissertation and my PhD studies . I admire his consistent and unyielding pursuit of interesting and meaningful science that is useful to community members . I would also like to thank my committee members Dr. Steven Gray and Dr. Michael Hamm for their patience and support especially when things proved difficult. Their perspectives were invaluable throughout the process. Thank you to my friend and research partner, Ren”e Wallace from Food Plus Detroit. Renee and I started this re search journey six years ago and I believe we share this dissertation as a joint achievement in what can be done in collaboration and shared values. This dissertation was written during a pandemic under a lockdown order. I would never have seen the light t hrough that darkness if not for the support of Dr. Jennifer Lawlor who quarantined with me and talked out every idea in this dissertation. I am forever in your debt Jenny . Thank you for all that you do and who you are Ña critical thinker with a dedication f or community engagement and an immense level of patience. This dissertation could not have been completed without support from the Sustainable Michigan Endowment Project (SMEP) , and the C.S . Mott pre -doctoral !vi!fellowship. Additionally, I Õd like to thank Jessica Brunacini, Dr. Alison Singer, Dr. Jason Snyder, Dr. Aniseh Bro, and Dr. Udita Sanga for their support, friendship and help thinking through the writing and research design process. Finally, thank you to my family for their unwavering support and l ove throughout my educational journey. Thank you to my mother Lynn, for teaching me empathy, kindness, and the importance of community. Thank you to my sister Kimberly for her support and always being the loudest one cheering . Thank you to my father Ric hard for inspiring and supporting a deep curiosity in me from a young age. !vii !TABLE OF CONTENTS LIST OF TABLES ................................................................................................................... ix LIST OF FIGURES .................................................................................................................. x Chapter 1 Introduction ........................................................................................................... 1 Introduction ......................................................................................................................... 1 Chapter 2 Participatory Modeling Network ........................................................................ 4 Introduction ......................................................................................................................... 4 Bibliometrics ........................................................................................................................ 8 Dataset ................................................................................................................................ 11 Results ................................................................................................................................. 12 Discussion .......................................................................................................................... 16 Conclusion .......................................................................................................................... 20 Chapter 3 Connectivity and Social -Ecological Resilience ............................................... 22 Introduction ....................................................................................................................... 22 Small Worlds ...................................................................................................................... 24 Stakeholder Mapping ....................................................................................................... 26 Translating Stakeholder Maps to Networks .................................................................. 28 Analytical Methods ........................................................................................................... 31 Results ................................................................................................................................. 32 Discussion .......................................................................................................................... 34 Conclusion .......................................................................................................................... 36 Chapter 4 Translating Narratives into Archetypes .......................................................... 38 Introduction ....................................................................................................................... 38 Archetypes .......................................................................................................................... 38 Qualitative Research and SD ........................................................................................... 41 Data Collection .................................................................................................................. 43 Data Extraction ................................................................................................................... 44 Major Stocks ................................................................................................................... 45 Central Actors ................................................................................................................. 45 Behavior ........................................................................................................................... 45 Sectors and System Boundaries .................................................................................... 45 Structure .......................................................................................................................... 45 Problem ............................................................................................................................ 45 Action ............................................................................................................................... 46 Intended Consequences ................................................................................................ 46 Unintended Consequences ........................................................................................... 46 Delays .............................................................................................................................. 46 Feedback loops ............................................................................................................... 46 Descriptive Results ........................................................................................................... 48 Examples ............................................................................................................................. 49 Underachievement ......................................................................................................... 50 Out -of-control ................................................................................................................. 52 !viii !Relative Control .............................................................................................................. 54 Relative Achievement .................................................................................................... 56 Discussion .......................................................................................................................... 58 Conclusion .......................................................................................................................... 61 Chapter 5 Conclusions ......................................................................................................... 62 Introduction ....................................................................................................................... 62 APPENDICES ........................................................................................................................ 66 APPENDIX A Stakeholder Mapping Protocol .............................................................. 67 APPENDIX B Archetype Codebook ............................................................................... 68 REFERENCES ........................................................................................................................ 69 !ix!LIST OF TABLES Table 2 -1 Detected Communities in Participatory Modeling ......................................... 15 Table 2 -2 Participatory Modeling Compared .................................................................... 16 Table 3 -1 Descriptive Statistics & Small -World Quotient .............................................. 33 Table 4 -1 Archetypes Descriptive Table ............................................................................ 48 Table 4 -2 Actors and Stocks by Frequency ........................................................................ 49 !!x!LIST OF FIGURES Figure 2 -1 Web of Science Search Terms ........................................................................... 11 Figure 2 -2 Tree Map of Participatory Modeling ............................................................... 12 Figure 2 -3 Participatory Modeling Publications Overtime ............................................. 13 Figure 2 -4 Co -citation Network ........................................................................................... 14 Figure 3 -1 Comparing Degre e Distributions Across Networks ...................................... 29 Figure 3 -2 Social Ecological Governance Network .......................................................... 34 Figure 4 -1 Comparing Generic and Semi -generic Archetypes ....................................... 47 Figure 4 -2 Generic Archetypes Underachievement .......................................................... 51 Figure 4 -3 Generic Archetype: Out of Control .................................................................. 53 Figure 4 -4 G eneric Archetype: Relative Control ............................................................... 55 Figure 4 -5 Generic Archetype: Relative Achievement ..................................................... 57 !1!Chapter 1 !Introduction !Introduction The economist Thorstein Veblen once wrote, "It is always sound business to take any obtainable net gain, at any cost and at any risk to the rest of the community." Though his intention is satirical, the Ôrest of the community Õ is often invisib le and difficult to consider. They are invisible not only to business leaders and decision -makers but also to other community actors. Even if we put selfish intent aside, the rest of the community goes unseen and unheard. Whether the rest of the community is downriver from an operation or scholars circling similar ideas or a generation yet to be born, it can feel impossible to calculate how actions today and here may impact them and there. By taking the study of interdependence seriously, as this dissertati on attempts, we can reveal the structure, or the links and connections, to the rest of the community and potentially learn to act with them in mind. A complex systems approach allows researchers to engage with the concept of interdependence. Interdependen ce occurs when autonomous elements of a system interact, enabling the emergence of an overall system and its behavior. Studying interdependence requires that we account for the whole and appreciate mutual interactions. Studying the patterns that underlie i ndependence can yield insights into causal dynamics revealing how structure can lead to innovation, novel system states, and problems resistant to intervention. In the social sciences , the concept of Òcommunity Ó can be used to describe geography, affili ation , family structures, and identities of faith, circumstance, or interest (Doberneck, Glass, and Schweitzer, 2010). Often, Òcommunity Ó is used to describe !2!physical locations or as a categor ical indicator used to explain variance in opinions or outcomes (Maqueen, et al 2011). Unlike traditional social sciences, complex systems and network science view community as more than an explanatory variable or even the context in which research happens. In this approach community is defined as the structural relati onal ties that exist in-between individuals. When represented as graphs, communities are present when there is high transitive closure, or when relations between three entities X, Y, Z show closure in that X is connected to Y and X is connected to Z , closu re would be when Z and Y are also connected. This approach to community considers the interdependent relationships of individuals, their environments and the social learning that transmits across cultures and spaces. In this dissertation, I present three studies to broaden the literature on how complex systems and systems thinking can unmask the structure and function of interdependence in place -based sustainability problems. This unifying theme of interdependence creates a space to think critically about the community structures that drive systems. The first study, Chapter 2, examines participatory modeling and uses document citation and network analysis to reveal communities of practice in this field. Participatory modeling (PM) provides various tools and techniques to represent empirical, theoretical, and experiential understanding using a semi -standardized language. There are many strands of PM literature, here defined as different approaches and methodologies for considering models, modelers, stakeholde rs, and issues, that span the environmental management, planning, and action research spectrum. Though recent contributions to the PM literature have been able to synthesize trends, there hasn't been a complete systematic review of this literature, and man y open questions remain. !3!Chapter 3 and 4 are scientific contributions that look closely at Flint, MI, a community still responding to the water crisis's social -ecological disaster. These studies examine how a community can use systems thinking to elevate a nd target their positive change efforts. Chapter 3 explores how interdependent connections in the network governing the food system in Flint, MI, can explain the community's capacity to foster social learning, innovation, and adaptation. It uses the small -world network model to assess social -ecological resilience as a function of a network's clustering and density. Using descriptive network measures examines how the Flint food system responds to crisis, how it can improve and harness the power of the commu nity's collective assets. Chapter 4 looks closely at the concept of system archetypes or system structures that produce characteristic patterns of problematic behavior due to the interdependence of components. Though system archetypes are a well -documented tool for communicating the structure and behavior of systems and have been applied across various contexts, their identification is often difficult. This Chapter is motivated by a desire to create accessible systems science employed in community contexts . This study demonstrates an explicit process for identifying system archetypes. It uses a qualitative coding scheme adapted from Wolstenholme's (2003) definition of isometric archetypes to elicit structure and behavior from purposive text data generated f rom a community visioning process. This process increases the narrative's connectedness to the model, enhancing the modeling process and giving specific insights into systems thinking pedagogy and practice By studying community as the result of interdepend ent relationships, we can begin to understand the intersecting environments and conditions that either promote or hinder health, environmental conditions, and general well -being. 