Theories of the deep: combining salience and network analyses to produce mental model visualizations of a coastal British Columbia food web

• Abstract
• Introduction
• Methods
  • Interviews
  • Salience analysis
  • Network analysis
• Results
  • Network analysis of imagined food web data: measuring and visualizing centralities
• Discussion
• Speculation
  • Government managers
  • Males versus females
  • First Nations versus non-First Nations
  • Potential uses of the method
• Conclusion
• Responses to this article
• Acknowledgments
• Literature Cited

INTRODUCTION

Considerable work has been done describing the potential for mental models research to enrich our understanding of complex social-ecological systems (Lynam and Brown 2011). In theory, such insights about how people make sense of complex systems can help improve multistakeholder management of shared resources. Specifically, studies in organizational management have found that it is precisely the development of shared mental models among multiple parties that serves as the lynchpin for successful negotiations (Brodt and Dietz 1999, Liu 2004).

However, actively constructing shared mental models among different parties in a resource conflict or management context is not necessarily straightforward. One challenge is that people’s innate beliefs about their social-ecological system are often tacit, and thus cannot be stated directly without adequate elicitation efforts (Jansen et al. 2006, Beratan 2007). Another is that, in many resource conflict or multistakeholder management contexts, actors constitute not only individuals, with individual mental models, but also loosely affiliated groups, or blocs, more usefully described to each be acting within the logic of a more nebulous, “aggregate” mental model. These group-level aggregate models (sometimes called “cultural models”; see Appendix 1) can prove uniquely challenging to elicit, let alone represent with meaningful validity.

In this paper we demonstrate a first iteration of an approach synthesized specifically to elicit and represent widespread tacit beliefs about a regional food web. In particular, we outline how it is possible to combine freelisting tasks, salience analysis (Smith and Borgatti 1997), and network analysis (Gephi 2014) to ultimately create tangible, empirically responsible visual representations of aggregate mental models that are unique in highlighting the effects of cognitive constraints on people’s thinking.

We report on our application of these techniques to one specific case study site: the Clayoquot Sound UNESCO Biosphere Reserve, on the west coast of Vancouver Island (see Appendix 2). There, a resurgence of the once extirpated sea otter (Enhydra lutris), combined with a regional decline in fisheries, had, at the time of our fieldwork in 2012, created an atmosphere of tension among locals and government officials over how best to manage marine resources in rapid flux. In the context of in-depth semistructured interviews, First Nations, in particular, had been lamenting the voraciousness of the sea otter, which is protected by law but perceived to be decimating edible shellfish populations once a staple of First Nations’ diets. Non-First Nations’ sentiments toward the otter were often more positive, with some welcoming the creatures for their endearing appearance, and potential ecotourism value. Nonetheless, there was considerable ambivalence among local residents as a whole regarding the sea otter’s dramatic recent resurgence. In stark contrast to the lay public, government managers were highly enthusiastic about otters’ return to the region, specifically citing positive knock-on effects (a “trophic cascade”) whereby otters’ reduction of the sea urchin population could lead to denser kelp beds, and hence more habitat for a wider range of biodiversity.

For the sake of methodological clarity, in this paper we focus mainly on analyzing the data elicited from our interviews with lay public participants as a whole. In so doing, we derive a visual model that reveals which trophic relationships are tacitly structuring popular conceptions of the local marine food web. Before concluding, we also present a series of models derived both from government manager interview data, and by dividing the lay public data into two pairs of contrasting demographic subgroups, First Nations vs. non-First Nations, males vs. females. Because of logistical constraints, these latter models are speculative, but demonstrate how the method, if refined, could be used to generate hypotheses regarding otherwise tacit cognitive drivers of management conflict between groups.

METHODS

Our method consisted of three phases: (1) in-depth structured interviews in which freelisting and related tasks were used to elicit participants’ beliefs about the local social-ecological system, and specifically their individual mental models of the local marine food web; (2) salience analyses of the collective freelist data at the group level using the software package ANTHROPAC (Borgatti 1996); and finally (3) network analyses of the group-level food web salience data using the software package Gephi (2014). Ultimately, we used Gephi to render these results visually, as food web models that highlight which species, and which trophic connections among which species, are most cognitively salient to participants (see Figs. 1 to 6).

Interviews

Over the course of several months in the spring of 2012, we completed in-depth structured interviews with 67 local residents of Clayoquot Sound, as well as with four key government resource managers responsible for conservation prerogatives in the region. Local residents were recruited primarily using popular community poster boards in the area, supplemented by word-of-mouth snowball sampling. For those participants who were interviewed on recognized First Nations-administered lands, contact was first established with elected community leaders who subsequently introduced us to individual community members, which was also followed by word-of-mouth snowball sampling. Government managers, as a separate sample, were identified and approached based on prior working relationships and job title.

