Modeling Boundaries of Concern among Conflicting Stakeholders

Both the utility and the limitations of group dialogue for envisioning alternative scenarios in a decision process became apparent to me in a June 2000 workshop on sustainable mobility. The workshop was hosted by the Strategy Support Center of General Motors, in the Renaissance Center of downtown Detroit, Michigan. The goal was to consolidate insight from approximately fifty stakeholders across the corporation, stimulated by presentations from a few invited third-party experts. As a new intern with expressed interest in the subject matter, I created a workshop primer on the topic of sustainable mobility by defining the constituent elements—the economic, ecological, and social dimensions of sustainability—and the role of mobility in providing access to geographically distributed amenities. If mobility is the ability to move (people, goods, information) at will from one place to another, then “sustainable mobility” is the ability to spread this freedom globally and equitably across current and future generations.

Decisions regarding the mix of mobility options available and in use are not made by any one entity—rather, they are propelled by the interdependent actions of the stakeholders involved. Interactions among stakeholders (individuals, industry, government) effectively shape what mix of mobility choices are available (see Figure 1). Choice is constrained, or bounded, by the purpose of the trip (e.g., physical exercise or commute to work), the available selection (as per technology, supply, regulation, and location), and the supporting infrastructure (e.g., highways without bike lanes constrain bicycle choice).

The two-day workshop involved a number of breakout sessions to brainstorm future scenarios of sustainable mobility envisioned for the year 2050. Corporate stakeholders struggled with the task of thinking fifty years ahead, let alone imagining how current modes of mobility could become sustainable. Nonetheless, the dialogue was lively and creative possibilities were generated within a short span of time. An independent writer worked with the stakeholder ideas to generate narratives corresponding to each of four scenarios. To be plausible for 2050, the scenarios were broadly defined in the geopolitical context so that the relative prospects for sustainable mobility could be considered in each. Visual representations of the scenarios were also created.

While the sustainable mobility workshop challenged stakeholders to think beyond the boundaries of their day-to-day concerns, the fifty-year horizon seemed out of reach to some. To address this gap, I spent the remainder of my internship developing a computer model to simulate customer adoption of alternative energy technologies such as fuel cells and hybrid vehicles (Metcalf 2001). This dynamic model enabled some flexibility of temporal extent, so that scenarios for 2010 were evaluated as part of a thirty-year simulation horizon. One advantage of this simulation model was to ensure the underlying assumptions were internally consistent. Moreover, the model was transparent and useful enough to be maintained for additional scenario creation after my internship was complete.

Working on sustainable mobility within the automotive industry sparked my interest in addressing aspects of sustainability beyond the vehicle per se. As I began my doctoral studies in geography in 2003, I expanded the scope of my research to encompass the social dynamics of urban sprawl as part of the land-use evolution and impact assessment model (LEAM). LEAM is an evolving, complex simulation model of urban sprawl that uses remotely sensed data to map regional land use and integrates different component models to create future scenarios. At its core, LEAM draws on economic drivers to project population growth for a region and distributes the population spatially by travel time proximity to attractors such as city amenities, major roads, and recreational areas. For the St. Louis region, if land were available close to the city center, it would have a high probability of being developed in the model. The principals at LEAM were concerned that this heuristic did not capture the realities of abandoned structures and spaces in central areas such as East St. Louis, located on the Illinois side of the Mississippi River in the floodplain. This area’s history has been marked by industrial investment and subsequent abandonment (Thiesing 2003). For example, the East St. Louis neighborhood of Rush City was filled with over three hundred households in the late 1950s and early 1960s, but now has less than fifty. As of 2003, the area was undergoing soil testing for pollution from nearby industrial activity. The city has discouraged reinvestment and would like to convert the neighborhood into an industrial park, but the residents have resisted displacement. They maintain community strength through such means as the local church.

