Appendix
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
Supporting Software for the Elicitation, Construction,
Analysis and Comparision of Causal Maps
In all but the very simplest of applications, the use of computer software systems can
greatly assist the researcher in all stages of the causal mapping process, from knowledge
elicitation and map construction to individual and comparative analysis. This is equally
true not only in the case of applications involving the detailed analysis of single
(Cossette & Audet, 1992) or small numbers (Clarke & Mackaness, 2001) of maps, but also
in much larger-scale comparative studies (Markóczy & Goldberg, 1995) of causal maps.
Clearly there are times when small-scale, complex ideographic studies, exploratory and
inductive in nature, conducted in the context of under-explored knowledge domains, are
invaluable. In this type of application, which can result in maps containing as many as
several hundred concepts (Eden & Ackermann, 1998a), it is impracticable to analyze the
structure and content of the maps using basic manual procedures. However, much of the
utility of causal mapping techniques in organizational research lies in their application
to larger numbers of individuals and/or groups, comparing their similarities and differences
in a range of contexts and/or over multiple points in time. As noted in the main body
of the chapter, such comparisons are potentially unwieldy, but fortunately recent
developments in mobile computing and associated software advances are paving the
way for new support systems that will rapidly resolve these difficulties.
Generic computer software tools such as ATLAS/ti (Jasinski & Huff, 2002) are enabling
ideographic researchers to tackle more demanding problems and extend their analyses
considerably further than would have been possible using manual coding techniques.
Moreover, software systems devised for the structural analysis of social networks, such
as UCINET (Borgatti, Everett & Freeman, 1992), are potentially also suitable for the
analysis of causal maps, having common mathematical roots in graph theory. Indeed,
many of the structural indices commonly employed by network analysts and routinely
available in software packages to support the analysis of social networks bear a strikingly
close resemblance to those devised by Eden et al. (1992) specifically for the analysis of
causal maps. There is no doubt that these software tools are extending the range of
computer technology broadly capable of supporting causal mapping. One other system
Table A1. Selected software supporting causal mapping
Name of
Progam
Approach to Knowledge
Elicitation/ Map Construction
Output Statistics Strengths Limitations Illustrative applications Source of Software
ATLAS/ti Though not developed
specifically for causal mapping
purposes, this knowledge-based
computer system (KBS) has
been used to support the
construction of maps based on
ideographic interview
transcripts and could equally be
applied to facilitate the
construction of maps from other
forms of documentary data.
Not Applicable (N/A) The software assists researchers in
identifying patterns among key
concepts in documentary data,
enabling them to be depicted
graphically in the form of nonhierarchical
network structures.
Data are readily exportable to other
software packages (
e.g. SPSSTM)
suitable for performing quantitative
analyses of the text patterns
identified.
Data must take the form of plain ASCII
or ANSI text, i.e. there is no recognition
of word processor formats (though data
can be internally translated).
The reliability and validity problems
associated with the coding of maps
constructed from ideographic data
gathered through indirect means (as
detailed in the main body of this chapter)
apply when using this software package.
Jasinski &
Huff, (
2002)
The package is available commercially
from Scolari, Sage Publications
A demo version is downloadable from:
http://www.scolari.com
CMAP2 This software was developed to
support the construction of maps
indirectly elicited from
documentary sources, including
interview transcripts.
Documentary data is coded post
hoc. In addition to modules for
the inputting of raw data and the
creation of a standard language
vocabulary for coding natural
language expressions and
achieving comparability over
research participants, the
program also contains a variety
of tools for editing the input
data and the generation of
analyzable databases.
Key numerical outputs from
the program include
measures of the distances
between the maps of
individual participants or
clusters of participants.
CMAP2 is based on the observation
that causal maps can be decomposed
into causal units, i.e. conceptconcept
pairs assumed to be causally
linked to one another. The
computer can process these causal
units as semi-independent data
entries.
The software links the computer’s
databases with the input data
elements and thus with the source.
As such, it promotes good
organization of data, and enables an
audit trail from the source to the
final standardized concepts and
causal units.
The software is currently written in DOS
mode. As acknowledged by the author,
this does not permit easy export of the
various output data into better-developed
user interface environments for
supplementary statistical analysis, using
standard software packages such as
SPSS.