The 67 local-resident participants ranged in age from 20 to 80. Within this group, there were 29 females, and 38 males. Of these participants 41 self-identified as non-First Nations, while the remaining 26 were self-identified First Nations, primarily from the Ahousaht and Toquaht nations, with a few participants from the Tla-oh-qui-aht and Yuułuʔiłʔath (Ucluelet) nations. Participants from the latter two bands granted interviews outside of their communities, in Tofino and Ucluelet proper, while interviews with the former were conducted on tribally administered lands. All participants were offered financial compensation for their time at a rate of CAD$15 per hour. The study was deemed low-risk, and subsequently approved, by the University of British Columbia Research Ethics Board as per institutional regulations.

The interviews consisted of a relatively lengthy structured protocol, conducted at a place of the participants’ choosing. The protocol involved a range of freelisting tasks, sorting tasks, and drawing tasks, as well as number of questions pertaining to participants’ demographics, their beliefs about the local ecology, and their otter-management preferences (see Appendix 3 for the full protocol).

The relevant segment of the interview for the purposes of this paper began with a freelisting task, wherein the participants were asked to verbally list “as many species and/or resources [you] can think of on the west coast of Vancouver Island.” As participants did so, they were requested to write the name of each species or resource down on small individual cards in the order that they came to mind. This ordering ultimately gave us a measurement of species’ relative “cognitive accessibility'” to participants, i.e., how readily a given species comes to a given participant’s mind, and, by inference, a sense of which species feature most prominently in participants’ mental models. We later conducted a salience analysis (Smith and Borgatti 1997) on this ordering data to explore species’ relative accessibility at the group level (see Appendix 4, Tables A4.1 to A4.6, for a relevant selection of this data).

Subsequent to the first freelisting task, the protocol aimed to elicit participants’ beliefs specifically regarding trophic interactions among species. To do this, we devised a task wherein participants were provided with a box full of a large volume of wooden popsicle sticks. On each popsicle stick we had written the word “eats” with a long arrow pointing in one of two directions. The interviewer then requested participants to begin connecting the pieces of paper on which they had written species’ names by using the popsicle sticks to show “what eats what” in the local ecosystem.

This technique is similar to the “diagram method” of mental model elicitation (see Green et al. 2003), but with a few consequential innovations. First is the relatively tangible nature of the exercise we implemented, whereby participants were free to physically manipulate “species” and interconnections with their hands, in space, to help them work through their implicit knowledge.

Next, participants were asked to verbalize their thinking as much as possible during this task, so the interviewer could clearly understand their reasoning. Crucially, as they did this, the interviewer recorded the order in which the participants thought of, and represented, the various trophic interactions. This preservation of ordering data is what enabled later salience analysis, and is thus a key feature of the elicitation method.

The participants’ task ended once they had connected as many species, one to the other, as possible. For each given participant, the end result was thus a visually tangible depiction of his or her mental model of the local food web, which was then photographed (see Appendix 5, Fig. 5A.1, for an example).

Salience analysis

Once all the species orderings and trophic web data had been collected, we began to analyze it at the aggregate level. Understanding this shift from eliciting data at the individual level, to analyzing it in the aggregate, is crucial because this is what allows inferences to be made regarding average salience.

With the aggregate level in mind, we first analyzed the freelisted species data using the salience analysis function of ANTHROPAC (Borgatti 1996). Salience analysis helped us approximate which species were most readily recalled by people, thus providing one measure of tacit cognitive tendencies that can affect reasoning about the local ecosystem (see Appendix 4, Tables 4A.1 to 4A.6 for relevant selections from these initial results).

To do this, we used ANTHROPAC to calculate a Smith’s salience index (Smith and Borgatti 1997), or simply “salience score” for each species mentioned by participants. Calculating salience indices is a cognitive anthropological technique for analyzing interviewee-generated lists of terms, which has since been adopted across multiple disciplines (Smith and Borgatti 1997, Sutrop 2001, Barg et al. 2006, Thompson and Juan 2006, de Morais 2009, Pradhan and Ram 2010, Ghorbani et al. 2011, Malan and Neuba 2011, Dongre and Deshmukh 2012, Gravlee et al. 2013). A given term’s Smith’s salience index (S) among a group of participants is a function of both the frequency with which the term is mentioned during a freelisting exercise, i.e., the number of participants who include a given term on their respective lists, as well as a function of the term’s average position on participants’ lists. In our case, terms’ positions on the list were equivalent to the order in which those terms were mentioned by participants. Terms that are mentioned relatively often and are positioned relatively highly on people’s lists, i.e., mentioned early on, obtain the highest saliency scores, and vice versa (Barg et al. 2006). Scores range from 1 (highest) to 0 (lowest; see Appendix 4 for a full description of the calculation).

