In search of a mental model-like concept for group-level modeling

System Dynamics Review Vol. 25, No. 3, (July–September 2009): 207–223Published online 21 August 2009 in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/sdr.422Copyright © 2009 John Wiley & Sons, Ltd

Abstract

In system dynamics, we care about mental models, because it is what we attempt to describe, understand, and improve using qualitative maps and computer simulation models. But if we are developing maps and models from a group of individuals, can we still call what we are trying to study a mental model? The main difference between individual and group-level phenomena is that in a group setting interaction and communication among individuals shape what is repre-sented in maps and models. This paper attempts to clarify the object of our group-level modeling by surveying existing literature searching for concepts similar to “mental model” at the group level. The survey leads to an insight that what we attempt to represent with our maps and models varies widely by our selection of modeling methods, and how we defi ne our object will draw attention to a different set of theoretical literature. Copyright © 2009 John Wiley & Sons, Ltd.

Syst. Dyn. Rev. 25, 207–223, (2009)

Relevance of the mental model concept in group settings

In system dynamics, we create qualitative maps and computer simulation models to describe, understand, and improve mental models. The popular use of the term “mental model” led to a need to defi ne the term more explicitly. The defi nition suggested by Richardson et al. (1994) integrated rich theories of human judgment processes involving the means/ends model, Brunswikian lens model, and designer/operator logic. Their study describes how one perceives a system, selects and interprets information, plans an action, and modifi es models. They argued that a mental model that underlies this process is “not directly accessible or observable, and efforts to elicit mental models distort what is elicited” (p. 182). Doyle and Ford (1998, 1999) defi ned mental model by emphasizing various attributes as identifi ed in system dynamics and cognitive psychology literature. With input from Lane (1999), Doyle and Ford (1999) defi ned the mental model of a dynamic system as “a relatively enduring and accessible, but limited, internal conceptual representation of an external system whose structure is analogous to the perceived structure of the system” (p. 414).

Two key assumptions are shared by the aforementioned studies and many other studies on mental models (e.g., Rouse and Morris, 1986): mental models are metaphysical entities residing in individual minds, and we need a method

Hyunjung Kim is a PhD candidate in the Rockefeller College of Public Affairs and Policy, University at Albany. Her dissertation involves use of system dynamics mapping tools to describe the Federal Reserve’s group decision-making process. She is a research staff member at the Initiative for System Dynamics in the Public Sector, participating in various public sector consulting projects. Recent projects include a systems study of a community mental health program implemented by the New York State Offi ce of Mental Health.

In search of a mental model-like concept for group-level modelingHyunjung Kim*

207

* Correspondence to: Hyunjung Kim, Rockefeller College of Public Affairs and Policy, University at Albany, State University of New York, 110 Milne Hall, 135 Western Ave., Albany, New York 12222, U.S.A. E-mail: [email protected] September 2007; Accepted November 2008

208 System Dynamics Review Volume 25 Number 3 July–September 2009

Published online in Wiley InterScience(www.interscience.wiley.com) DOI: 10.1002/sdr

or tool to access the mental models. The methods employed by system dynamicists are qualitative mapping (i.e., causal loop diagramming and stock-and-fl ow diagramming) and computer simulation modeling. However, what is represented in our maps and models does not always come from an individual. Recognizing the importance of group decision making, system dynamics modeling and systems thinking interventions have been actively carried out in group settings, and techniques for group model building have been developed and elaborated (Richardson and Andersen, 1995; Vennix, 1996; Andersen and Richardson, 1997; Rouwette et al., 2002; Luna-Reyes et al., 2006).

Modeling at the group level raises many questions: if we are developing qualitative maps or simulation models from a group of individuals, what are we representing with our maps and models? Should we still call it a mental model? The main difference between individual and group-level phenomena is that in group settings interaction and communication among individuals play an important role in determining what is captured through representation methods used by the modeler. In other words, in group-level modeling, what the modeler attempts to portray using maps and models may be quite different from what we call a mental model.

This paper attempts to clarify this issue by surveying existing literature, looking for concepts similar to “mental model” at the group level. This survey leads to an insight that the group-level concepts defi ned in the literature make different assumptions about the group decision-making process, and the paper introduces a framework to identify such differences. The framework is then applied to the system dynamics context to illustrate how different modeling methods implicitly make different assumptions about what is represented in the maps and models at the group level, and how those assumptions connect the different modeling methods to different sets of theoretical literature.

The goal of this paper is to promote a discussion about the nature of our maps and models, and to introduce relevant literature to the system dynamics community that will enrich our theoretical understanding of what is generated from our modeling efforts. Linking our maps and models to the theoretical literature will facilitate our communication with scholars from other disci-plines and promote a cross-fertilization of research methods and theory-building efforts. The paper also contributes to conceptual understanding of the group learning process, especially by examining the “mental model” con-cept at the group level.

