Download pdf - Reinterpreting ‘generic structure’: Evolution, application and limitations of a concept

Reinterpreting ‘generic structure’: evolution, application and limitations of a concept David C. Lane and Chris Smart

This paper traces the evolution of the generic structure concept in system dynamics and discusses the different practical uses to which they have been put. A review of previous work leads to the identification of three different views of what a ‘generic structure’ is and, hence, what transferability means. These different views are distinguishable in application as well as in theory. Examination of these interpretations shows that the assumptions behind them are quite distinct. From this analysis it is argued that it is no longer useful to treat ‘generic structure’ as a single concept since the unity it implies is only superficial. The conclusion is that the concept needs unbundling so that different assumptions about transferability of structure can be made explicit, and the role of generic structures as generalisable theories of dynamic behaviour in system dynamics theory and practice can be debated and clarified more effectively.

The typical history of a concept, whether it be “chemical element”, “atom”, “the unconscious”, or whatever, involves the initial emergence of the concept as a vague idea, followed by its gradual clarification as the theory in which it plays a part takes a more precise and coherent form.-Chalmers (1982)

From its inception, a clear aspiration of the field of system dynamics was to offer an integrative theory of the separate processes of management which would make it possible to evaluate and reinterpret specific cases and experi- ences, so producing generalisable insights.’ In the last three to four decades the creation and evolution of generic structures as vehicles for storing and applying these insights has been one of the ways in which this aspiration has advanced.

Despite the important role that generic structures are seen to have within system dynamics there is no agreed or even precise definition. Paich in his 1985 review of the status of generic structures offered as a tentative working definition that generic structures are “dynamic feedback systems that support particular but widely applicable behavioural insights” (Paich 1985, 127). Senge suggested that “generic structures are relatively simple models of dynamic processes that recur in diverse settings and that embody important management principles” (Senge 1985, 791). However, generic structures also appear to be closely related to several other concepts in system dynamics which we might list. They are linked with generic models, “case studies converted to explicit dynamic form” (Forrester 1980, 18), or “self-contained behavioural theor[ies] of the dynamic processes it illustrates. They are a way of storing knowledge and feedback structure of social and business systems” (Morecroft 1988, 314). The literature also speaks of elementary structures, simple feedback structures, which can be used “to approach understandings of real-world problems” (Andersen and Richardson 1980,101), and of systems archetypes, which Senge identifies as being generic structures, or; “patterns of structure that recur again and again” (Senge 1990, 94). On a markedly less serious note, at the 1993 System Dynamics conference in Cancun, Mexico, John Sterman confessed to a profound uncertainty in defining generic structures but associated them with the U.S. definition of pornography, “I know one when I see one”. Lane extended this conceit by proposing that the

The authors would like to offer their thanks to Jay Forrester, Michael Goodman and George Richardson. Their observations on generic structures, particularly those offered to DCL during his visit to Boston and Albany in November 1993, were a valued contribution to our work. We would also like to thank John Morecroft who, in a number of discussions with DCL in 1995, shared his developing ideas about the role of archetypes in teaching. We also benefited consider- ably from comments by two referees. We are grateful for the time and interest given by all of these individuals. Naturally, the authors must take responsibility for the opinions expressed in this paper.

System Dynamics Review Vol. 12 , no. 2, (Summer 1996): 87-120 Received June 1995 0 1996 by John Wiley & Sons, Ltd. CCC 0883-7066/96/020087-34 Accepted December 1995

87

88 System Dynamics Review Volume 12 Number 2 Summer 1996

David Lane is currently a lecturer at the London School of Economics. He has mathematics degrees from Bristol and Oxford Universities and a Doctorate in mathematical biology also from Oxford. He was a consultant in Shell International (where he led the system dynamics and decision analysis groups) and a marketing manager in Shell UK. His research, teaching and consultancy focus on problem structuring techniques for strategic thinking, in particular system dynamics, and on the validation of these group modelling processes. He is also interested in the philosophical similarities between system dynamics and ‘soft’ operational research. Address: Operational Research Department, London School of Economics and Political Science, Houghton Street, London WCZA ZAE, UK

Chris Smart teaches business computing and management information technology at City

equivalent British view-material liable to deprave and corrupt-might also be apt.