4 Chapter 2 !Participatory Modeling Network Introduction Participatory modeling (PM) is an increasingly popular approach to including stakeholders in defining, analyzing, and managing socio -environmental issues. Modeling provides a variety of tools and techniques to represent empirical, theoretical, and experiential understanding using a semi -standardized language. Scholars have acknowledged that there are many strands of PM literature, here defined as different approaches and methodologies for considering the role of models, modelers, stakeholders, and issues that span the environmental manag ement, planning, and action research spectrum (Gray et al., 2018; Lyna m, Jong, Sheil, Kusumanto, & Evans, 2007; Voinov et al., 2016) . The approaches may also differ in how structured or prescribed the participatory process tends to be. They pull from different lines of participatory literature and have lineages in different types of modeling. However, there is no broad level systematic synthesis that examines these different strands analytically. PM is unified across broad approaches in its epistemological orientation and appreciation of knowledge integration across concept ual, expert, and experiential dimensions. This orientation provides a framing in that explicit representation of knowledge (or models) is an important way to learn about knowledge and to create objects that facilitate dialog and further insights (Boundary Objects). PM approaches hold that knowledge can be attained and understood by exploring diversity and convergence of thought related to complex issues. The practices are also united in their problem orientation and in their application to complex or wicked problems, which are dependent on context -specific circumstances with interacting forces, and are resistant to singular solutions (Rittel & Webber, 1973). 5 Participatory modeling approaches are also aligned in their orientation and application of systems thinking. Systems thinking is a way to view a problem and its causes as a whole system, recognizing that the patterns and cycles of behavior are a result of interrelated components and how they change over time (Meadows, 2008). Using systems thinking skill s can add perspective to wicked problems and address the helplessness often associated with them. It can take abstraction and provide an understanding and ability to find solutions to the root causes of problems (D. H. Meadows, 2008; Stave, 2003) . Be cause of these features, all strands of participatory modeling are appropriate and well suited for conducting sustainability science research --or place -based, problem -oriented form of inquiry with the goal of linking knowledge creation to actions that adva nce ecological and social wellbeing (Miller, 2013) . Scholars in sustainability science and ecological modeling have attempted to synthesize some of the PM literature, though not in a systematic way (Gray et al., 2018; Naivi na, W., Le Page, M., Thongoi, M., Trebuil, n.d.) . In a review by Gray et al. (2018) they show that PM does not share a consensus on how participation is framed, though they reference how authors who have attempted to clarify how processes work while main taining the flexibility to adapt to constraints of problem, sector, and modeling tool kits. Voinov & Bousquet (Voinov & Bousquet, 2010) describe PM as the use of an assortment of tools to have participants create formalizations of knowledge. These formalizations (or models ) can take the form of collective diagrams, rich -pictures, or individual representations of mental models. Beyond representation, PM may use participation to inform or interact with simulation models, and graphically represent a distributional understand ing of populations under high uncertainty using local or indigenous knowledge. Voinov & Bousquet (2010) contend that because of the human dimensions of PM, there can be no 6 unique guidance or methodology to inform PM on how to create meaningful engagement o f all participants. Though they briefly give an overview of participatory action research and other methodologies for engaging with stakeholders, they contend that the human dimensions require more flexibility for modelers. They do, however, provide an ada ptable and detailed analysis of the necessary components and principles for a participatory modeling process. Voinov et al. (Voinov et al., 2016) argue that there remains a gap in guidance articles for practitioners regarding the tools, methods, and processes used in PM. They write that the "current lack of guidance is, in part, the result of our highly diverse human society that retains a heterogen eous distribution of knowledge and highly localized believe systems." Voinov & Bousquet (2010) offer two summary objectives that often motivate PM. The first being to increase and share knowledge of a system, and secondly, to identify and clarify the impa cts of a solution to a given problem to support decision making, policy, or management. Voinnov et al. (2016) expand on the ambiguity of the participatory process and motivations in PM by noting, "É articles document the development of new tools and method s in a particular case study rather than critically assessing the stakeholder engagement process per se. This is not a trivial issue..." Answering some of these issues of guidance, Voinov et al. (2018) outline the methods and tools used in PM and Gray e t al. (2018) provide the 4P framework to "Éhelp design and assess all cases of PMÉ." The 4P framework includes the Purpose (1) for selecting PM, the Process (2) by which the public was involved, the Partnerships (3) formed, and the Products (4) from the ef forts. The 4 - frameworks intended purpose is to assist in the Synthesis and reporting of PM projects across tool -based paradigms, subfields, or publishing outlets. The 4p framework can be helpful in understanding what is similar in these practices, what is different, and what can be learned across the 7 emerging field of practice. In the 4P Framework, purpose is specifically related to two dimensions of why a PM process was chosen, it includes the justification for why PM is used and the defining issue that the model hopes to elucidate. Clarity is needed in these separate respects to understand when participation is necessary and when modeling is necessary. As with many aspects of applied social -ecological research, beginning with a rich description of the p roblem statement directs and informs the research direction, methods, and theories employed in the work. Beyond the 4P Framework, other authors in the PM literature have put emphasis on clarifying the purpose of PM. Voinov and Bousquet (2010) contend that ÒStakeholder engagement, collaboration, or participation, shared learning or fact -finding, have become buzzwords and hardly any environmental assessment or modeling effort today can be presented without some kind of reference to stakeholders and their invo lvement in the processÓ (p. 1268). They refer to two major objectives specific to environmental modeling with stakeholders, one being to increase and share knowledge of a system under a variety of conditions, and the other to increase stakeholder buy -in of potential solutions. However, the range of goals of a PM process can vary and may explain differences in practice and orientation. Though recent contributions to the PM literature have been able to synthesize some of these trends, there isn't a complete review of this literature, and many open questions remain including what literature informs the practice, what types of problem spaces is PM helpful, and what practices guide the participatory processes. This emerging interdisciplinary field could benefit from a clear examining of the PM literature through the lens of the 4 P's framework in a systematized process. A synthesis can help explain how these divergences in practice manifest in conceptualized and actualized purposes employed. 8 Bibliometrics Biblio metrics is the quantitative analysis of the nature and course of scientific discovery and disciplines. It uses the documents of science and inquiry (books and research articles, etc.) to understand their bibliographic content. Pritchard (1969) first introd uced the idea of bibliometric analysis as "the application of mathematics and statistical methods to books and other media of community" (Pritchard, 1969 p 348 -349). Bibliometric methods include areas of citation analysis, content analysis, and network science. While these tools are often used in the field of information and library science there has been an expanded use in applying it to other areas. Emerging fields have used bibliometrics to explore and explain the prominence of certain works, scholars, a nd ideas. These techniques are increasingly used to map science as a structure of knowledge and view discovery of knowledge as multifaceted communication (Pritchard, 1969) . De Solla Price (1985) coined the concept of the research front in his seminal paper on the topic. In PriceÕs conceptualization, the tendency for scientists to cite the most recently published articles on a topic creates citation networks that are very dense and relevant to specific aspec ts and contributions of research. Research fronts can be seen as the pockets of science in a given domain that describe specific knowledge creation that is being communicated through scholarly products. Citation Analysis (CA) is a method developed by bib liometric scholars to identify areas of scholarship and has been used in areas of interdisciplinary research to understand relationships and trends in the literature (Leydesdorff, 1998; Trujillo & Long, 2018; Yan, 2012) . It allows the researcher to construct a network based on features of the citi ng and cited literature of each document. This process reveals patterns in the 9 document dataset and has been used to identify research fronts and emerging communities of practice. Citation analysis uses citations as a way to understand the evolutionary, ve rsus historical context of knowledge development. It views science as a knowledge object being continuously reconstructed through the reflexive rewriting of histories in the light of new empirical findings (Leydesdorff, 1998). Co-citation analysis, a form of citation analysis, constructs networks of documents where edges are based on shared citing literature. The associations here are inferred based on the level at which other documents cite a set of works (Yan, 2012). Bibliographic Coupling (BC), like co -citation analysis, is a method for understanding fields of research that specifically looks at the development of a field or research front (Kessler, 1963) . It looks to the past and constructs relationships based on documents co -citing work. This allows us to consider what literature is central to the framing of the scholarship, what methods are being used, and what topic areas are relevant to a particular group. Though similar, BC and CCA construct networks with different structures and have different purpo ses. BC can tell us more about the development of the research front by explaining what literature informs scholarship. CCA is more forward looking, and captures the relationships based on how documents are being co -cited. Boyack (2010) empirically tested CCA and BC to determine which network allows for the most accurate detection of front of fields. They find that though BC most accurately finds research fronts in fields with a long history, and that CCA can be helpful when identifying new or emerging fiel ds (Boyack, 2010) . For the purposes of PM it is necessary to understand where the strands of practice come from and if new fields of practice are emerging. In network science, Com munity Detection Algorithms (CDA) or modularity 10 assessments, are a set of algorithms that consider connections and the presence of transitive relationships to identify sub -groups (clusters) within a larger network (Fortunato, 2010; Girvan & Newman, 2002; Porter, Onnela, & Mucha, 2009) . It is a method used to identify modules an d hierarchical structures based on the topographic network information. Community structure is key to understanding how networks function and to analyze patterns of connection (Porter et al., 2009) . For the purpose of this research, these communities are groups that tend to cite documents in similar patterns. Newman (2004) describ es modularity as the fraction of edges (or links) in a network that connect nodes (here documents) controlling for the expected value of the same quantity in a network when random connectivity is assumed between nodes in the same cluster. CDA can be used w ith networks created with BC and CCA to identify the research fronts in PM and create the context for the cross -comparative analysis. Within bibliometrics research co -citation clusters within a coherent field represent research foci and specializations (McLevey & McIlroy -Young, 2017) . To understand and ascribe meaning to t he different groups we found in the networks we followed a similar protocol outlined in Trujillo & Long (2018). Where we identified important works in the detected modularity groups through an analysis of centrality. Here centrality refers to the top frequ ency of co -cited works by degree. The high centrality score of weighted edges indicates that a document received recognition among scholars in that identified community. The top three works in each modularity group will be evaluated for inclusion in a qual itative analysis. To be included in the final analysis the article needs to (1) be a paper that deals with participatory modeling in a way that deals with stakeholder groups creating formal to semiformal representations of systems and (2) describes in empi rical terms the process at which the modeling took place. 11 Dataset To create the corpus data -set we included as many documents within the domain of PM as possible. As noted above, PM is an interdisciplinary form of scholarship that includes many disciplin ary homes and venues. Web of Science was the search platform used for its wide coverage of scientific fields and capability to provide citation information in meta -data. The proposed dataset was created by Web of Science Core Collection using the ÒAll Fiel dsÓ search with the following criteria for all languages, though only the English variants of search terms were used. ÒModelingÓ and ÒmodellingÓ were used in all variants due to regional spelling differences. This search resulted in 1,117 records after fil tering to include articles, reviews, books, conference papers, and structured abstracts. The corpus largely consists of articles (813), conference/meeting papers (363) and book chapters (58). Figure 2-1 Web of Science Search Terms 12 Results Figure 1 displays a tree map of the field categories classified by Web of Science for the PM corpus. Here we see that the interdisciplinary nature of PM is represented with computer science, environmental sciences, and business representing over a third of the corpus. Integrative social sciences, including sociology, public administration, psychology, make up another third. Figure 2-2 Tree Map of Participatory Modeling The corpus contains 1,117 documents representing the range of disciplines in PM. The source documents have a varied distribution of publication dates with a strong trend beginning in the mid 1990s reflecting a the fairly recent growth of this field in rec ent years. Table 1 shows the top authors, journals, and documents in the corpus based on citations. This description of the corpus serves as a preliminary check that no subfields of PM are being excluded systematically by the query or the indexing of the database. 