Committed to work on the question of why some areas were neglected despite proximity to attractors, in May 2003 I observed a workshop with a variety of stakeholders from the St. Louis region, including planners, activists, ministers, and mayors. After a brief update on the state of the model, the director of LEAM asked these stakeholders to brainstorm drivers of development in a morning session and scenarios for the future in an afternoon session. Nearly one hundred stakeholders broke into smaller discussion groups, brainstormed ideas on flip charts and then consolidated them into broader categories. When reassembled into the larger group, each subgroup presented their results and each participant assigned votes from a set of adhesive dots. In this process of aggregation, richness from the original discussions was lost, but connections among ideas were strengthened.

Back in the LEAM lab, I used two approaches to try to understand what had been learned from this activity. One was to simply tally the votes into broader categories. With this approach, a striking contrast emerged between perceived drivers and suggested scenarios. Social factors were frequently mentioned as drivers—school quality, fear of crime, safety, lifestyle, class and race segregation—with 27 percent of votes, but rarely mentioned as scenarios, with only 5 percent of votes (see Figure 2). Recognition of social phenomena was key to understanding regional development, but appeared to be outside the domain of leverage for most participants.

The second approach I used involved my own intuition in examining the original, disaggregated ideas on flip charts for the groups in which I had served as scribe, and try to extend the ideas deeper. I employed the analytical technique of language processing, moving down the “ladder of abstraction” to get at the richness of individual ideas by creating complete sentences from these ideas (Hallowell 2005; Shiba 1994). I then clustered the idea-sentences using intuition (this is part of the process, moving back up the ladder of abstraction), and used the resulting insights to develop hypotheses of sprawl and decline. For example, residents who move to outlying areas because of school quality change the property-tax base, and thereby enhance the likelihood that the suburban school continues to be higher quality (see Figure 3).

Subsequent spatial analysis of poverty rates in the St. Louis region revealed that while regional poverty rates stayed about the same, the spatial isolation of poverty increased from 1990 to 2000 (Metcalf 2004). While high-poverty clusters remained focused in the central city and East St. Louis areas, new low-poverty clusters emerged at the fringe. At the same time, urban sprawl increased in the form of low-density fringe development. Since the ability to move to a new home usually requires an income well above the poverty level, new development serves to separate those who can afford to choose from those who cannot. In this way, sprawl dynamics exacerbate spatial disparity between socioeconomic classes.

This analysis demonstrated that LEAM’s expansive spatial boundary of concern was appropriate for the problem of urban sprawl, but implicitly discounted the problems within such that the problem boundary excluded relevant social dynamics. When I presented the LEAM social model to the St. Louis stakeholders, one of them informed me privately that there are three social drivers of sprawl: racism, racism, and racism. My poster displayed maps of regional change in terms of variables such as housing, population, vacancy, and income. Race could be mapped as well as any other variable, but doing so would be insufficient to capture the boundaries of human concern that are implied by racist beliefs. The existing map-based structure of LEAM made it difficult to include the mental models of racial bias that would help to explore the role of racism in driving urban sprawl. As my tenure at LEAM was nearing completion, I resolved to pursue the simulation of mental models in future research.

While I considered the social drivers of urban sprawl, my colleagues at LEAM examined the consequences of urban development, particularly in the St. Louis floodplain (Pinter 2005). In August 2005 I observed a participatory modeling workshop in which stakeholders of the Upper Mississippi River (UMR) convened at the National River Museum and Aquarium in Dubuque, Iowa, to model the dynamics relevant to the floodplain ecosystem functionality in the face of encroaching development (Metcalf 2005). Organized by The Nature Conservancy (TNC), the three-day workshop was dense with alternating subgroup and plenary discussions to ensure a balance of divergent and convergent thinking (Anderson and Richardson 1997). The conference organizer, on loan from the U.S. Geological Survey (USGS) to TNC, had developed an appreciation for the utility of computer modeling to promote individual learning, and envisioned the compounded benefits that could be achieved for interagency collaboration (Lubinski et al. 2008).

Two subgroups of a dozen stakeholders each worked in parallel to create models capturing ecological, economic, and social dynamics of floodplain land use along the UMR. Experienced computer modelers worked to translate stakeholder ideas into model elements and causal relationships. An independent facilitator oversaw the process and kept stakeholders engaged through poetry and musical analogies. Many of the participants discovered their inner balance through an afternoon Tai Chi session. Some even chose to sleep on the river in the historic William M. Black dredge boat.