Such uncertainty cannot be healthy for a term that is used so frequently by our professional community and which has recently become very popular with a wider audience. Does the label ‘generic structure’ make sense if it is so poorly defined? Is the pursuit of generic structures a Snark hunt, the essential triviality of the endeavour obfuscated by the confusing variety of definitions? Or does this variety cloak the fact that the notion of generic structures is inherently treacherous and fraught with danger because it harmfully promises something which can never properly be delivered, or indeed validated? To address these questions a careful review of the relevant literature is called for. The purpose of this paper is therefore to trace the evolution of the ‘generic structure’ concept and to review the practical application of the idea in order to tease out the different strands of thought that have been associated with the concept. In so doing we will examine the different roles that have been proposed for generic structures and the extent to which generic structures have lived up to expectations. We will find that the concept has been overburdened with so many subtleties of meaning and has been used in so many diverse ways that it is no longer helpful to regard ‘generic structure’ as a unified concept. We believe that it is time to try to unbundle the concept and we suggest a way that this might usefully be achieved.

The evolution of ‘generic structures’

Generic structures were not invented but evolved gradually. The twists and turns of this evolution are such that it may be helpful to identify at the outset the three different facets of the concept that we have identified as being emphasised in different works. Generic structures are models, and as such are theories of dynamic structure and behaviours. Beyond this central concept, system dynamicists have variously stressed: the relationship between the dynamic structure and the application domain in which it was found, which we have termed canonical situation models; the mathematical structures which generate a dynamic behaviour, abstracted micro-structures; and the dysfunctional characteristics which are commonly apparent in complex systems, counter-intuitive system archetypes (Figs 2, 3 and 4 respectively). These different emphases have created three different interpretations of generic structure and we highlight them throughout the evolution story (Fig. 1).

First, we will discuss the earliest interpretation of generic structures; canonical situation models. Second, we will summarise a debate on the basis for classifying systems and what is meant by the term generic or general in the

Lane and Smart: Reinterpreting ‘Generic Structure’ 89

University. He holds a B.Sc. in psychology and an M.Sc. in business systems analysis and design. He has held a number of managerial and consulting posts in the area of information technology and his current research interests focus on the impact of IT on work systems.

context of generic structures. Third, we will discuss abstracted micro- structures, an interpretation which arose out that debate. Fourth, we will revisit Paich’s (1985) review of generic structures. Fifth and last, we will discuss the latest interpretation of generic structures, counter-intuitive system archetypes.

Canonical situation models

The term generic structure itself evolved from the term “general model” which was first used by Forrester in 1961 to describe the generalisation and simplification of case study models so as to, “represent a wider class of industrial situations” (Forrester 1961, 208). The early emergence of the concept of a generally applicable model and the need to transfer behavioural insights from one situation to another are bound up with the original ambitions for system dynamics. It is this interpretation of generic structure that we call “canonical situation models” (Fig. 2).

Clearly and forcefully stated in the first publication on the subject, the basic ambition of system dynamics was “to become a basic theory of behaviour” (Forrester 1958, 37). The way the theory would be applied was to be by using “case studies to enhance theory, to use theory as a guide for solving problems” (p. 37). The means by which case studies and theory would be interrelated in this way are not described but a clearer picture emerges with the publication of Industrial Dynamics (Forrester 1961). In this book Forrester enlarges on the ambition of system dynamics to become a theory of behaviour as a “common frame of reference to enable transference of experience” (Forrester 1961, 3), setting out the need not only for a general theory and method for understanding and modelling dynamic systems but also for some means by which the lessons learnt studying one situation could be interpreted and applied in other situations. Although Forrester does not explain what he means by “experience” nor explicitly identify ways by which transference of experience is to be achieved, he extracts general principles from the case studies, e.g. “worse before better” (pp. 348-349) and repeatedly asserts that a dynamic model of the fundamentals of a class of situations, one of our canonical situation models, can be built that is suitable for any member of the class being represented (pp. 311-343). The examples given of situation classes are described in terms of the application domain, e.g. market dynamics (p. 313) and product life-cycle (p. 320). The customer-producer-employment model, which Forrester discusses in depth, has been simplified and generalised from an actual case study to represent a wider class of industrial situations without “any major effect on the nature of dynamic behaviour nor on the conclusions drawn” (p. 208). The inference which can be made is that both dynamic

