Feedback Loop Analysis in the Encyclopedia Project
Anthony Judge and Nadia McLaren
Extract from the final report on Information Context for
Biodiversity Conservation (2000).
Vicious cycles and loops (1995) and Strategic ecosystem: Feedback loops and dependent co-arising (1995)
In parallel with the work on individual data elements in the Strategies and
Problems files (Section 9.2: Web module: Issues, actions, treaties and organisations, Activities
of the workpackage), UIA undertook to advance its explorations of feedback
loops (both self-reinforcing and self-damping), specifically for this project
in the fields of environment and conservation. The specifications for this
work were to:
Develop, refine and seek to dynamically display the self-sustaining,
interlocking loops of conservation issues and solutions. In the event that
on-the-fly generation and visualisation of loops is feasible during web server
access, such dynamic displays would be developed as a means of shifting the
level of analysis beyond seemingly isolated environmental Problems and Strategies. The
visualization tools would then be adapted to assist editorial and error detection
processes. The key issue here is speed of detection and generation of loops.
This will be explored as a combination of machine capacity, algorithm logic
and display design.
The significance of this work is that there has long been recognition of how
one problem can aggravate another and of how several Problems can reinforce
each other. The UIA data register many relationships between Problems in complex
networks. Clearly such relationships may form chains or pathways linking many
Problems. But hidden in the data as presented is also the existence of chains
that loop back on themselves. The UIA data offer a unique opportunity to identify
such feedback relationship loops or cycles through which several Problems constantly
reinforce one another.
The notion of "loops", and its relevance to this project, requires
some further explanation. As defined by the authors of Making Strategy
in describing the value of Decision Explorer (Section 16.2.4: Decision
Explorer): a loop represents a description of a chain of consequences
that produces a dynamic outcome by feeding off itself (positive feedback = "vicious" or "virtuous" loops),
or by controlling itself (negative feedback). Typically a feedback loop will
be an important strategic issue in its own right. The purpose of detecting
feedback loops is to raise the level of analysis of individual issues to a
higher, systematic level. It is a technique which has the potential to add
extra meaning to basic data, particularly relevant for policy makers (one significant
user group for this product) and others concerned with understanding the interrelationships
and root causes of environmental problems, notably those relevant to biological
conservation. This is one perspective on the title of this project: Information
Context for Biodiversity Conservation.
A self-reinforcing ("vicious") problem loop, then, is a chain of
Problems, each aggravating the next, and with the last looping back to aggravate
the first in the chain. An example is:
Man-made disasters > Vulnerability of ecosystem niches > Natural
environment degradation > Shortage of natural resources > Unbridled
competition for scarce resources > Man-made disasters.
Such cycles are "vicious" because they are self-sustaining problem
organisational strategies and programmes that focus on only one problem in
a chain may fail because the cycle has the capacity to regenerate itself. Individual
"vicious problem cycles" also tend to interlock, forming tangled
skeins of interlinked global Problems which implicate single environmental
problems in chains and complexes of multi-sectoral issues. Without the means
to untangle the relationships, the response to a conservation challenge may
be ineffective, self-defeating or, even, harmful.
Before commenting on the project work in detecting vicious cycles, it is important
to recognise that it is precisely through the detection of such loops that attention
can be drawn to defects in the pattern of relationships in the data. It is possible
for some loops to be the result of incorrect relationships rather than being
representative of genuine feedback, and so "accidental" loops appear. Detection
of loops is therefore in the first place an editorial tool for hyperlinkage
within a relational database. It raises questions as to the appropriateness
of certain links which otherwise may go unquestioned. It also sharpens the discussion
on how distinctions are made, using verbal categories and definitions, and how
system boundaries are drawn grouping what is represented in this way. The results
indicate this is a very interesting area for further exploration.
Prior to the Definition Phase of this project, experiments in 1995 gave rise
to the results in Column 1 of the Table (below).
It was concluded that the procedure had promise but needed refinement, notably
to detect Problems erroneously excluded from loops, as well as loops excessively
connected to a single problem.