We then proceeded with a methodological innovation, in which we took participants’ lists of trophic interactions, preserving the order in which participants mentioned each interaction (a key step), and treated these orderings as freelists in and of themselves. The first trophic interaction a participant listed was taken to be the most cognitively salient, i.e., most readily thought of, for that participant, the second, the second-most salient, and so forth. Group-level analyses of this data then provided a general sense of which interspecies relationships were, on average, relatively cognitively prominent for participants at the aggregate level. In other words, for each demographic subgroup, the salience analysis produced an ordered list of “what eats what,” indicating by salience index which trophic relationships, e.g., otters eating urchins, or killer whales eating salmon, were most readily recalled by people, suggesting which ecological relationships most directly structure people’s mental models of the system. (See Appendix 6).

The second crucial feature of our synthesized method is that, because this data specifically characterizes perceived relationships between species, it can also then be further analyzed as a network of species connected by trophic relationships, with each connection weighted by salience index (see Appendix 6, Tables 6A.1 to 6A.6 for relevant sample data). This second innovation is what ultimately enabled us to produce a uniquely data-rich visualization of local residents’ aggregate mental model of the Clayoquot food web (see Fig. 1), the primary methodological contribution of this paper.

To test the hypothesis-generating capacity of the method, we also twice experimented with dividing the local-resident sample into preidentified (see Appendix 7) contrasting halves: once into First Nations versus non-First Nations, and once into males versus females. The results thereof are purely demonstrative, not statistically derived, and so are listed under “Speculation” (see Figs. 3-6), alongside a similarly demonstrative model derived from the responses of our small n (4) of government managers (see Fig. 2).

Network analysis

Having used ANTHROPAC’s salience analysis function to produce lists of trophic relationships with salience indices (see above for a description, and Appendix 6 for a sample of the data), we then took that data and entered it into the free network analysis software package Gephi (2014).

Once this data had been entered into Gephi, we could then proceed with a range of network analyses that calculated a number of “network-centrality” measures for each species. Ultimately, this enabled us to produce data-rich visualizations of aggregate mental models of the local food web (see Figs. 1 to 6), depicting both which connections (“what eats what”) appear to be most cognitively accessible to people, as well as which species feature as the most prominent predators, prey, or ecological connectors as determined by network-centrality measures.

RESULTS

We outline some of the most basic measures that network analysis allows us to calculate that are relevant to mental models of food webs. For clarity, we describe how these features are represented in our chosen visualization scheme. We then follow with a visual presentation of the results of the network analysis itself (i.e., Fig. 1).

Network analysis of imagined food web data: measuring and visualizing centralities

Network analyses consist of two core components: nodes and edges. Nodes simply constitute the objects or “things” that are related to one another in some way. In the case of our Clayoquot Sound food web data, these nodes are the plants and animals that participants mentioned during their trophic connection task (see Appendix 5). Edges are the connections between two nodes. Edges can vary in a number of ways, including by their relative strength (in our case, their “weighting,” as determined by salience index), and their directionality (in our case, the direction of predation).

There is a plethora of ways to measure the relative importance of nodes to a network, i.e., in our study, the importance of species to the wider food web. Such metrics are referred to in the network analysis literature as “centrality measures.” Given that our present study comprises merely a first iteration of a new technique, for simplicity’s sake, we have chosen to highlight only three of the most basic and seemingly relevant centrality measures in our visualizations: “in-degree” centrality, “out-degree” centrality, and so-called “betweenness” centrality.

Out-degree centrality: bigger text depicts a more voracious predator

We chose the size of each species’ name to correspond to that species’ relative out-degree centrality. In this case, out-degree centrality equates specifically to the number of other individually named plants or animals that participants believe a given species preys upon. Thus, the more diverse range of prey a given species is believed to have, the larger its name appears in the network, and vice versa.

In-degree centrality: bigger circles depict more widely targeted prey

The size of a species’ circle, inversely, was selected to represent that species’ relative in-degree centrality. In our study, this equates specifically to the number of individually named predators that participants believe prey on the given species. Thus, the greater the diversity of predators a species is believed to have, the bigger its corresponding circle appears in the network.

Betweenness centrality: darker circles depict more crucial connectors

We chose the darkness of a given circle to correspond to that species’ perceived betweenness centrality in the food web. Betweenness centrality is one of the most basic measures of how pivotal a node is as a connector among other nodes in a network. Gephi calculates this measure based on an algorithm refined by Brandes (2001).

