Introduction

Research on causal mapping has been an active area in information systems (IS) research.

Causal mapping has been applied in information systems requirements analysis (Montazemi

and Conrath, 1986) and for planning network services (Dutta, 2001). Boland, Tenkasi &

Teeni (1994) argue that causal mapping can be used to capture subjects’ perspectives

for use in decision making, system design, and other activities. Several information

systems have been designed to support causal mapping (Eden, 1989; Zhang, Wang &

King, 1994; Boland, Tenkasi & Teeni, 1994; Kwahk and Kim, 1999; Hong and Han, 2002).

Sheetz, Tegarden, Kozar & Zigurs (1994) proposed a group support system as an aid in

uncovering causal maps of users.

Developing and evaluating group or collective causal maps has been the subject of

several recent IS studies (Tegarden and Sheetz, 2003; Lee, Courtney & O’Keefe, 1992;

Vennix, 1996; Kwahk and Kim, 1999). Ackerman, Eden & Williams (1997), Massey and

Wallace (1996) and Vennix (1996) have maintained that collective maps can be used to

broaden problem solvers’ perspectives by taking alternative views into account in the

definition of a messy problem situation. In their view collective maps can be viewed as

a means to access multiple perspectives for problems with no definitive formulation. This

is a particularly important application, because several researchers (Checkland, 1981;

Courtney, 2001; Linstone, 1984; Mitroff and Linstone, 1993; Senge, 1990) have proposed

that systems-based multiple perspective approaches are required to deal with problems

in organizations and society today. Different perspectives are assumed to hold different

models, and it is through the juxtaposition and combination of models that perspectives

can be mediated. To gain the precision necessary to compare, contrast, and combine

multiple perspectives, it is necessary to build models of the situation. However, it is by

no means straightforward to develop group or collective maps. Tegarden and Sheetz

(2003) found that merging causal maps of individuals into a collective causal map has

been problematic and argued that the creation of collective causal maps is impractical for

many organizational situations.

The research reported in this chapter focuses on the use of modeling to mediate multiple

perspectives on problems through the development of group causal maps. We focus on

the comparison of existing modeling approaches that are capable of representing multiple

perspectives through collective maps. Our research questions are: What approaches are

available to formulate collective causal maps based on multiple perspectives? Are some

approaches superior to others?

This chapter has three objectives. First, it discusses and compares at a conceptual level

three methods of constructing collective causal maps. These three approaches represent

fundamental distinctions in methods of collective causal mapping and each has some

track record of success. Second, it reports the results of two studies conducted to

compare the three methods in terms of various criteria. The criteria include objective

measures such as map complexity, density, and map distance ratio and subjective

measures such as user perceptions of the adequacy of problem representation, solution

implications, stakeholder implications, and degree to which the collective maps capture

different perspectives. These studies test two hypotheses regarding map complexity and

the perceived utility of collective maps for the aggregate, congregate, and workshop

mapping procedures. The results of these studies should help illuminate the relative

strengths and weaknesses of the different approaches to building collective causal maps

and suggest guidelines for selecting the best procedure to fit the modeler’s situation.

This chapter starts with a conceptual comparison of three methods for constructing

collective maps. We then detail a research design for two studies that compare the three

methods. In Study 1 we built collective causal maps to facilitate problem formulation for

a sales problem in a Vietnamese company and in Study 2 we built collective maps to

support the development of a research model of the impacts of infrastructure and

infrastructure projects on a large urban area. In each study we empirically compare

models derived with the three methods on both objective and subjective measures.

Methods for Constructing a Collective

Map/Model

A map is an aggregation of “interrelated information” (O’Keefe and Nadel, 1978). Maps

help represent people’s perceptions about their environments (Weick and Bougon,

1986). A causal map consists of nodes and links (or arrows) that one may use to

understand a situation (Axelrod, 1976). Nodes stand for factors, labels, concepts, or

variables. Links represent relationships or associations. If causal relationships are used,

the maps are called causal maps. Most researchers (Eden, Jones & Sims, 1981; Hart, 1977)

do not differentiate causal maps from cognitive maps. However, Weick and Bougon

(1986) believe that the concept of a cognitive map is broader than a causal map as the

former may have other relationships than causal—such as “contiguity, proximity,

resemblance, and implication.” For consistency, we use the term causal map throughout

the chapter.