The boundaries of this study: decomposing maps and models

This paper will begin by setting boundaries for theoretical discussion. We use maps and models to access what we loosely call “mental models”. If the process of accessing a mental model is decomposed, there is an elicitation of information about the mental model and a representation of the elicited

H. Kim: A Mental Model-Like Concept for Group-Level Modeling 209


product. The object of interest, or what we often refer to as a mental model, is portrayed in the maps and models, but it is fi ltered—and likely it is distorted—by the elicitation and representation methods used as well as the modeler’s own mental models and assumptions (Scheper and Faber, 1994; Doyle, 1997). In other words, the product of our modeling only has a partial relation-ship to what we intend to study, because it is infl uenced by (1) the true object of interest, or what we have been calling “mental model”, (2) the elicita-tion and representation methods being used, and (3) the modeler’s mental model.

The fi rst part of the paper will discuss how the existing literature defi nes the object of interest of our maps and models in a group setting. If the group-level phenomenon is different from that of the individual level and if the term mental model does not appropriately capture the object of interest at the group level, we need a group-level alternative for the mental model concept. The second part of the paper will discuss how different lenses used by the modeler infl uence how the modeler defi nes his or her object of interest. A modeler’s lens is a combined product of his or her prior experience and a selection of research methods. Different lenses will permit a view of different aspects of the object and thus infl uence what is represented in the maps and models.

Who studies mental model-like concepts in group settings?

Researchers interested in the mental model-like concept for group environ-ments come from disciplines such as cognitive psychology, organizational behavior, sociology, artifi cial intelligence, and decision sciences. Their research topics include, but are not limited to, perception, interpretation, attention, memory, knowledge representation, learning, problem solving, and social cognition (Huff, 1990). Despite the common interest in defi ning and analyzing mental model-like concepts at the group level, there has been little effort toward cross-fertilization among these disciplines (Klimoski and Mohammed, 1994). One barrier for interdisciplinary communication is the use of different terms to describe the concept of interest. Researchers are interested in the black box that lies between somewhat observable information inputs and decision outputs, but they use different names for the black box. Therefore, in order to discover a group-level concept like mental model that is grounded in the established literature, it is necessary to explore various terms and defi nitions used in the literature that refer to the black box.

For analytical convenience, the black box will be called “the processor” in this study. The processor includes all aspects of cognition, information processing, and judgment building in a group. The processor generates one or more decisions or actions in a group. It does not refer to physiological constructs, such as brain or neurons. Rather, it encompasses all the mental representations and social processes that transform information inputs into decision outputs. The term



processor is comprehensive and general enough to be used as a higher-level concept for various mental model-like concepts used in the literature. It is like a bin where all the similar concepts can be collected for further analysis. However, the term is used only for analytical purposes in this study, and it is not proposed as a mental model-like concept at the group level.

Lists formulated by Schneider and Angelmar (1993), Klimoski and Mohammed (1994), and Walsh (1995) give a good overview of some of the terms used to describe the processor in the literature. The three studies together list over 200 variations in the terms used. There are many more if we consider those identifi ed by other studies. Table 1 shows some of the words used by researchers to formulate a descriptor for the processor, usually by com -bining a qualifi er describing group-ness of the phenomenon with one or two words describing the processor. Sometimes, a qualifi er that emphasizes spe-cial attributes of the processor is used with a descriptor for the processor. If one reviews the previous research on the processor, he or she is likely to come across many of the combinations of the listed words, as we will see in the next section. The defi nitions of these terms vary even more.

Having diverse terms and defi nitions has its benefi ts. It demonstrates the fact that researchers have explored various attributes of the processor and shows that what they are discussing is in fact not one concept but different concepts with possible overlaps among them. The problem is that there has been little effort to identify how these concepts differ from one another. When a cross-referencing occurs, do the authors mean the same thing? In order to highlight such conceptual differences, this study proposes a framework to organize different processor terms.

Qualifi ers for group-ness: Name for the processor:Shared Knowledge structureTeam Cognitive structure/system/frameworkSocial CognitionCollective MemoryGroup Mental modelOrganizational SchemaNegotiated Culture. . . Minds

RepresentationQualifi ers emphasizing other attributes of the processor:

Interpretation systemBelief structure/system

Managerial Social orderSituated Perceptual fi ltersTransactive . . .. . .

