К оглавлению
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 
119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 
136 137 138 139 140 141 142 143 144 145 146 

Beliefs shared by A, B, C, and D

Beliefs shared by C and D only



















maps at the group and organization levels of analysis. One limitation results from

individuals who use different terms for the same concept (i.e., synonyms). Another

limitation results from the use of the same term for different concepts (i.e., connotations).

Such differences have been found to lead to disagreements over appropriate descriptions

of particular ideas. Similar difficulties arise when individuals have not developed

“a sufficiently comprehensive body of shared language” or cannot agree on “the most

appropriate language with which to describe a particular idea” (Langfield-Smith, 1992).

Moreover, these concepts and word uses may change over time as the individuals learn

about the environment around them, presenting additional difficulties in relating causal

maps elicited across long periods of time. The nuances of human language and thought

and the idiosyncrasies in the styles and rates of individual learning make it difficult to

compare the causal assertions of multiple individuals over time and thus construct

collective causal maps (Langfield-Smith, 1992).

Linguistic, learning, and other social phenomena may not be properly considered when

causal maps are simply linked together. That approach treats node and arc elements as

(in the words of Boland, Tenkasi & Te’eni, 1994) “unproblematic, predefined, and

prepackaged” rather than “subjective” and “interpretive,” and ignores the need to

“provide the conditions for surfacing and challenging important assumptions…, for

complicating their thinking…and for enabling significant change when it is required.” To

adequately represent the linguistic and learning dimensions of the social causal reasoning

problem space, collections of causal maps should exhibit characteristics of “good”

representations. “Good” representations exhibit the following characteristics (Winston,


1. “make the important things explicit”

2. “expose natural constraints, facilitating some class of computations”

3. “are complete…[they] say all that needs to be said [about the problem space at


4. “are concise”

5. “are transparent” (i.e., easy for users to understand)

6. “facilitate computation…[they] can store and retrieve information rapidly”

7. “suppress details” unless requested

8. “are computable by an existing process”

Causal maps exhibit many of these aspects in representing individual causal reasoning.

They are explicit, concise, and relatively transparent. However, they are not so good at

representing social causal reasoning. They do not explicitly reveal nuances in human

language and cognition, which are important considerations in the social construction

of meaning. It is also difficult to represent opposing views of causality within a single

causal map. This means that a single causal map cannot represent this important feature

of social learning. This limitation also limits the computational power of causal maps in

understanding multiple node-arc-node segments or feedback loops (i.e., multiple nodeObject-

arc-node segments that form a chain that returns to the point of origin, thus creating a


Fortunately, there are other types of cognitive maps that more fully represent the

linguistic, temporal, and conceptual dimensions of the problem space of social causal

cognition. Again, the term cognitive map refers to a general class of maps that represent

cognition, understanding, and beliefs, of which causal maps are only one type. Other

types of cognitive maps that can assist in social causal mapping include categorical

cognitive maps showing how concepts and words are related linguistically across

vocabularies and knowledge domains. Associative maps describe patterns of word use,

such as those found through content analysis. Argument cognitive maps represent an

individual’s assumptions, evidence, and reasons that underlie beliefs. Collections of

cognitive maps are thus better able to meet the requirements of “good” representations.

Long-term collections of categorical, associative, and other cognitive maps can facilitate

deeper insight into how organization members reason and develop patterns of causal

belief over time, and provide important tools for making the representation of social

causal thinking more complete. Thus, collections of cognitive maps are better though

nonetheless imperfect representations of the problem space of social causal cognition.

Descriptions of cognitive maps and related terms are summarized in Table 1.

Representing Social Causal Cognition

with Information Technology

The use of multiple types of cognitive maps can address issues of representational

completeness in social causal cognition, but unfortunately that approach exacerbates

the already challenging problem of analyzing large numbers of causal maps. Processing

large numbers of causal maps is difficult enough. For example, Axelrod (1976) found that

Table 1. Definitions

Term Definition Source

Argument Map Represents assumptions, evidence, and reasons that underlie beliefs. Huff (1990).