Causal maps were originally devised to elicit mental models for individuals (Axelrod,

1976; Eden, 1989). A number of researchers (Landfield-Smith, 1992; Bougon, 1992; Weick

and Bougon, 1986; Schneider and Andgelmar, 1993; Nicolini, 1999; Laukkanen, 1994;

Lant and Shapira, 2001) have extended the application of the concept to a group,

collective or organization. Three common methods found in the literature for constructing

collective mental models are aggregate mapping, congregate mapping, and workshop

mapping. A conceptual comparison of these approaches is provided in Table 1.

In the aggregate mapping approach, the focus is on representing all individual maps

as fully as possible in the collective map. All labels and links from each individual causal

map are included in the collective map. As a result, the aggregate map may become quite

complex. The aggregate approach does not emphasize the causal loops in the collective

map. For this reason, it may not provide a full representation of social systems, because

typically social systems consist of many actors with significantly diverse viewpoints.

Aggregating is also referred to as “merging” or “overlaying” (Eden, 1989; Eden, et al.,

1983), or the “structural/relational join” operation (Lee et al., 1992), or “combination”

(Kwahk and Kim, 1999).

The congregate mapping approach centers on the identification of key causal loops that

drive system dynamics (Bougon, 1992). The study of causal loops or cycles in causal

mapping and causal modeling has been emphasized by systems dynamics researchers

(Bougon et al., 1990; Forrester, 1961; Senge, 1990). To these researchers, if a causal map

or model is used to represent a social system, causal loops are essential elements that

are responsible for the system’s identity and change (Bougon, 1992) and for the system’s

complex behaviors (Forrester, 1961). In the congregate approach, only labels and links

that contribute to forming loops are entered into the collective map. As a result, the

congregate map may be simpler than the sum of individual maps.

In the workshop mapping approach, the focus is on consensual model building at the

group level. Group members exchange their perceptions of a problem situation to foster

consensus (Vennix, 1996). Workshops are group meetings where the group as a whole

builds a model aided by a facilitator. The purpose of the workshop is to reach agreement

on what elements should be entered into the collective map. As a result of group

discussion and interaction in the workshop, the workshop collective model is expected

to be shared among group members. In some cases, individual maps are not used in the

workshop method, but the facilitator leads the group in building a collective map from

scratch.

Hypotheses

With respect to comparing the three methods for developing collective maps we propose

two hypotheses regarding map complexity and the perceived utility of collective maps.

Researchers in organizational cognition tend to use some simple analyses to measure

causal map complexity. The simplest form of map analysis is based on the number of

nodes. This approach suggests that the more nodes (or constructs) in a map, the more

complex is the map. Eden and Ackermann (1992) note that the number of nodes should

be treated with great care as a measure of complexity, because the number of concepts

surfaced depends on the interviewing skills of the map builders and the length of

interviews. The major weakness of this measure is that it does not include the total number

of links in a map, which is associated with the density of the map. Eden and Ackermann

(1992) suggest that links-to-nodes ratio (L/N) better represents a map’s density. A higher

links-to-nodes ratio “indicates a densely connected map and supposedly a higher level

of cognitive complexity.”

Hart (1977) proposed an alternative measure of map complexity—map density— as a

measure of degree of interconnection. It is measured by dividing the total number of links

by the maximum possible number of links (L/N(N-1)). Klein and Cooper (1982) use map

density to measure the cognitive complexity of decision makers. In a set of maps they

constructed, they found that maps with largest densities were also the three smallest

maps. In smaller maps the concepts tend to be of central importance to the situation, thus

the decision makers acknowledge many relationships between them, making the maps

dense.

Building a map by aggregation involves combining individual maps so that all concepts

and links in the individual maps are included. When the number of group members

increases, complexity of aggregate maps will increase enormously. Thus, the aggregate

mapping method is expected to produce collective maps with the highest degree of

complexity. In contrast, congregate mapping samples a set of links and nodes from

individual maps by identifying key loops or cycles in one or more individual maps, which

should result in a simpler representation than the aggregate map yields. When the number

of group members increases, the complexity of congregate maps will increase slowly.

Thus, the congregate mapping method is expected to produce collective maps with the

Table 1. A comparison of different approaches to building a collective map

Aggregation Congregation Workshop

Work that has

used the

approach

Lee et al. (1992),

Kwahk and Kim (1999),

Eden, et al. (1981), and

Eden (1989)

Bougon (1992)

Hall (1984),

Diffenbach (1982)

Langfield-Smith (1992),

Massey and Wallace (1996),

Vennix (1996)

Core processes Joining, merging

individual maps through

common concepts.

Looking for

congregating labels,

forming loops.

Workshop mapping, group

meeting/discussion, consensus

building, group facilitation.

