Philosophical Preamble

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 

It is important to note at the outset that there is a wide spectrum of views concerning the

ontological status of causal maps (and cognitive maps more generally). In this section

we outline some of the main perspectives and clarify our own position.

In their attempts to capture information systems expertise, Nelson, Nelson and Armstrong

(2000b, p.1) point out that it is not possible to literally “open the expert’s head” and extract

Figure 1. Schematic overview of the principal stages of the causal mapping process,

as reviewed in the chapter




of cause maps

Analysis of

cause maps



comparison of

cause maps

domain knowledge as represented directly in the human brain. To the extent that such

a true one-to-one correspondence is unattainable, it follows that methods are required

that can represent knowledge in ways that capture the essence of actors’ thoughts and

belief systems. This philosophical distinction between “causal maps” and “revealed

causal maps” is an important one, reflecting fundamentally different schools of thought.

In the context of his work on political elites, Axelrod (1976, p.10) maintained that a valid

map does not necessarily have to be consistent with a person’s private beliefs. Indeed,

the overall research strategy advocated by him in his seminal volume was “to base what

is being measured on what is being asserted rather than what is being thought by a

person.” In keeping with this stance, a number of organizational researchers (e.g., Eden,

1992; Laukkanen, 1998) maintain that causal mapping need not necessarily be linked with

the cognitive map construct — as developed in the field of psychology — to be a useful

tool for summarizing and communicating information. Viewed from this perspective,

causal maps are a meaningful way of representing elements of the thoughts (rather than

the thinking) of an individual (or group), expressed in the form of a system of causal

relations. For others, however, causal maps are viewed as more than a mere methodological

tool and/or decision-aiding technique, being capable of representing an individual’s

literal beliefs concerning a particular domain at a given point in time (Langfield-Smith &

Wirth, 1992), with the potential to have the same essential characteristics as thought

itself (Huff, 1990).

Our own position falls somewhere between these philosophical extremes. We view causal

mapping techniques (and other forms of cognitive mapping procedure) as one method

for accessing the thinking of individuals in applied settings, adding to the general stock

of knowledge elicitation and knowledge representation techniques — such as those

discussed in Hodgkinson and Sparrow (2002), Shadbolt and Milton (1999), J. Sparrow

(1998) and Schraagen et al. (2000) — more widely available for use in a variety of contexts.

The overall degree of literal correspondence between the data generated by such

procedures and the human information processing system that ultimately underpins

cognition is of secondary importance, relative to the insights they yield into organizational

life. As expressed by Nelson et al. (2000b, p.1): “Theory building is a cumulative

rather than exhaustive process.” To the extent that cognitive mapping procedures (of

whatever form) give rise to findings as predicted by rigorously derived hypotheses

grounded in well-supported management and organization theory, all well and good. To

the extent that such predictions are also supported by theory and research from the

cognitive sciences, even better (cf., Scheper & Faber, 1994).

Another important issue is that of how actors’ collective belief systems might be

captured most appropriately. To what extent is it meaningful to represent “shared beliefs”

and how? Again, theorists and empirical researchers are divided on this issue, reflecting

fundamental differences not only regarding the ontological status of cognitive maps but

also the status of collective cognition. According to Scheper and Faber (1994), while

certain forms of causal map are able to represent meaning at the individual level, this is

not the case at the collective level. In respect to the latter, they advocate an alternative

approach, based on semiotic analysis. In the words of Fiol (1989, p.278), citing Eco (1979):

“Semiotic analysis is a formal mode of analysis used to identify the rules that

govern how signs convey meanings in a particular social system…semiotics

assumes that diverse signs or expressions can convey shared meaning because

they are grounded in a common set of underlying values.”

We view Scheper and Faber’s stance as premature at this stage in the development of

the managerial and organizational cognition field. As noted by Cannon-Bowers and Salas

(2001), in a discussion of shared cognition in the context of team functioning, there are

a number of pressing issues upon which researchers have yet to reach basic agreement,

not least questions concerning what it is that is actually shared, what sharing means, how

sharing might most appropriately be measured and the nature of the outcomes that might

be expected as a result of shared cognition.