Assertion An individual’s statement concerning their thoughts or beliefs about

the world, environment, etc.

Axelrod (1976).

Associative Map Inventories concepts and their complexities, and describes patterns of

word use.

Huff (1990).

Cognitive map A general class of physical representations of thoughts or beliefs.

These maps can represent individual assertions, or those elicited from

a group.

Huff (1990);

Montazemi and

Conrath (1986).

Causal map A sub-class of cognitive maps that focuses on the representation of

causal beliefs; a network of causal relations embedded in an

individual’s explicit statements, an explicit representation of the deeprooted

cognitive maps of individuals.

Huff (1990); Nelson,

Nadkarni, Narayanan,

and Ghods (2000).

Categorical map A sub-class of cognitive maps that focuses on relationships of

similarity (e.g., a map linking word synonyms).

Huff (1990).

elite foreign policy decision makers “employ rather large structures in presenting their

images of their policy environment,” citing three maps whose sizes were 31 nodes and

43 arcs, 53 nodes and 84 arcs, and 73 nodes and 116 arcs. Analyzing dozens of similarly

sized causal maps is a daunting task. This problem would be exacerbated by including

association, categorical, argument, and other cognitive maps into such analyses.

Set-Theoretic Representations of Causal Mapping

The increased computational complexity of the social causal mapping problem space

requires a concurrent increase in the representational power of the corresponding

information system. Accordingly, Winston’s (1984) characteristics of “good” representations

(i.e., make important things explicit, expose natural constraints, completeness,

etc.) can be applied to information systems representing cognitive maps, just as they

have been applied to cognitive maps representing human cognition.

While computer programs have been available for some time to create and evaluate

individual causal maps, most of those systems store individual maps in separate files

whose contents are not easily integrated. This shortcoming severely restricts their use

for social causal mapping. Other information systems that have been used to represent

cognitive maps are set-theoretic relational databases, i.e., those based on the axioms and

mathematical theories about sets (Codd, 1970). Such databases exhibit many of the

characteristics of “good” representations: they are explicit about their important features

(e.g., the membership relationship between an element and a set) and clear about their

constraints (e.g., they cannot express operations that produce transitive closure over

binary relations) (Ullman, 1988). They are also concise, transparent, efficient, and

abstract (i.e., details can be hidden unless requested). Some processes (e.g., inventorying

organizational knowledge) can be addressed efficiently via set-theoretic algorithms

similar to those now employed in some knowledge management systems, e.g., “knowledge

Yellow Pages” (Davenport & Prusak, 1998).

Unfortunately, set-theoretic approaches do not fully reflect the graph-theoretic nature

of causal and other cognitive maps. That is, the relationships between elements in settheoretic

approaches correspond to membership in a set, while those in graph-theoretic

approaches correspond to one element “pointing to” another. For example, determining

the shortest path between two non-adjacent nodes in a causal map cannot be done with

set-theoretic approaches because set theory does not represent the logic necessary to

travel across—or traverse—multiple node-arc-node segments. This limitation is important

because it restricts the types of analysis that can be performed (i.e., it contradicts

the representational requirement for the existence of computational procedures). These

analyses include: tracing paths along multiple nod-arc-node segments, summing the

effects of a series of direct and inverse relationships, and identifying feedback loops.

These graph-theoretic procedures cannot be performed efficiently—if at all—by settheoretic

relational databases on long series of node-arc-node segments. Differences

between set and graph theories are summarized in Table 2.

Graph-Theoretic Representations of Causal Mapping

Computer-based systems designed upon graph-theoretic representations and algorithms

(Smith & Smith, 1977a, 1977b; Ullman, 1988) are a more promising approach. The

cognitive maps in Figure 1 can again serve as an example. The map in Figure 1 represents

assertions from two individuals about their conceptualizations and perceptions of risk.