Moreover, the absence of methods for
direct raw data acquisition and input by
the participants themselves, in
conjunction in with primary-level
standardization facilities and
opportunities for concurrent feedback,
renders the whole process extremely
labor intensive on the part of the
researcher.
Laukennan (1994, 1998) This software is available on a noncommercial
basis from the developer.
For details, see Laukennan (1998: 189)
Decision
Explorer
This software was developed to
permit the direct elicitation and
construction of cause maps.
Typically, maps are constructed
iteratively, in face-face meetings
in situ
This software enables the
graphical representation of
maps as well as the
calculation of a
variety of
quantitative indices of a
structural nature.
This software has proved to be of
immense benefit in areas where
interactively generated maps have
helped to build up a comprehensive
qualitative map or model, which is
then explored and analysed to help
develop strategy, decision making
and business problems
Decision Explorerallows
researchers to manipulate the data
and hence view it from a
variety of
perspectives. This is not only
beneficial from an analytical
standpoint, but enables the
researcher to actively gain and
maintain the interest of participants
in the research process.
Data can be readily imported from
and exported to a
number of
standard software packages for
supplementary qualitative and/or
quantitative analysis (e.g. QSR
NVivo and SPSSTM).
Overly simplistic or complex and
‘messy’ maps may be constructed at the
original stage of data input.
As with ideographic methods more
generally, the highly idiosyncratic nature
of the data renders problematic the
systematic comparison of maps,
particularly where large numbers of maps
are involved.
There are no features to support the
analysis of inter-coder and code-recode
reliability. (
For an explanation of these
terms see the section entitled
‘Psychometric Issues’ in the main body
of this chapter).
The software does not permit the links
between variables to be formally
weighted numerically, ,
although basic
polarities (positive and negative) can be
depicted and important variations can be
identified through the use of contrasting
color codes and/or variations in the
relative thickness of the lines
interconnecting the concept nodes.
Cropper, Eden & Ackermann
(1990)
Eden & Ackermann (1998a,
1998b).
This software package is available
commercially from Banxia Software
Ltd.
A demonstration copy can be
downloaded from:
http://www.banxia.co
Table A1. Selected software supporting causal mapping (continued)
Name of
Progam
Approach to Knowledge
Elicitation/ Map Construction
Output Statistics Strengths Limitations Illustrative applications Source of Software
Distrat/
askmap
suite of
programs
Not applicable.
This suite of programs was
devised facilitate the comparative
analysis of cause maps elicited
using the hybrid procedures
devised by ________
Goldberg
(1995)
The programs perform several
of the analytical tasks
associated with the ________-
Goldberg approach.
As stated in Goldberg’s (
1996) analysis
guide, the software is inflexible and of little
benefit to anyone not using the particular
methods described. The input and output
data are configured such that they are not
readily transferable to and from other
systems, especially of the larger, fully
integrated, interactive variety.
__
_
____
_
_
_
__ _
__
_____
__
_
____
_
Goldberg (1995)
The complete collection of programs is
available to researchers on a noncommercial
basis. For details see
Goldberg (
1996), down loadable from:
http://www.goldmark.org/jeff/programs/d
istrat/software/drdoclet.ps.gz.
KNOT
(Knowledge
Network
Organizing
Tool)
Not applicable.
This software package was not
developed specifically for the
purpose of causal mapping. Its
purpose is to implement the
Pathfinder network algorithm.
The Pathfinder algorithm seeks to
represent the information
contained in the input proximity
matrix in as few links as possible,
consistent with the parametric
values set by the investigator. The
aim is data reduction to facilitate
comprehension of the resulting
network.
The software performs a
similar range of functions to
programs for the analysis of
social network structures, i.e.
the input distance measures
are represented as nodes, with
links representing the relations
between objects. Weights
associated with the links,
derived from the original
proximity data, reflect the
strength of the relations.
The KNOT system includes several
programs and utilities to facilitate
Pathfinder network analyses of
proximity data. The system is oriented
around producing graphical
representations of the solutions, but
representations of networks and other
information are also available in the
form of text files, which can be used in
conjunction with other software.
Although it was designed for in DOS
mode with minor modifications it runs
in a Windows environment.
This software does not support the initial
elicitation of data for the construction of
cause maps.