Table 1. Examples of words used to formulate a descriptor for the processor



A framework to capture conceptual differences in the terms

The framework proposed in this study is based on the idea of a continuum (see Figure 1). The continuum recognizes that it is not always easy to identify different terms and defi nitions with clear-cut categories. One dimension of the continuum is the location of the processor. At one extreme, the entire processor could be located at the level of an individual. Thinking is done at the individual level only, and even in a group setting the locus of study is on individual mind processes. At the other extreme, the entire processor could be assumed to be located at the collective level. The collectivity may refer to a group, an organization, or even an industry. When the processor is located at the collective level, it is a collectivity that receives and processes information and generates action. A similar criterion is briefl y mentioned by Lant (2002) when she explains the “locus of organizational cognition” (p. 355). It must be noted that the location of the processor is different from the level of analysis. While the former is about where the processing takes place, the latter is about where the processor is observed. The location of the processor explicates the researcher’s assumptions about the locus of thinking, while the level of ana-lysis is relevant to research design.

The second dimension of the proposed continuum is the form of the proces-sor. At one extreme, researchers are interested in the static product form, such as memory and knowledge. At the other extreme, researchers are interested in the process that changes in the course of group dynamics. Examples would be communication and interaction. There are studies that defi ne the processor in terms of both products and processes. Therefore, the continuum idea will be useful to deal with such problems.

Positioning studies on the continuum

In this section, different processor terms used in the literature will be posi-tioned in the two-dimensional space described in Figure 1. The positioning

Fig. 1. Framework of location–form continuum



involves a close study of defi nitions of the processor terms in order to identify assumptions made in relation to the processor’s location and the form.

Cognitive psychologists have been inclined to use the term mental model or cognition. In general, mental model refers to a representation of reality that allows one to make sense of the world and anticipate its future paths. For a more formal defi nition, many scholars, including Doyle and Ford (1998), refer back to Johnson-Laird’s book Mental Models (Johnson-Laird, 1983). When used in a group setting, a mental model is often referred to as a shared mental model or a team mental model (Cannon-Bowers et al., 1993). Salas and Cannon-Bowers (2001) defi ne a shared mental model as “knowledge held by a team member that is either compatible, complementary, and/or overlapping with teammates” (p. 87). Similarly, Klimoski and Mohammed (1994) defi ne a team mental model as “what is being shared and operating among team members as a collectivity” (p. 414). In the same study, they use the term shared cognition interchangeably with team mental model, and emphasized that a team mental model or shared cognition is one’s knowledge about his or her team members’ knowledge and is different from unarticulated subconscious knowledge. In other words, their defi nition of team mental model is narrower than that of Salas and Cannon-Bowers, because to qualify as a team mental model knowl-edge has to be shared between team members, and must also be known to all members that the knowledge itself is being shared. In both studies, the proces-sor is in product form (i.e., knowledge) and the location of the processor is within the individuals. In sum, these studies are focused on the part of individual knowl-edge that is shared among the team members. Therefore, the position of their defi nitions in the continuum framework is at the individual–product end (see Figure 2).

In contrast to the above defi nitions, shared mental model defi ned by Levesque et al. (2001) moves slightly towards the collectivity–process ends of the con-tinuum. Their interest is in overlapping cognition which they associate with enhancement in team performance. Since overlapping can only occur though coordination and communication, they study interactions leading to conver-gence of mental models over time. In other words, while mental models are primarily located in individuals, the process of sharing modifi es the mental models of the individuals.

When the location of the processor is at the individual end of the con-tinuum, the same concept may be used to describe a phenomenon at both the individual and group level. One example would be the concept of a schema. Schemas are defi ned as cognitive templates or simplifi ed representations of knowledge that are used to identify elements of a situation and relationships between these elements (Fiske and Taylor, 1984; Walsh, 1995; Elsbach et al., 2005). Because schemas are individually held, relatively stable, and seldom observed, schema discussions rarely differentiate between the individual and group/organizational level processes. A concept similar to schema is know-ledge structure. According to Walsh’s (1995) defi nition, “a knowledge structure



is a mental template that individuals impose on an information environment to give it form and meaning” (p. 281). If positioned in the location–form framework in Figure 2, schema and knowledge structure would be located at the individual–product ends of the continuum.

In contrast to these individual-level perspectives, there is a group of re-searchers who use the term “cognition” at the collective level. Nicolini (1999) identifi es organizational cognition as “social process of cognition and thinking at organizational level” and differentiates it from “cognitive process at individual level in organizational setting” (p. 834). He makes it clear that organizational cognition is a social process and the locus of the processor is at the collectivity. However, Tegarden and Sheetz (2003) use the same term to mean “shared under-standing that managers have in common with each other” (p. 114). Compared to Nicolini’s perspective, Tegarden and Sheetz’s organizational cognition closely resembles the shared mental model defi ned by Salas and Cannon-Bowers (2001). There are others who prefer to use the term collective cognition. Gibson (2001) defi nes collective cognition as the four phases of accumulation, interac-tion, examination, and accommodation that take place at the group level. With

Fig. 2. Positions of the terms used in the literature to describe the processor



an exception for Tegarden and Sheetz’s, the defi nitions of organizational or collective cognition can be positioned near the collectivity–process ends of the continuum in Figure 2.