Procedure Unique concepts are

merged directly to the

composite map while

common concepts are

merged taking care not

to introduce conflicts.

Merge two maps at a

time until the group

maps are exhausted.

Use common concepts

as coupling device to

combine two maps.

Individual maps are

connected to form loops

through “cryptic” labels,

which are repeatedly

used by the subjects.

Individual maps remain

separate, intact in the

congregate map.

Group members add concepts

(or labels) and relationships (or

connections) between concepts,

discuss and decide whether

they should be included in the

group map under the guidance

of a group facilitator.

Applications Organizational memory

(Lee et al., 1992),

Business Process

Reengineering (BPR)

(Kwahk and Kim,

1999). Distributed

decision making (Zhang

et al., 1994).

Strategic planning,

organizational identity

analysis (Bougon, 1992)

Understanding the

dynamics of

organizations (Hall,

1984).

Group decision support

systems (GDSS) (Eden, 1989),

solving messy problems

(Vennix, 1996).

Advantages Merging is simple and

straight forward; It can

be automated with an

algorithm; Conflict

detection.

Loops help understand

the dynamics of the

system; Better at

capturing multiple

perspectives on the

problem situation.

May have more beliefs than

individuals’ maps; Individual

biases can be overcome with

group interaction.

Disadvantages A simple merging of all

maps may not be a

“shared” map;

There is no chance to

mitigate biases in

individual maps.

Difficult to identify the

congregating labels;

Complex and difficult to

automate; Hard to apply

as it requires

perspectives of all

stakeholders.

Premature consensus

(groupthink) due to dominant

perspectives; Unresolvable

conflicts prevent convergence

on single map.

lowest degree of complexity. Workshop mapping relies on the efforts of a group to

identify nodes and links. In view of the limited information processing capacity of group

discussion, a workshop map should also be simpler than an aggregate map. The

workshop mapping method is expected to produce maps with intermediate degrees of

complexity, which depends on the nature and skill of the facilitator. Therefore, we

propose the first hypothesis:

Hypothesis 1: Aggregate maps will be more complex than either congregate or workshop

maps.

Subjective measures have been used in the literature to evaluate and compare maps. For

example, Nicolini (1999) used the subjects’ feedback/knowledge to compare maps.

Massey and Wallace (1996) used the knowledge of a panel of experts to judge the maps

using a Multi-Attribute Value (MAV) model (Massey and O’Keefe, 1993; Massey and

Wallace, 1996; Sakman, 1985). The MAV model consists of five attributes (or criteria):

structure, stakeholders, solution implication, level of focus, and clarity. The aggregate

measure was the weighted sum of the scores on these attributes. In this chapter, we use

three attributes: representation of the problem situation, solution implication or direction,

and representation of stakeholders or multiple perspectives to evaluate the

effectiveness of derived group maps.

Workshop mapping relies on a high level of member involvement. By contrast, aggregate

and congregate maps can be built through analysis of individual maps, and thus require

much lower levels of participant involvement. Even in cases when subjects play a role

in developing the aggregate and congregate maps, they must follow a well-codified set

of rules and procedures for developing the map and this will restrict their degree of

involvement in the process compared to workshop mapping. The second hypothesis is

based on the expectation that the degree to which members participate in building a map

is positively related to their evaluation of it:

Hypothesis 2: Subjects will evaluate a map developed using the workshop approach more

favorably than with those developed using the congregate or aggregate approaches.

We conducted two studies to compare the three mapping methods and to test the two

hypotheses.

Research Design

Two studies were conducted to carry out the comparison of the three approaches.

Research designs for the two cases were similar in general respects, but they differed

in some details. Some of the differences between the two cases are highlighted in

Table 2.

As the detailed designs were different for the two cases, we did not expect that the two

cases would produce exactly similar results, but we expected that the patterns fund in

the two cases would be comparable. We believe that to the extent similar patterns emerge

from different studies we can have more confidence in our conclusions. Conversely, if

we find different patterns for the two studies, this highlights areas that require more

nuanced judgment and further research. Study 1 commenced before Study 2. Some of the

lessons learned (such as the procedure, questionnaire design, etc.) from Study 1 were

incorporated into Study 2. It happened that Study 2 was completed before Study 1. Thus

some insights gained from Study 2 were also fed back to the conducting of Study 1.

Study 1

The objective in Study 1 was to develop an understanding of the causes of a problem

in HALONG, a Vietnamese retail organization, through building a collective causal map.