In summary, contemporary theorists and empirical researchers are divided on two

fundamental issues: (1) the nature and purpose of causal and other forms of cognitive

mapping techniques, and (2) the nature of collective belief systems. Further consideration

of these issues is beyond the scope of the present chapter, but sufficient detail has

been provided to serve as a useful backdrop for understanding the range of alternative

choices confronting would-be users of causal mapping techniques.

Over the years organizational researchers have devised a variety of alternative methods

for the elicitation, analysis, and comparison of actors’ individual and collective causal

belief systems. We turn now to provide a summary of the many developments that have

occurred in relation to these key, non-mutually-exclusive activities, each of which is

fundamental to the mapping enterprise, commencing with the process of knowledge


Approaches to Knowledge Elicitation

Despite the widespread popularity of causal mapping techniques, there is currently no

consensus within the literature concerning the most appropriate way(s) to elicit actors’

causal belief systems (Hodgkinson & Sparrow, 2002; Jenkins, 1998). Following Hodgkinson

(2001) and Mohammed et al. (2000), we shall consider two broad classes of elicitation

procedure: indirect and direct (see Figure 2).

Indirect elicitation techniques entail processes whereby maps are constructed from

secondary data sources, typically extant written documents (including interview transcripts

and letters to shareholders) derived initially for some other purpose then

subsequently analyzed using causal mapping procedures (e.g., Barr &Huff, 1997; Barr,

Stimpert & Huff, 1992), or primary sources in situations in which the data are elicited

specifically for the research project but not in a manner that requires the participant to

reflect on their causal beliefs in an explicit fashion. An example of the latter would the

use of interview transcripts generated in narrative form by the researcher and subsequently

converted into causal maps through a process of post hoc coding (for representative

examples, see Calori, Johnson & Sarnin, 1992, 1994; Jenkins & Johnson, 1997a,

1997b; and Nelson et al., 2000a). The common defining feature of indirect approaches to

knowledge elicitation, regardless of whether the data is gathered from primary or

secondary sources, is that the process of map construction is undertaken without the

active involvement of the research participant. In contrast, direct elicitation methods

require the active involvement of participants in the map construction process from the

outset. Direct elicitation methods include structured questionnaires requiring participants

to evaluate causal relations among predefined sets of variables — also referred to

as elements or nodes1 (e.g., Roberts, 1976; Swan & Newell, 1998) — and the use of

computerized systems such as Decision Explorer(Eden, Ackermann & Cropper, 1992)

that enable maps to be constructed dynamically, in real time, through an iterative

interview process. As we shall see, there is no such thing as a perfect method. Each

approach is characterized by particular strengths and weaknesses.

Indirect Elicitation Procedures

In point of fact, the initial approach to causal mapping entailed the use of secondary data

in conjunction with indirect methods of elicitation. (By definition it is impossible to

combine direct elicitation methods with secondary data, unless the researcher is reanalyzing

pre-existing maps from earlier studies.) Axelrod’s (1976) preference was that

cognitive maps be derived from whatever materials are left behind in the normal course

of the decision-making process, on the grounds that although this was potentially

problematic in terms of issues of authenticating the researcher’s interpretation, documentary

evidence is non-intrusive and therefore unlikely to influence participants’

thought processes. Working with documentary evidence also allows the investigator to

gain access to busy individuals who might otherwise be unwilling to participate using

more intrusive, interactive forms of data generation procedures (Huff, 1990). However,

documentary sources are beset with a number of potentially severe limitations in that the

data contained within them is often only of tangential relevance to the investigator’s

purpose(s). Moreover, the fact that secondary source documents, such as letters to

Indirect elicitation and


Direct elicitation and




Ideographic Hybrid

Figure 2. Taxonomy of principal methods for the elicitation and construction of cause


shareholders, by definition, are prepared for particular audiences renders it difficult if not

impossible for the researcher to ascertain the extent to which any biases contained within

them are genuinely a product of the originator’s sensemaking processes and/or a

deliberate attempt to influence the perceptions of the stakeholders to whom they were

initially directed, a problem which is compounded by the fact that the data emerging from

the use of causal mapping in this way can rarely be checked for “accuracy” and validated

against comparable data from objective, independent sources (Hodgkinson & Sparrow,


In a number of respects, the tasks associated with the coding of primary data originating

from interview transcripts in narrative form using causal mapping techniques are similar

in nature to the process of coding secondary documents. However, a major advantage

of the former is that the data are obtained specifically for the researcher’s own purposes,

thus circumventing, to a certain extent at least, the authenticity problem alluded to above.