Refinements of Problem Loops
||Prior to Project
|Loop display for the Problem "Deforestation". Each cell is a
|Display of all Problems involved in loops concerning "Deforestation"
During the Definition Phase for the project, the data file on loops in the
Problems data was critically reviewed. The programme previously written to
detect aggravating pathways in the data and identify loops was re-run. Loops
were identified for selected types of problems only and for a maximum of 7
problems per loop (since chain searching requires many days of computing time,
even with 133 MHz PCs available to us at that time). Two weeks of judicious
source editing of aggravating links between Problem profiles reduced the size
of the file from 19,000 problem loops to around 7,000. A preliminary detection
analysis was also made of cycles of facilitating strategies in the Strategies
database. At the completion of the Definition Phase, there were 200 loops recognised
to contain environmental issues relevant to biodiversity conservation. They
were presented on the prototype CD. Work was also done in improving the display
of loops, using popups, from single data records on the prototype CD.
It was anticipated that the number of loops detected in the data would increase
significantly following the editing work on content and hyperlinks during the
Implementation Phase of the project (Section 9.2.1: Substantive activities). One need for
this work was to acquire a faster computer (to avoid having to segment the
This was done with project funds.
However, loops are relatively rare in chains of problems. It was anticipated
that the technique to detect loops (to this point explored in a DOS environment
in batch mode and delivered online in a static mode) would need to be improved
to make it easier for an online user to explore them dynamically. The algorithm
through which such loops are detected was referred to a mathematician, but
with no breakthrough. However, our objectives were achieved in other ways.
The next step in this work was enabling the contextual listing of Problems
and Strategies on user request relating to a selected node. The list was provided
"on the fly" at the request of the visitor to the UIA website. This
facility provided a rich pattern of information in which loops were indicated
if detected in the data.
During 1999, some 15 million chains of Problems
were searched to detect those that looped back on themselves within 7 links
maximum. Some 6,000 loops were detected and were scanned for potential errors
and hyperlink redundancy. The results of this work were integrated into the
facilities offered to web users via the UIA website. Loops were flagged in
the hit index and so became available to webusers for the first time in beta
test mode in October 1999. This web module was delivered in beta mode in early 1999 and has been continuously upgraded
on numerous occasions since that time. Later in the year, as a result of third-party
contribution to Java graphics capability, loops were visualised as an
integral feature of the bespoke Java spring mapping facility (Section
16.2.2: Spring maps (Java applet)).
This was a major step in the online graphic presentation of feedback loops
in the data and is described below in Section 11.2.2: Loop patterns.
The non-web standard editing software was augmented to detect local link redundancies
and errors in the cross-relationships of Problems and Strategies. This meant
that the daily work of editing database profiles was also directly contributing
to the refinement of loops. Of particular interest in any loop analysis were
Problem profiles indicated as being the most highly connected to loops, namely
members of the largest numbers of loops. These were treated as indicative of
inappropriate hierarchical clustering and encouraged redistribution of functional
relationships to broader or narrower Problems.
The final stages of the loop detection and visualization activity were completed
towards the end of 1999. First, the loop detection program was run in batch
mode for all Problems (except the most detailed and minor). The result was
15,958 feedback loops. Loops that involved the more detailed and minor Problems
were edited at source to correct anomalous linkages
This activity reduced
the number of loops to 5,873. To further eliminate errors, entries involved
in more than 80 loops were carefully edited; redundant and anomalous links were
corrected. This activity reduced the number of loops to 2,675. It is believed
that this list effectively represents "keystone" biodiversity issues
-- problems that are implicated in many negative feedback systems concerning
the natural environment.
Finally, the loop detection program was run again for the entire Problems database.
The resulting 51,555 loops were subjected to the error tests developed in previous
iterations. A total of 9,315 Problem loops were uploaded onto the website.
A similar procedure undertaken on the Strategies database produced over 2,000
loops, but of inconsistent quality. No project time remained to invest in refinement
of these links and the loops were not uploaded at this time.
It is expected that procedures developed in this project for working with loops
will become the standard for future reiterations as the databases evolve in
In order to give users a sense of the pattern that multiple loops formed, a
colour-coded tabular presentation is generated on-the-fly from database entries
with loops. An example for "deforestation" is shown.
|Java spring map display of loops for the Problem "Deforestation"
Users are able to manipulate both of the above displays. The elements of
the Java spring map can be dragged to redispose them. The same is true
of the three-dimensional virtual reality display, which in this case is dynamic.
