1992
Atlas of International Relationship Networks
Status
-- / --
Annex 2 of Visualization
of International Relationship Networks
(1992)
Summary
In the light of information and software packages acquired during the visit
to North America in February 1988, two distinct approaches can be envisaged
(possibly with some common elements):
-
(a) Arrange for a comprehensive package to be specially produced, possibly
under corporate sponsorship (e.g. Apple Corporation or IBM).
-
(b) Link together selected features of existing software packages to
produce a prototype marketable Atlas, which could be improved for subsequent
editions (possibly via Approach (a)).
In summary, the approach (a) is dependant on negotiation with corporations
for the necessary funds and might require some 12 months to complete once
such funds had been received. Approach (b) is dependant on UIA being able
to interlink and use the software. This needs further investigation.
A. PRODUCTION PHASES AND PROBLEMS
In the case of either approach, the following distinct phases (and problems)
can be distinguished:
1. Reformat UIA data: In order for the available software to process
UIA network data the latter must be converted to a different format. This
should not present any major difficulties and can be done using Revelation.
2. Segmenting UIA data: Only one of the software packages detected
is suited for networks in excess of 400 points (and a revised version will
only be available in September 1988). Most have been developed for 255
or less. Even such small networks can require very large amounts of computer
time on micro-computers. (Some large networks are now being analyzed by
methods necessitating the use of the so-called super-computers). It is
therefore necessary to investigate various techniques for splitting the
UIA data into a number of separate segments, preferably corresponding to
distinct "maps" in the Atlas. Two approaches to this are: (a)
use of some network analysis facilities to detect suitable segments (despite
the computer time problems) or (b) use of UIA subject codes or similar
devices to split up the data.
3. Network analysis: Given a suitable number of points there does
not appear to be any problem in analyzing the network. The initial problem
is rather one of deciding which combination or sequence of software routines
to use in the many available packages. This would require some preliminary
investigation and testing. The following groups of packages may be usefully
distinguished.
3.1 Group A: Social network analysis: The packages detected and
obtained include:
-
(a) UCI-NET (University of California Irvine)
-
(b) AL (University of California Irvine)
-
(c) NETPAC (University of California Irvine)
-
(d) STRUCTURE (Burt, University of Columbia)
-
(e) GRADAP (University of Amsterdam)
-
(f) NEGOPY (Richards, Simon Fraser University, Canada)
-
(g) FATCAT (Richards, Simon Fraser University, Canada)
-
(h) WORLDNET (Levine, University of Dartmouth)
A summary of the key characteristics and restrictions of these packages
is presented in Annex A. In contrast to the usual difficulty in such situations,
in a number of cases it is possible to switch flexibly between routines
in different packages to benefit from particular advantages. The constraints
on this, and learning how to benefit from such advantages, need to be discovered.
A key factor is the efficiency of the software, which may depend on the
computer language and the algorithms used. It is appropriate to note that
the level of academic competition amongst the producers of the above packages
is high, even though they are not normally distributed on a commercial
basis.
3.2 Group B: Statistical analysis packages: The specialized packages
in Group A in many cases incorporate routines selected from the large range
of statistical programmes commercially distributed in well-known standard
statistical packages such as:
-
(a) SAS/PC
-
(b) SPSS/PC
-
(c) STATPAC
-
(d) STATGRAPHICS
Indeed it is common practice to prepare some analyses with Group B packages
for use by routines in Group A packages.
3.3 Group C: Special application packages: A completely different
approach, necessitating a form of network analysis, is taken by some graphics
or design packages.
Of special relevance are:
-
(a) PCB: Printed circuit board design (of which a demonstration package
has been obtained.)
-
(b) CASE: Computer-aided software engineering packages provide facilities
for mapping relationships between entities and ensuring their appropriate
placement.
-
(c) CAD: Computer-aided design (e.g. AUTOCAD) of which a copy has been
obtained.
-
(d) CAL: Computer-aided learning has been developed to the point that
knowledge elements and relationships can be represented with appropriate
text and graphs of conceptual "entailment meshes" for a given
subject area. It is pedagogically useful to present the user with a visual
representation of any subject area so that the scope of the conceptual
system and each topic of sub-system - and the relationships between topics
- can be readily seen. These networks can be explored by the user in an
interactive manner analogous to browsing a cybernetic encyclopedia through
linked cross-referenced, self-referential and recursive information. A
prototype of such a system has been developed in an Apple environment in
Montreal.