With this intense mix of activity in the agenda, some stakeholders acknowledged their initial trepidation: “I thought, ‘This is absolutely insane!’” In the end, however, something was accomplished. Two models were created. Different perspectives were shared: “People disagreed and listened at the same time.” Perspectives were enlarged for nearly everyone. Some final thoughts included: “I still think it’s insane. But the process is more important than the outcome.” And: “I was concerned that the objective was too ambitious, but I was wrong.” (Quoted statements were recorded from stakeholder comments during workshop dialogue.)

As with the sustainable mobility workshop described earlier, the UMR stakeholders experienced tension between modeling the system as it “should be” (normative) and as it currently stands (descriptive). This tension between present reality and future vision is an integral part of systems thinking (Senge 1990)—understanding how to articulate and achieve goals. From a modeling perspective, both status quo and alternative futures could be explored with a common set of structural dynamics.

By the end of the first day, the modelers planned to complete a “bulls-eye” diagram of intrinsic, extrinsic, and excluded variables (Ford 1999). To do so, they spent the afternoon engaged in subgroup dialogue about model elements. This involved revisiting both the driving question and the geophysical scope of the model. This experience was constraining, as one stakeholder related, “I was sitting in the uplands, and had to force myself to stick to the floodplain.”

The process of bounding the model through its scope and elements was intensified by both a sense of urgency and uncertainty, as articulated later by a participant: “We knew we had a limited time, so we were overwhelmed at the beginning. We needed to cut to the chase. We felt unclear about what the outcome would look like.” As with identifying driving questions, the process of identifying possible model elements (divergent thinking) was easier than identifying which ones should be central to the system dynamics (convergent thinking).

The process of identifying which variables should be “excluded” as beyond the scope of the model was quite challenging for stakeholders. It was easier to identify “extrinsic” variables that might be incorporated as input parameters (see Figure 4). But incorporating everything as an external influence would render the intrinsic dynamics meaningless. One stakeholder related the challenge of choosing an appropriate scale for the system: “There was a tension between looking so broadly that we have to lump and aggregate, versus going so much into the minutiae that we couldn’t get things done.” This difficulty was resolved by setting explicit system boundaries: “We placed bookends on our thoughts so we could move forward, rather than spinning on one idea.”

Before the end of the second day, stakeholders in each subgroup articulated scenarios for testing the models. That night, the modelers worked behind the scenes to complete the models, even though they were still highly abstract. The proposed scenarios reflected a creative and relevant mix of ideas, but would be difficult to implement in models that were not yet calibrated. Therefore, in lieu of policy testing, the final day was spent in an open plenary dialogue. This collective meeting allowed greater transparency about how each model developed, opportunity for further feedback, and some establishment of consensus to move forward.

One stakeholder explained: “[E]veryone felt that we had accomplished a lot by increasing our understanding of the modeling process, but everyone also agreed that we had not created anything usable yet. We all expanded our awareness and knowledge base of both the challenges of developing a relevant modeling system, and the need for collaborative water resource management information and strategies. The group seemed to collectively agree that this project should be taken to the next step.”

Dialogue about the “next step” was particularly rich. Some wanted to use the model as an educational tool, and emphasized that it wouldn’t take much refinement to be superior to “seat of the pants” decision making. Several mentioned the importance of engaging local communities along the UMR floodplain. At the same time, some experts cautioned about the risk of misperception from presenting a model prematurely or with insufficient transparency.

As the UMR stakeholders revisited the problem on the final day, the workshop ended in much the same way that it began: grappling with problem formulation and model boundaries. “Have we asked the wrong question from the get-go?” one wondered. This “problem of problem formulation” is consistent with Sterman’s (2000) overview of the modeling process as a feedback loop that returns to the step of problem articulation after struggling to evaluate a completed model (see Figure 5). For an individual, this iteration may reflect increased insight from previous results. For a group, this iteration may reflect a more complex struggle to converge individual learning experiences into a collective “a-ha.” Time is needed to move toward convergence of perspectives on the problem through the modeling process. Such convergence is unlikely to be attained through a single iteration. In this way, group model building is still at least as much an art as it is a science (Andersen and Richardson 1997; Andersen et al. 1997; van den Belt 2004).