Clicking on the nodes in either display will open a window with the corresponding
profile text from the database.
|3-D display of loops for the Problem "Deforestation". Each node is
a Problem in the loop.
[Clicking on a node opens the Problem profile in the interactive variant]
At the close of this project work, the loop detection program successfully
explored 1,239,769,768 chains of relationships in the Problems file to detect
51,555 loops (of up to 9 elements) in 12,397 Problem profiles (types B,C,D,E,F
-- the more general categories). This took only 500 hours of computer time.
The results were as indicated in Column 4 of the table above.
Having proved the possibility of identifying such cycles, the question raised
was how this information could be best portrayed through various mapping techniques.
One attractive option is to map the circles around the surface of a sphere
with whatever interlocking the data implies.
In anticipation of such expertise, during the final phase of the project,
several approaches were taken to facilitate user visualization of loops from
the profile of any selected entry that was part of a loop). This work drew
from complementary work in the concerned with "Graphics and Visualization" (Section
16: Multimedia visualization).
As presented online at the culmination of this project, users can choose between
the following presentations:
Tabular presentation: Users have a choice of two sets of tables:
problems: Presents an array of cells in a first table, coloured to give the
user a sense of the pattern of links from the selected problem, with
a second table as a key to the first. Users can click from there to a
similar table focusing on any selected problem in the array. Unfortunately
because of technical constraints in the construction of the HTML page,
the number of loops presented in this way is currently limited.
problems: Since problem relationships are reciprocal, a second set of tables
allows users to explore the reverse set of relationships in a similar
Java spring map: Users have the option of selecting a spring
map presentation, generated in response to the user request, restricted to
problem loops. This can be used as an access interface to other such maps or
to individual profiles of problems identified as nodes in the display.
Virtual reality display: Users have the option of selecting a virtual
reality display, generated in response to the user request, restricted to problem
loops. This presents a limited number of loops with which the selected problem
is associated. The concern here was to demonstrate the viability of allowing
users to generate and explore loops in this mode. As with Java map,
users could then click into a text profile of any problem in a loop. Of special
interest is the relative ease with which different display metaphors could
be designed. The challenge is effectively switched to the question of which
design metaphors provide the most additional meaning.
Identification of future activities
Partnerships and supportive coalitions of organisations
Both functionally and conceptually, such vicious cycles may offer a better
way to approach complex networks of problems. Indeed they serve to make clear
that any organisation with projects focusing on a single problem needs to be
aware of any vicious cycle of which that problem is a part. Unless that organisation
coordinates its activities with any bodies focusing on other problems in that
cycle, its work may be undermined. Despite apparent success in responding to
a particular problem in the short term, this may not affect the sustainability
of the vicious cycle of which it is a part. Pressures building up in other
parts of the cycle may regenerate the problem.
Ideally a coalition of organisations would form in response to each vicious
problem cycle. Information passed between organisations should then match the
pattern of impacts between the problems with which they are concerned. The
strategic issue may be less one of how to 'break' the cycle and more
one of how to reverse it, exploiting the fact that problems are functionally
linked in this way.
Also, there are not just "vicious problem cycles". A
significant number of problems also alleviate other problems, although this
information is less easy to obtain and such links are consequently less frequent
in the UIA data. There may therefore be "beneficent problem cycles" through
which problems constrain each other. It could prove strategically advantageous
to locate such cycles. Correspondingly, in the Strategies data there would
be "facilitative strategy cycles" and "constraining strategy
Using the techniques developed through this project, UIA is in a good position
to now identify the families of self-reinforcing and self-damping cycles in
international activities. From this additional work it could identify both
(a) networks of organisations working on the same issues and (b) supportive
coalitions of organisations involved with problems and strategy cycles.