4. Calculation of map coordinates
-
4.1. Group A: Social network analysis: Following the network
analysis, or as part of it, it seems to be necessary to calculate an "adjacency
matrix" of the points. This may be done with a number of the social
network analysis packages. From the adjacency matrix the coordinates of
the map can be produced using a multi-dimensional sealing programme (of
which a form known as MINISSA was frequently cited.)
-
4.2. Group B: Statistical analysis approach: It appears that
some statistical packages provide extensive plotting capabilities and therefore
built-in options for coordinate calculation.
-
4.3. Group C: Special application approach: Such packages appear
to by-pass the difficulties of this problem for the user.
5. Map plotting
-
5.1. Group A: Social network analysis: Only one of the packages
(NETPAC) provides for a very crude form of network plot. Surprisingly little
need is felt by analysts to map the networks they analyze. The keynote
speaker at the 1988 Sunbelt Social Network Conference (San Diego) specifically
indicated the importance of moving further in this direction.Two packages
have been located:
-
(a) The above-mentioned keynote speaker, Charles Kadushin (City University
of NY) provided a demonstration package for 3-D mapping of small networks
and will provide to the UIA a package for 2-D mapping of larger networks.
-
(b) Joel Levine (University of Dartmouth) has developed a special package
for plotting intercorporate links for his Corporate Interlock Atlas. Some
of this software is being provided to the UIA.
In both cases the interface to coordinate calculation has to be investigated
in terms of UIA data. Both conclusively demonstrate the viability of the
whole approach.
-
5.2. Group B: Statistical analysis plotting: Information received
from those concerned with social network analysis did not indicate that
the facilities for plotting network graphs were extensive, if they existed
at all. There is some indication that packages such as STATVIEW (on Apple)
could be of considerable value.
-
5.3. Group C: Special application packages: Such packages have
built in plotting capabilities.
6. Labelling points
The problem of labelling points on a network graph has not yet been clarified.
Various options are however available. The challenge is to do this in a
way which enhances the quality of a map.
B. HARDWARE CONSIDERATIONS
1. The quantity of computer power required to analyze and plot large networks,
especially in an interactive environment, has forced many such applications
onto large computers and, just recently, onto super-computers. Thus Levine's
Corporate Interlock Atlas was produced using a main frame, and Klovdahl's
sophisticated graphics were produced using a specialized Evans and Sutherland
graphics terminal. These are not feasible options for the UIA, nor are
they economically viable.
2. Over the past few years, much network analysis and graphics has been
developed on microcomputers compatible with those of the UIA (i.e. IBM
compatible). The problem here is two-fold:
-
the constraints of 640 K memory in the current hardware/software
generation,
-
the relatively slow speed of such machines.
Whilst major increases in speed can be obtained from faster machines of
this type, little can be done about the memory constraint without (considerable)
adaptation to the software.
3. An alternative route is through the use of the latest generation of
Apple computers (e.g. the Mac II), which can use much larger memory, are
very fast and have good graphics capacity. The problem is that relatively
little network analysis software has been written for this environment
(with the possible exception of STATVIEW) and it is not clear whether the
IBM-based packages could be easily recompiled in this environment.
4. Whichever approach is selected, an appropriate plotting device is required,
whether in the form of a plotter or a laser-writer.
C. PROVISIONAL CONCLUSIONS
1. There is every possibility of producing some kinds of maps from UIA
data, as indicated by the presentation already available.
2. The IBM compatible packages should be investigated to determine more
clearly the possibilities and constraints and whether or not they are adequate
for a prototype product possibly in conjunction with routines from packages
such as STATGRAPHICS.
3. Parallel investigation should be made of special application packages
such as PADS-PCB, CASE and CAL which may prove more attractive or offer
alternative modes of representing or interacting with the data.
4. Parallel consideration should be given to obtaining corporate sponsorship
for a specially written software package to facilitate this approach, preferably
in the light of the above investigations and the need to maximize comprehensibility
of the individual maps.