Specifically, analyzing for this measure of centrality identifies which nodes in a network are most frequently located along the most direct path between all other pairs of nodes in the network. In the case of our food web data, this equates to how directly a given species functions as a connector in the energy transfer among all other species in the food web. To our knowledge, this is the first instance of such a use of betweenness centrality in the social-ecological mental models literature.

Edge weight: bigger arrows depict greater cognitive accessibility

Finally, because our networks represent not an actual food web, but rather participants’ aggregate cognitive representation of that food web, we also included in our visualizations the relative cognitive accessibility of each trophic interaction. This greatly increases the inferential relevance of each visualized aggregate mental model because it enables the viewer to immediately grasp which relationships between which species feature most prominently in which peoples’ minds. This is the most crucial contribution of the method because it makes immediately accessible otherwise tacit patterns in participants’ ecological cognition.

Seeing these weightings immediately allows the reader to infer which relationships among which species are likely to most actively structure people’s thinking about the ecosystem. We chose to depict this relative cognitive accessibility of trophic relations using the size of the arrows that connect species in the network one to another. In other words, while the direction of each arrow represents the direction of predation, the size of each arrow represents the cognitive weight of that trophic connection for participants, as calculated by salience index. Thus, the more cognitively salient a given trophic connection is on average among a given group of people, the larger the arrow connecting the two relevant species in the visualization of that group’s aggregate mental model.

For simplicity’s sake, and to reduce visual clutter, we capped the number of trophic connections included in a given visualization. The metric we used as a threshold was once again the salience ranking we had calculated earlier using ANTHROPAC: only the 50 most salient trophic connections among species were included in this analysis for each of the demographic subgroups we studied. See Figure 1 for the visualized results of local residents’ responses.

DISCUSSION

The relative size of the words in Figure 1 indicate that for the lay public in Clayoquot Sound, the most central predators, ranked in descending order by out-degree centrality (noted numerically in parentheses) include: human (10); bear (8); killer whale (6); sea otter (6); and wolf (4). It is these five species, in other words, that locals implicitly believe dominate the ecosystem. Conversely, the relative size of the circles associated with each word indicate that the generic category salmon (6), and the even more vague, but quite commonly cited, category of fish (6), both share the position of being the most cognitively salient prey “species” (in fact, species groups) to feature in lay people’s mental representation of the ecosystem.

Bears appear especially important in people’s web of beliefs, in that it is bears (with a betweenness centrality of 12) that emerge as primary conduits of caloric energy in this visual representation of people’s responses. Salmon (7) and sea otters (5) also emerge as important in this regard, as do killer whales to a slightly lesser degree.

Note, however, that each of these features described above outlines people’s implicit understanding of the system without accounting for the effects of time or effort, both of which actively constrain people’s thinking in actual decision making contexts (Levine et al. 2015). To understand what comes to people’s minds most easily, with the least amount of effort, in the least amount of time, note the relative size and density of the arrows. Namely, when we consider such cognitive accessibility, people’s thinking about the system appears dominated by a specific handful of individual trophic relationships: bear→salmon, sea otter→urchin, wolf→deer, cougar→deer, sea otter→clam, killer whale→salmon, and killer whale→sea lion.

Together, this paints another picture, in which the general public’s thinking about the local ecosystem revolves primarily around those trophic relationships that are most viscerally noticeable to the unaided human senses, i.e., without diving equipment, microscopes, or theoretical training in ecology. Thus, even if, for instance, wolves and cougars are not implicitly believed to be closely connected to the rest of the ecosystem, the eating habits of these charismatic megafauna nonetheless loom large in people’s thinking. Crucially, because human attention and processing power is limited (Levine et al. 2015), this popular “habit of mind” (Atran 1998, Medin and Atran 2004) could be crowding out a conscious understanding of less obvious, yet structurally important, aspects of the regional food web.

SPECULATION

Government managers

To further explore the potential of our method, we also created a visualization based on interviews with four government managers (Fig. 2). The small sample size limits the figure’s inferential power, but here we demonstrate how, even with a limited amount of data, this method can create an accessible image for use in hypothesis generation, or as a conversation starter in multistakeholder management contexts.

For instance, note that, although local residents identified a suite of predators as significant (Fig. 1), our government manager model suggests the sea otter (10), is by far the most central predator in managers’ thinking. If borne out by further investigation, this could well signal a different locus of ecological attention among managers relative to the wider local population, a difference well worth considering in the course of any public consultation over the management of marine resources in the region.