Weick and Roberts (1993) call the processor collective mind or group mind. They defi ne a collective as individuals in a group who interrelate their actions with care, and mind as an activity as opposed to an entity. They explained that collective mind is located in the process of “heedful interrelat-ing” (p. 361), and their defi nition can be interpreted as individually located processors which operate during an interrelating process among the indi-viduals. This defi nition thus falls on the individual–process side of the con-tinuum in Figure 2. A similar term used is organizational mind. Sandelands and Stablein (1987) defi ne organizational mind as an ideational process that is carried out by organizational behavior. They contrast their view with others who regard mind as “a substance or static pattern of relationships” (p. 138), and emphasize that complex interaction of ideas is at the heart of the organi-zational mind. According to Sandelands and Stablein, organization is mind, and the criteria for it to qualify as mind should be different from that of the human mind. Their defi nition of organizational mind is clearly on the collectivity–process end of the continuum. As a theory of group mind, Wegner (1986) proposed the idea of transactive memory. He defi ned it as “a set of individual memory systems in combination with the communication that takes place between individuals” (p. 186). It is cognitive interdependence among team members; a way that individuals use other people’s memory as an external memory. In organizations, transactive memory is a “shared system for encoding, storing, and retrieving information” (Wegner et al., 1991, p. 923) and Wegner explains that it is neither memory of individuals nor the process of interaction among these individuals, but a combination of these two. There-fore, the position of transactive memory would be somewhere in the middle of the product–process and the individual–collectivity continuum in Figure 2.

Walsh and Fahey (1986) and Walsh et al. (1988) focus more on the political process within a group. They posited that what is represented in a group’s col-lective knowledge structure is infl uenced by individual power differences in the group. They call the aggregated model a negotiated belief structure. The position of the negotiated belief structure in the continuum framework is somewhat less clear. Although the negotiated belief structure is a knowledge structure, i.e., a product, what generates the product is a political process of negotiation and infl uence. The location of the processor is also a combination of individual and collectivity. While negotiation takes place in the collectivity, there is a substantial infl uence by powerful individuals in creating negotiated belief structures.

There is a group of researchers who emphasize the interpretation and mean-ing creation aspects of the processor. Daft and Weick (1984) called the proces-sor organizational interpretation systems, and defi ned it as “the process of translating events and developing shared understanding and conceptual schemes



among key managers” (p. 286). They explain that organizations differ in their attitudes towards interpretation, and these attitudes are determined by the managers’ assumptions about the environment, and by the organization’s intrusiveness into the environment. A similar term used is organizational sensemaking (Weick, 1995). Sensemaking is a process of creating meaning from selected information, and is infl uenced by one’s previous experiences. Organizational interpretation systems and sensemaking are both process- oriented concepts. In terms of the location of the processor, these concepts assume interpretation at both the individual and collective level, and recog-nize that a powerful individual can infl uence interpretation at the collective level. Therefore, these terms are positioned in the middle of the location continuum as illustrated in Figure 2.

Levitt and March (1988) use the term organizational memory to describe the organizational learning process. It is defi ned as routines, rules, and proce dures in organizations where organizational members’ experience and know ledge are accumulated and maintained. This organizational memory is an artifact of organizational behaviors and is separate from members of the organization. Therefore, organizational memory may be positioned at the product–collective ends of the continuum in Figure 2.

Modeling methods and the processor

The survey of terms and defi nitions for the processor reveals that, while re -searchers appear to discuss similar concepts, there are subtle differences among the concepts discussed. When placed on the continuum framework, the differences become explicit and clear. The literature review and the position-ing exercise in the previous section lead to the conclusion that the mental model-like concept at the group level may not be one concept but may likely be different concepts along the continuum. Then how do the conceptual sub-tleties described in the previous section allow system dynamicists to clarify what is represented in our maps and models?

Our maps and models are not “the processor” per se (Weick, 1990). They are created from observable (verbal or other) behaviors of individuals or a group which only partially refl ect the processor. Depending on the elicitation and representation methods used, modelers implicitly make assumptions about the nature of the processor, and such assumptions will infl uence what is represented in the maps and models (Scheper and Faber, 1994; Doyle, 1997; Nicolini, 1999). In this section, fi ve ideal types of group-level modeling modes are identifi ed to highlight their different assumptions in respect to the proces-sor’s location and form.