Established as a private company in 1986, HALONG has about 120 employees and

manufactures and distributes construction products in Vietnam. It has three plants, near

Ho Chi Minh City, in DaNang and in Can Tho and its annual revenue is 10 million USD.

According to the President of HALONG, the company grew steadily for the period from

1990 to 1996, but sales had been declining from 1996 to 2001. The objective of the mapping

Table 2. Comparison of the two studies

HALONG (Study 1) HOUSTON (Study 2)

Problem/issue A particular problem was identified

based on discussion with the

management team. The problem we

arrived at was: “Sales situation and

factors that affect sales at the

organization.”

The problem was developing a model of

the relationship of infrastructure growth

to quality of life of Houstonians based on

the expertise of the research team, which

was composed of scholars of different

disciplines.

Groups involved

in the study

Six groups were formed. Two groups

were assigned to each of three

treatments. Each treatment was one of

the methods of building collective

causal maps. Four subjects were in

each group.

There was one group, which was the

research team, but it was composed of

scholars with different perspectives on the

problem.

Limitations of

factors and

relationships

The number of factors in individual

maps was limited to 15 and the number

of relationships was limited to 35. We

wanted subjects to focus on important

factors and their relationships.

There was no attempt to limit the number

of factors and relationships in individual

maps and group maps. We wanted a rich

picture of the issue.

Data sources for

the models

Only subjects’ reports of their cognitive

constructs were used to build the

models.

A variety of data sources were used to

build the models, including relevant

literature, interview transcripts with

infrastructure decision makers, and the

subjects’ interdisciplinary knowledge.

Treatments The aggregate and congregate methods

were designed to include group model

construction. For all three methods,

group members worked together to

construct their collective map.

The aggregate and congregate methods

were designed not to include group model

construction. The aggregation and

congregation of individual maps were

carried out by the researcher. The subjects

only participated in the workshop

method.

process was to develop a collective understanding of the causes underlying the decline

in sales. Different members of HALONG were expected to have different degrees of

familiarity with the issue and to have different perspectives on the problem. After some

discussion with managers in HALONG, the issue was framed in terms of creating a causal

map depicting the “sales situation and factors that affect sales at HALONG.” The

relevance of this problem to the livelihoods of the subjects ensured their involvement

in the study.

All three methods were employed to construct the collective maps. The mapping process

consisted of two stages: individual causal mapping and group mapping. Individual maps

were obtained for three reasons. First, they provided the basis for comparison with the

collective maps. Second, the individual stage provided the subjects with an opportunity

to learn and become accustomed to the mapping method and the researchers. Finally, the

individual maps improved the quality of the group mapping process by encouraging

members to advance their own individual ideas.

The specific process was incorporated into a five-step experimental procedure: (1)

Subjects were asked to identify lists of factors important to understanding the problem;

(2) Based on these lists, subjects created their own causal maps; (3) The groups built

maps using the particular method assigned to them; (4) Subjects were then asked to

update their own causal maps if they wished; and (5) Subjects completed the questionnaire

and were debriefed. These steps will now be described in more detail.

(1) List of factors. Subjects were first asked to identify variables that may be used to

describe the problem. In the HALONG case, the problem variables most commonly

identified were sales, profit, and customer satisfaction. Subjects were asked next

to identify causal factors that have impacts on the problem variables and to identify

consequent factors that the problem may have impacts on. This was facilitated with

an open questionnaire that had blanks for the subjects to list factors in.

(2) Individual causal maps. Individual causal mapping sessions were held with all

subjects in the conference room at their workplace. We followed the procedure

described in Markoczy and Goldberg (1995) to capture individual causal maps1.

This procedure is well documented and easy to follow for the subjects. Subjects

were asked first to identify possible sensible influences among factors and specify

the type, the sign, and the strength of influences. After that they were asked to

review their maps and revise them wherever applicable. Specific guidelines for map

revision (also included in the questionnaire) were provided. Finally, subjects were

asked to identify important feedback loops in their maps. During the individual

mapping process, the researcher and two colleagues were available for questions

and guidance. The result of this step was a collection of individual causal maps

about the problem.