Government managers also appear to display at least two key differences from the local civilian population with respect to their conceptions of prey. One, managers think at a higher phylogenic resolution than do locals. Note, for instance, the diversity of individual salmon species that emerge as salient in the managers’ aggregate model, relative to local residents’. The second difference is that the most central prey species for managers is kelp (6), followed by urchin (3), and the category forage fish (3). This alerts us to the fact that government managers appear acutely cognizant of a distinct ecological subsystem that is entirely absent from locals’ aggregate model: kelp beds.

Thus, by comparing the central prey species in managers’ versus local residents’ respective aggregate models, it quickly becomes clear that the two are characterized by a focus on largely different ecological subsystems and relationships. Although local civilians’ aggregate mental model pivots largely around the consumption of fish species (particularly salmon) by a range of relatively easily observed, often charismatic megafauna, the managers’ aggregate mental model is characterized by an awareness of several distinct ecological subsystems, none of which surfaces with the same degree of detail among locals’ aggregate model. The human→salmonid⇓bear subsystem, the sea otter→invertebrate subsystem and vague awareness of the baleen whale→microscopic prey subsystem do feature in both managers’ and locals’ models. However, the former’s model also includes a kelp-centered subsystem, a shark-centered subsystem, and a keen awareness of the role of two relatively unassuming predators that receive little to no attention in the locals’ model (sunflower starfish and a shorebird known as an oystercatcher). The single species that links any of these subsystems together in the managers’ model is the urchin, which receives the highest betweenness centrality score (1.583) of any species in their model.

This difference in cognitive accessibility of different features of the same shared ecosystem seems to us at least partly an artifact of the kinds of interactions and observational experiences that local residents, as opposed to managers, tend to have with their environment and vice versa. In the context of the often fraught relations between government resource managers and local residents in the Clayoquot Sound region, this apparent gulf in cognitive focus, specifically (as distinct from any simultaneous difference in environmental values), is a key point worthy of further testing, and consideration in management (Liu 2004, Van den Bossche et al. 2011).

Males versus females

Comparing the aggregate mental models of local resident men versus local resident women, respectively, also yields hypothesis-generating insights. Note that the most central predators in male participants’ aggregate model (Fig. 3) are, in descending order: bear (7); killer whale (7); and sea otter (6); followed by the relatively generic category salmon (3); as well as cougar (3); wolf (3); and halibut (3). For women, however, the most centrally prolific predator to emerge by far is human (12; see Fig. 4).

Although men did note, with a relatively low degree of saliency, that humans tend to eat “everything,” it is the women’s attention to the variety and specificity of human-ecosystem trophic relations that dominates the latter group’s aggregate mental model. Although women did identify salmon as the most central prey species in the ecosystem (with an in-degree centrality score of 6), unlike men, they did not identify salmon, nor halibut, nor any other fish species, as important predators (see Fig. 4).

Although our data suggest there are certainly important similarities between men and women’s mental models of the Clayoquot ecosystem, key differences in salience and content suggest that for men, salmon emerge as far more central, with a betweenness centrality score of 22.5, whereas for women, it is sea otters (11), that take the relative, albeit considerably dimmer, spotlight. Although both genders appear cognizant of the same general array of relationships (e.g., bear→salmon, sea otter→invertebrates, human→many species, wolf and cougar→deer), there do appear to be consequential differences in the relative focus and resolution of men’s ecological reasoning (i.e., fish-centered), versus women’s (i.e., human- and otter-centered). Such evidence merits further investigation, and suggests the importance of acknowledging potential biases that could develop should either gender come to dominate consultative processes on marine resource management in the region (e.g., Johnson et al. 2004).

First Nations versus non-First Nations

Finally, dividing the local resident sample into First Nations and non-First Nations participants suggests even starker contrasts. In First Nations’ aggregate mental model of the food web (Fig. 5), sea otter is by far the most central, prolific predator, represented as preying on eight different species of invertebrate, as well as on the general categories of shellfish and everything, for a total out-degree centrality score of 10. Otters’ perceived tendency to predate on clams, and urchins, in particular, appears especially salient for First Nations participants. In non-First Nations’ aggregate model, however (Fig. 6), humans are by far the more prolific, although not particularly salient, predator, boasting an out-degree centrality score of 12. Sea otter, in contrast, receives a relatively low out-degree centrality score of five among non-First Nations, while otters’ trophic connections also appear considerably less salient for non-First Nations than they do for First Nations.

