The GMB sessions for the HKHC evaluation were intended to introduce systems thinking at the community level by identifying the essential parts of the system and how the system influences policy and environmental changes to promote healthy eating and active living and to prevent childhood obesity through causal mapping of feedback loops for each of the 49 HKHC CPs. Using an inductive approach, GMB participants identified the essential parts of the system through variables produced during a behavior-over-time graph exercise (see companion article in this supplement1).
The purpose of this article is to describe the methods, results, and implications associated with a synthesis of the causal maps, or causal loop diagrams, for each of the 49 HKHC CPs. Specifically, this article addresses the following evaluation questions:
- What were the most prominent variables in the causal loops diagrams across communities?
- What were the major feedback structures across communities?
- What implications from the synthesized HKHC causal loop diagram can be translated to policy makers, practitioners, evaluators, funders, and other community representatives?
From 2010 to 2014, the evaluation team worked collaboratively with HKHC CPs to design and implement 49 community-based, half-day GMB sessions as part of HKHC site visits as well as to develop customized community reports to support local dissemination efforts (see systems thinking in communities' storybooks, www.transtria.com/hkhc.php).
Evaluators worked with HKHC project directors and project coordinators to host GMB sessions on these visits. The half-day sessions followed a structured protocol involving a sequence of GMB scripts4 , 21 , 22 that are available in the Healthy Kids, Healthy Communities Group Model Building Facilitation Handbook (www.transtria.com/hkhc.php).
The GMB sessions had 2 main activities designed to gain insight into groups' common understanding of the policy, system, and environmental work going on in their community related to healthy eating, active living, and childhood obesity. The first activity was a 60-minute behavior-over-time graph exercise, in which participants individually created and shared graphs of things that affect or are affected by policy, system, and environmental changes in their community using a nominal group technique (ie, all participants described their top-ranked graph, followed by the second-ranked graph, and so on, until all graphs were shared or time ran out). The second activity was a causal loop diagram, or structural elicitation, 60-minute exercise, in which participants collectively shared their perceptions of causal relationships among variables generated from the first exercise to develop a causal loop diagram, or system map, illustrating the community's theory of change.
A wide range of community participants were recruited by HKHC project directors and project coordinators, including residents, elected officials, representatives from government agencies and community-based organizations, businesses, and university-based researchers. Most sessions were conducted in English, with exception of 4 sessions in communities using interpretation and translation services. All sessions were recorded and transcribed to add further context and clarification postsession to the interpretation and analysis of variables and causal relationships identified in the participants' stories.
Following the behavior-over-time graph exercise, facilitators selected approximately 9 to 12 variables to use as “seed” variables to start the causal loop diagraming exercise. Variables were primarily chosen to represent active living or healthy eating policy and environmental strategies for each HKHC CP; in addition, evaluators selected variables to reflect a range of health behaviors or outcomes, partnership or community capacity efforts, or social determinants of health, if these were identified. During training, the GMB modeler-facilitator received instructions on how to code the behavior-over-time graphs (see Hoehner et al1 in this supplement) into the referenced categories and to incorporate the variables receiving the greatest amount of attention or discussion during the session into the causal loop diagram. The final set of variables selected by the modeler was approved by the facilitator and, subsequently, reviewed and approved by the session participants. See Table 1 for examples of seed variables in each category.
The seed variables were written on white board paper posted to a wall prior to the sessions so that participants were able to make modifications to anything written on the paper during the sessions (eg, change to a variable name, addition of new variables). Participants were instructed to identify causal connections among the seed variables or to generate new variables to be added to the white board indicating causal relationships. As participants nominated links, facilitators drew the causal relationships using the conventions of system dynamics (see Table 2 and scripts from the handbook referenced previously) and highlighted simple balancing and reinforcing feedback loops as they emerged. The resulting causal loop diagrams were translated into Vensim software (www.vensim.com) as a product for further refinement and analysis.
