Conclusions and Recommendations

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 

The central message of this chapter is that, as with any other cognitive mapping/

knowledge elicitation technique, the would-be user of causal mapping procedures faces

a series of inter-related issues and choices that have a direct bearing on the type of data

that can be gathered, the sorts of analyses that can be conducted, and what inferences

that can be drawn. These issues apply equally regardless of whether the work is being

undertaken for policy-making/intervention purposes, or in an attempt to capture actors’

beliefs in the context of theoretically driven empirical research.

The question as to what constitutes the most appropriate methodological choices in

causal mapping research can only be answered by carefully considering the precise

nature of the inquiry being undertaken and the context(s) in which the investigation is

taking place (cf., Daniels & Johnson, 2002; Hodgkinson, 2002). As we have seen, causal

mapping procedures have been adapted in a variety of ways over the years, particular

approaches having evolved in response to demands for data in forms suitable for

addressing particular sorts of research questions, taking due account of the practical

constraints imposed by specific research settings. Clearly, however, these developments

represent more than a set of mere pragmatic reactions to prevailing circumstantial

contingencies. As noted at the outset, researchers are divided along clear ontological

faultlines regarding the fundamental nature and status of causal maps (and other forms

of cognitive map) and collective cognition. The particular approaches we have reviewed

are as much a manifestation of the underlying ontological assumptions of their advocates,

upon which they are predicated, as they are solutions to what are essentially

mundane practical problems, such as the need to gain site access with minimal intrusion,

the requirement for robust data, and so on. It is the combination of methodological

differences in underlying ontology and the non-trivial pragmatic issues such as access

requirements that are the main determinants of which particular research questions are

pursued and how they come to be formulated in the first place (cf., Easterby-Smith,

Thorpe & Lowe, 1991; Gill & Johnson, 1991; Jenkins, 1998). Given this complex state of

affairs, what concrete recommendations are we able to make for the IS and IT research

communities that might assist the potential user of causal mapping techniques?

In the final analysis, we ourselves are advocates of a Pragmatic Science approach to

knowledge production, which entails the pursuit of research questions directed toward

the development of insights that are both theoretically and methodologically robust on

one hand, but also of high practical relevance on the other (Anderson, Herriot &

Hodgkinson, 2001; Hodgkinson & Herriot, 2002; Hodgkinson, Herriot & Anderson,

2001). Skillfully adopted, this philosophy will yield actionable knowledge (Argyris,

1999), i.e., knowledge that is both academically rigorous and contributes directly to the

enhancement of employee well-being and organizational effectiveness. Use of the term

‘science’ in this connection is not meant to imply that we are advocating the wholesale

abandonment of in-depth, qualitative approaches in favor of larger-scale hypotheticodeductive

ones. Nor should the ‘pragmatic’ element of our approach be taken to imply

the adoption of sub-standard theory and methods in order to generate immediate

solutions to the most pressing practical issues of the day. On the contrary, as explained

by Anderson et al. (2001): “there is a need to broaden our search for, and the acceptance

of, methodological alternatives that meet the twin imperatives of rigour and relevance.”

‘Scholarly consulting,’ as advocated by Argyris (1999), major elements of which have

been termed action research, fall within this definition of pragmatic science, as potentially

do all of the approaches to causal mapping reviewed in this chapter.

The overriding necessity, from our point of view, is that researchers using causal

mapping methods make choices that are both internally consistent with one another and

commensurate with the requirements of the research question under investigation. As

researchers we are trying to get as close to the worldviews of participants as our

(imperfect) techniques will allow. Techniques that impose too much structure will stifle

participants, whereas procedures that fail to provide sufficient structure will yield overly

elaborated data. Both are potentially problematic, but the extent to which each is actually

a problem in practice is a function of context and the nature of the research question to

be addressed (Hodgkinson, 2002; Hodgkinson et al., 2004). For instance, if the aim of

research were to try and capture the dynamics of cognition in real time, such as in

applications seeking to sample the causal beliefs of IT users on the Internet, how would

one set about studying this? One way would be to go down the ideographic route, as

championed by Eden and his associates. This would require the researcher to take

repeated snap shots of small numbers of participants as their maps evolved. Another

approach would be to have them make decisions then immediately try and capture the

complexity of their thinking as fully as possible, using highly structured elicitation

techniques, such as the Pathfinder network approaches reviewed in Gillan and

Schvaneveldt (1999). In this context, as with all applications of causal mapping (and any

other knowledge elicitation and knowledge representation procedure), the adopted

choices must depend on what one is trying to do with the data.