Bear and killer whale appear as relatively central predators in both demographic groups’ models, as does the general category salmon emerge as a shared central prey “species.” However, on average, First Nations appear to find sockeye salmon, in particular, quite central (3), whereas non-First Nations appear less likely to think of salmonids at that high a resolution. For First Nations, clam, as a broad taxonomic group, is in fact the most central prey, with an in-degree centrality score of five. Non-First Nations, in contrast, find the general category fish (6), as well as oyster (3) and plankton (3), to be more central as prey species than they do clam (2).

Remarkably, although sea otters are clearly the most central predator in First Nations’ aggregate model, because they are apparently not readily thought of by First Nations as prey, it is instead salmon, seal, and sockeye salmon that feature as the most salient “connector” species for the food web as a whole, with betweenness centrality scores of 9.5, 5.5, and four, respectively. In non-First Nations’ model, inversely, sea otter has a relatively high betweenness centrality (4), topped only by bear (8.5). Meanwhile salmon, although important, does not emerge as a particularly central connecting species in non-First Nations’ model, partially because it is not readily regarded by non-First Nations as a predator. This latter point suggests that, on average, First Nations have much clearer, more salient beliefs about what salmon feed on than do non-First Nations.

In sum, although First Nations’ and non-First Nations’ models do suggest important areas of overlap, non-First Nations appear to think of humans, once they do think of them, as the central source of predation in the system. For First Nations, however, it is sea otters that loom cognitively as the most significant marine predator. And although both groups appear to think readily about salmon as a key prey species, and killer whales as an important predator, First Nations appear to think of the trophic interactions of both animals at a significantly higher phylogenetic resolution than do non-First Nations.

Together, these apparent differences hint at a number of possibilities. One, First Nations’ comparative discounting of humans as predators, but relative stress on sea otters as predators, suggests there may be a subtle epistemic difference in terms of how First Nations and non-First Nations each tend, on average, to tacitly frame the human relationship to nature . Such effects have been reported in other cognitive anthropological studies contrasting the epistemic stance of indigenous and settler colonial peoples (Ross et al. 2007).

A second possibility is that differences between First Nations and non-First Nations participants’ models of the local ecosystem may well constitute reflections of different habits of mind (Atran 1998, Medin and Atran 2004) acquired by virtue of differences in the two groups’ economic and subsistence activities in the area. For instance, although our interview data suggest that, on average, non-First Nations spend their outdoors time in largely touristic, industrial, or recreational capacities, First Nations reported spending relatively more outdoors time devoted to subsistence food collection, which they are uniquely permitted to do by law.

This difference in habitual interaction with the ecosystem would help account both for the relatively higher phylogenetic resolution at which First Nations appear to think about key marine food species, e.g., sockeye salmon or nearshore edible invertebrates, as well as a higher degree of awareness of the various trophic interactions that support, or threaten, the particular food species themselves. From this perspective, it is also unsurprising that non-First Nations, conversely, seem more inclined to focus on the trophic relationships most readily apparent to the eye, and to the imagination, of a visitor, e.g., the feeding habits of charismatic species such as killer whales, bears, and wolves, without much need for attention to high resolution phylogenetic distinctions.

A third alternative is that both these above factors reinforce one another to create the sorts of observable differences among First Nations’ and non-First Nations’ aggregate mental models suggested by this speculative analysis. Given evidence for the pivotal role of shared mental models in negotiation processes (Liu 2004, Van den Bossche et al. 2011), we argue marine resource management negotiations among First Nations and non-First Nations actors could well stand to gain from actively investigating and considering the implications of such potential group-level differences.

Potential uses of the method

The above comparisons are based on demographic groups selected a prior. As such, they are speculative, overlap with one another, and thus do not necessarily constitute statistically robust axes of difference. However, in future applications, this limitation could be overcome by combining the method described here with one of several factor analyses. By employing, for instance, one or more forms of consensus analysis (e.g., Romney et al. 1987, Weller 2007), researchers could explore whether given preidentified groups do in fact exhibit significantly different shared mental models (informal consensus analysis). Alternatively, the data could be rotated on various factors to determine if there are any tacit groups with significantly different response patterns present in a single sample. Both approaches could help reveal significant tacit drivers of conflict between groups, or, inversely, help debunk essentialist stereotypes that may be negatively affecting cooperation. Given this potential array of applications, as well as the hypothesis-generating capacity of the simpler approach demonstrated above, we suggest the outputs of this method could be well-suited for direct use in deliberative processes among stakeholders (e.g., Parkins and Mitchell 2005, Rodela 2012), or in participatory action research contexts (e.g., Johnson et al. 2004).