For details of the general
Pathfinder method and its
application see Schvaneveldt
(1990)
KNOT is a commercially distributed
software package, available from
Interlink. For details see
http://www.interlinkinc.net/Pathfinder.ht
ml
UCINET
N/A.
This package was devised for the
analysis of social network data
(typically questionnaires or
documentary sources).
A number of basic output
statistics for the identification
of social network structures
including, for example,
centrality measures, predicated
upon elementary graph theory
are equally suited to causal
mapping applications. In
addition, the package has a
number of matrix analysis
routines useful in the context
of cause map analysis.
The software program NetDraw is
integrated with UCINET for drawing
diagrams representing social networks,
but is equally suitable for drawing
networks of concepts, as in causal
mapping applications.
This software does not support the initial
elicitation of data for the construction of
cause maps.
Borgatti, Everett, & Freeman
(1992)
A demonstration copy of this
commercially distributed software
package is available at
http://www.analytictech.com
Cognizer
This software
package is
currently being
developed by
Mandrake
Technology
This software enables direct
elicitation of cause maps.
Building on __
_
____
_
Goldberg (1995), data elicitation
and map construction are achieved
by participants first selecting from
a predefined pool of variables
shown on screen (randomised on
an individualized basis, so as to
minimize potential order effects in
variable selection and pairwise
evaluation tasks). Participants
then perform on-screen pairwise
evaluations of their chosen
variables (including explicitly
detailing the strength of
relationship between each pair of
variables). The resulting cause
map can next be immediately
available for viewing and can be
edited by direct manipulation of
the on screen graphical outputs.
Many basic analytical
functions are incorporated,
including, a number of map
content measures (e.g.
indegree, outdegree and
reachability values) and
structural measures (e.g. linkto-
node ratio and map density)
(see Table 1 in the main body
of the chapter for a description
of these measures). Distance
ratios, reflecting the degree of
overall dissimilarity between
pairs of cause maps
(Langfield-Smith & Wirth,
___ __
_
___
_
_
_______
_
_
1995) can be readily computed
and employed in order to
investigate patterns of
similarity and difference
among subgroups of
participants.
Cognizer is essentially being
developed to permit the elicitation of
cause maps in a manner that is
meaningful to participants and
amenable to mass-comparison.
However, the software also allows
cause maps to be elicited in other
ways, each of which may be more or
less acceptable in particular contexts
of application. For example, maps can
be constructed directly, by drawing a
weighted digraph (a
sophisticated
variant of the basic influence diagram,
as discussed in the main body of this
chapter under ‘basic metrics for the
analysis of individual cause maps’).
The results are easily exportable to
other appropriate analytical packages,
such as SPSSTM.
The limitations of the software will depend
upon the chosen method of elicitation and
analysis (see the main body of the chapter
for a
detailed discussion regarding the pros
and cons of each method).
Clarkson, Hodgkinson &
Fearfull (2001)
.
For additional details and/or a
demonstration copy of Cognizer, e-mail:
mandraketech@
fsbusiness.co.uk
worthy of brief mention in this connection, before turning to consider more specialist
software tools specifically devised for the analysis of cause maps, is the general
approach known as Pathfinder (Schvaneveldt, 1990; Schvaneveldt & Durso, 1981;
Schvaneveldt, Dearholt & Durso, 1988, 1989). In a similar vein to UCINET, the Pathfinder
algorithms, as implemented in software systems such as KNOT (The Knowledge Network
Organizing Tool) (http://www.interlinkinc.net/Pathfinder.html), are used to explore network
structures derived from proximity data (i.e., distance matrices reflecting the degree
of overall (dis)similarity, or some other proximity measure, between concepts). Within
the specific domain of information technology, Pathfinder has been successfully applied
to a variety of problems concerning the design of user interfaces (e.g., Gillan, Breedin &
Cooke, 1992; Roske-Hofstrand & Paap, 1986). As observed by Gillan & Schvaneveldt
(1999), in general, applications in this context (typically involving the analysis of
relatedness ratings) have demonstrated that users are more effective in using interfaces
derived from their revealed models of the system, as identified by the Pathfinder
algorithm, in comparison with existing interfaces.