The variations along the continuums are mainly determined by when and how a modeler or facilitator intervenes with the subject or client group. The modeling process can be divided into the elicitation, representation, and refl ection/



analysis phases. In the elicitation phase, the raw data about the processor is collected from the subject of the study. If a system dynamics modeler attempts to understand a mental model of an individual, the modeler would interact with the subject or the client in order to obtain data that tell something about the person’s perception of the system. Typical methods for elicitation are interview, survey, observation, and content analysis (Luna-Reyes and Andersen, 2003). In the representation phase, the elicited part of the processor is repre-sented using causal loop diagrams, stock-and-fl ow diagrams, or formal simula-tion models. In the refl ection phase, the represented maps or models are analyzed for insights. The modeling process is reiterative, and the refl ection phase may be followed by another round of elicitation and representation. Group-level modeling adds another layer to the process. There is a phase where individual-level processors are integrated into a group-level processor. When and how this integration takes place shapes what ends up being elicited from the client group and represented in the maps and models.

This study identifi es fi ve ideal modes of group-level modeling (see Table 2). Each modeling mode is characterized by (1) a different sequence of the four phases of elicitation, integration, representation, and refl ection, and by (2) the main player, either modeler or the client group, in each phase (indicated in parentheses in Table 2). Figure 3 summarizes the location and form of the processor studied by each of the ideal modes of modeling.

• Mode 1: Interview with individuals in a group. A modeler can create a map or a model from the data collected from interviews with individuals in the client group. In this case, the elicitation phase comes before the integration phase. In order to understand the group-level phenomenon, the modeler integrates the data collected at the individual level. When data are collected at the individual level and later combined by the modeler, there is no group interaction taking place among the individuals during the data elicitation or

Table 2. Five ideal modeling modes

Modeling sequence

1. Interview with individuals in a group

Elicitation Integration Representation Refl ection(modeler) (modeler) (modeler) (both)

2. Content analysis of data collected from individuals in a group

Elicitation Integration Representation Refl ection(client) (modeler) (modeler) (both)

3. Content analysis of data collected from a group process

Integration Elicitation Representation Refl ection(client) (client) (modeler) (both)

4. Non-participant observation of a group process

Integration Elicitation Representation Refl ection(client) (modeler) (modeler) (both)

5. Modeler-leading group process Integration Elicitation Representation Refl ection(both) (modeler) (modeler) (both)



integration period. In both phases, it is the modeler who takes the leading role. In this modeling mode, the modeler is studying the processors located at the individual level and the form of the processor is static knowledge or memory related to a problem (see Figure 3).

• Mode 2: Content analysis of data collected from individuals in a group. A modeler can create a map or a model from the data collected by individuals in the client group. As in Mode 1, the elicitation phase comes before the integration phase, but the elicitation phase is now controlled by the client group. In order to understand the group-level phenomenon, the modeler integrates the individual-level data provided by the clients. Again, there is no group interaction taking place among the individuals during the elicitation or integration period. The position of the processor being studied is similar to Mode 1, at the individual–product end of the continuum as shown in Figure 3. Integration is done by the modeler identifying overlapping struc-tures from individual mental models or by combining all the structures mentioned by every individual. If the former is the case, what the modeler is trying to capture is similar to that of a shared mental model or team mental model defi ned by Salas and Cannon-Bowers (2001) and Klimosky and Mohammed (1994). If the latter is the case, the term aggregate, congregate, or composite used by Bougon (1992) may replace the term shared.

Unlike Modes 1 and 2 where data are collected at the individual level and then integrated by the modeler to study the group-level processor, data can be collected during or after the interaction and communication process among individuals in a group. In this case, integration of the individual processors to the group-level processor takes place prior to, or at the same time as, the elicitation

Fig. 3. Positions of the processor represented in SD maps and models



phase. The degree of the modeler’s participation in the integration phase determines the nature of the elicited product as described in Modes 3, 4, and 5:

• Mode 3: Content analysis of data collected from a group process. In Mode 3, a modeler does not observe interactions among individuals during the integration process. Rather, the integrated output from the group process is compiled by the subject group, and is presented to the modeler afterwards. In other words, the modeler does not have infl uence over the elicitation or integration process, and it is the client group that takes the leading role in these modeling phases. In this case, the processor studied by the modeler is at the product–collective end of the continuum in Figure 3. It is a product, because the modeler can only study an artifact of an organizational process. The processor is located at the level of collectivity, because the modeler is forced to see the data as they are collected from the group without having access to individual mental models.

• Mode 4: Non-participant observation of a group process. In some cases, the modeler has access to the integration process. By observing group interactions, the modeler is able to collect data during the observation. The degree of modeler participation in the integration phase differentiates Mode 4 from Mode 5. When a modeler observes the group process and collects data during the observation, it is a non-participant observation method. The processor stud-ied is located somewhere between the individual and the collectivity as shown in Figure 3. It is because the interaction of individuals within the collectivity is where the processing takes place. In terms of the form of the processor, the modeler as a passive observer will focus more on the process than the product, because the modeler is paying attention to conversation taking place in the meeting instead of exploring what one may know. What is being said in the group process is often shaped by one’s motivation to infl uence others, and a substantial part of this is inseparable from the political and social process.