13 Figure 2-3 Participatory Modeling Publications Overtime These core documents cite a total of 28,940 unique documents. However, because we are interested in the subfield of PM and not uncovering t he larger mapping of the scientific community, we focus on the connections within the internal core of the corpus. Within the core we have 405 documents co -cited with 587 links. To further refine the data for visualization purposes, we removed weak connect ions of co -citing >3 times. This leaves 143 uniquely cited documents and 579 links between them. This network is visualized at this trim level in Figure 2. This graph represents the co -cited documents as nodes and the frequency of co -citation as weighted l inks or edges. Size of the node represents degree and coloring reflects detected communities. 14 Figure 2-4 Co-citation Network Table 2 reports bibliographic information for the three most frequency co -cite d document by degree across each identified community groups. Data is reported as follows: ÒcitedÓ refers to the number of source documents citing this document, and degree is the number of other documents jointly cited within the core PM documents. For in stance , ÒVoinov A, 2010 Ó represents the largest degree and cited document in community 1, being least three times with 82 other documents in the core. !"#$%&''()*(+,-... !/#0(1*(+,,2*(3... 4"567589(:*(+,,... ;&<9/'8=>?&'9"6... @"<<&A/%A0"'B(C... ?#"99(!D*(+,-E*... F8''/G(H&%*(-22... I<&J(?&*(+,-E*(... 1K8<6(;*(+,-+*(... L8(M&B8()*(+,-+... N'68O%8(4*(+,-,... P"K%&'Q(M6*(+,-... ;8'Q"=&(I&*(+,,... R&/K/'/9(@*(+,-... I5<5'B(S<*(+,,T... H"'86(R&*(+,,2*... M&O%8<(!'*(+,-E... LJ'&%(S*(+,,+*(... !&K8<&(:*(+,--*... )&K&'&(!J*(+,,U... :&O#"'/(?%*(+,-... 16$/'&6/(;*(+,,... V"<)&<<8<&(L... L&'B6Q&O8(?*(+,... C<58B8<(S*(+,-+... ;#<"W8<96(C#*(+... F8''/G(H&%*(-22... N$$8<%&'6(P*(+,... ?/%"'()*(+,-,*(... N;"<5&(N*... M<8OO()*(+,,U*(... 4&<<898&5(^*(+,... Z_N75/'"(M*(+,,... ;&'8=(;*(+,,U*(... P"%&(L*(+,-]*(N... 4"5@"69O()*(+... F/Q8/<&(R*(+,-X... !"5Y8998(1&D&*(... I<&J(?*(+,-[*(1... ?#"99(!D*(+,-T*... F&'(Z8'(48O9(;*... NOO8'Q8<(?*(+,-... S0"%&6(3%*(+,-]... !"5Y8998(1&D&*(... S/QY8OO(F#*(+,,... I"5<%8O"'(:*(+,... L&'B6Q&O8(?%*(+... !"5Y8998(1*(+,-... I<&J(?&*(+,-]*(... !/#0&@"69O()*(+... F"/'"K(N*(+,-X*... N'95'86(M*(+,-]... F/Q8/<&(R*(+,,2... I&QQ/6(1DW*(+,,... 4&<'&5Q()*(+,-E... ?"5#08<8(F*(+,-... L5'&>!8J86(L\*(... F/668<(;*(+,,U*... F/Q8/<&(R*(+,-+... N9$/'6"'(H&%*(+... )&698OO899/(N*(... N#$8<%&''(:*(+,... ?9&K8(C&*(+,,+*... L&'8(Z#*(+,,[*(... ?#"99(!D*(+,-]*... I<&J(?@"69O()*(+... ?8QO&#$"(;*(+,-... P8'OJ>?08A&?&'9"6... ?8'Q=/%/<(H*(+,... ?&'Q"K&O>?"O/6(... R8O6"'(Z&*(+,-]... N'Q8<68'(Z\*(-2... 15 Table 2-1 Detected Communities in Participatory Modeling 16 Table 2-2 Participatory Modeling Compared !Discussion By inspecting the network graph, detected communities, and top degree papers in the PM core some generalizations can be made about the content of the d ifferent sub -fields of PM. Group 0 represents a small cluster of documents written in the mid to late 2000s representing 14% of the documents. This group is dominated by an approach to PM called Mediated Modeling. Mediated Modeling (MM) is an applicatio n of GMB aimed 17 especially at building consensus and conflict resolution in social -environmental management issues. The approach uses visually oriented system dynamics modeling software to iteratively and collaboratively construct a model (or models) of a s ystem in which conflict about alternative policy or management decisions is existing or anticipated (Van den Belt, 2004). Its conflict mediation approach prescribes a structured engagement attuned to disagreement and utilizing explicit representation of m odels to represent and understand perceived conflict. This orientation necessitates the inclusion of processes of deep reflective process and constructive dialog to address and remove tension and includes momentum towards acting in difficult situations. MM finds its participatory lineage in the design -oriented action research field and uses an action -reflection cycle to understand the MM process as it is initiated and proceeds with each group. MM pulls from Zuber -Skerrit (1992)Õs CRASP framing of design -oriented action research, in which ÒCritical collaborative enquiry by reflective practitioners, who are Accountable in making the results of their enquiry public, Self -evaluative of their practice, and engaged in Participative problem solving and continuing professional development .Ó Group 1 is the largest group in the network and contains 50.35% of documents , all of which are from journals dealing with ecology, environmental management, or socio -environmental issues. Due to its larger breadth, it is more di fficult to draw specifics, therefore we can consider it the Generalized Environmental Modeling group. GEM is situated more fully in sustainability science and ecological modeling than the other strands, and (in some cases) shares some of the formalized me thodologies of GMB and ecological economic approaches (Gray et al. 2018; Naivina et al., 2012; Voinov et al. 2018). Often, GEM is a tool for adaptive management and adaptive co -management. This strand does not share a consensus in how participation is fram ed, though some 18 authors have attempted to clarify how their processes work while maintaining the flexibility to adapt to constraints of problem, sector, and modeling tool -kits. Voinov & Bousquet (2010) describe GPM as the use of an assortment of tools to h ave participants create formalizations of knowledge. These formalizations (or models) can take the form of collective diagrams, rich pictures, or individual representations of mental models. Beyond representation, GPM may use participation to inform or int eract with simulation models, and graphically represent a distributional understanding of populations under high uncertainty using local or indigenous knowledge. Group 2 is the smallest group representing only 5% of the documents in the network and seems to be dominated by documents describing agent -based modeling and role -playing games oriented towards environmental decision making. Documents in this group seem to be associated with the Companion Modeling (ComMod) approach associated with researchers with the Agricultural Research for Development Agency (CIRAD) in France. The approach requires crossing disciplinary boundaries and views modeling as an intermediary object to facilitate collective and interdisciplinary thought (Barreteau, Bots, & Daniell, 2009; Kelly et al., 2013) . This approach is novel in that it requires p rocesses for understanding, confrontation, and shared analysis. In describing this posture, architects of the approach stress that that modelers discard all assumptions backing models after each interaction, to have no a priori implicit hypothesis, and to pay critical attention to issues and processes for validation. ComMod is described as having two main objectives: Understanding complex environments, and to support collective decision -making processes. They base their work in iterative fieldwork Ð model ing Ð simulation cycle that produces a diversity of models and processes that each contribute to the main objectives. Group 3 represents 31% of the network and can be thought of as Group Model 19 Building, an approach to modeling with stakeholders that origi nates in the 1980s from collaborative work by researchers in the Netherlands and the SUNY Albany system dynamics group (Eskin asi, Rouwette, & Vennix, 2009; Richardson & Andersen, 1995; Rouwette, Vennix, & Van Mullekom, 2002; Vennix, 1999) . GMB builds off of early work in system dynamics and client -based modeling (John D Sterman, 1992) . GMB is one of the first facilitated processes developed and systematically studied that looks at the effects of stakeholder involvement in the development, parameterizing, and scenario testing of (mostly system dynam ics) models. The larger methodology, however, has been used with agent -based -modeling, concept mapping, network simulations, and a variety of combinations of integrated modeling. The method, though flexible and amenable to many contexts, finds its roots in the business and organizational behavior literature. Vennix (1999) outlines GMB as a practice of involving stakeholders in the modeling practice that introduces social dynamics that can affect the model quality, stakeholder buy -in, and the likelihood t hat actions are taken based on modeling insights. Richardson and Anderson (1995), also developers of GMB, distinguish their approach in narrower terms, as a process with "the intent to involve a relatively large client group in the business of model formul ation, not just conceptualization Ó (Richardson and Anderson ,1995). GMB is united in that it employs structured processes that involve the use of facilitation scripts to illicit causal system -level understanding from stakeholders (Andersen & Richardson, 1997; Hovmand et al., 2012) . This design choice is to increase the empirical and testable nature of the model building process, with an understanding that the social dimensions of model building have an impact on the resulting models and insights (Hovmand et al., 2012) . Also, GMB largely focuses on top dow n modeling and documenting and uncovering the presence of feedback loops, 20 and attempts to come to group consensus about system structure. Also , within Group 3 we find Community -Based System Dynamics (CBSDM). CBSDM is an approach to GMB that provides a met hodological framing rooted in the literature of Community -Based Participatory Action Research. The developers of this approach provide a highly structured and community -centered methodology to GMB that emphasizes long -standing community partnerships and co mmunity ownership of models (Hovmand, 2010.). It diverges from GMB, in its purpose of Òinvolving participants to create a community of practice around a model that can be used to design innovations that the community will advocate for and implementÓ (Hovma nd, 2010. p. 26 ). The processes of CBSD are prescriptive beyond general modeling of GMB to provide tools to define the community clearly, and places participants in the role of researcher, modeler, and interpreter of modeling results and process. Rooted i n action research, this approach is ontologically and epistemologically tied to how the community is framing the problem. Conclusion The findings from this analysis provide a novel understanding of the field of participatory modeling and contributes a bro ader understanding about the nuances in the different ways modelers approach practice . By identifying PM subfields through quantitative means we contribute to the PM review literature (van Brugg en, Nikolic, & Kwakkel, 2019; Voinov & Bousquet, 2010; Voinov et al., 2018) while integrating across disciplinary and modeling frameworks . PM often requires an interdisciplinary understanding of the problem the modelers are investigating. By this same pr inciple , modelers could benefit from learn ing from other 21 literatures and traditions of PM. This study shows that there are divergences in the specificity and formal processes used in participatory modeling and the practices of one tradition may enhance the practice in other traditions. For instance, though the sub field s of environmental modeling and public and community health modeling have developed within specific traditions , there is significant overlap s in the type of problem s being modeled and the role of models play in system scale decision making under uncertainty. This work provides a new cross -tradition framework for considering PM as a 22 Chapter 3 !Connectivity and Social -Ecological Resilience Introduction Connectivity is an important component in the heuristic of SES Resilience; however, it is difficult to measure or approximate. Using Stakeholder Mapping and Small World measures can give prac tical insight into what this means for food systems. In this paper, we review the concept of connectivity and how it has been used in SES resilience and offer the measure of Small World effect to understand one aspect of connectivity that has been overlook ed in empirical studies. We then present a case study using this measure to understand the SES resilience of an urban food system. We then contrast the SME with robustness measures and propose the creation of a metric to articulate the tradeoffs between ro bustness and small worldliness . The concept of resilience has multiple meanings related to the scholarship of sustainability. Originating in the field of ecology (Holling, 1973) and now extending to many interdisciplinary scholarship branches, resilience is universally seen as a property of a system. Quinland et al. (2015) outline three definitions from the ecological literature. These are: engineering resilience Ðor the speed a system returns to a (particular or singular) equilibrium after experiencing a shock; ecological resilience Ñor the magnitude of disturbance that a system can absorb before shifting to an alternative regime (multiple e quilibria); and social -ecological resilience which extends the ecological definition to include the amount of disturbance that a system can absorb and remain within a domain of attraction, the capacity of a system to learn and adapt, and the degree to whic h a system is self -organizing (Carpenter et al., 2001; Quinland et al., 2015). 23 As a system property, social -ecological resilience offers a rigorous appreciation for the emergent and complex properties of social -ecological systems (Berkes & Folke, 1998; Westle y et al., 2011 ) Folk e et al. (2010) illustrate that aspects of social -ecological systems inform this definition of resilience, namely persistence, adaptability, and transformability. Adaptability refers to any social -ecological system's capacity to adjust to or respond to changes in both exogenous drivers and endogenous processes and remain within its current stability domain. Transformability, however, is the capacity to create new stability regimes once critical thresholds are crossed. The key to these features is the fundamental property of complex systems to self -organize. Connectivity is a construct used in social -ecological resilience to refer to the strength and structure in which resources, actors, or species interact across geographies, ecosystems , and social domains (Biggs, Schlter, & Schoon, 2015) . It has been theorized that connectivity, operating across multiple spatial and temporal scales, can increase resilience to the provisioning of ecosystem services, system governance, and to fa cilitate recovery after a disturbance or shock (Bodin & Prell, 2011; Col ding & Barthel, 2013; Janssen et al., 2006) . However, it has been observed that highly connected systems increase the potential for disturbances to spread and that densely connected systems lose the ability to adapt and appear "locked" into their current system structure (Bodin & Crona, 2009; Bodin & Prell, 2011; Janssen et al., 2006) . This tension exists because the relationship between resilience and connectivit y is complex and multi -dimensional. Network analysis has been proposed as a method to understand the structure of SES and to assess the level at which connectivity is responsible for specific outcomes. A network approach to SES can provide a way to compa re cases with a topology of network properties relevant to SES. Though empirical studies on the effect of connectivity on SES resilience are increasing in number, they are still quite rare. Bodin 24 and Prell (2011) show how densely connected networks facilit ate the governance of ecological resources . Bodin and Noreberg (2005) show how densely connected networks lower diversity of management strategies increasing risk. Janssen et al. (2006) describe two dimensions of connectivity in its importance to SES gov ernance and resilience. First, they describe connectivity as primarily a function of network density, or the total number of connections in a network divided by the total possible connections a