Since it contains the causal assertions of multiple individuals, it can be considered as a

social causal map. Consider the problem of identifying and comparing the individuals’

causal assertions between IS security procedures and risk. The traversal algorithm must

identify two and only two paths of node-arc-node chains:

1. IS security procedures __ “acceptable IS use” training −→litigation

settlement costs __ risk

2. IS security procedures __ updating firewalls−→damage from virus attacks

loss of data __ risk

where __ represents synonymous (i.e., categorical) relationships, →and−→

represent causal relationships of direct and inverse proportionality, respectively.

Graph theory facilitates the analysis of cognitive maps because it has long supported the

development of algorithms that can efficiently and effectively travel across the repetitive

“node-arc-node” structures of graphs (i.e., node-arc-node-arc-node-arc-node-etc.).

Other programs can build upon these traversal algorithms to assess the total effects of

chains of node-arc-node segments. For example, a computer program can be developed

to support the query whether two individuals agree if IS security procedures and risk are

directly or inversely proportional (e.g., the parallel paths 1 and 2 above). The program

can be conceptualized as follows: 1) use a traversal algorithm to identify parallel nodearc-

node chains between IS security procedures and risk, regardless of whether the arcs

are causal and linguistic (i.e., word categorical); 2) sum the “+” and “-” values of the

Characteristic Set Theory Graph Theory

Relationships between


Based on membership in a set Based on one element “pointing to”


Pictorial representations Venn diagrams Directed graphs, workflow diagrams

Examples of computer


Relational databases


Microsoft® Access)

Query languages such as SQL

Work-flow simulation programs



Object-oriented languages



Some object-oriented languages (e.g.,

Java) can be written to draw data from

relational databases.

Table 2. Set theory versus graph theory

causal arcs in iterative fashion during traversal; and 3) compare the sums of causal values

of the two paths. If the two final sums are equal, the individuals share the same belief.

If the final two sums are unequal, the individuals have contradictory beliefs.

The previous discussions about social causal reasoning and mapping in Sections 2 and

3 can be summarized in the three following design criteria:

1. Social causal mapping systems must represent the multiple types of cognitive maps

that reflect social causal cognition.

2. Social causal mapping system must integrate and process large numbers and many

types of cognitive maps that will result from working with large numbers of

organization members.

3. Social causal mapping system should retain the maps in such a way that incompatibility

problems within and between long-term storage units are eliminated.

These three design criteria should be met in the context of providing “the conditions for

surfacing and challenging important assumptions…, for complicating their thinking…and

for enabling significant change when it is required” (Boland et al., 1994).

Enhanced Representation of Cognitive Maps: An Object-

Oriented Approach

The directed cause-and-effect structure of causal maps is more consistent with graphtheory

than with set theory. Relational databases, being based on set theory (Codd,

1970), are thus ill-suited to the task of representing the complexity of causal mapping.

Object-oriented systems, on the other hand, are better suited because of four major

elements of object-oriented modeling: abstraction, encapsulation, modularization, and



Abstraction is perhaps the most important element in object-oriented development.

Abstraction has been defined in many ways, but perhaps the most succinct is a “selective

emphasis on detail” (Shaw, 1984). Abstraction is particularly useful in IS analysis and

design because many of the problem spaces faced by IS professionals are complex and

messy, and abstraction allows the analyst to subdivide a complex problem into workable

segments. Abstraction facilitates this strategy by permitting the analyst to focus on

those facets of the problem segment that are important, and ignore those that are not.

In the domain of social causal mapping, abstraction permits the designer to focus on one

type of cognition at a time (e.g., causal assertions or linguistic relationships between

words) and to derive specifications based on that type of reasoning for its corresponding

cognitive map. One particular form of abstraction, called aggregation, is particularly

useful in social causal mapping because it supports compositional views of nested

objects (i.e., nodes and arcs are nested within feedback loops, which in turn may be

nested within a collection of individual causal maps).