CONCLUSION

Mental models are receiving a growing amount of attention in the field of resource management (Jones et al. 2011, Lynam and Brown 2011, Lynam et al. 2012). However, their importance has yet to be fully integrated into multistakeholder management practice. This lack of full integration is relevant, because empirical research (Brodt and Dietz 1999, Liu 2004, Van den Bossche et al. 2011) suggests that it is actually the act of moving toward a convergent mental model itself that is most important in fostering good outcomes in negotiation contexts.

The method we detailed above helps facilitate this by combining salience analysis with network analysis. The result is an accessible but quantitatively robust representation of participants’ collective mental model of a system, with an emphasis on relative cognitive salience. As suggested above, such outputs could have potential applications in both research and practice. There are however, some downsides to this method as it stands. One, although limited sample sizes can produce outputs that help generate hypotheses, small sample sizes can lead to visualizations that amplify the visceral impact of individual responses that are not necessarily reflective of shared cognitive terrain.

Two, because of time and cost limitations, we conducted only one iteration of the method with participants. Because people provided freelists of different lengths, this limited our ability make robust inferences about differences among individuals and subgroups. In future applications of this method, two iterations could be carried out to first generate a list of terms, and then use that same set of terms with each participant.

Three, without the integration of one or more forms of factor analysis into the method, it is not possible to statistically differentiate between groups of participants that have significant patterns of similarity and difference from one another. Time and financial constraints prevented us from doing so in the context of this first iteration of the method, but future applications would benefit from experimenting with various approaches to addressing this challenge.

Finally, we developed this method specifically for application to food webs, an important but limited domain. That said, with minor changes, the method could indeed be adapted to eliciting group-level mental models of any system that can be conceptualized as a network of nodes and edges. We thus encourage readers to consider adapting our method to a range of social-ecological contexts, iteratively modifying it to further develop the quality and inferential power of such mental-model visualizations.

LITERATURE CITED

Atran, S. 1998. Folk biology and the anthropology of science: cognitive universals and cultural particulars. Behavioural and Brain Sciences 21(4):547-569. http://dx.doi.org/10.1017/s0140525x98001277

Barg, F. K., R. Huss-Ashmore, M. N. Wittink, G. F. Murray, H. R. Bogner, and J. J. Gallo. 2006. A mixed-method approach to understanding loneliness and depression in older adults. Journal of Gerontology B: Psychological Sciences 61(6):S329-S339. http://dx.doi.org/10.1093/geronb/61.6.s329

Beratan, K. K. 2007. A cognition-based view of decision processes in complex social-ecological systems. Ecology and Society 12(1):27. [online] URL: http://www.ecologyandsociety.org/vol12/iss1/art27/

Borgatti, S. P. 1996. ANTHROPAC 4.0 [Computer Software]. Analytic Technologies, Natick, Massachusetts, USA.

Brandes, U. 2001. A faster algorithm for betweenness centrality. Journal of Mathematical Sociology 25(2):163-177. http://dx.doi.org/10.1080/0022250x.2001.9990249

Brodt, S. E., and L. E. Dietz. 1999. Shared information and information sharing: understanding negotiation as collective construal. Pages 263-283 in R. J. Bies, R. J. Lewicki, and B. H. Sheppard, editors. Research in negotiation in organizations Volume 7. Elsevier Science/JAI Press, Bingley, UK.

de Morais, F. F. 2009. Traditional ecological knowledge about plants cultivated by fishermen at Community Estirão Comprido, Pantanal matogrossense, Brazil. Boletim do Museu Paraense Emîlio Goeldi--Ciências humanas 4(2):277-294.

Dongre, A. lR., and P. R. Deshmukh. 2012. Farmers’ suicides in the Vidarbha region of Maharashtra, India: a qualitative exploration of their causes. Journal of Injury and Violence Research 4(1):2-6. http://dx.doi.org/10.5249/jivr.v4i1.68

Gephi. 2014. The Open Graph Viz Platform. The Gephi Consortium, Paris, France. [online] URL: https://gephi.org

Ghorbani, A., G. Langenberger, L. Feng, and J. Sauerborn. 2011. Ethnobotanical study of medicinal plants utilised by Hani ethnicity in Naban River Watershed National Nature Reserve, Yunnan, China. Journal of Ethnopharmacology 134(3):651-667. http://dx.doi.org/10.1016/j.jep.2011.01.011

Gravlee, C. C., H. R. Bernard, C. R. Maxwell, and A. Jacobsohn. 2013. Mode effects in free-list elicitation: comparing oral, written, and web-based data collection. Social Science Computer Review 31(1):119-132. http://dx.doi.org/10.1177/0894439312455312

Green, D. W., S. J. Muncer, T. Heffernan, and I. C. McManus. 2003. Eliciting and representing the causal understanding of a social concept: a methodological and statistical comparison of two methods. Papers on Social Representations 12:2.1-2.23.