Hurricane Katrina hit New Orleans less than two weeks after the UMR workshop on floodplain use. While levees were discussed in the workshop, the bulls-eye diagram in Figure 4 placed levees in the “extrinsic” category of external influences on the central dynamics of land-use change. However, scientists and engineers have long studied the reinforcing dynamics of levee height on extreme flood vulnerability and public perception of flood protection from levees with a calculable risk of failure (Pinter 2005). The extent of destruction was not a surprise to scientists who used computer simulations to anticipate levee failure (Travis 2005), but the slow emergency response was unprecedented for the history of New Orleans (Kates et al. 2006). In the post-Katrina aftermath, the responsibility for levee failure and flooding has fallen on the U.S. Army Corps of Engineers, though the legal system has been unable to penalize the Corps for failing to protect New Orleans (Nossiter 2008). While reconstruction efforts are proceeding in a pace consistent with the historical record, stakeholders seek to pursue conflicting goals of rapid recovery, safety, equity, and “betterment” (Kates et al. 2006). Participatory modeling for decision-scenario creation could help to expose underlying beliefs that lead to different objectives for the outcome of the system.

In 2007, a different crisis of water management emerged in Atlanta as extremely warm temperatures in the southeastern United States induced a severe drought (Shelton 2008). The water shortage coincided with a series of workshops in Georgia’s statewide water planning process. As part of this process, Hirsch (2007) conducted a survey of stakeholder values, attitudes, and goals that shape the boundaries of concern for shared water resources in Georgia. Data from this survey were analyzed to identify significant correlations between variables. When analyzed through the lens of organizational affiliation, distinct differences between stakeholders emerged, particularly between agricultural and environmental stakeholders. A partial construction of these conflicting mental models reveals that the environmental model emphasizes government regulation for environmental protection, while the agricultural model emphasizes the importance of private property rights in allocating water for food supply (see Figure 6). A constructive dialogue would focus on the needs of both environmental and agricultural stakeholders, as opposed to blaming any one party for their role in shaping the shared problem.

Mental models may be simulated using neural networks that consist of computational nodes that are connected with varying strengths. The human brain contains around one hundred billion neurons (nodes), with as many as ten thousand synaptic junctions per neuron. As the number of neurons alone approximates the number of stars in the Milky Way, Kosko (1992, p. 13) describes the brain as “an asynchronous, nonlinear, massively parallel, feedback dynamical system of cosmological proportions.” Fortunately for computational purposes, a smaller set of nodes is usually adequate to explore the boundaries of concern implicit in mental models.

A mental model may be embedded in a software agent that may be mobile or stationary, with uniquely defined attributes and behaviors. Mobile, interactive software agents may employ richly structured decision rules. Montague (2006) highlights the foundational role that mobility plays in enabling choice, from the simple decisions of a single cell bacterium to the complex heuristics of human beings. Individuals exercising choice do so by interacting with their present context.

Mental models may be updated through new information, either directly perceived or socially communicated (see Figure 7). Understanding involves updating connections in a conceptual network in response to an observation or communicated signal. In contrast, learning expands the network to include new ideas (nodes). The stimuli for such changes may derive from social influence or direct perception of one’s environment.

Montague (2006) notes that an element of self-deception enables us to be fully human. Because an unexpected contingency may at any time intervene with a person’s life path, humans have evolved to accept uncertainty in predicting our own future states of being. Dynamics under uncertainty may be understood through careful experimentation, simulating and evaluating the plausibility of alternative scenarios. Such simulation occurs via our own mental models all the time (Montague 2006). Simply becoming more aware of mental models and their intrinsic ability to simulate scenarios will help to create collective understanding with or without computers per se. By anticipating surprise in dialogue, conflicting stakeholders will be better able to transcend prior boundaries of concern in the pursuit of tolerance and understanding.

Work on this paper was supported in part by NSF Award #0433165, “Ecological Boundary-Setting in Mental and Geophysical Models.”