network could potentially have. Reachability or the ability fo r any node to reach another node in a network is the second dimension. Like many network scientists before them, they show how these two dimensions are independent, and it is possible to have high density and low reachability and vice versa. In this stu dy, we explore how extending these dimensions further to include small -world effects can yield insights into a network's ability to foster social learning, innovation, and adaptive capacity. Our analysis will contribute to the growing literature on network s and SES resilience and will further our empirical understanding of how small -world networks can explain resilience outcomes. Small Worlds The concept of small -world networks can be critical to understanding how SES structure is responsible for outcomes essential to resilience. Sometimes referred to as the Small World Effect (SWE), based on the inherent characteristics and outcomes that these structures enable, we argue that it is related to the concepts of adaptive capacity, innovation, and creative pro blem solving all vital to understanding how connectivity operates within SES. Small -world networks are in a class of mathematical models in network science 25 that describe topographical patterns observed in various networks (Uzzi, 2014; Watts & Strogatz, 1998) . The literature on the small -world pheno menon is varied and includes a diverse collection of popular work. Milgram (1967) was the first to study communication chains and discovered that even in extreme geographic and social distances, strangers are connected by no more than six degrees of separa tion (Milgram, 1967). Further described by Watts and Strogatz (1998) as networks that are Òhighly clustered, like regular lattices, yet have small characteristic path lengthsÓ (Watts and Strogatz, 1998) . This provides networks that allow for highly specialized clusters while simultaneously being able to reach or communicate with all parts of the network quickly and efficiently. Small -world properties have been implicated in social capital studies, demonstrating both bonding and bridging capital (Inkpen & Tsang, 2005) . Thus, small -world networks combine structures that support the close group bonding associated with local cooperation and trust, with broad reachability to transmit resources and information throughout the entire network effectively . SWE can be an important indicator of a networkÕs adaptive capacity. Writing about adaptation and adaptive capacities, Eakin et al. (2014) describe adaptation as being contextual and related to the specific capacities needed to act and respond to increa sing vulnerability and climate risks. Eakin stresses that adaptation is necessary to Òmanage environmental variabilityÓ and that these actions are taken in the pursuit of meeting and enhancing human needs, speaking to the foresight, and intentionality of t he actions taken (Eakin, Lemos, & Nelson, 2014) . Here the capacities necessary for adaptati on reflect the conditions that promote and reflect learning, experimentation and encourage innovative solutions (Berkes & Folke, 1998; Walker et al., 2002) . 26 Stak eholder Mapping Eckert and Vojnovic (2017) demonstrate how the path of many Midwest U.S. cities, disinvestment and decline, can lead to spatial disparities in characteristics of food system outcomes. They explain that these spatial disparities in cities like Flint, MI are associated with behavioral preferences, the availability of consumer choices, and the relative distance consumers travel for meals (Eckert & Vojnovic, 2017) . However, residents and stakeholders of Flint, MI, exist in a dynamic culture of activism and experimentation around reimagining their food systems through collaborative community -based exploration. Like many social -cultural problems, redefining food environments is often considered community -centric systems change. Frameworks on community systems change have demonstrat ed the need for many stakeholders to work towards common goals across scales (Foster -Fishman & Watson, 2012) . Understanding the diversity of these actors and how they are connected within the system can be a first step in org anizing that knowledge towards collective action. Stakeholder mapping is a tool to assess a social -ecological system's features by way of its connections and flows of resources. It falls into a long history of rapid appraisal methods developed to assess current and past conditions when baseline data is not available or too costly or difficult to collect (Chambers, 1994) . Like many rapid appraisal methods, it relies on the experiential knowing of knowledge holder s with unique information about the system. These methods and tools are designed to be deployed in the field settings, with low technology, and to capture accurate information. The technique also draws from Cognitive Social Structures (CSS), a network scie nce approach that usually prompts individuals to describe their e go-networks and inform about perceived relationships between other actors (Brands, 2013) . 27 Stakeholder m apping (SM) as a resilience assessment method has been detailed in the Resilience Assessment manual and used in various cases to understand the structure of social -ecological systems ( Resilience Alliance, 2010 ). A modified version of this protocol was deve loped to understand the flow of resources and connectivity in the Flint food security system. This protocol was designed to include food system actors and experts from different sectors of the food system using the conceptualization by (Ericks en, 2008). (SM) exercises were conducted with stakeholder groups representing the Flint food system (Stakeholder Mapping Procedure in Appendix). SM was conducted with ten groups in Flint representing consumers, the supplemental and emergency food system, neighborhood leaders, food processing, governance, and philanthropic organizations. In total, 64 individuals participated in the SM exercises . This is a slight departure from the Ericks en (20 08) conceptualization but focuses on what community stakeholders view as the central components of their localized context. Ericksen (2008) states that food systems Òincorporate multiple and complex environmental, social, political, and economic determinants encompassing availability, access, and utilizationÓ which exist along diff erent temporal, spatial and governance levels (Ericksen 2008, p 234). The decision to focus on the localized context of food security for residents of Flint, is not a reconceptualization of the food system, but focuses on components that are accessed and u nderstood by participants. The SM workshops focused on connections in the food system, as defined as flows of material resources, or information about the food system , specifically aspects of the food system dealing directly with resident food security. Therefore, from here forward we will refer to this network as the food security governance network. Participants were asked to first free -list food system actors on to post -it notes 28 individually. Then the post -it notes were aggregated on a central board for combining redundancies and sorting by theme. Participants were then asked to indicate the resource flows between food system actors, indicated by directed edges. For instance, if participants know that or perceive that a local food pantry receives inf ormation about food need from a specific church group, they would draw a line indicating a flow. The resulting diagrams from the individual workshops were then digitized using Kumu (Kumu, 2020 ) software and then translated into adjacent matrices for furthe r analysis. Translating Stakeholder Maps to Networks Like all network assessment methods, SM is sensitive to missing data, especially the edges that can drastically change the network's topographic characteristics (Wasserman & Faust, 1994) . To address this limitation, we aggregated the individual SM diagrams into a single network, which may address the discrepancies and limited network knowledge of individual workshops , similar to consensus based cognitive social structures (Brands, 2013) . Node standardization procedures were developed t o construct the aggregated network. The primary research aim was to understand the localized food system structure, so we standardized it to the closest meaningful unit for grocery stores, restaurants, and organizational and governance actors. This process was essential to reduce noise in the data while preserving the structural integrity of the network. Figure 1 compares the nonstandardized aggregated network to the standardized aggregated network. What is evident in the sociogram and degree distribution i s that the unstandardized network has many peripheral nodes with a degree of 1. This may be an artifact of the data collection process because consumers receive material resources (food, namely) from many different restaurants and grocery 29 stores. However, participants were less likely to know of connections between individual restaurants or grocery stores. However, they were aware of broader connections, such as a connection from "grocery stores" to the food bank. This example further illustrates the choic e to collapse nodes instead of extending edges to all grocery stores prevents adding many more edges that may overstate an actor's participation in programs or relationships. Figure 3-1 Comparing Degree Dist ributions Across Networks Furthermore, the SM protocol largely had "consumers" start with a central node "resident"/"me"/"consumer". This resulted in a highly centralized node, with a degree count double that of the next highest degree node. This also created a fully connected network through the individual. Though it is important to understand the consumer role in the food system, this created an artifact in the network where consumers lay on "#$!%!#&&'()*+),(!)-,! .,$*,,!.#()*#/')#01! +1.!()*'2)'*,!03!)-,! 1,)40*5!)-0'$-!)-,! ()+$,(!03!)-,! ()+1.+*.#6+)#01! 7*02,((8! ! 30 the shortest path between organizations, nonprofits, and gov ernance actors and were largely artificial. For instance, though it makes sense to combine instances of "The Food Bank" from different workshops, as the intended meaning of this node is the same, the presence of "resident"/"me"/" consumer" does not represe nt a single entity or node. To address this, we removed the aggregated consumer node from the network, while leaving representations of specific consumer or resident groups deemed vulnerable (ie, seniors, Latin -X, children). This largely allows us to view the network as the social -ecological governance network of the food system, and more closely analyze its structures, clustering, and capacities without distortion created by this artifact. Because this network represents the aggregated diagram of multipl e workshops, we had to deal with parallel edges. Parallel edges or multiple edges between two nodes are treated differently in networks to yield different insights into the properties of a network, depending on the meaning of, or reason for the parallel ed ge (Wasserman & Faust, 1994) . For instance, in a citation network, multiple edges mean multiple citations of a given au thor to another. These can be summed, averaged, or transformed to yield some metric of the strength of a particular edge. In aggregated SMs, parallel edges indicate that a particular edge was identified in multiple workshops and may give insight into the v alidity of a particular edge. However, establishing this metric of edge validation is beyond the scope of this research but could be a future direction in SM following similar research in CSS in consensus representations (Brands, 2013; Freeman, Romney, & Freeman, 1987; J. W. Neal, 20 08). For our purposes, the minimum edge value is used. It creates a binary value indicating whether an edge is present (1) or not present (0) and provides the minimum threshold for a tie to be represented. 31 Analytical Methods To conduct our analysis, we used the networkx v 2.3 package (Hagberg, Sch ult, & Swart, 2008) in the Python 3. 7 computing environment (Van Rossum & Drake , 2009) to calculate all network measures. Our analysis is mostly descriptive and focuses on examining networks' structural characteristics to understand connectivity and SES resilience. Janssen et al. ( 2006) describe three important network metrics for understanding a system's social -ecological resilience. We have calculated these metrics using the following formulas. We calculated the following descriptive statistics for the network: radius, diameter, density, degree centralization, average clustering coefficient, and average path length. We calculated standard centrality scores (degree centrality, betweenness centrality, eigenvector centrality, and closeness centrality), whi ch will enable us to consider localized properties of the network (Wasserman & Faust, 1994) . Density is the most straig htforward measure of connectivity and represents the fraction of observed ties over the maximum number of possible ties (Wasserman & Faust, 1994) . The average clustering coefficient is a measure of triadic closure. It is calculated by averaging the local clustering of each node and the fraction of that node's connected neighbors (Wasserman & Faust, 1994) . Average path -length is the mean number of edges on the geodesic path between any two nodes in the network. The diameter of a network is the maximum geodesic distance in the network. It gives the number of steps that are sufficient to go from one node to any other node. A small diameter means that it is possible to traverse the entire network in only a few steps. There are multiple ways t o compute small world quotients provided in the literature (Z. P. Neal, 2015) . For computing the small -world quotient we used Omega as 32 it appropriately compares the clustering coefficient to a lattice -based reference and the mean path length against a rando m graph reference. It offers a fixed scale for comparison across other networks (Watts & Strogatz, 1998) . !"#$#%&&' C and L are the average clustering coefficient and average shortest path length of the network. Lr and Cl are the average shortest path length and average clustering coefficient of an equivalent lattice graph. This coefficient range s between -1 and 1. Values close to 0 mean that the grap features small -world characteristics. Values close to -1 mean that the network has a lattice shape, whereas values close to 1 means the network resembles a random graph. The equivalent networks use d to calculate Lr and Cl were created using networkx reference network generator. The metrics Lr and Cl were sampled from 1 0,000 respective equivalent lattice and random networks. We used the default setting for the rewiring coefficient to be consistent wi th other works (Telesford, Joyce, Hayasaka, Burdette , & Laurienti, 2011) . Results Tab le 2 shows the general descriptive for the SM network. Figure 4 shows the degree distribution of the aggregated, minimum edge value network with residents removed. The food security governance network contains 87 nodes and 174 edges. It has a density of 0.047, radius of 4 and diameter of 6. The observed average path length is 3.060 and is very similar to the expected average path length of 3.107. The observed average clustering of 0.195, lower than the expected average clustering of 0.315. The SMQ is 0.397. As noted earlier, the SMQ is scaled between -1 and 1. A score around 0 indicates a perfect small -world structure. Scores closer to one indicate that the network resembles more of a random structure and scores closer t o -1 a structure similar to a 33 lattice network. The network has small -world features but tends towards the random network side of the scale. Table 3-1 Descriptive Statistics & Small -World Quotient Descriptive Statistics and Small -World Quotient Major Component Nodes 87.00 Edges 174.00 Diameter 6.00 Density 0.05 Radius 4.00 Average Shortest Path 3.06 Average Clustering 0.19 Expected Average Shortest Path 3.11 Expected Average Clustering 0.32 Small World Quotient 0.40 34 Figure 3-2 Social Ecological Governance Network Discussion The SWQ for the network is 0.397, meaning it has some features and characteristics of small -world networks but also tends t owards a random network. We can turn to other descriptive statistics to explain this outcome. The network has a relatively small average path length of around 3 steps and a diameter of 6. This means that the reachability of the network is quite high. Infor mation can travel through this network with relative ease though there are peripheral outliers, it resembles the average path length of the expected random network equivalent. Recalling that the SWQ is a ratio of ratios, it compares the average path length of the network to a random network and "#$8!9!"00.!:,2'*#);!