Although software systems such as ATLAS/ti, UCINET and the Pathfinder algorithm are
proving generally useful as basic support mechanisms in the conduct of causal mapping
research, fully integrated software systems, dedicated to the elicitation, construction,
analysis and comparison of causal maps are ultimately required, if causal mapping is to
fulfill its true methodological and substantive potential. To this end, there have been a
number of advances over the past decade or so and in the remainder of this appendix we
highlight what we consider to be the most significant of these. Due to space limitations
we shall confine our attention to a brief consideration of just three of the more popular
software packages presently available for the dedicated analysis of causal maps, namely,
CMAP2 (Laukkanen, 1994), Decision Explorer(Eden et al., 1992) and the suite of
programs developed by Goldberg (1996), known as distrat/askmap, in addition to
reporting some ongoing developments of our own. Clearly, all of these systems are
constrained (albeit to varying degrees) by virtue of the underlying assumptions and
concomitant choices that their developers have made in relation to the various issues
discussed in the main sections of this chapter.
CMAP2 (Laukkanen, 1994, 1998) was developed for the comparative analysis of causal
maps derived through interview transcripts and/or documentary sources. A data-basedorientated
PC program, it is intended specifically for use in settings where the input data
take the form of natural communication and key parameters such as the number of
concepts explored, the number of mapped relationships and indeed the number of
participants must be flexible (Laukkenan, 1998). Unfortunately, as observed by Jenkins
(1998), CMAP2 is limited in several important respects. First, no research has been
undertaken to assess the reliability of the processes by which the input data are
transformed into comparable units of analysis. As noted in our discussion concerning
the relative merits of direct vs. indirect elicitation procedures, this is clearly not a problem
unique to CMAP2 but is common to a number of applications of causal mapping
procedures more generally, where the maps have been inferred from interview transcripts
and/or other indirect documentary sources. Clearly, however, if the practice of causal
mapping and the associated application of particular procedures such as CMAP2 are to
gain credence in terms of their scientific legitimacy, there is an urgent need to increase
the volume and quality of research addressing these and other equally pressing issues
concerning their psychometric efficacy.
Decision Explorer(Eden et al., 1992), a re-launch of Graphics COPE, the system
developed several years earlier by Eden and his colleagues for use in the context of group
decision support (e.g., Ackermann, Eden & Cropper, 1990; Eden & Cropper, 1990), has
proven to be of immense benefit in the context of building comprehensive cognitive maps
of complex organizational problems. Decision Explorerallows the researcher to manipulate
data in ways that enable it to be viewed from a variety of perspectives (Eden &
Ackermann, 1998b). This is helpful not only from an analytical standpoint, but also in
enabling the researcher to actively gain and maintain the interest of participants in the
research process. However, Decision Explorer, as with Laukkanen’s software package,
was designed primarily for use in the context of local settings, where the focus of
attention is on the intensive analysis of ideographic data, gathered from small numbers
of individuals. It is less suitable for use in the context of larger-scale studies.
In contrast, Goldberg’s (1996) computer programs were designed to perform several of
the tasks associated with the Markóczy-Goldberg approach to causal mapping (Markóczy
& Goldberg, 1995). As discussed in the main body of the chapter, this approach is
potentially very useful in situations that demand the comparative analysis of large
numbers of maps. Unfortunately, however, a number of the statistical procedures as
devised and implemented by Markóczy and Goldberg (1995), including those building on
the earlier work of Langfield-Smith and Wirth (1992), have no accompanying software
provision within the distrat/askmap system, thus rendering their implementation difficult,
if not impossible, using these programs. Nevertheless, the fact remains that the
Markóczy-Goldberg approach to causal map elicitation, analysis and comparison – and
the earlier work of Langfield-Smith and Wirth, which laid the foundations for these
innovations – represents a major methodological breakthrough. However, if the ultimate
potential of this approach is to be realized, there is an urgent need for further developments
in the provision of user-friendly software, capable of readily implementing the full
range of associated procedures, from elicitation through analysis to comparison, in realtime
environments. At the time of writing, the present authors are in the advanced stages
of actively evaluating such a system. To date, this Windows-based system, known as
Cognizer, has been successfully implemented in the elicitation, analysis and comparison
of well over 200 maps, all gathered in the context of face-to-face interviews, in situ, with
busy employees. (Further details of all of the individual software systems discussed in
this Appendix, including a summary of their main strengths and limitations, together with
information concerning their availability, are presented in Table A1.)