Evaluators then reviewed each causal loop diagram against the transcripts to ensure the range of variables and causal relationships generated through the behavior-over-time graph exercise and the causal loop diagram exercise were represented. Evaluators placed an emphasis on making sure the diagrams characterized the mental models as expressed during specific conversations as well as throughout the session. For example, the use of a term may shift during a conversation, with the group coming to an agreement by the end of the session. Therefore, the transcripts were used to identify and resolve any ambiguity. In some cases, participants also nominated links that did not appear in the diagram because the conversation was moving too quickly or there was too much crosstalk. Sometimes the direction of the relationship between variables was mistakenly recorded on the diagram, or the nature of the relationship between variables was misrepresented. This happened when arrows were drawn in the wrong direction or the associated polarities, or positive and negative signs, did not reflect the quantitative relationship between variables. In causal loop diagrams, a plus sign (“+”) from x to y means that as x increases, y increases, and equivalently, as x decreases so does y. Similarly, a minus sign (“−”) from x to y means that as x increases, y decreases, and as x decreases, y increases. See Table 2 for basic terminology and symbols used in the causal loop diagrams. These situations called for modifications to the original causal loop diagram.
After reviewing and cleaning the causal loop diagrams in Vensim, the evaluators identified the feedback loops associated with each CP's primary strategies (ie, partnership and community capacity building as well as healthy eating and active living) and then created systems thinking storybooks for each CP. A total of 50 causal loop diagrams were produced for 49 communities (1 community had 2 causal loop diagrams representing different geographic regions).
To develop the synthesized causal loop diagram across communities, evaluators conducted a content analysis of the variables across all 50 causal loop diagrams. Variable names were independently coded into 5 major subsystems: healthy eating policies and environments, active living policies and environments, partnership and community capacity building, social determinants of health, and health and health behaviors. All variables from the causal loop diagrams were then entered into a database according to these 5 major subsystems in order to identify common variables across communities. While there were a number of variables that appeared in only 1 causal loop diagram, 80% of the variables across the 50 causal loop diagrams were accounted for by variables that appeared in 20% or more of the causal loop diagrams. Thus, evaluators chose to use 20% as the threshold for including a variable in the synthesized causal loop diagram. That is, if a variable appeared in 10 or more causal loop diagrams, it was included in the synthesized causal loop diagram.
The synthesized causal loop diagram was developed by taking the aggregation or union of the links between these variables across the 50 causal loop diagrams. This was initially done in an incremental fashion by adding links from one causal loop diagram to another to yield the union of 2 diagrams and then adding a third causal loop diagram to that for the next iteration until the final diagram represented all the links between the identified variables. This diagram was simplified by focusing on the feedback relationships and then compared against community diagrams to ensure that the synthesized diagram had the capacity to retell the stories from each community.
A total of 590 individuals participated across 49 communities, with an average of 12 participants per session. Table 3 provides session characteristics and selected HKHC CP characteristics by HKHC CP.
A synthesis of all 50 casual loop diagrams is presented in Supplemental Digital Content Figure 1 (available at: http://links.lww.com/JPHMP/A147), reflecting subsystems, feedback structures, and structural elements corresponding to policies, environments, local collaborations, and social determinants that influence healthy eating, active living, and, ultimately, childhood obesity. As illustrated in this figure, the causal loop diagram provides a way to visualize all the elements of the system and their interactions, with a focus on causal relationships as opposed to associations. The causal loop diagram represents a holistic perspective of the system and several subsystems. To digest the depth and complexity of the diagram, it is helpful to examine it in terms of the subsystems of influence, including healthy eating policies and environments (red), active living policies and environments (blue), health and health behaviors (orange), partnership and community capacity (purple), and social determinants (green).
What were the most prominent variables in the causal loop diagrams across communities?
Participants' causal loop diagrams included a total of 2399 variables extracted from the transcripts for the behavior-over-time-graph and causal loop diagram exercises; this represented a total of 227 unique variables across all CPs. Common variables for each major subsystem are identified in this section; other less common variables are included in Supplemental Digital Content Table 1 (available at: http://links.lww.com/JPHMP/A148).