<0=,*1+12,!>,)40*5!'(#1$!30*2,.!(7*#1$!&+;0')8 !>0.,!(#6,!#(!/+(,.!01!*,&+)#=,!/,)4,,11,((!2,1)*+&#);8!?0&0*! #(!@0.'&+*#);! $*0'78! ! 35 compares the average clustering to that of a lattice network. The network has a low average clustering of 0.19 compared to the expected average clustering of an equivalent lattice network, which is 0.315. This low c lustering could be due to the need for the food security governance network actors to act in unison while not specializing in any aspect of the food security system. This clustering is often credited for the development of trust, reciprocity, and specializ ed modes of action. This is not to say that the network does not have clustering, but it is more integrated across the groups which can have positive effects for sharing information and efficiency. The degree distribution can also explain why this networ k tends to have low small -world features. There are highly central actors in the network that most nodes must communicate through or with to take any actions. These high degree actors can be thought of as gatekeepers or boundary spanners. In the case of ga tekeeping, these actors may be inhibiting the development of specialized groups or clusters by controlling the flow of resources making the network highly dependent on the actions of these high degree nodes (Bodin & Prell, 2011) . This is not to discount the small world features of this network which is still agile and able to adapt, innovate, an d change focus to work on specific issues due to the high global efficiency of the network . However, the capacity to do so could be highly reliant on or directed by the most central nodes. The lack of clustering could also be an artifact of how the SM pr otocol was developed. The SM protocol prescribed that workshops be conducted with stakeholders with similar roles in the food security system. The workshops were designed to capture expertise in the localized networks of the specific groups that the partic ipants represented. It was expected that similar stakeholder grouping would provide more accurate information about the group's mental models of the network, but be biased 36 towards their connections and expertise (like in CSS). However, it could be the case that these inter -group connections were underrepresented. These actors may exchange ideas, information, and even collaborate on projects within the food security system, but because they were all in the room together, the ties may have been implied. Fur ther research into the effect of workshop group homogeneity on SM accuracy is needed, especially if the within -group ties lack. In our analysis of community detection, we did not find tight clusters around similar types of actors that would be unexpected. The largest modularity or group contained actors from all sectors of the food security system. Future work could compare workshop diagrams to one another to find if the group composition and diversity affect the accuracy of the modeled network, similar to how CSS has been analyzed in Freeman et al. (2013), Neal (2008) and Brands (2013) (Brands, 2013; Freeman et al., 1987; J. W. Neal, 2008) . Conclusion The purpose of this work was two -fold. One to examine how outcomes of SM can be analyzed quantitatively as networks. Secondly, to demonstrate how the SWQ can be used to understand aspects of SES resilience, namely adaptive capacity and robustness. The resulting aggregate network of the f food security governance network has small -world features but has lower clustering than expected in a small -world graph. We discussed how this graph may be efficient at transmitting information and resources throughout the network, but that there may be a lacking component of trust and intergroup connectivity. Specifically, i nformation can travel through this food security network with relative ease . When faced with a shock this efficient information spread among different 37 actors is vitally important. Different actors are also likely aware of similar information and each otherÕs actions . However, due to the lack of clustering in the net work , trying to coordinate these efforts may be difficult ( Lawlor & Neal, 2016) . Furthermore, consensus on actions and other governance decisions may be costly to implement (Chavis, 2001) . This c ould lead to disagreement about how to respond to shocks or even redundant, uncoordinated actions (J. Lawlor, Metta, & Neal, 2020) . This food security governance network would likely benefit from a network intervention in the form of a coordinating cou ncil, food policy council, or coalition effort to build intergroup connections and trust (Chavis, 2001; Schiff, 2008) . These types of entities can assist in distributing power and build coordinating capacity (Harper, Shattuck, Holt -Gim”nez, Alkon, & Lambrick, 2009) . Methodologically this paper contributes a novel way to compare the adaptive capacity of SES network structure s going further than prior works (Bodin, Ramirez -Sanchez, Ernstson, & Prell, 2011; Janssen et al., 2006) . It also contributes a process of using a mixe d-methods, rapid approach to collecting and analyzing network structure in complex social -ecological settings. 38 Chapter 4 !Translating Narratives into Archetypes Introduction The topic of using system archetypes in the modeling process has been raised repeatedly in the system dynamics literature. Many authors have pointed to the communication and implementation stages of the modeling process in their evaluation of archetypes' usefulness. However, few works detail the exact process used for identifying the archetypes . In this paper, we examine explicitly how archetypes can be identified in qualitative data and how they can be used as a boundary concept to translate modeling concepts. We then extend on Wolstenholme's (2003) work on archetypes and Kim and Anderson's (19 98) framework for analyzing qualitative data in model building by demonstrating a qualitative coding schema adapted from Wolstenholme's definition of totally generic systems archetypes (Wolstenholme, 2003) . We demonstrate the usefulness of this process by using focus group data that was designed to elicit future visio ns of the food system. We illustrate how this process can retain the narrative form that the data originated while being useful enough to provide generic modeling structures to the modeler. By increasing the narrative's connectedness to the model, we will show how this can enhance the modeling process and give specific insights into systems thinking pedagogy and practice. Archetypes Archetypes are system structures that produce characteristic patterns of behavior. System archetypes are a well -documented tool for communicating the structure and behavior of systems and have been applied across various contexts (Kim & Anderson, 39 1998; Senge, 2006) They are useful both as a communication heuristic and as an initial step towards building a model that reflects a system of interest. Kim and Anderson (1998) describe system archetypes as recurring narratives or stories that help build an understanding of system structure by being attuned to systems' behavio r over time. Like many in the field of system dynamics (Newell, 2012; Senge, 2006; E. Wolstenholme, 2004), Kim and Anderson (1998) find that archetypal structures promote systems thinkin g by creating a communicative environment to express intuitive observations of familiar systems. Newell (2012) points out the value of metaphors in establishing shared understanding. He argues that metaphors must be easily understood across various knowl edge backgrounds and that system archetypes can be a particularly powerful metaphor because they are simple, easily understood, and provide relevant representations of systems. This is critical when communicating in a community context around systems and s ystem behavior, for example, when engaged in community -based modeling. There have been divergences in the systems literature on what constitutes a systems archetype (Lane & Smart, 1996; Paich, 1985) and how many genuinely exist (Senge 1990, Wolstenholme and Coyle, 1983; Kim, 1992). Meadows (2008) building on early work of Forrester (1968), Goodman, Kemeny, and Roberts (1994) and Senge (1990) present eight archetypes for learning systems thinking. These eight referred to as semi -generic archetypes by Wolstenholme (2003) and here throughout, are quite descriptive in the problem space one might observe and how stakeholders may experience an archetype but are arguably imprecise in their description of the underlying system structure. Writing on the importance of boundary setting, Wolstenholme (2003) identified 40 four generic two -loop archetypes (Underachievement, Out of Control, Relative Achievement, Relative Control) to address complex intra and inter -organizational challenges. Wolst enholme argues that these four archetypes represent the truly generic structures that capture the system's observed dynamic behavior. These generic two -loop archetypes build off of the isometric properties of feedback loop polarity and demonstrate how two feedback loops in different combinations can create different behavior. Wolstenholme provides the fundamental characteristics of a two -loop archetype. First, it is composed of an intended consequence (ic) feedback loop representing the initial action of an organization or group. Secondly, it contains an unintended consequence (uc) feedback loop resulting from the reaction from within/or outside the organization. Thirdly, it contains a delay before the uc manifests or is known. Furthermore, that there are or ganizational boundaries that mask the uc from actors initiating the ic action. These characteristics allow for a precise description of the structure of an archetype. They can help identify the proper archetypes and solutions that may be useful in addres sing the problem. In the following sections, we will show how we adapted the Wolstenholme Generic Archetype Criteria to create a qualitative data analysis scheme to identify archetypes in qualitative focus group data. Then we will compare the Wolstenholme Generic Archetypes to the semi -generic archetypes as a framework for extracting and analyzing qualitative data. We will then demonstrate how the process can retain the narrative form of the collected data and aid in multiple steps of the modeling process. 41 Qualitative Research and SD The foundational literature on creating system dynamics models has stressed the iterative processes necessary to create, test, and evaluate models. Part of the iterative process of building models has been the conversion of of ten rich qualitative data into numerical models that can be used for decision support through the use of simulation modeling. Often this rich qualitative data has been described as living within the minds of system experts or managers, often referred to as mental models. These mental models have been recognized as a vital source of system information. Forrester, Sterman, and Vennix all discuss the importance of capturing these expert mental models in different modeling stages (Forrester, 1991; J.D. Sterman, 2001; Vennix, 1999). However, guidance on how to analyze and interpret these data has lacked in the foundational literature. In the next sections of this paper, I will review the literature that explains the integration of qualitative data into system dynamics models. Luna -Reyes and Andersen (2003) establish that there is an agreement in the field of system dynami cs that qualitative data is vital to the development of models. They stress that the field lacks rich documentation of how these processes should be integrated, obtained, and analyzed when used to build quantitative models (Luna (Reyes & Andersen, 2003) . They claim that this creates a gap between the problem modeled and the model of the problem. They document that meta -physical variables are challenging to measure and create difficulty integrating them into qua ntified models. They believe that the development of qualitative system dynamics practice (Wolstenholme 1990) is a reaction to this difficulty and a desire to preserve the integrity of these data. Without engaging in the debate, they argue that understandi ng qualitative social science could enhance the modeling process across all stages. 42 Luna -Reyes and Andersen (2003) outline the main areas of qualitative social science methods and illustrate how these methods can be used in system dynamics modeling. They then turn to grounded theory to collect, extract, and analyze qualitative data to build, conceptualize, and formulate model representations. Turner, Kim, and Anderson (2013) demonstrate using grounded theory and textual data to create shared representat ions of a groupÕs mental model by analyzing purposive focus group data to create diagrams of the system in question. Kim and Anderson (2012) use grounded theory and purposive text data to demonstrate a technique to map mental models as causal loop diagrams . Building of this prior work Eker and Zimmerman (2016) introduce an approach that synthesizes qualitative techniques with a focus on causal relationships, creating simplified maps, and maintains links to the data and causal map choices. All of these sch olars have also stressed how costly and time -intensive the process can be. They use grounded theory as a way to build a theory with the data. One aspect of qualitative data that has not been demonstrated in SD processes well, but is likely used, is the use of directed qualitative coding. This involves using a theoretical framework from the beginning, as opposed to the in vivo method of grounded theory. This can allow coders to direct their attention and efforts to understand generic structures in the narrat ives of experts, and code across observation to aggregate and merge an understanding of the dynamics at play. In this paper, we will demonstrate the use of a directed coding procedure that focuses on the generic structures of systems, or system archetype s, to analyze data in the conceptual and formalization of the model. We will first review the system archetype literature and show how archetypes can bridge the gap from purely qualitative representations to rapid prototyping of formalized system dynamics models. 