(3) Group maps. The subjects met for about 60 - 90 minutes to build group maps. Each

group member was given a detailed instruction sheet that they were expected to

follow. The researcher and two of his colleagues were present to answer questions

and to make sure subjects followed the instructions. Summaries of the group

process for each method are as follows:

The Aggregate method. The subjects took turns: i) introducing a factor with a brief

description if needed (determined by other members); ii) a member in charge wrote

down this factor on a sheet and asked other members whether they agreed to

include this factor in the group map; iii) the group then decided whether the factor

should be included (if all members agreed, the factor was entered into the group

map; if a majority did not favor it the factor was left out; if there were mixed opinions,

the factor was marked on the group map with a different color for a second round

of discussion); and iv) the process was repeated until all new factors were

exhausted. During the factor entry process, relationships were entered into the

group map in a similar manner. At the end, the groups were asked to revise their

maps by discussing the marked elements. These elements remained in the group

maps if group members agreed (including if they agreed that these elements were

relevant from other perspectives that they had no knowledge about). Otherwise,

the marked elements were left out. Groups were encouraged to add new information

into the group map as a result of group discussion and interaction.

The Congregate method. The subjects took turns: i) introducing a causal loop with

a brief description if needed (determined by other members); ii) a member in charge

recorded this loop on a blackboard and asked other members whether they agreed

this loop should be included in the group map; iii) the group then decided whether

the loop should be included (if all members agreed, the loop was entered into the

group map; if a majority did not favor it the loop was omitted; if there were mixed

opinions, the loop was marked on the group map with a different color for a second

round of discussion) and; iv) the process was repeated until no more loops were

identified. During the loop entry process both factors and relationships were

entered into the group map at the same time. At the end, the groups were asked to

revise their map by discussing the marked loops or elements. These loops remained

in the group maps if group members agreed (including if they agreed that these

elements were relevant from other perspectives that they had no knowledge about).

Otherwise these loops were omitted. Groups were encouraged to form new loops

in the group map as a result of group discussion and interaction.

The Workshop method. The subjects took turns describing the problem situation

by identifying problem variables, consequence factors, and causal factors that

affect the problem variables, and causal relationships between them. The facilitators

recorded these factors/variables and their relationships on blackboards and

asked other members whether they agreed to include these elements in the group

map and agreed with the story being told. If members all agreed, these elements were

put on the group map. If a majority did not advocate inclusion, they were left out.

If there were mixed opinions, they were marked with a different color for a second

round of discussion. The process was repeated until element entries were exhausted.

At the end, the groups were asked to revise their map by discussing the

marked elements and considering reducing the number of zero in-degree and outdegree

nodes. The agreed-upon elements remained in the group maps and elements

on which there was disagreement were left out.

(4) Individual map update. After their group meeting, the subjects were asked to

revise/update their individual maps (either on the questionnaire or on the diagram/

map) to include any insights they had from the group meeting and from viewing the

group map.

(5) Post experiment. The subjects were asked to give feedback on the experiment via

a questionnaire. This questionnaire gathered information that was used to calculate

the subjective measures to evaluate the collective causal maps, described

below.

The treatments for this experiment were the three methods of deriving collective maps:

aggregate mapping, congregate mapping, and workshop mapping. Subjects were selected

from employees of HALONG who work in the sales, production, and accounting

departments. They were randomly assigned to groups A, B, C and D. Groups A and B

were assigned to the aggregate method, and groups C and D to the congregate method.

Due to time restriction, sales reps were randomly assigned to groups E and F (the

workshop method).

Measures

The collective maps were compared in terms of both objective measures and subjective

measures. Objective measures consist of map complexity and distance ratios between

collective models and individual models. These were calculated by the researchers based

on comparison of the maps. Subjective measures were based on participants’ ratings.

These included the degree to which the maps gave a full and accurate representation of

the problem representation, the degree to which the maps suggested effective solutions

to the problem, and the degree to which the map fairly represented different stakeholders’

views. These subjective attributes are based on a Multi-Attribute-Value (MAV) model

developed for evaluating causal maps (Massey and O’Keefe, 1993; Massey and Wallace,

1996; Sakman, 1985). These responses were elicited from subjects with the questionnaire

that was filled out in the final step of the study. Indices calculated from these responses

were used as criteria to measure the effectiveness or utilities of the derived collective

maps. Because the two studies used quite different samples and addressed different

problems, the questionnaires and procedures for administering them differed, and details

of this are given in the description of Study 2.

To analyze the results of Study 1, we used ANOVA, nesting subjects within groups to

evaluate the impact of grouping and the impact of the method used on the effectiveness

of the composite map in understanding the problem at the individual level. At the group

level we were not able to employ statistical tests, but looked for patterns in results.

Table 3. Grouping and treatments

Treatments Aggregate method Congregate method Workshop method

Group A (5 subjects) Group C Grouping (5 subjects) Group E (5 subjects)

Group B (5 subjects) Group D (5 subjects) Group F (5 subjects)