Active living policies and environments
For the active living policies and environments subsystem, a total of 30 different variables were generated, with 14 represented in at least 20% of the CPs' causal loop diagrams (see Table 1). Of the 14 variables, several supported active transportation (eg, access to public transportation, Complete Streets) or recreation (eg, access to parks, access to trails). Some referenced community design and land use (eg, urban sprawl, school siting) or motorized transportation (eg, traffic safety, car dependence). Two broadly referred to policy adoption and enforcement or maintenance of environments, and 1 was specific to school and child care policies and environments for active living.
Healthy eating policies and environments
In the healthy eating policies and environments subsystem, participants identified a total of 48 different variables, with 20 represented in at least 20% of the CPs' causal loop diagrams (see Table 1). Of the 20 variables, some referenced access to healthy or unhealthy foods and beverages generally and then many of the others depicted settings for purchase or consumption of foods and beverages (eg, fast food restaurants, farmers' markets, child care). Several focused on food production settings, such as gardens and farms, yet distinguished the setting (school vs community), scale (small farms vs agribusiness), or approach (organic vs sustainable). A couple referred to the affordability of foods and beverages as well as nutrition assistance. Similar to active living, policy adoption and enforcement and land use (eg, zoning for urban agriculture/produce sales) were identified. And, finally, unhealthy food and beverage marketing and advertising were included.
Partnership and community capacity
With respect to the partnership and community capacity subsystem, participants produced a total of 27 different variables, with 17 represented in at least 20% of the CPs' causal loop diagrams. Of the 17 variables, many corresponded to community organizing and advocacy, including political will, youth or civic engagement, partnership and collaboration, advocacy, and youth or community leadership. One also highlighted support from policy makers and decision makers. A couple referred more generally to sense of community and community empowerment. Some described programs and promotions, such as health education or sports and recreation. And, one specifically identified the affordability of recreation programs.
Social determinants of health
For the social determinants of health subsystem, participants discussed a total of 82 different variables, with 19 represented in at least 20% of the CPs' casual loop diagrams. Of these 19 variables, several referred to harmful social conditions, beliefs, or practices, such as crime, poverty, and segregation. In addition to poverty, a handful of others also focused on financial or economic-related circumstances, including employment, affordable housing, economic development, and funding for healthy eating and active living. A handful of other variables also referenced schools or education related to curricula and standardized testing, educational attainment, and vocational training. Others identified access to health care, families spending time together, and socially and environmentally responsible policies.
Health and health behaviors
Finally, in the health and health behaviors subsystem, participants identified a total of 40 different variables, with 15 represented in at least 20% of the CPs' causal loop diagrams. Of these 15 variables, many referenced physical activity, or complementary sedentary behaviors, including active transportation, screen time, outdoor recreation, and driving. Several focused on healthy eating and consumption of unhealthy foods and beverages as well as food preparation and purchasing healthy foods and beverages. The remaining referred to overweight and obesity or chronic diseases.
What were the major feedback structures across communities?
Through the model, specific types of causal relationships, or feedback loops, underlying the behavior of the dynamic system, can be identified to provide insights into what is working or not working to support the intended outcomes (in this case, increases in healthy eating and active living and decreases in childhood overweight and obesity).
Active living policies and environments
An example feedback loop representing active living policy and environmental variables in the synthesized causal loop diagram is shown in Figure 1. This illustration is a reinforcing loop and may be interpreted as follows:
With more safe, quality parks and recreation facilities, more children are outside playing. In turn, this can stimulate greater youth civic engagement and collaboration across youth and other community organizations. With collaboration, more funds and resources can be generated to support healthy eating and active living. Some of these funds can be used to increase the safety or quality of parks and recreation facilities.
In a reinforcing loop, the effect of an increase or decrease in a variable continues through the casual pathway and reinforces the increase or decrease in the initial variable. In isolation, this reinforcing loop can be a “virtuous cycle” when all of these assets positively support one another, but the same feedback loop can also be a “vicious cycle” when a decrease in one variable is perpetuated around the loop into a downward spiral. This reinforcing loop is only one part of the larger causal loop diagram (see Supplemental Digital Content Figure 1, available at: http://links.lww.com/JPHMP/A147), with a total of 1555 feedback loops that incorporate safe and quality parks and recreation facilities and “compete” for influence over the variables in the loop. Some of these influences further reinforce the direction of change, whereas others balance or counteract the direction of change. At some point, these reinforcing influences (good or bad) level off as balancing feedback loops ultimately limit the upward or downward trends. For example, communities may become saturated with safe and quality parks and recreation facilities, reducing the added value of new safe and quality parks and recreation facilities. Similarly, there may be so many unsafe, poor quality parks and recreation facilities that kids are unable to play outside.