43 Data Collection The analysis for this paper builds on existing system dynamics research using qualitative data to construct models and maps (Luna -Reyes and Anderson 2003, Kim, 2007) and extends it by creating a coding scheme that aids in the iden tification of generic feedback structures and system archetypes. The codebook allows for directed qualitative content analysis and the development of the system's theory and models based on the purposive text data from focus groups. The categories for the codebook represent necessary components of a system archetype, but the codes within the categories are generated inductively through the emergence of essential variables and concepts. A food system conceptualization informed the data collection design fro m Ericksen et al. ( 2008). The Visioning Prot ocol is found in the Appendix . The intention was to engage with knowledge holders from a variety of perspectives: Consumers, Producers, Emergency Food Delivery, Philanthropy, and Governance sectors. In total, sev en workshops were conducted with these community knowledge holders, with two workshops dedicated to consumers, two to the emergency food sector, and one each for the philanthropy and governance sectors. Participation in focus groups ranged from 2 participa nts to 10 participants, in all representing 64 participants. The focus groups were designed to follow a visioning protocol, asking participants to share their experiences with the past, present, and future of the food system in their community. Because t he data was collected as a visioning exercise, the data is not explicitly about system dynamics or using known scripts from Group Model Building (Andersen & Richardson, 1997; Hovmand et al., 2012) . However, the visioning protocol does prompt discussion abou t the dynamics of the food system over time. This data is 44 relevant for our task as it demonstrates the strength of the coding process to identify system archetypes to data that is not specifically designed to elicit them. It also provides case examples of when there was not enough data to determine an archetype or causal structure, allowing us to reflect on the development of future facilitation scripts aimed specifically at eliciting archetypical structures from knowledge holders. Data Extraction The foc us group audio was transcribed into verbatim text data. It was first necessary to scope and create a codebook to guide the extraction and initial review of the qualitative data. Working with guidance from Turner, Kim and Anderson (2013) , in the use of soci al science techniques in system dynamics modeling, it is clear the goal of this research process is based on a grounded theory approach, with the intended purpose of constructing causal theories from the data at hand. We also wanted to utilize system arc hetypes as a boundary concept to link the original data to causal models or theories. For this, we utilized the generic structures identified by Wolstenholme (2003). These structures have precise components that guide the modeler and participant in the con struction and identification of the archetype involved in the described dynamic behavior. Wolstenholme (2003) also provides criteria for identifying archetypes, as noted in the previous section. The two coders (first and second authors, respectively) then read and re -read the transcripts to identify instances of dynamic behavior over time. The full quote containing the dynamic behavior was then the unit of analysis that we applied our broad coding scheme. 45 Major Stocks The data were coded for the potenti al of major stocks that may be important to capture in the final model Central Actors System dynamics always include actors and decision -makers. Coders were instructed to extract any information about central actors mentioned or inferred from the data. Behavior In some cases, participants describe the dynamics of important stocks overtime. Behavior was captured as both graphs drawn by the coders and the participant's descriptive language. Sectors and System Boundaries As Wolstenholme discusses, syst em boundaries often mask the effects of unintended consequences in a system. It was important for the data extraction to include system boundaries or sectors when necessary or apparent. Boundaries were only included if it was explicitly mentioned or flagge d for follow up with experts with unique insights into the system in question Structure The coders also needed to infer through careful analysis, what structure led to this outcome. These are represented as dynamic hypothesis/reference modes or archetypes . The coders used the components of Wolstenholme archetypes to guide this extraction. Problem This was lifted directly from the transcript and referred to something in the system that 46 is either not working, underperforming, working well, or in need of el imination. Action Something that actors in the system have tried as a way to remedy the problem Intended Consequences The expected outcome of the action Unintended Consequences The unexpected outcome of the action, often happening with delay or outside of the sectoral boundaries of the major actors conducting the action. Delays Delays could either be explicitly referenced in the text or inferred by the coder from the described system behavior Feedback loops Feedback loops were rarely referenced direc tly in the text but were inferred by the coders when there was language representing feedback. In facilitation or interview informed by a modeling process, we would expect this to be more explicitly represented in the data and more accessible to code/tag/f lag. Then the extracted texts were categorized into both a Wolstenholme and semi -generic archetype. Though in Wolstenholme (2003), it is shown that semi -generic archetypes map on to Wolstenholme generic archetypes, it was of interest for testing of commu nication strategies to identify both sets of characterizations. Archetypes were identified using Wolstenholme's characterization and by analyzing the cases across the extracted and inferred data. For instance, if the extracted data revealed a Problem of growth of a sector, with an Action of increased investment, and Intended Consequence of more services provided, and an Unintended Consequence of increased complexity 47 and difficulty coordinating services, the archetype was identified as "Underachievement" as the generic and "Limits to Success" as the semi -generic archetype , as shown in Figure 1. Figure 4-1 Comparing Generic and Semi -generic Archetypes Finally, the extracted data were classified into categorie s of "It Worked," "It Did Not Work," and "Not an Archetype." These categories specifically refer to whether or not the archetype identification process was successful for these individual cases. These determinations were primarily based on there being eno ugh data in the quote, or within the context of the focus group transcript to determine the archetype category. This final classification will allow for further analysis of what type of data was missing and how to design future workshops with archetype dat a in mind. Each coder independently read and extracted data and met to discuss each instance of extracted text to gain consensus on the classification and categorical data. The open coding process within the categories was then classified into categories that fit 48 the scale of a regional food system. For instance, in Actors, specific grocery stores or organizations were converted into standardized versions based on their scale of influence. Descriptive Results This section outlines the results of the codi ng process and the archetype analysis. It discusses the different archetypes identified in the data, gives examples of how the data was structured, and the archetypes identified. There were over seven hours of transcribed audio. From those transcripts, 2 08 instances of dynamic behavior were identified and extracted for analysis. They deal with 70 stocks, 35 types of actors and contain examples of all generic and semi -generic archetypes. We were able to identify semi -generic archetypes for all instances an d fully generic archetypes for 179. Table 3 illustrates the general descriptive results of the extracted data. Table 4-1 Archetypes Descriptive Table Archetypes Descriptive Table Classification Count Semi Ge neric Archetypes Almost Worked 17 Out of Control 55 It Worked! 184 Fixes That Fail 31 Not Archetype 2 Shifting the Burden 12 Grand Total 203 Seeking the Wrong Goal 8 Eroding Goals 1 Generic Archetypes Squeaky Wheel Gets 1 Out of Control 55 Accidental Foes 1 Relative Achievement 8 Rule Beating 1 Relative Control 12 Relative Achievement 7 Underachievement 93 Success to the Successful 7 Unknown 17 Relative Control 12 Eroding Goals 11 49 Table 4 -1 (contÕd) Semi Generic Archetypes Escalation 1 Limits to Growth 97 Underachievement 93 Fixes That Fail 32 Limits to Growth 93 Shifting the Burden 12 Unknown 17 Eroding Goals 12 Rule Beating 7 Seeking the Wrong Goal 11 Limits to Growth 4 Success to the Successful 8 Seeking the Wrong Goal 3 Table 4 -1 (contÕd) Rule Beating 8 Fixes That Fail 1 Squeaky Wheel Gets the Grease 2 Success to the Successful 1 Accidental Foes 1 Squeaky Wheel Gets the Grease 1 Escalation 1 Table 4-2 Actors and Stocks by Frequency Actors and Stocks by Frequency Top 10 Stocks Frequency Top 10 Actors Frequency Capacity to Act 15 Food Pantries 33 Food Quality 15 Consumers 26 Knowledge of Food 12 Vulnerable Peop le 22 Funding 9 Grocery Stores 13 Capacity to Collaborate 9 Flint Residents 13 Time 7 Non -Profit Organizations 10 Capacity to Provide Services 7 Farms/Nonprofit 10 Social Capital 6 Farmers Market 5 Food Prices 6 Gardeners 5 Food Waste 6 Corner stor es 4 Total 92 Total 141 Examples ! Though it is beyond the scope of this paper to present the context -specific findings and uncovered themes, below, we provide examples of each generic archetype. We demonstrate how the purposive text was translated into important stocks, actors, actions, intended consequences ( ic) and unintended consequences ( uc), and how they 50 were used to determine the appropriate generic archetype structure. The context for the exercise was focused on various stakeholders in a Midwest community and their perspectives on the past, present, and future of their food system. Underachievement Wolstenholme describes the Underachievement archetype as having a composition of a reinforcing ic loop and a balancing uc loop with delay. In these i nstances, the ic is trying to achieve a successful outcome from the action but is dampened by the result of a resource constraint, or a balancing uc loop. In our analysis, this was the most prevalent identified archetype. For semi -generic archetypes, we id entified limits to growth, limits to success, growth and underinvestment , and fixes that fail (see Discussion). 51 Figure 4-2 Generic Archetypes Underachievement An example of Underachievement can be found in the following quote of a participant describing the growth of the local food system being restricted by its own rising complexity: "Our state has a lot of associations and networks and groups like that, since our community is a key community in the stat e, they are part of these networks of information and resources, food bank networks, community action network, statewide organizations, really provided help and best practices... more now. They are everywhere and it begins to get to the point we can't even act as one" 52 Another related example that we found repeatedly is illustrated in the following quote with a participant describing the increased complexity of the food system makes continued coordination and possibly management and success more difficult t o define or achieve: "É when I first started here, it was simpler, there is an oversaturation of things that are happening that makes coordination difficult fo r us in organizations and difficult for consumers, there feels like there is a lot going on, I do n't want to say too much, but you know we probably lose sight of where we are going". In these examples, it is evident that the dynamics are playing out on different scales, and it is necessary to think carefully about system boundaries. At the most indiv idualized level, we can see that the major actors are individual organizations or consumers. The major stock is the count of different organizations in the system. The dynamics at play are that the food system is growing, offering more services and choices , creating the ic loop. The uc loop plays out on the individual level, where actors find it more difficult to navigate these resources or collaborate with other organizations in which connections could be made. As it becomes more challenging to navigate or collaborate in this system, it happens less so, limiting the system's growth. If this happened without delay, the system would likely tend towards an equilibrium size or complexity, but because delays are likely present, oscillatory patterns are likely. Out -of-control Wolstenholme describes the Out -of-control archetype as having a composition of a balancing ic loop and a reinforcing uc loop with delay. In these instances, the ic is trying to control the extent of a problem through an action. However, th is creates a reaction (possibly from another sector), resulting in a worsening of the problem causing the problem symptom to become more and more out of control. 53 Figure 4-3 Generic Archetype: Out of Control In our analysis, this was the second most prevalently identified archetype. For semi -generic archetypes, we identified fixes that fail, shifting the burden and accidental adversaries. An example of Out -of-Control can be found in the following quotes o f two participants describing the problem of food insecurity with respect to free food distributions: 54 First Person: "Another thing i hear a lot from, two agencies in particular, but this is more of a general thing that doesn't work well, but there is ther e is a fact that there is so much free food distribution means that people are less likely to support urban agriculture with dollars. If people expect it to be, maybe not expect, but if you have a free source of produce you are less likely to purchase it and these urban agriculture folks need to eat too" Second Person: "We have seen that with the numbers and the people applying for food assistance, declining and changing, because of the reaction to the increase of free food. And then there is the concern wh at if that goes away, or if it goes way, then what? so finding different avenues to getting people signed up for these assistance programs even though it may look like they don't need it right now". The main stock in the above example is Food Security, wi th actors being food pantries and urban agriculture entrepreneurs. The action is food distributions at food pantries, which creates the ic loop, balancing the level of food insecurity. The uc loop is played out in increased dependency, and the decrease of willingness to pay for urban agriculture products as the distributions crowd out the market. Relative Control Wolstenholme describes the Relative Control archetype as having a composition of two balancing feedback loops, both the ic and the uc loops. The ic consists of a loop with an action intended to control a relative outcome. However, this action signals to another sector or part of the system to compromise the outcome of the action. In this archetype delays can be present on both loops or only one. I n our analysis this was the third most prevalently identified archetype. For semi -generic archetypes, we identified eroding goals and escalation. 