Nevertheless, there are still significant risks of bias, not least due to the potential

influence of demand characteristics arising from the research situation during the

elicitation process (cf., Hodgkinson, 1997b). Moreover, as with maps derived from

secondary source documents, when using primary interview transcripts the researcher

must face the vexed question as to how the maps so derived are to be subsequently

validated. This validation problem is compounded in the case of unstructured documents,

including unstructured interview transcripts, by the associated problems of poor

data quality that often result from using such sources, not least the fact that these

documents typically contain sentence fragments, incomplete thoughts, and overelaborate

explanations (Kemmerer, Buche & Narayanan, 2001).2

Direct Elicitation Procedures

Increasingly, direct methods of knowledge elicitation are being employed by organizational

researchers in the field, both prescriptively, as a basis of intervention through

‘action research’ (Cropper, Eden & Ackermann, 1990; Eden & Ackermann, 1998a; Eden

et al., 1992), and for descriptive purposes, where the object of the exercise is to better

understand the extent to which and in what ways actors’ mental representations of

organizational phenomena are similar to and/or different from one another and isolate the

correlates of such similarities and differences (Markóczy, 1995, 1997, 2001; Markóczy &

Goldberg, 1995). A primary advantage of direct methods over their indirect counterparts

is that they obviate the need for cumbersome coding procedures for map construction

— the maps being constructed in situ, directly from the raw data — and enable the

researcher to focus the data collection on issues of immediate concern to the investigation.

Used in this fashion, causal mapping techniques are akin to knowledge elicitation

techniques employed more generally within the cognitive and organizational sciences.

Direct elicitation procedures can usefully be sub-divided in terms of the extent to which

the elicitation process requires participants to identify the variables to be causally

mapped, using their own everyday natural language, or whether the subject matter is

supplied by the researcher, on the basis of extant theory and research or an a priori

conceptual analysis of the domain to be mapped. In the case of the first approach, known

as ideographic elicitation, the primary concern of the researcher is to ensure that

valuable richness and detail in individual cognition are not lost or threatened by

researcher bias. This approach can be traced to the personal construct theory of George

Kelly (1955), which asserts that individuals are inherently unique in the ways in which

they construe their worlds. Accordingly, if we are to gain insights into participants’

beliefs, it is vital that the elicitation process does not impose concepts that are alien in

meaning. In Kellyian terms the elements involved in any mapping exercise must fall within

the participants’ “range of convenience.” Kelly devised a particular approach to

cognitive mapping, the repertory grid technique, which lies beyond the scope of this

chapter (for details and applications, see Daniels et al., 2002; Fournier, 1996; Huff, 1990;

Reger & Huff, 1993). Within the realm of causal mapping, Eden and his colleagues (e.g.,

Eden & Ackermann, 1998a, 1998b; Eden, Jones & Sims, 1979, 1983) have devised a system

of elicitation that is derived ultimately from personal construct theory. Laukkanen (1994,

1998) also strongly advocates that causal maps should be elicited in a manner that

enables participants to express their thoughts using their natural language. In this

connection, a prime strength of documentary sources, particularly interview transcripts

gathered in situ, is that they are expressed in their natural language form. The same is

true of certain archival sources. While it is undoubtedly the case that maps in their natural

language form are inherently more meaningful to the individual participants, a major

drawback of this approach is the problems this poses for comparative analysis purposes,

an issue to which we shall return in due course.