https://www.researchgate.net/publication/238759226_Industrial_Dynamics_A_Major_Breakthrough_for_Decision_Makers?el=1_x_8&enrichId=rgreq-fd303bc0-77a0-4359-a313-65bc5a17e02a&enrichSource=Y292ZXJQYWdlOzQ0NzExNjc3O0FTOjE5NTI1Nzc0MTM4NTczM0AxNDIzNTY0NDgxNTQy

Fig.

1. R

epre

sent

atio

n of

the

evol

utio

n of

the

gene

ric s

truc

ture

con

cept

with

in th

e lit

erat

ure

of s

yste

m d

ynam

ics.

To

illus

trat

e th

e pr

imar

y em

phas

es o

f the

ir a

rgum

ents

, som

e of

the

mai

n w

orks

on

the

subj

ect h

ave

been

loca

ted

with

in th

e th

ree

diff

eren

t str

ands

of

inte

rpre

tatio

ns th

at h

ave

been

iden

tifie

d. A

rrow

s be

twee

n re

fere

nces

indi

cate

sim

ilari

ties

in a

rgum

ent.

Not

e th

at a

rrow

s em

anat

ing

from

and

/or c

onne

ctin

g to

refe

renc

e bo

unda

ries

indi

cate

a re

latio

nshi

p co

mm

on to

the

who

le w

ork

whi

lst a

rrow

s cr

ossi

ng b

ound

arie

s in

dica

te li

nkag

es to

spe

cific

sub-

argu

men

ts.

cn 6 B 2


Fig. 2. A “canonical situation model”. These are general models of an application domain, perhaps inspired by a particular case study. The system structure is a fully specified simulation model which generates a range of significantly different modes of dynamic behaviour depending on the parameters and policies employed. This example is based on the market growth model for the launch of a new product (Forrester 1968a).

~~ ~~

Canonical Situation Model: Market Growth

Structure Sales

Revenues Revenues Revenues

Structure The system structure is shown here without the accompanying equations. The loops have been labelled ‘R’ and ‘B’ to indicate reinforcing and balancing effects. As the Sales Force generates orders these flow into Order Backlog and are subsequently dispatched, thus generating Revenues and a Budget which allows for the adjustment of the Sales Force. Order Backlog influences Delivery Delay, which limits the orders generated by each member of the Sales Force. Levels of Delivery Delay are compared with an Operating Goal and Production Capacity is adjusted in consequence. Historical values of Delivery Delay can erode this Operating Goal. Behaviours Only three examplesfiom the range of behaviours have been illustrated, values of the variable Revenues being plotted. From left to right: growth and stagnation due to Production Capacity constraint; restoration of growth via Production Capacity expansion and self- liquidation as the Operating Goal adjusts to past performance.

92 System Dynamics Review Volume 1 2 Number 2 Summer 1996

behaviours and the important lessons to be learnt are significant when they are common to a class of application domain situations rather than being unique to a particular instance.

In 1968, Forrester presented his market growth model, a general model of the launch and growth in sales of a product (Forrester 1968a), which has subsequently been regarded as an exemplar of generic structures (Fig. 2). For- rester’s conclusion from this study is that market interactions are so complex that an understanding of the underlying feedback structure is necessary to organise knowledge about the system and he calls for increased research on “conceptualising the structures which produce typical classes of market behaviour” (p. 102). This conclusion highlights that within an application domain the situations that a general model can represent may be classified according to the underlying dynamic structure and its associated behaviour modes. The presentation of the market growth model is far simpler than that of the models in Industrial Dynamics, tending to concentrate on the causality and interactions of the feedback loops and the nature of the dynamic behaviour generated rather than on the detail of the underlying quantified and operational model. This distilled form of general model emphasises the dynamic fundamentals of the class of situations and so reinforces the explanatory qualities of canonical situation models.