Healthy eating policies and environments
Another example feedback loop representing healthy eating policy and environmental variables in the synthesized causal loop diagram is shown in Figure 2. This illustration is a balancing loop and may be interpreted as follows:
Increased access to healthy foods and beverages provides more opportunities to purchase and consume these products. Healthier eating behaviors can help reduce rates of childhood obesity. Yet, with declines in rates of childhood obesity, this may also reduce the perceived need for childhood obesity advocacy initiatives. Thus, there is likely to be a subsequent decrease in political will to address this issue, resulting in fewer new or modified policies along these lines. As attention to childhood obesity may have stimulated policy discussions in related topics, such as reduced crime and violence to increase safe trips to local food vendors, these declines in political will may be met with increases in crime and violence. With more crime and violence, there is a corresponding decline in the local economy and funds that are designated to healthy eating and active living initiatives. With a declining economy and reductions in funding, neighborhood food stores may also have to shut down, as small businesses have a difficult time thriving in this climate. Therefore, this sequence may lead to a reduction in access to healthy foods and beverages.
In a balancing loop, the effect of changes in variables within the loop is to counteract or balance the direction of change. Rather than accelerating the direction of change (reinforcing loops), balancing loops tend to slow down the rate of change so that, in addition to counteracting the initial change, they also tend to push a system toward some stable goal. In Supplemental Digital Content Figure 1 (available at: http://links.lww.com/JPHMP/A147), this loop is disconnected visually (eg, the connection from childhood obesity to community and youth advocacy is not a direct connection). To improve the readability of the diagram and minimize the links from crossing over other links, these figures use “shadow” variables indicated by an open and closed bracket (eg, “<childhood obesity>”).
What implications can be translated to various audiences?
In addition to the examples provided in Figures 1 and 2, many other feedback structures emerged from the HKHC communities' causal loop diagrams and several exemplary themes are presented in Table 4. Looking at the synthesized causal loop diagram (see Supplemental Digital Content Figure 1, available at: http://links.lww.com/JPHMP/A147), partnership and community capacity structural elements landed in a central position in the synthesized causal loop diagram, suggesting the critical role of these elements in fueling the change process in communities. Likewise, social determinants appear to have a cascading influence on all of the other subsystems and feedback structures in the diagram. The diagram also highlights several places where active living and healthy eating subsystems intersect (eg, automobile use with air, water, and soil quality with potable water; public transportation with access to healthy foods).
Along with the practical implications presented in Table 4, the HKHC communities' causal loop diagrams also suggested many questions for assessment and evaluation of this work (Table 5). In some communities, the GMB sessions helped build participants' skills and knowledge related to the various subsystems and their interactions and these may be transferred to other health topics or community conversations. Several community examples are included in this supplement.23–26
The causal loop diagram ties together the behavior-over-time graphs, the participants' stories and dialogue, and feedback loops to understand the common behaviors of systems affecting health across communities and to stimulate greater conversation related to an HKHC theory of change, including places to intervene in the system and opportunities to reinforce what is working. This article only begins to uncover the many subsystems, feedback structures, and system insights from the HKHC GMB sessions.
Through causal loop diagrams, people can share mental models and enrich them. Causal maps assist in improving the process of thinking about the structure of a problem by having community members describe the feedback loops and associated effects as well as recognize more immediate versus delayed responses, self-reinforcing side effects, and possible sources of policy resistance, which are often ignored when they are not mapped in a causal diagram.
In comparing computer simulation versus causal mapping, Homer and Oliva20 describe that the latter is useful for describing the possible causes and solutions for a problem situation. Causal loop diagrams are used not just to create simulation and quantified models but also to provide detailed system description and stand-alone policy analysis.20 Coyle27 reviews several examples of qualitative models and defines the role of such models in finding policy insights. For example, causal loop diagrams describe the complex problem in a limited space in contrast to narratives that take larger space.27 Also, causal loop diagrams are helpful reminders to distinguish causal from associational relationships during discussions so that these conversations lead to the identification of feedback loops to explain behavior or insights.