• Mode 5: Modeler-leading group process. Finally, there is a modeler-leading group process, as in group model building (Richardson and Andersen, 1995; Vennix, 1996; Andersen and Richardson, 1997). In this mode, the modeler plays an active role in the integration process. An active modeler becomes a leading facilitator and induces the client group to discuss parts of their mental models that are more relevant or important to the problem at hand. The modeler also explores where the mental models confl ict, and by promoting group communication the modeler allows the group to come to an agreement. As in the non-participant observation, the processor is located somewhere between the individual and collectivity as illustrated in Figure 3. However, an active modeler may have access to a product form of the processor as well as a process form, because the modeler can probe knowledge and memory from individuals in the group.



The above analysis shows that the processor we want to represent and under-stand with our maps and models may vary depending on the data collection and integration methods. To make the issue more complex, we rarely resort to one type of modeling mode in real modeling practices, and rather we use a combination of different methods. A modeler using multiple methods will move along the location–form continuum exploring different aspects of the processor. What are the implications of this in relation to the processor terms reviewed in the fi rst part of the paper?

Implications for system dynamics group-level modeling

This paper attempts to fi nd an alternative concept to “mental model” in a group environment. The concept is called “the processor” for the convenience of the analytical discussion. In order to ground the alternative term in the literature, this study has surveyed terms and defi nitions that have been used to describe the processor at the group or organizational level. The main fi nding from the survey is that different terms used in the literature to describe the group- level processor differ in their underlying assumptions about location and form of the processor. This fi nding leads to an insight that the processor at the group level is not a single concept that parallels a mental model at the individual level. Depending on the modeling mode used, system dynamics researchers and practitioners in a group-level modeling may portray different aspects of the processor with their maps and models. This study suggests that rather than loosely using the term “mental model” in the context of group-level modeling, we, as system dynamicists, must carefully select the processor name that shares our assumptions about the processor. The location–form framework makes this task easier. Clarifying assumptions about the processor has several practical benefi ts:

• Deepens theoretical understanding of system dynamics modeling. Grounding our modeling practice in social theories enriches our understanding of what we do (Bell and Bell, 1980; Lane, 2001a, 2001b). The location–form framework makes it explicit that the processors we study with our maps and models are tied to a different theoretical literature depending on our modeling modes. For example, if a modeler is interviewing individuals in a group, the proces-sor is relevant to the theories in the literature on the individual–product end of the continuum. Schema (Elsbach et al., 2005), knowledge structure (Walsh, 1995), and collective cognition (Tegarden and Sheetz, 2003) would be a few examples. In contrast, if the modeler is conducting a modeler-leading group process, he or she might want to look at a different group of literature such as Walsh and Fahey’s (1986) work on negotiated belief structure. By relating ourselves with the appropriate theoretical studies, we can deepen our understanding of modeling practice.



• Facilitates interaction and communication outside the system dynamics community. This study explains that we are not always talking about the same construct when we refer to “mental model” in group-level modeling practices. The location–form framework suggests that many alternative terms are available to describe the processor. By selecting a processor name that shares similar assumptions about the processor, we can communicate better with scholars and practitioners from other disciplines studying the similar aspects of the processor.

• Enhances sharing of different processor elicitation and representation methods. There are methods developed outside the system dynamics community that may supplement our methods for eliciting and representing the processor. The location–form framework makes it easy to locate the methods that study the similar aspects of the processor. For example, system dynamics modelers using the fi rst modeling mode (an interview with individuals in a group) may benefi t from other elicitation methods of the individual-product type of processor, such as a Self-Q interview (Bougon et al., 1990) or Card-Sorting technique (Smith-Jentsch et al., 2001). Likewise, other disciplines studying the processor may benefi t by using our maps and models as a new way to elicit, represent, and analyze the processor of interest.

• Avoids inappropriate use of modeling methods. Different methods will measure different constructs (Nicolini, 1999). The location–form framework suggests that we look at different aspects of the processor depending on our modeling modes. Therefore, it is important to select the appropriate modeling mode that is aligned with our assumptions about the processor. For exam-ple, if we are interested in the processor as a social process of creating a collective understanding of the policy system, building our maps and models based on the data collected from individual clients, as in the fi rst modeling mode, would be less helpful. It is because the processor as a social process assumes that the phenomenon at the collective level is inherently different from the sum of individuals’ knowledge.