Display modes for feedback loops
During the Definition Phase of this project, Nadia McLaren visited the Millennium
Institute in Washington DC for a demonstration of their software applications
of system modelling and displays, which include feedback loops. The bespoke Java applet for spring maps,
designed for this project, has the capacity to display feedback loops. Experiments
with VRML showed that feedback loops could be represented in 3-D, notably as
structures of loop clusters.
|Screenshots of virtual reality configurations generated from problems, organizations, strategies databases
Late in the project, the new features in the software Decision Explorer were
found to display and analyse feedback loops. It is important to stress that
this product is a tool for exploring information in a feedback model, not just
a 'page layout' tool where the concepts are neatly placed. It is
intended to be a dynamic modelling tool working with a database of concepts
and links where areas of the model are extracted to focus on using a "map"
command to "bring" further concepts onto the display. The producers
of the software have suggested to the UIA that it may be preferable to have
an initial map layout stored with the model data. If this is appropriate, the
UIA could work with them to add a new section to the data with X/Y coordinates
for the concepts. This would present people with a neater view of the data
Another possibility suggested by the producers is their development of a 'Viewer'
version of Decision Explorer that will be restricted to displaying only
views that have been saved in a model, with no analysis or other capabilities.
It would be free. In this way UIA users could distribute 'finished'
models more widely.
The interesting possibilities for the development of more sophisticated displays
of feedback loops depend on the identification and implementation of algorithms
capable of positioning a multiplicity of loops (several hundred) over the surface
of a sphere. To be useful, this has to be organised so the more detailed loops
are positioned "locally" whereas the intersectoral loops exploit
the global properties of the sphere.
As discussed elsewhere, it is interesting
to hypothesise the existence of major 'pathways' formed by elements
of different loops. Such pathways can be envisaged as being like rivers from
which local loops break off (as whorls). A number of such pathways may intersect.
It would then be the interlocking of these pathways which ensured the integrity
of the system of problems as a whole.
This approach becomes especially interesting if it is hypothesised that such
pathways are themselves necessarily circular. The question can then be formulated
in terms of the nature of the surface onto which the pattern of loops can be
usefully projected or mapped so as best to bring out the systemic integrity.
A flat surface is not especially interesting for such an exercise (although
this has been done for metabolic pathways). More interesting in comparison is
a sphere or a torus. In the case of a sphere, any such major pathways could
emerge as 'great circles' in a geometrical sense. It would then be
the interlocking of such great circles that provided the integrity of the structure.
Lesser circles could encompass portions of the sphere. Local circles could then
be positioned appropriately in relationship to them. Graphically displayed on
computer, they could be accessible by zooming when detail was required, but
the overall pattern would not be lost.
Clearly the projection of the loops onto such a surface constitutes a special
challenge. What mathematical operation is necessary to effectively 'massage'
the loops into the positions on the surface that most effectively bring out
the integrity of any great circle-type phenomena? How should this question be
formulated in mathematical terms?
Clearly the number of loops of the type detected can be considered very large,
or even infinite, when the number of elements in a loop is increased. What then
is the mathematical constraint that usefully excludes certain loops, so limiting
the number that can be detected? This question would seem to have something
to do with maximising the number of tangential relationships between loops,
notably by 'nesting' the maximum number of loops within other loops
-- minimising the number of loops that need to be represented non-tangentially
as crossing other loops.
It might be argued that this approach to the analysis of the data is more complex
than other more conventional forms available from graph theory. The assumption
made here is that the constraint of representation on a surface comprehensible
to the human mind is of immediate relevance to the ability to make informed
decisions on such matters at a policy level. The existing ability to provide
specialised analysis of what amounts to local loops in isolation has been well
demonstrated, as has the inability to act on the larger loops to which these
may contribute. A more comprehensive approach is required to 'thinking
globally and acting locally', whether in the geographical or the systemic
sense (as suggested here).
As in the case of studies currently supported by computer graphics (CAD, PCBD,
molecular chemistry, etc), and even dependent on them, it is suspected that
this approach might offer an entirely new grasp of global system properties
of very large
Note: In some cases the relationship may be a feedback loop in its own right
where A influences B and B influences A -- creating a nested negative feedback
loop within a positive feedback loop.
The logic behind this is that it is inappropriate for broad and detailed Problems
to be directly aggravating narrow and minor Problems, and vice versa;
this connection is best shown through hierarchical relationships where "suites" of
problems are so aggravating. The most common remedy was to remake the link higher
up the hierarchy of the detailed or minor Problem.
Colin Eden and Fran Ackermann. Making Strategy: the journey of strategic management. Sage
Publications, London, 1998.
Anthony Judge. Clarification of a Mathematical Challenge for Systems Science,