Group model building is a powerful method because it actively involves a wide range of participants in modeling a complex system. Decision makers, community partners, and trained modelers each take part in causal loop diagram development. This process leads to deeper and shared insights among participants while they create the causal loop diagrams that are grounded in community experience. Because of the broad involvement in creating the causal loop diagrams, this process promotes “buy-in” to high-leverage prevention policy recommendations.
The causal loop diagrams complement each CP's work plan by mapping how the partnership's goals influence what is happening in the community and how the resources interact within the system. These diagrams can provide insight into the ways that changes in various strategies, policies, and activities are related and may synergistically impact the community. Furthermore, these diagrams represent data about the feedback structures within a community from the perspective of partners living and engaged in the community. Resulting diagrams can be used by community partners in several ways; for example, communicating prevention strategies and programs; revising or designing policy, system, and environmental strategies; and designing evaluation efforts.
An example of the limitations of qualitative models and their use in finding policy insights is the obesity system map published by the Foresight Program of the UK Government Office for Science.28 This model was developed through engagement of various stakeholders, including scientists, private sector parties, and government departments. The qualitative model has 108 variables and more than 300 causal links. Finegood et al28 described this map as a useful tool to convey the complexity of the obesity challenge to the field; yet, the complexity of the map may lead to perceptions that it is not feasible to tackle this problem. Causal maps often display dense information that makes it difficult to comprehend the details. In essence, these models may be useful for describing the complexity as opposed to discovering further system insights. This was a major challenge to synthesizing all 50 HKHC causal loop diagrams.
In addition, GMB draws on knowledge and skills from system dynamics and systems thinking. This requires some introductory training in systems thinking or system dynamics that includes the use of behavior-over-time graphs and causal loop diagrams, identifying feedback loops, and distinguishing reinforcing and balancing feedback loops. In system dynamics, the goal is to identify and understand the system feedback loops, or the cause-effect relationships that form a circuit where the effects “feed back” to influence the causes. There are many different feedback loops interacting simultaneously to influence or to be influenced by each of the variables. Some variables may strengthen or increase values for variables they influence, whereas other variables may limit or decrease these variables. However, causal loop diagrams cannot be used for dynamic behavior inference but only for describing structure.29 Determining the feedback loop or loops that dominate the system's behavior at any given time and validating the dynamic hypothesis are more challenging problems and, ultimately, require the use of computer simulations.
The application of systems thinking tools combined with GMB techniques creates opportunities to define and characterize complex systems, with multiple interfacing subsystems (eg, healthy eating, active living, social determinants), in a manner that draws on the authentic voice of residents and community partners. Benefits at the community level include the development of a shared language and a common understanding of the system across various disciplines and sectors as well as the identification key leverage points in the system for intervention. For the field, insights derived from a synthesis of findings across communities may highlight community assets and resources that yield the greatest return on investment as well as root causes of poor health outcomes, disparities, and inequities perpetuating resistance to interventions. While these conversations and methods are highly complex, they are also necessary to move toward positive systems change.
1. Hoehner C, Brennan L, Sabounchi N, Hovmand P, Kemner A. Behavior-over-time-graphs: assessing perceived trends in healthy eating
and active living
environments and behaviors across 49 communities. J Public Health Manag Pract. 2015;21(3 Suppl):S45–S54.
2. Andersen D, Vennix J, Richardson G, Rouwette E. Group model building: problem structuring, policy simulation and decision support. J Oper Res Soc. 2007;58(5):691–694.
3. Hovmand P. Community Based System Dynamics. New York, NY: Springer; 2014.
4. Hovmand P, Andersen D, Rouwette E, Richardson G, Rux K, Calhoun A. Group model building “scripts” as a collaborative tool. Syst Res Behav Sci. 2012;29:179–193.