The location–form continuum framework proposed in this study has some limitations. Because it uses the concept of a continuum, determining relative positions of different terms requires subjective interpretation of the defi nitions given in the literature. While some defi nitions include a clear statement of location and form of the processor, there are defi nitions that are conceptually vague about these factors. If this is the case, a more in-depth study of the literature should be carried out for clarifi cation. The scope of literature reviewed in this paper may pose another problem—there are terms and defi nitions relevant to the processor not included in this study. However, with such a limitation, this study demonstrates the diversity of terms used to describe the processor, highlights the subtle differences among these terms, and proposes a way to organize them. Finally, it must be noted that the co-location of a modeling mode and a processor term on the location–form framework does not mean that the two are referring to the same processor. The co-location is a loose categorization, since



location and form are not the only criteria defi ning the processor. Similarity in location and form does not necessarily guarantee that the two processors share exactly the same characteristics.

Conclusions and future research

The study concludes without providing one alternative term that can be used like “mental model” in a group environment. However, this study highlights that the substance of our study may differ despite the fact that we are all engaged in system dynamics group intervention. We cannot have one single term that describes the mental model-like concept at the group level, because sometimes we are studying an interaction process within a group, and at other times we are looking for knowledge held by individuals in a group. It is important to explicitly recognize what is represented with our maps and models and relate ourselves to the relevant literatures with similar theoretical orientations so that we can enrich our modeling practices.

Several important topics for future research emerge from this study. First, we need to deepen our investigation on mental models and the group learning process. Group learning and decision making have an important role in system dynamics, yet much needs to be done in terms of developing constructs of group learning (Wilson et al., 2007; Rouwette and Vennix, 2008). This study contributed to the line of research by clarifying a number of assumptions underlying the concept of mental model. Second, we need to examine when and how two processors (i.e., client processor and modeler processor) are integ-rated during modeling interventions. The second part of the paper (see Table 2) briefl y discusses this issue, but a more thorough study on this issue will enhance our understanding of what is captured with our maps and models. Third, we need to promote interdisciplinary efforts to develop tools and tech-niques to study the processor. Sharing assumptions about the processor facilitates this effort. The recent work by Eden et al. (2008) is an excellent example in this area. Finally, the discussion in this study can be extended to a wider range of social theories. For example, linking mental model discussion to the philo-sophical debates introduced by Lane (2001a, 2001b) will strengthen the theo-retical development of system dynamics.

References

Andersen DF, Richardson GP. 1997. Scripts for group model building. System Dynamics Review 13: 107–129.

Bell JA, Bell JF. 1980. System dynamics and scientifi c method. In Elements of the System Dynamics Method, Randers J (ed.). Productivity Press: Cambridge, MA; 3–21.

Bougon MG. 1992. Congregate cognitive maps: a unifi ed dynamic theory of organiza-tion and strategy. Journal of Management Studies 29: 369–389.



Bougon MG, Baird N, Komocar JM, Ross W. 1990. Identifying strategic loops: the self-Q interviews. In Mapping Strategic Thought, Huff AS (ed.). Wiley: Chichester; 327–354.

Cannon-Bowers JA, Salas E, Converse S. 1993. Shared mental models in expert team decision making. Individual and Group Decision Making, Castellan NJ (ed.). Erlbaum: Hillsdale, NJ; 221–246.

Daft RL, Weick KE. 1984. Toward a model of organizations as interpretation systems. Academy of Management Review 9: 284–295.

Doyle JK. 1997. The cognitive psychology of systems thinking. System Dynamics Review 13: 253–265.

Doyle JK, Ford DN. 1998. Mental models concepts for system dynamics research. System Dynamics Review 14: 3–29.

——. 1999. Mental model concepts revisited: some clarifi cations and a reply to Lane. System Dynamics Review 15: 411–415.

Eden C, Ackermann F, Bryson JM, Richardson GP, Andersen DF, Finn CB. 2008. Integrating modes of policy analysis and strategic management practice: requisite elements and dilemmas. Journal of the Operational Research Society (advance online publication 26 March 2008) 60: 2–13.

Elsbach KD, Barr PS, Hargadon AB. 2005. Identifying situated cognition in organiza-tions. Organization Science 16: 422–433.

Fiske ST, Taylor SE. 1984. Social Cognition. Addison-Wesley: Reading, MA.Gibson CB. 2001. From knowledge accumulation to accommodation: cycles of collec-

tive cognition in work groups. Journal of Organizational Behavior 22: 121–134.Huff AS. 1990. Mapping strategic thought. In Mapping Strategic Thought, Huff AS

(ed.). Wiley: Chichester; 11–49.Johnson-Laird PN. 1983. Mental Models: Towards a Cognitive Science of Language,

Inference, and Consciousness. Cambridge University Press: Cambridge, U.K.Klimoski R, Mohammed S. 1994. Team mental model: construct or metaphor? Journal

of Management 20: 403–437.Lane DC. 1999. Friendly amendment: a commentary on Doyle and Ford’s proposed

re-defi nition of “mental model”. System Dynamics Review 15: 185–194.——. 2001a. Rerum cognoscere causas. Part 1: How do the ideas of system dynamics

relate to traditional social theories and the voluntarism/determinism debate? System Dynamics Review 17: 97–118.