Janssen, M. A., Ö. Bodin, J. M. Anderies, T. Elmqvist, H. Ernstson, R. R. J. McAllister, P.Olsson, and P. Ryan. 2006. Toward a network perspective on the resilience of social-ecological systems. Ecology and Society 11(1):15. [online] URL: http://www.ecologyandsociety.org/vol11/iss1/art15/

Johnson, N., N. Lilja, J. A. Ashby, and J. A. Garcia. 2004. The practice of participatory research and gender analysis in natural resource management. Natural Resources Forum 28(3):189-200. http://dx.doi.org/10.1111/j.1477-8947.2004.00088.x

Jones, N. A., H. Ross, T. Lynam, P. Perez, and A. Leitch. 2011. Mental models: an interdisciplinary synthesis of theory and methods. Ecology and Society 16(1):46. [online] URL: http://www.ecologyandsociety.org/vol16/iss1/art46/

Levine, J., K. M. A. Chan, and T. Satterfield. 2015. From rational actor to efficient complexity manager: exorcising the ghost of Homo economicus with a unified synthesis of cognition research. Ecological Economics 114:22-32. http://dx.doi.org/10.1016/j.ecolecon.2015.03.010

Liu, L. A. 2004. Shared mental models in negotiation. Graduate School of Vanderbilt University, Nashville, Tennessee, USA.

Lynam, T., and K. Brown. 2011. Mental models in human-environment interactions: theory, policy implications, and methodological explorations. Ecology and Society 17(3):24. http://dx.doi.org/10.5751/ES-04257-170324

Lynam, T., R. Mathevet, M. Etienne, S. Stone-Jovicich, A. Leitch, N. Jones, H. Ross, D. Du Toit, S. Pollard, H. Biggs, and P. Perez. 2012. Waypoints on a journey of discovery: mental models in human-environment interactions. Ecology and Society 17(3):23 http://dx.doi.org/10.5751/es-05118-170323

Malan, D. F., and D. F. R. Neuba. 2011. Traditional practices and medicinal plants use during pregnancy by Anyi-Ndenye Women. African Journal of Reproductive Health 15(1):85-93.

Medin, D. L., and S. Atran. 2004. The native mind: biological categorization and reasoning in development and across cultures. Psychological Review 111(4):960-983. http://dx.doi.org/10.1037/0033-295x.111.4.960

Parkins, J. R., and R. E. Mitchell. 2005. Public participation as public debate: a deliberative turn in natural resource management. Society and Natural Resources 18(6):529-540. http://dx.doi.org/10.1080/08941920590947977

Pradhan, M. R., and U. Ram. 2010. Perceived gender role that shape youth sexual behaviour: evidence from rural Orissa, India. Journal of Adolescence 33(4):543-551. http://dx.doi.org/10.1016/j.adolescence.2009.10.014

Rodela, R. 2012. Advancing the deliberative turn in natural resource management: an analysis of discourses on the use of local resources. Journal of Environmental Management 96(1):26-34. http://dx.doi.org/10.1016/j.jenvman.2011.10.013

Romney, A. K., W. H. Batchelder, and S. C. Weller. 1987. Recent applications of cultural consensus theory. American Behavioral Scientist 31(2):163-177. http://dx.doi.org/10.1177/000276487031002003

Ross, N., D. Medin, and D. Cox. 2007. Epistemological models and culture conflict: Menominee and Euro-American hunters in Wisconsin. Ethos 35(4):478-515. http://dx.doi.org/10.1525/eth.2007.35.4.478

Smith, J. J., and S. P. Borgatti. 1997. Salience counts-and so does accuracy: correcting and updating a measure for free-list-item salience. Journal of Linguistic Anthropology 7(2):208-209. http://dx.doi.org/10.1525/jlin.1997.7.2.208

Sutrop, U. 2001. List task and a cognitive salience index. Field Methods 13(3):263-276. http://dx.doi.org/10.1177/1525822x0101300303

Thompson, E. C., and Z. Juan. 2006. Comparative cultural salience: measures using free-list data. Field Methods 18(4):398-412. http://dx.doi.org/10.1177/1525822x06293128

Van den Bossche, P., W. Gijselaers, M. Segers, G. Woltjer, and P. Kirschner. 2011. Team learning: building shared mental models. Instructional Science 39(3):283-301. http://dx.doi.org/10.1007/s11251-010-9128-3

Weller, S. C. 2007. Cultural consensus theory: applications and frequently asked questions. Field Methods 19(4):339-368. http://dx.doi.org/10.1177/1525822X07303502