5. Vennix J. Group Model Building. New York, NY: John Wiley & Sons; 1996.
6. Vennix J. Group model-building: tackling messy problems. Syst Dyn Rev. 1999;15(4):379–401.
7. Fallah-Fini S, Rahmandad H, Huang T, Bures R, Glass T. Modeling US adult obesity trends: a system dynamics model for estimating energy imbalance gap. Am J Public Health. 2014;104(7):1230–1239.
8. Johnston L, Matteson C, Finegood D. Systems science and obesity policy: a novel framework for analyzing and rethinking population-level planning. Am J Public Health. 2014;104(7):1270–1278.
9. Bures R, Mabry P, Orleans C, Esposito L. Systems science: a tool for understanding obesity. Am J Public Health. 2014;104(7):1156.
10. Mabry P, Bures R. Systems science for obesity-related research questions: an introduction to the theme issue. Am J Public Health. 2014;104(7):1157–1159.
11. Strunk S, Bussel J. The Healthy Kids, Healthy Communities national program. J Public Health Manag Pract. 2015;21(3 Suppl):S1–S3.
12. Brennan L, Kauh T, Leviton L, Brownson R. Evaluation
of Healthy Kids, Healthy Communities. J Public Health Manag Pract. 2015;21(3 Suppl):S4–S7.
13. Brownson R, Brennan L, Kemner A. Applying a Mixed-Methods evaluation
to Healthy Kids, Healthy Communities. J Public Health Manag Pract. 2015;21(3 Suppl):S16–S26.
14. Fawcett S, Boothroyd R, Schultz J, Francisco V, Carson V, Bremby R. Building capacity for participatory evaluation
within community initiatives. J Prev Interv Community. 2003;26(2):21–26.
15. Richardson G. Reflections on the foundations of system dynamics. Syst Dyn Rev. 2011;27(3):219–243.
16. Luna-Reyes L, Martinez-Moyano I, Pardo T, Cresswell A, Andersen D, Richardson G. Anatomy of a group model-building intervention: building dynamic theory from case study research. Syst Dyn Rev. 2006;22(4):291–320.
18. Stave K. Using system dynamics to improve public participation in environmental decisions. Syst Dyn Rev. 2002;18:139–167.
19. Stave K. Participatory system dynamics modeling for sustainable environmental management: observations from four cases. Sustainability. 2010;2(27):2762–2784.
20. Homer J, Oliva R. Maps and models in system dynamics: a response to Coyle. Syst Dyn Rev. 2001;17(4):347–355.
21. Ackermann F, Andersen D, Eden C, Richardson G. ScriptsMap: a tool for designing multi-method policy-making workshops. Omega. 2010;39:427–434.
22. Andersen D, Richardson G. Scripts for group model building. Syst Dyn Rev. 1997;13(2):107–129.
23. Keane P, Ortega A, Linville J. Healthy Kids, Healthy Cuba: Findings from a group model building process in the rural Southwest. J Public Health Manag Pract. 2015;21(3 Suppl):S70–S73.
24. Nelson D, Simenz C, O'Connor S, et al. Using group model building to understand factors influencing childhood obesity
in an urban environment. J Public Health Manag Pract. 2015;21(3 Suppl):S74–S78.
25. Thomas I, Reilly S. Group model building: a framework for organizing healthy community program and policy initiatives in Columbia, Missouri. J Public Health Manag Pract. 2015;21(3 Suppl):S79–S83
26. Wieczorek J, Forsyth J. Systems thinking
: a Healthy Kids, Healthy Communities partnership in Denver, Colorado. J Public Health Manag Pract. 2015;21(3 Suppl):S84–S87
27. Coyle G. Qualitative and quantitative modelling in system dynamics: some research questions. Syst Dyn Rev. 2000;16(3):225–244.
28. Finegood D, Merth T, Rutter H. Implications of the Foresight Obesity System Map for solutions to childhood obesity
. Obesity. 2010;18(suppl 1):S13–S16.
29. Richardson G. Modelling for Management: Simulation in Support of Systems Thinking
. Aldershot, England: Dartmouth Publishing Company; 1996.
active living; childhood obesity; evaluation; healthy eating; partnership and community capacity; social determinants; systems thinking
Supplemental Digital Content
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.