Ultimately, researchers must make a tradeoff between depth and richness of insight on

one hand and comparability and generalizability on the other hand (Hodgkinson, 2002).

In situations where there is a fundamental requirement for greater depth and richness of

insight into the thoughts of individual participants, ideographic approaches to elicita30

tion and map construction are the order of the day. Clearly, however, these are not

suitable for use in situations where large-scale comparisons and generalizability of the

findings are fundamental prerequisites, not least due to the unreasonable coding

burdens placed on the researcher, leading in turn to fundamental concerns with regard

to reliability and validity. While the nomothetic alternative of providing all participants

with an a priori standardized list of variables has been criticized on the grounds that

potentially this might yield less salient data (Eden et al., 1992), the implication being that

the researcher’s subjectivity rather than that of the participant overly determines the

nature of the data obtained, it is clear that data transformation processes as employed

by ideographic researchers also entail a considerable amount of researcher subjectivity,

despite the development of techniques to enhance inter-coder reliability (c.f., Huff,

Narapareddy & Fletcher, 1990).

In sum, as observed by Jenkins (1998), there needs to be some level of tradeoff between

fully capturing data which is meaningful to participants and ensuring that data is elicited

in such a manner as to ensure sufficient commonality, so that comparisons of causal maps

are meaningful. Hybrid elicitation procedures, such as those devised by Hodgkinson et

al. (1999) and Markóczy and Goldberg (1995), were developed in an effort to strike a

balance between these competing requirements. As we have seen, they are especially

promising in a number of research contexts, since by allowing choice within pre-specified

limits (participants choose variables to be mapped from a menu) the data is not only more

meaningful for the individual participants concerned, but also comparable across

multiple levels of analysis, without the necessity for elaborate coding procedures of

dubious reliability and validity (Hodgkinson, 2002).

While in principle the Markóczy and Goldberg (1995) procedure could prove highly

suited to the collection and comparative analysis of much larger-scale datasets than has

been possible hitherto, a number of software limitations have prevented its wider

adoption and all applications within the extant literature having been authored by its

originators (e.g., Markóczy, 1995, 1997, 2001). Fortunately, however, as discussed in the

Appendix, software currently being evaluated by the present authors looks as if it will

rectify these limitations. Repeated trials have shown that the Windows-based system is

capable of performing all aspects of the Markóczy and Goldberg procedure — and the

earlier approaches to the comparison of cause maps devised by Langfield-Smith and

Wirth (1992) — in real time, within highly demanding workplace settings.

As this review of methodological advances in causal mapping has demonstrated, the

study of managerial and organizational cognition is complex, but it is this very complexity

that makes it such a challenging and exciting endeavor. The introduction of causal

mapping techniques to the IS and IT communities at this particular juncture is highly

fortuitous. The large volume of work that has been undertaken in the fields of strategic

management, and management and organization studies more generally, means that IS

and IT researchers are inheriting a rich legacy. The gathering and analysis of large-scale,

multi-level longitudinal datasets — much needed for the scientific advancement of many

areas of application, but which have thus far eluded all but a handful of scholars — is

now within our wider methodological capabilities. Much has been accomplished, yet

there is still much to do, and researchers in the IS and IT fields are eminently well placed

to contribute to the advancement of cognitive mapping theory, method and practice.

Notes and Acknowledgments

The financial support of the UK ESRC/EPSRC Advanced Institute of Management

Research (AIM) in the preparation of this chapter (under grant number RES-331-25-0028)

is gratefully acknowledged. We are also grateful to V.K. Narayanan and two anonymous

referees for their helpful and constructive comments on an earlier version of our

manuscript.

1. This lack of agreement over basic nomenclature is unfortunate, leading to frequent

confusion.

2. Recently, Maule et al. (2003) have reported a laboratory experiment in which

participants were required to record their thoughts by writing free-text narratives

immediately following a decision task. The narratives were subsequently coded

into causal maps. While this approach shares the problems identified by Kemmerer

et al. (2001) in respect to the coding of free-response source documents, it

circumvents the potential problems associated with face-to-face interviews, arising

from the demand characteristics of the social situation, which can result in

overly elaborate or impoverished maps, as noted by Hodgkinson (1997b).