55 Figure 4-4 Generic Archetype: Relative Control An example of Relative Control can be found in the following quote with a participant discussing the quality of food found at grocery stores in his community: "I would say that when Save -A-Lot came, the quality went down all around the board. Like, I feel like becaus e Save -a-lot was here, and they had this low -quality product, and the people bought it, so the rest of the stores started doing it too. Everything started lowering quality. Then you had all of this "great value" here, and "great value" there. Next thing yo u know, and it is lower and lower quality and a steeper price, and it is actually not good." 56 The main stock of interest is Food Quality, with actors being Grocery Stores and Budget Grocers. As is described, the action or ic loop is created by Save -a-Lot, a budget grocer, who enters the market with what the participant describes as lower quality food. This creates a uc loop in the rest of the Grocery Stores, which lowers their quality, presumably staying competitive on prices. The participant describes an o verall decrease of quality over time and multiple sectors of the food system trying to achieve relative control of the market. This example also illustrates how the same two -loop structure can explain escalation and Drifting Goal dynamics. Escalation betwe en the competing grocers and Drifting Goals with the slow shift of acceptable quality foods in the community. This dynamic happens when the acceptable quality is compared to recent memory, allowing quality to decrease, with a lowering of standards incremen tally. Relative Achievement Wolstenholme describes the Relative Achievement as having a composition of two reinforcing feedback loops, both the ic and uc loops. The ic consists of a loop with an action intended to achieve a relative advantage from an ac tion. However, this action and resulting achievement is at the expense of other sectors or parts of the system. Here the ic loop magnifies the relative outcome in a zero -sum game. In our analysis, this was the Fourth most prevalently identified archetype. For semi -generic archetypes, we identified success to the successful. 57 Figure 4-5 Generic Archetype: Relative Achievement An example of how Relative Achievement was identified in our data is found in the quote below with a participant describing the dynamics of the farmers market in their community. They describe how prepared food vendors attract more attention, have higher -priced products, and can pay for multiple spots. They discuss how this has changed the farmers market to be more of a prepared food vending area than a market geared towards the direct sale of produce. 58 "Spots at farmers market used to be cheap, food trucks driving up booth price, driving farmers away. Less about people buying fresh pr oduceÉ.Right now its more people making better, when it was outside it was better, now it is much more commercialized. You now have more restaurants... it went commercial and got more expensiveÉright now you have BBQ trucks that are taking up three spots t hat a farmer can't afford to pay. You go down there right now, they have a perfect spot, that takes up three spots with a BBQ truck., selling they BBQ which sell for a high price. And it have nothing to do with no fruits or vegetables, but they have the mo ney to pay it. I have a buddy right now that his and his wife got a spot in there and they started paying about $70 dollars, and right now they are up to around 300 dollars. Éand now it ain't about the food, it is about the money, it ain't about the vegeta bles, it's not about the market. You can take the title of Farmers market, you can take that title off and that place will still surviveÉ " The main stocks in this example are the relative market share of vendors at the farmers market and farmer's access to the market. The actors are Prepared Food Vendors and Fruit and Vegetable Vendors. The ic loop is reinforcing the Prepared Food Vendors' action using their higher profit margins and sales to purchase more spaces in the farmers market at the detriment to the Fruit and Vegetable Vendors, which is captured in the uc loop. Discussion We coded each instance for both the generic and semi -generic archetypes. Comparing the two types of archetypes for their usefulness can inform modeling practice and systems thi nking education. Though it largely depends on what the archetype analysis will be used for and who will be engaged in the process. We found that the generic archetypes were exact and precise in articulating system structure, causing certain problem behavio rs. With this precision, however, came a necessity for higher resolution data that examined the causal structure of the observation. In the majority of instances, the participants described these causal structures with enough detail that we were able to id entify the generic archetype. This precision provided by 59 the generic archetypes also led to unexpected findings related to the semi -generic archetypes. For instance, Fixes That Fail, the semi -generic archetype, was identified in many instances that were ei ther Underachievement, Out of Control, or Relative Achievement. The story -like features of semi -generic archetypes may focus on the narrative and place less emphasis on the structure, something valued by systems thinking. Examining our case of Fixes That F ail, it was often in instances where someone described a situation in which a solution or policy action is taken and but it does not create the intended results. Though this meets the criteria of the codebook for Fixes That Fail, it overlooks what kind of problem they are describing and what sort of feedback loop is dominant over it. It also does not consider why the action failed: due to a limit of another resource or stock (similar to a Limits to Growth), or because of unintended consequences, or path dep endent behavior. This example illustrates the benefits of the generic archetypes and the codebook approach. However, the generic archetypes require a more advanced understanding of feedback loops, the behavior they cause, how they interact, and how feedb ack dominance operates. Largely, the text data provided examples of generic archetypes with rich enough descriptions to identify generic archetypes and provide a starting point for model conceptualization. Our success in this process could be due to the structure of the visioning protocol to think about the past, present, and future. This is consistent with other system dynamics literature around using visioning and envisioning processes as an effective technique in modeling (Van den Belt, 2004) . Considering the instances where the data was insufficient to identify a generic archetype, it is not to say that it is not useful in the modeling process. Broadly, these were instances of reference modes or behavior overtime that lacked a causal explanation for what was happening. These can 60 be referenced again during the model ing process as touchstones for places to seek more information. However, it should be noted that best practices in facilitation could make this data more useful to the modeling process. Namely, facilitating a process that seeks causal explanations for des cribed behavior with a caveat that ÒI donÕt knowÓ is acceptable. In our examples, it was unclear if there was an implied causal explanation, or if the participant knew why something happened, if they were unsure, or lacked any knowledge within their role t o answer because they were not asked. More empirical analysis is required on the use of archetypes in systems thinking practice. Though our findings suggest that the generic archetypes allow for participants to identify causal structures and feedback loo ps governing their system, the practice of doing so takes sophisticated systems thinking skills. We would hypothesize that semi -generic archetypes are useful in getting new system thinkers to identify when dealing with a complex system and can serve as a w arning or draw attention to a systemÕs complexity. It would be interesting to determine, through the use of systems thinking learning scale, or systems thinking self -efficacy scale, which approach works best for the different intended outcomes. Another a rea where these findings could be useful is developing an archetype script (scriptopedia) for Group Model Building processes. Here, instead of purposive text data, we would want a group of people who are expert knowledge holders to engage in the process of conceptualizing and identifying the archetypes themselves. Here we believe the codebook we presented could be used as a guiding framework for this script. 61 Conclusion In this paper, we demonstrated two things. 1. That system archetypes can be useful in mu ltiple stages of modeling and help retain a link to the original data as abstraction and simplification processes dominant the modeling translation. 2. We demonstrated the utility of using archetypes in a coding schema for analyzing qualitative modeling. This study demonstrates that a systematic approach to identifying generic archetypes can be used for analyzing data and provide consistency throughout the process. Being explicit about the structures the modelers are seeing provides a way to be in constant communication with data, addressing the old adage that modeling is more of an art than a science. Furthermore, these links to the data can aid in client or community -centered modeling by creating a boundary object to connect the physical structure of the model to the data that generated its structure. While the coding process illustrated in this study was precise in thinking about generic causal structures in the data and provided insights over the use of semi -generic archetypes, it was not without its c hallenges. Like any qualitative analysis, it was labor -intensive and took many iterations of analysis to determine the archetypes and relevant units of analysis. Developing future work to understand system boundaries, especially given their importance in m asking the unintended consequences, is needed. Though this is not unique to the purposive text as boundary work in many aspects of system dynamics has been overlooked, and further research could yield important findings. The coding process also needs to b e replicated to collectively test its usefulness in identifying and mapping structures operating in systems. Findings from these efforts could inform not only the formal modeling process but build on the pedagogy of systems thinking, and client and communi ty-centered practice. 62 Chapter 5 !Conclusions Introduction This dissertation included three studies that examine d interdependence in community issues and contributes to the literature on applying participatory modeling to place -based problems . The unifying theme of interdependence create d a space to think critically about the structures that drive systems in communit ies . This dissertation contributes to the field of sustainability science, defined as place -based problem -oriented inquiry that focuses on applying knowl edge to action (Miller, 2013) . This dissertation shows how s ustainability science can interface with c omplex system features of interdependence, path dependent time effects (Meadows, 2002) , with participatory research which requires an appreciation for community held, experiential knowledge (Greenwood & Levin, 2006; Whyte, Greenwood, & Lazes, 1991) . The primary goal of community -based sustainability research is to understand ende mic sustainability problems with community partners (Deakin & Reid, 2014) . It is to integrate a diversity of information in a knowledge creation p rocess based on reciprocal learning (Wittmayer & Sch−pke, 2014) . As this dissertation demonstrates, c ombining a participatory method with complex systems research requires an interdisciplinary methodology , that interfac es with network science, informatics, participatory inquiry and mathematical modeling. This dissertation makes methodological contributions in the integration of different types of data to understand community problems. The studies in Chapters 3 and 4 were largely completed during the 2020 Covid 19 pandemi c. T he resulting restrictions to research and data collecti on required adapting and using different kinds 63 of data in combination to understand the research questions. Though this research context was unique , it reflects the changing environment in which community -based research is often conducted , including accelerated timeline s which rarely accommodate lengthy data collection and analysis process es (Chambers, 1994) . By integrating community knowledge with commonl y used modeling modes, we avoid ed an extensive data collection and model formulation process which may have prevent ed the use of systems methods to address these community problems. As show n in Chapter 3, data collected primarily to understand important stakeholder s in the inclusion in the broader project design, was used in a new way and treated as a representation of the network structure in a community. This study used the resulting structures as a object to understand aspects of the food security syste mÕs connectivity and resilience. For the first time in community -based research , the small -world -quotient was used to examine the structural capacity for simultaneous global and local efficiency. This demonstrates the utility and necessity of relooking at data in different ways and from a plurality of methods. This dissertation also furthers our understanding of how to research these types of problems in ways that are explicit, open to communicative opportunities, and links any and all modeling efforts to the data of lived experience. In Chapter 4, a process for explicit translation of community narratives into generic feedback structures was demonstrated . This study utilized secondary data that was originally collected as part of a food system visioning wo rkshop. This shows how different kinds of data, can inform modelers and researchers of the underlying structures driving systems. This chapter also illustrate d how necessary it is to be transparent about claims made with community held data, and how slight discrepancies in meaning can lead to vastly different interpretations of the causal structure behind a particular system level 64 outcome. It provides a case study on the importan ce of being explicit about structure in understanding the causal mechanisms beh ind what participants are experiencing. In Chapter 2, the field of participatory modeling is mapped to show its characteristics and subfields that approach participation and modeling differently. By understanding the connections and communities of schola rs in this work, the emergence of separate but related research fronts was revealed. These research fronts , examined as networks , explain s how practitioner s approach both participation and modeling . At the broadest levels, there are practitioners in the fi eld that use PM as a tool to enhance participatory processes. This subfield addresses the challenges of shared understanding and conflict with different modeling tools. This is contrasted with the how others in the field of PM use participation and collect ive intelligence as an asset in creating better, more complete, models of contested systems. This partition of the field is defin ed in the approach and purpose of PM. Though there can be some significant overlap in how modelers approach their work, these f undamental goals should be clear to participants and those evaluating the modeling effort. These chapters together point to future research directions in the area of applied systems science and participatory modeling. What remains uncertain across these studies is how participants (or community members ) respond to the modeling process and its intermediate and end products . An evaluation of how PM accomplishes goals for participants is necessary. One area of limited understanding across all threads of PM that should be expanded on is the area of systems thinking self -efficacy. As shown in this dissertation, there are many modes of model building with participants. Often a modeler engages with community partners to tackle a complex of wicked problem. It is unclear, however, how participants in these various settings engage with the modeling tools and if it increases their perceived ability to act as problem owners. 