In the following decade a number of canonical situation models of much greater complexity were published. In 1969 Forrester published a model of development processes in urban areas (Forrester 1969). Unlike previously published general models which were distilled from real-world case studies and data, the urban model was a hypothesis based on carefully argued assumptions and in publishing the model Forrester was inviting the methods, assumptions and results to be evaluated. In fact, Forrester had always advocated building a general model or theory first, and then modifying it to fit the particular situation under study as the preferred method for building any system dynamics model (e.g. Forrester 1961, 318), but he had usually reported that a canonical situation model had been evaluated against a real-world case study before publication. The scale and complexity of urban situations, however, does not allow for easy evaluation of a theory of urban growth and For- rester’s purpose was to present a method for social analysis, in which canonical situation models have a central role. A similar approach was taken with the “World Model”, a model of the basic factors which determine growth on this planet (Forrester 1971). This model was adopted by The Club of Rome, purportedly eminent domain experts, who modified the model and attempted to validate its assumptions and findings and has subsequently been revisited (Meadows et al. 1972; and Meadows et al. 1992). While these models are of far greater complexity than earlier general models, they are conscious


attempts to define and understand the fundamentals of dynamic structure and behaviour of a class of situations.

A debate on the meaning of ‘generic’

In the same year that he published the market growth model, Forrester reviewed the first decade of system dynamics (Forrester 1968b). The tasks ahead that he identified for the field included the cataloguing of “system examples”. Forrester comments that system dynamicists approach problem situations by drawing on a mental library of previously studied systems. He advocates assisting others to do the same by putting these examples into writ- ten form, which as a series would then identify common relationships in industry. These examples should “concentrate on the minimum structure necessary to create a particular mode of behaviour” (p. 412) and be supplemented by that which would indicate the circumstances under which a particular subsystem will be dominant. These “system examples” are clearly a form of general model and a distinction is made between the dynamic structure which generates the behaviour modes and the application domain data, both of which being needed to help users to reason about the causation of the problem being experienced.

During the debate with Ansoff and Slevin, Forrester asserts that general models are theories of dynamic structure and behaviour and that “particular applications of industrial dynamics could become theories of behaviour of particular systems” (Forrester 1968c, 604). Up to this point descriptions of general models have included features of the application domain. However, later in the same paper Forrester asserts that “an industrial dynamics model is a theory of structure and dynamic behaviour for a particular class of systems” and that “systems belong to the same class if they can be represented by the same structure” (p. 606). Forrester illustrates the meaning of a class of systems by comparing a simplified employment-inventory system with a swinging pendulum. He shows that both systems have the same underlying structure and generate the same dynamic behaviours, and are therefore both members of the same class. The commonality between these two examples rests not on features of the application domains but purely on the commonality of the components of dynamic structure, i.e. feedback loops, levels, rate equations, etc., and the relationships between them. This dynamic structure when abstracted from any application domain data defines the class of system. A development of this interpretation of generic structure is discussed in the sub-section on abstracted micro-structures.

In 1980 Forrester looked to the future of system dynamics and among the “missing links” called for a “library of fundamental dynamic structures that


generate the difficulties with which managers must cope” (Forrester 1980, 18). By way of example he cited the production-distribution model (Forrester 1961) and suggested that the model could be simplified and generalised further. These structures he terms “generic models’ and describes them as “case studies converted to explicit dynamic form” and-in an oft repeated phrase-he asserts that “probably twenty basic structures would span 90% of the policy issues that most managers encounter” (Forrester 1980, 18).’ In the same collection of papers, Bell and Senge define generic models as models that “attempt to provide a general theory of the behaviour of a class of systems” (Bell and Senge 1980, 66). It should be noted that a “class of systems” here is described in terms of a class of application situations, e.g. urban development, unlike the “class of systems” defined earlier by Forrester purely in terms of abstract dynamic structure (Forrester 1968~). Bell and Senge advance the cause of generic models as having greater potential refutability because each new specific model, as a member of the class, tests the assumptions of the generic model. A test for “family membership” of a class of systems is described by Forrester and Senge (1980).