The second approach to direct elicitation, nomothetic elicitation, entails the use of

standardized lists of variables supplied by the researcher. A variety of approaches to the

basic task of map construction have been adopted by researchers using this type of

procedure, ranging from highly structured questionnaires involving the pairwise evaluation

of all possible combinations of causal relations (Roberts, 1976; Swan, 1995; Swan

& Newell, 1998) to more basic methods, entailing the hand-drawing of causal maps (Green

& McManus, 1995). Systematically considering all pairwise effects (Swan & Newell,

1998) involves assessing causality by reviewing every possible combination of variables

and should significantly diminish the possibility that important effects are omitted (Hart,

1976). Pairwise comparison is also seen as being particularly helpful in overcoming the

potential problem of coding errors with respect to loops, which tend to be common with

causal maps because of the problematic nature of determining the interviewee’s view

about what is cause and what is effect (Eden et al., 1992).

A major criticism leveled against researcher-standardization of variables for elicitation

purposes by the advocates of ideographic approaches (e.g., Eden & Ackermann, 1998b)

is that researchers run the risk that the basic map construction task might prove

meaningless for participants. However, as we shall see later, there are also some major

advantages to nomothetic approaches, particularly in relation to comparative analysis

in situations involving large numbers of participants, where the aim is to statistically

analyze the maps in order to identify patterns of belief similarities and differences and/

or identify factors that explain such patterns. Moreover, there are a number of strategies

that can be readily adopted to minimize the dangers of lack of meaning alluded to by those

favoring ideographic approaches to elicitation, not the least of which is ensuring that the

final list of variables forming the focus of the mapping exercise are carefully formulated

by recourse to relevant literature and/or the use of expert panels, the members of which

are highly representative of the participant sample involved in the main mapping exercise.

Careful piloting of the requisite elicitation task is also invaluable in this respect. Use of

standardized variables for elicitation purposes not only overcomes difficulties associated

with post hoc coding schemes (which are considered in further detail later in this

chapter), but also minimizes the impact of demand characteristics associated with semistructured

interviews, as discussed above. Ultimately, however, the use of fixed sets of

variables, by definition, limits the extent to which the resulting maps can capture

individual differences in terms of both map content and map structure. The requirement

that participants work with a common set of variables eliminates the possibility of the

detection of individual differences in terms of what is considered to be sufficiently salient

to warrant incorporation into the maps, the inclusion or exclusion of particular variables

not being permitted. Clearly this type of approach suppresses a potential source of

significant variation.

Fortunately, in recent years researchers have begun to develop new approaches that

seek to combine the major strengths of ideographic and nomothetic approaches to

knowledge elicitation, while dispensing with their associated weaknesses (e.g.,

Hodgkinson, 2002; Hodgkinson et al., 1999; Hodgkinson & Maule, 2002; Markóczy &

Goldberg, 1995). We shall consider these developments in detail in a later section when

we review ‘Recent Advances in the Large-Scale Comparative Analysis of Cause Maps’


Basic Metrics for the Analysis of

Individual Cause Maps

The previous section identified the principal methods for eliciting data for the construction

of causal maps and considered their relative strengths and limitations. Having

acquired such data, the researcher must then set about the task of map construction and

analysis. In this section we consider some of the major approaches that have been

devised for these purposes. We shall confine our discussion to a brief consideration of

the various indices that have been derived over the years for analyzing the structure and

content of individual causal maps, as a precursor to a more detailed treatment of issues

concerning the aggregation and comparative analysis of such maps.

In their most basic form, causal maps can be depicted graphically, using the medium of

the influence diagram (Diffenbach, 1982). Adopting this approach, variables are depicted

as nodes in a network, interconnected by a series of arrow-headed pathways, terminating

in each case on the dependent variable(s). The simplest forms are restricted to a

consideration of positive (increases in one variable cause corresponding increases in

one or more other variables), negative (increases in one variable cause corresponding

decreases in one or more other variable(s)), and neutral (no causality implied) relationships.

More sophisticated variants of the technique enable these relationships to be

differentially weighted, on the basis of the participant’s belief strength, for example, or

the degree of certainty/uncertainty surrounding each causal assertion.