——. 2001b. Rerum cognoscere causas. Part II: Opportunities generated by the agency/structure debate and suggestions for clarifying the social theoretic position of system dynamics. System Dynamics Review 17: 293–309.

Lant TK. 2002. Organizational cognition and interpretation. In The Blackwell Companion to Organization, Baum JAC (ed.). Blackwell: Oxford; 344–362.

Levesque LL, Wilson JM, Wholey DR. 2001. Cognitive divergence and shared mental models in software development project teams. Journal of Organizational Behavior 22: 135–144.

Levitt B, March J. 1988. Organizational learning. Annual Review of Sociology 14: 319–340.Luna-Reyes LF, Andersen DL. 2003. Collecting and analyzing qualitative data for

system dynamics: methods and models. System Dynamics Review 19: 271–296.Luna-Reyes LF, Martinez-Moyano IJ, Pardo TA, Cresswell AM, Andersen DF, Richardson

GP. 2006. Anatomy of a group model-building intervention: building dynamic theory from case study research. System Dynamics Review 22: 291–320.

Nicolini D. 1999. Comparing methods for mapping organizational cognition. Organization Studies 20: 833–860.



Richardson GP, Andersen DF. 1995. Teamwork in group model building. System Dynamics Review 11: 113–137.

Richardson GP, Andersen DF, Maxwell TA, Stewart TR. 1994. Foundations of mental model research. In 12th International System Dynamics Conference, Stirling, U.K. System Dynamics Society.

Rouse WB, Morris NM. 1986. On looking into the black box: prospects and limits in the search for mental models. Psychological Bulletin 100: 349–363.

Rouwette EAJ, Vennix JAM. 2008. Team learning on messy problems. In Work Group Learning: Understanding, Improving and Assessing How Groups Learn in Organiza-tions, Sessa VI, London M (eds). Erlbaum: New York; 243–284.

Rouwette EAJ, Vennix JAM, Mullekom T. 2002. Group model building effectiveness: a review of assessment studies. System Dynamics Review 18: 5–45.

Salas E, Cannon-Bowers JA. 2001. Special issue preface. Journal of Organizational Behavior 22: 87–88.

Sandelands LE, Stablein RE. 1987. The concept of organization mind. In Research in Sociology of Organizations, Vol. 5, Bacharach S, DiTomaso N (eds). JAI Press: Greenwich, CT; 135–162.

Scheper WJ, Faber J. 1994. Do cognitive maps make sense? In Advances in Managerial Cognition and Organizational Information Processing, Vol. 5, Stubbart C, Meindl JR, Porac JF (eds). JAI Press: Greenwich, CT; 165–185.

Schneider SC, Angelmar R. 1993. Cognition in organizational analysis: who’s minding the store? Organization Studies 14: 347–374.

Smith-Jentsch KA, Campbell GE, Milanovich DM, Reynolds AM. 2001. Measuring teamwork mental models to support training needs assessment, development, and evaluation: two empirical studies. Journal of Organizational Behavior 22: 179–194.

Tegarden DP, Sheetz SD. 2003. Group cognitive mapping: a methodology and system for capturing and evaluating managerial and organizational cognition. Omega 31: 113–125.

Vennix JAM. 1996. Group Model Building: Facilitating Team Learning Using System Dynamics. Wiley: Chichester.

Walsh JP. 1995. Managerial and organizational cognition: notes from a trip down memory lane. Organization Science 6: 280–321.

Walsh JP, Fahey L. 1986. The role of negotiated belief structures in strategy making. Journal of Management 12: 325–338.

Walsh JP, Henderson CM, Deighton J. 1988. Negotiated belief structures and decision performance: an empirical investigation. Organizational Behavior and Human Deci-sion Processes 42: 194–216.

Wegner DM. 1986. Transactive memory: a contemporary analysis of the group mind. In Theories of Group Behavior, Mullen B, Goethals GR (eds). Springer: New York; 185–208.

Wegner DM, Erber R, Raymond P. 1991. Transactive memory in close relationships. Journal of Personality and Social Psychology 61: 923–929.

Weick KE. 1990. Cartographic myths in organizations. In Mapping Strategic Thought, Huff AS (ed.). Wiley: Chichester; 1–10.

——. 1995. Sensemaking in Organizations. Sage: Thousand Oaks, CA.Weick KE, Roberts KH. 1993. Collective mind in organizations: heedful interrelating on

fl ight decks. Administrative Science Quarterly 38: 357–381.Wilson JM, Goodman PS, Cronin MA. 2007. Group learning. Academy of Management

Review 32: 1041–1059.

In search of a mental model-like concept for group-level modeling

Documents

Transcript of In search of a mental model-like concept for group-level modeling