65 In the case study in Chapter 3 on stakeholder mapping , it remains unclear how participants wou ld engage with and understand the network maps. This leads to open questions for stakeholder mapping and diagraming: Are these diagrams useful for individual actors and how the collaborate? Are structural holes evident to participants? How much training is necessary for participants to glean actionable information from these diagrams? Chapter 4 took a different approach to describing system dynamics as archetypes and narratives. It remains unclear how accessible the fully generic archetypes are for communi ty members, and if this approach enhances their understanding or is confusing. Future research focused on assessing the utility of this approach for participants is necessary, but this preliminary work demonstrates the near universality of structure in com plex community problems. 66 APPENDICES 67 APPENDIX A Stakeholder Mapping Protocol 2019 0206 Flint LP stakeholder mapping protocol Who are the important people/groups in Flint, Beecher and Burton Food System (including those based outside but that affect the food system)? Yellow post -it notes Note: May have to clump people/organizations to be feasible given the scale i.e. food pantries, churches. Note: Identify whether if outside Flint on post -it with (E) How are these groups linked to each other ? Flow of finance ($) = red Flow of information = blue What direction is the relationship? 68 APPENDIX B Archetype Codebook !Fully Generic System Archetype Codebook Code Description Major Stocks A stock is a variable of interest that can increase or dec rease. These are usually discussed as important quantities or qualities of the system. Major Actors Actors can be types of individuals, organizations, or agencies that are involved in the system (explicitly or implicitly) Problem referred to something i n the system that is either not working, underperforming, working well, or in need of elimination. Action Something that actors in the system have tried as a way to remedy the problem Intended Consequences Intended outcome of an action. Expressed as a f eedback loop Unintended Consequences The unexpected outcome of the action, often happening with delay or outside of the sectoral boundaries of the major actors conducting the action. Delay Delays could either be explicitly referenced in the text or infe rred by the coder from the described system behavior Food System Outcome This describes what primary food system outcome the extracted data is referring to: Access, Availability, Utilization, Social Welfare Environmental Capital 69 REFERENCES ! 70 REFERENCES Andersen, D. F., & Richardson, G. P. (1997). Scripts for group model building. System Dynamics Review , 13(2), 107 Ð129. https://doi.org/10.1002/(sici)1099 -1727(199722)13:2<107::aid -sdr120>3.0.co;2 -7 Barreteau, O., Bots, P. W. G., & Daniell, K. A. (2009). A framework for clarifying ÒparticipationÓ in participatory research to prevent rejection of participation for bad reasons. Ecology And Society , 15(3), 1 Ð29. https://doi.org/http://dx.doi.org/10.332 9/jesnr.v6i2.22113 Berkes, F., & Folke, C. (1998). Linking social and ecological systems for resilience and sustainability. Linking Social and Ecological Systems , 1, 13 Ð20. Biggs, R., Schlter, M., & Schoon, M. L. (2015). Principles for building resilien ce: Sustaining ecosystem services in social -ecological systems . Principles for Building Resilience: Sustaining Ecosystem Services in Social -Ecological Systems . https://doi.org/10.1017/CBO9781316014240 Bodin, –., & Crona, B. I. (2009). The role of social n etworks in natural resource governance: What relational patterns make a difference? Global Environmental Change , 19(3), 366 Ð374. https://doi.org/10.1016/j.gloenvcha.2009.05.002 Bodin, –., & Prell, C. (2011). Social networks and natural resource management : uncovering the social fabric of environmental governance . Cambridge University Press. Bodin, –., Ramirez -Sanchez, S., Ernstson, H., & Prell, C. (2011). A social relational approach to natural resource governance. In Social networks and natural resource management !: uncovering the social fabric of environmental governance (1st ed., pp. 3 Ð28). https://doi.org/10.1017/CBO9780511894985 Boyack, K. W. (2010). Co -Citation Analysis , Bibliographic Coupling , and Direct Citation !: Which Citation Approach Represe nts the Research Front Most Accurately !?, 61(12), 2389 Ð2404. https://doi.org/10.1002/asi Brands, R. A. (2013). Cognitive social structures in social network research: A review. Journal of Organizational Behavior , 34(S1), S82 ÐS103. Chambers, R. (1994). Pa rticipatory rural apraissal: analysis of experience. World Development , 22(9), 1253 Ð1268. https://doi.org/10.1016/0305 -750X(94)90003 -5 Chavis, D. M. (2001). The paradoxes and promise of community coalitions. American Journal of Community Psychology , 29(2) , 309 Ð320. Colding, J., & Barthel, S. (2013). The potential of ÒUrban Green CommonsÓ in the resilience building of cities. Ecological Economics , 86, 156 Ð166. https://doi.org/10.1016/j.ecolecon.2012.10.016 71 Deakin, M., & Reid, A. (2014). Sustainable urban d evelopment: Use of the environmental assessment methods. Sustainable Cities and Society , 10, 39 Ð48. https://doi.org/10.1016/j.scs.2013.04.002 Eakin, H. C., Lemos, M. C., & Nelson, D. R. (2014). Differentiating capacities as a means to sustainable climate change adaptation. Global Environmental Change , 27(1), 1 Ð8. https://doi.org/10.1016/j.gloenvcha.2014.04.013 Eckert, J., & Vojnovic, I. (2017). Fast food landscapes: Exploring restaurant choice and travel behavior for residents living in lower eastside Det roit neighborhoods. Applied Geography , 89(September), 41 Ð51. https://doi.org/10.1016/j.apgeog.2017.09.011 Eskinasi, M., Rouwette, E., & Vennix, J. (2009). Simulating urban transformation in Haaglanden, the Netherlands. System Dynamics Review , 25(Spring), 182Ð206. https://doi.org/10.1002/sdr Forrester, J. W. (1991). System Dynamics and the Lessons of 35 Years by. A SystemsBased Approach to Policymaking , 3(2), 1 Ð35. https://doi.org/10.1002/sdr.4260100211 Fortunato, S. (2010). Community detection in graphs. Physics Reports , 486(3Ð5), 75 Ð174. https://doi.org/10.1016/j.physrep.2009.11.002 Foster -Fishman, P. G., & Watson, E. R. (2012). The ABLe change framework: A conceptual and methodological tool for promoting systems change. American Journal of Community Ps ychology , 49(3Ð4), 503 Ð516. Freeman, L. C., Romney, A. K., & Freeman, S. C. (1987). Cognitive structure and informant accuracy. American Anthropologist , 89(2), 310 Ð325. Girvan, M., & Newman, M. E. J. (2002). Community structure in social and biological networks. Proceedings of the National Academy of Sciences of the United States of America , 99(12), 7821 Ð7826. https://doi.org/10.1073/pnas.122653799 Gray, S., Voinov, A., Paolisso, M., Jordan, R., Bendor, T., Bommel, P., É Zellner, M. (2018). Purpose, proc esses, partnerships, and products: Four Ps to advance participatory socio -environmental modeling: Four. Ecological Applications , 28(1), 46Ð61. https://doi.org/10.1002/eap.1627 Greenwood, D. J., & Levin, M. (2006). Introduction to action research: Social r esearch for social change . SAGE publications. Hagberg, A. A., Schult, D. A., & Swart, P. J. (2008). Exploring Network Structure, Dynamics, and Function using NetworkX. In G. Varoquaux, T. Vaught, & J. Millman (Eds.), Proceedings of the 7th Python in Scien ce Conference (pp. 11 Ð15). Pasadena, CA USA. Harper, A., Shattuck, A., Holt -Gim”nez, E., Alkon, A., & Lambrick, F. (2009). Food policy councils: Lessons learned . Food First/Institute for Food and Development Policy 72 Oakland, CA. Holling, C. S. (1973). RES ILIENCE OF ECOLOGICAL SYS TEMS. Annu.Rev.Ecol.Syst. , 4, 1Ð23. https://doi.org/10.1146/annurev.es.04.110173.000245 Hovmand, P. S., Anderson, D. F., Rouwette, E. A. J. A., Richardson, G. P., Rux, K., & Calhoun, A. (2012). Group Model -Building ÒScriptsÓ as a Collaborative Planning Tool. Systems Research and Behavioral Science , (29), 180 Ð193. https://doi.org/10.1002/sres Inkpen, A. C., & Tsang, E. W. K. (2005). Social capital networks, and knowledge transfer. Academy of Management Review , 30(1), 146 Ð165. https ://doi.org/10.5465/AMR.2005.15281445 Janssen, M., Bodin, –., Anderies, J., Elmqvist, T., Ernstson, H., McAllister, R. R. J., É Ryan, P. (2006). Toward a Network Perspective of the Study of Resilience in Social -Ecological Systems. Ecology and Society , 11(1). https://doi.org/10.5751/ES -01462-110115 Kelly, R. A., Jakeman, A. J., Barreteau, O., Borsuk, M. E., ElSawah, S., Hamilton, S. H., É Voinov, A. A. (2013). Selecting among five common modelling approaches for integrated environmental assessment and manag ement. Environmental Modelling and Software , 47, 159 Ð181. https://doi.org/10.1016/j.envsoft.2013.05.005 Kessler. (1963). Bibliographic Coupling Between Scientific Papers. American Documentation , 14(1), 10 Ð25. Kim, D. H., & Anderson, V. (1998). Systems ar chetype basics. Waltham, Mass, Pegasus Communications Inc . Lane, D. C., & Smart, C. (1996). Reinterpreting Ôgeneric structureÕ: evolution, application and limitations of a concept. System Dynamics Review: The Journal of the System Dynamics Society , 12(2), 87Ð120. Lawlor, J. A., & Neal, Z. P. (2016). Networked Community Change: Understanding Community Systems Change through the Lens of Social Network Analysis. American Journal of Community Psychology , 426 Ð436. https://doi.org/10.1002/ajcp.12052 Lawlor, J. , Metta, K., & Neal, Z. (2020). What is a coalition? A systematic review of coalitions in community psychology. Leydesdorff, L. (1998). Theories of Citation? Scientometrics , 43(1), 5 Ð25. Luna (Reyes, L. F., & Andersen, D. L. (2003). Collecting and analyzing qualitative data for system dynamics: methods and models. System Dynamics Review , 19(4), 271 Ð296. https://doi.org/10.1002/sdr.280 73 Lynam, T., Jong, W. De, Sheil, D., Kusumanto, T., & Evans, K. (2007). A Review of Tools for Incorporating Community Knowledge, Preferences, and Values into Decision Making in Natural Resources Management, 12(1). McLevey, J., & McIlroy -Young, R. (2017). Introducing metaknowledge: Software for computational researc h in information science, network analysis, and science of science. Journal of Informetrics , 11(1), 176 Ð197. https://doi.org/10.1016/j.joi.2016.12.005 Meadows, D. (2002). Dancing with Systems. The Systems Thinker , 13(2), 2 Ð6. Meadows, D. H. (2008). Think ing in Systems: A Primer . (D. Wright, Ed.) (Vol. 89). Susainability Institute. https://doi.org/10.1080/09644016.2011.589585 Milgram, S. (1967). The small world problem. Psychology Today , 2(1), 60 Ð67. Miller, T. R. (2013). Constructing sustainability scie nce: Emerging perspectives and research trajectories. Sustainability Science , 8(2), 279 Ð293. https://doi.org/10.1007/s11625 -012-0180-6 Naivina, W., Le Page, M., Thongoi, M., Trebuil, G. (n.d.). Participatory Agent -Based Modeling and Simulation of Rice Far ming in the Rainfed Lowlands of Northeast Thailand, 1 Ð10. Neal, J. W. (2008). Ò Kracking Ó the Missing Data Problem !: Structures to School -Based Social Networks. Sociology of Education , 81(2), 140 Ð162. Neal, Z. P. (2015). How Small Is It? Comparing Indic es of Network Small Worldliness. PLOS , 48824, 1 Ð12. Newell, B. (2012). Simple models, powerful ideas: Towards effective integrative practice. Global Environmental Change , 22(3), 776 Ð783. Paich, M. (1985). Generic structures. System Dynamics Review , 1(1), 126Ð132. Porter, M. A., Onnela, J. -P., & Mucha, P. J. (2009). Communities in Networks, 56(9). Retrieved from http://arxiv.org/abs/0902.3788 Pritchard, A. (1969). Statistical Bibliography or Bibliometrics. Journal of Documentation , 25(4), 348 Ð349. https: //doi.org/10.1108/eb026542 Richardson, G. P., & Andersen, D. F. (1995). Teamwork in group model building. System Dynamics Review , 11(2), 113 Ð137. https://doi.org/10.1002/sdr.4260110203 Rouwette, E. A. J. A., Vennix, J. A. M., & Van Mullekom, T. (2002). G roup model building effectiveness: A review of assessment studies. System Dynamics Review , 18(1), 5 Ð45. https://doi.org/10.1002/sdr.229 74 Schiff, R. (2008). The role of food policy councils in developing sustainable food systems. Journal of Hunger & Environ mental Nutrition , 3(2Ð3), 206 Ð228. Senge, P. M. (2006). The fifth discipline: The art and practice of the learning organization . Currency. Stave, K. a. (2003). A system dynamics model to facilitate public understanding of water management options in Las Vegas, Nevada. Journal of Environmental Management , 67(4), 303 Ð313. https://doi.org/10.1016/S0301 -4797(02)00205 -0 Sterman, J.D. (2001). System Dynamics Modeling: Tools for Learning in a Complex World. California Management Review , 43(4), 8 Ð25. https://doi. org/10.1111/j.1526 -4637.2011.01127.x Sterman, John D. (1992). System Dynamics Modeling for Project Management. System , 1951, 286 Ð294. https://doi.org/10.1109/SOCA.2007.45 Telesford, Q. K., Joyce, K. E., Hayasaka, S., Burdette, J. H., & Laurienti, P. J. ( 2011). The ubiquity of small -world networks. Brain Connectivity , 1(5), 367 Ð375. Trujillo, C. M., & Long, T. M. (2018). Document co -citation analysis to enhance transdisciplinary research, 1 Ð10. Uzzi, J. S. B. (2014). Collaboration and Creativity !: The Small World Problem. American Journal of Sociology , 111(2), 447 Ð504. van Bruggen, A., Nikolic, I., & Kwakkel, J. (2019). Modeling with Stakeholders for Transformative Change. Sustainability , 11(3), 825. https://doi.org/10.3390/su11030825 Van den B elt, M. (2004). Mediated Modeling: A System Dynamics Approach to Environmental Consensus Building . Island press (Vol. 21). https://doi.org/10.1007/s10980 -005-5569-5 Vennix, J. A. M. (1999). Group model -building: Tackling messy problems. System Dynamics Re view , 15(4), 379 Ð401. https://doi.org/10.1002/(SICI)1099 -1727(199924)15:4<379::AID -SDR179>3.0.CO;2 -E Voinov, A., & Bousquet, F. (2010). Modelling with stakeholders. Environmental Modelling & Software , 25(11), 1268 Ð1281. https://doi.org/10.1016/j.envsoft.2 010.03.007 Voinov, A., Jenni, K., Gray, S., Kolagani, N., Glynn, P. D., Bommel, P., É Smajgl, A. (2018). Tools and methods in participatory modeling: Selecting the right tool for the job. Environmental Modelling and Software , 109(April), 232 Ð255. https:// doi.org/10.1016/j.envsoft.2018.08.028 Voinov, A., Kolagani, N., McCall, M. K., Glynn, P. D., Kragt, M. E., Ostermann, F. O., É Ramu, P. (2016). Modelling with stakeholders - Next generation. Environmental Modelling and Software , 77, 196 Ð220. 75 https://doi.o rg/10.1016/j.envsoft.2015.11.016 Walker, B., Carpenter, S., Anderies, J., Abel, N., Cumming, G., Janssen, M., É Pritchard, R. (2002). Resilience management in social -ecological systems: a working hypothesis for a participatory approach. Conservation Ecolo gy, 6(1). Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications (Vol. 8). Cambridge university press. Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of Ôsmall -worldÕnetworks. Nature , 393(6684), 440 Ð442. Westle y, F., Olsson, P., Folke, C., Homer -Dixon, T., Loorbach, D., Vredenburg, H., É Thompson, J. (2011). Tipping Toward Sustainability: Emerging Pathways of Transformation. Ambio , 40(7), 762 Ð780. https://doi.org/10.1007/s13280 -011-0186-9 Whyte, W. F., Greenwoo d, D. J., & Lazes, P. (1991). Participatory action research: Through practice to science in social research. Participatory Action Research , 19 Ð55. Wittmayer, J. M., & Sch−pke, N. (2014). Action, research and participation: roles of researchers in sustaina bility transitions. Sustainability Science , 9(4), 483 Ð496. Wolstenholme, E. (2004). Using generic system archetypes to support thinking and modelling. System Dynamics Review , 20(4), 341 Ð356. https://doi.org/10.1002/sdr.302 Wolstenholme, E. F. (2003). Tow ards the definition and use of a core set of archetypal structures in system dynamics. System Dynamics Review , 19(1), 7 Ð26. https://doi.org/10.1002/sdr.259 Yan, E. (2012). Scholarly Network Similarities !: How Bibliographic Coupling Networks , Citation Net works , Cocitation Networks , Topical Networks , Coauthorship Networks , and Coword Networks Relate to Each Other, 63(7), 1313 Ð1326. https://doi.org/10.1002/asi