As noted earlier, the focus of Axelrod’s (1976) initial work was to explore in detail the

causal influences within individual participants’ maps. As his basis of analysis, Axelrod

used the theory of directed graphs (Harary, Norman & Cartwright, 1965) and represented

each cognitive map as a valency or adjacency matrix. Building on these foundations,

researchers over the years have devised a great many indices for the assessment of map

structure and content and a detailed consideration of these is not possible within the

confines of the present chapter. Given Diesner and Carley’s (2005) extended treatment

in this volume of the relative strengths and weaknesses of particular causal map indices,

we confine our discussion to a highly selective overview of some of the more commonly

employed structure and content measures applicable to most, if not all, forms of causal

maps, as widely used by researchers in an attempt to capture the essence of actors’ causal

belief systems.

Basic measures to assess the content and structure of causal maps have ranged from

simply counting the number of occurrences of particular variables and associated links

(i.e., arrows connecting constructs), through the link-to-node ratio (i.e., number of links/

number of nodes), to map density (i.e., the number of observed links/total number of links

theoretically possible, given the number of variables in the participant’s map). As shown

in Table 1, each of these measures is characterized by particular strengths and weaknesses

of which the would-be user needs to be aware. These measures are foundational

to the comparative analysis of causal maps, to which we turn in the next section.

Aggregation and Comparative

Analysis of Cause Maps

We noted at the outset that theorists and empirical researchers are divided not only

regarding the ontological status of cognitive maps at the individual level analysis, but

are also divided as to the nature and significance of collective beliefs and cognition.

Given the lack of basic agreement concerning the extent to which collective beliefs are

theoretically meaningful as a construct, it will come as no surprise to learn that

researchers are also divided as to how such beliefs might best be elicited and represented

(Mohammed et al., 2000). At the risk of over-simplification, in the present context the

faultline in respect of this issue centers broadly on the relative merits of the aggregation

of actors’ causal belief statements and/or evaluative judgments of causality versus the

systematic, comparative analysis of individuals’ causal maps in the search for patterns

of homogeneity and/or heterogeneity.

As observed in the previous section, individual causal maps can be analyzed along two

principal dimensions: content and structure. Content measures when used for the

purpose of comparative analysis capture key differences in terms of which constructs

individuals perceive as more or less relevant to a given domain and the ways in which

these constructs are perceived to relate to one another. Structural differences, in

contrast, are used to ascertain the relative complexity of the various maps under

consideration. The four basic types of difference that can be identified in the comparison

of cause maps are shown in Table 2.

Table 1. Nature and purpose of some commonly used metrics for the assessment of cause


Measure Description Purpose

Indegrees (

Axelrod, 1976)

The number of links to a given node/variable. Reveal the extent to which the construct in question is

influenced directly by the other constructs in the actor’s

causal belief system.

Measures of Map


Outdegrees (Axelrod, 1976) The number of links emanating from a given


Indicate the extent to which a

construct exerts a direct

causal influence on those other constructs within the

actor’s causal belief system.

Reachability (

Harary et al.,


The sum of all direct and indirect influences exerted

by a given node/variable on the system of perceived

causal relations as a


Indicate the direct and indirect impact of a construct

within the actor’s causal belief system.

Link-to-node ratio (Eden et al.,


The proportion of links to nodes ÷ variables within a

given map. Higher scores denote greater structural


Provides some evidence of cognitive complexity,

individuals whose maps are more interconnected having

a greater understanding of the relationships impacting

upon an issue.

Measures of Map


Map density (

Hart, 1976)

The number of observed links ÷ total number of

links theoretically possible, given the number of

nodes/variables in a given map. Higher scores

denote greater structural complexity.

As above.

Average chain length (Jenkins &

Johnson, 1997a)

A chain refers to a sequential set of perceived causal

links. Average chain length is derived by

calculating the mean length of all complete chains

within a given map. For example, in the case of a

map containing two chains, one comprising four

links, the other two, the average chain length would

be three.

The greater the average chain length, the greater the

elaboration in terms of explanations for the patterns of

events or actions depicted in the map.

Aggregation Procedures for the Analysis of Collective