Forrester’s assertion that a limited number of dynamic structures could explain the majority of significant dynamic behaviours in such a large and complex application domain has had an enormous impact on system dynamics. It has given generic structures widespread prominence within the field and restates the claim that system dynamics is a “general systems theory to serve as a unifying framework capable of organising behaviour and relationships in areas as diverse as engineering, medicine, management, psychology and economics” (Forrester 1968a). In principle, this claim might be substantiated by identifying a defined set of generic structures that do explain the significant observed dynamic behaviour in an application domain. Certainly the search for these structures has become the system dynamics equivalent of the quest for the Holy Grail.

Abstracted micro-structures

In 1983 Forrester once again looked to the future and-in a paper capable of many interpretations-identified generic models as a concept that needed further exploration (Forrester 1983). In describing generic models here, Forrester appears to suggest that a deeper, more fundamental level of generic model is possible than has previously been discussed. Forrester states that “a model is a theory of the system that the model represents” and that “the primary utility of a theory lies in its generality and transferability” (p. 6). The examples he gives of useful theories are Ohm’s law and Newton’s laws of motion; fundamental theories indeed. Forrester then goes on to suggest that “we should be


seeking general theories of behaviour” and notes that “a model can represent theories within theories” (p. 7). He concludes that simpler more widely transferable structures (theories), which he terms ‘generic structures’, would strengthen dynamic theories of application domains and that applying system insights to theories would strengthen the ties with other fields. From the line of argument taken by Forrester in this paper it is difficult not to infer that some sort of link is being made between system insights and generic structures.

In developing these “abstracted micro-structures”, much progress had already been made. As the theory of system dynamics became established, explanations of the field were made by elaborating from first principles. In Principles of Systems Forrester explains how basic behaviour modes are generated from simple dynamic structures and their components (Forrester 1968d). This kind of treatment was extended by Goodman (1974), Coyle (1977) and Richardson and Pugh (1981) to cover the principle types of dynamic behaviour explained in terms of the two types of feedback loop. These structures were expressed solely in terms of abstract system dynamics constructs (see Fig. 3).

Andersen and Richardson (1980) mapped examples of generic models to a similar sequence of theory-generated structures, or elementary structures as they termed them. Elementary structures are then the building blocks of all dynamic systems including the generic models discussed earlier, and can be seen as one interpretation of Forrester’s “theories within theories”. Again, these are abstracted micro-structures and this interpretation of generic structure may be seen to have its roots in Forrester’s early assertion that system dynamics models are theories of dynamic behaviour for classes of systems, where the class is defined in terms of the underlying abstract dynamic structure (Forrester 1968~). iThink and its user manuals (Richmond et al. 1990) provide an excellent implementation of this idea.

Paich’s review

In 1985 when Paich reviewed the status of generic structures within system dynamics, he attempted to synthesise the different views on what is “generally applicable” about feedback structures in his working definition of generic structures as “dynamic feedback systems that support particular but widely applicable behavioural insights” (Paich 1985, 127). Within this definition he distinguishes between two main viewpoints. The first viewpoint “that generic structures are feedback mechanisms that are transferable to new situations within a particular field” (p. 126), is a direct descendant of the generic model concept, and the examples given, e.g. production/distribution (Forrester


Fig. 3. An Abstracted Micro-structure: ‘abstracted micro- Overshoot and Collapse structure’. These are relatively abstract pieces of system

a fully-specified simulation model Renewable Populatlon which generates a single, commonly observed behaviour mode. This example concerns ‘overshoot

structure supporting Structure

1 Inflow Outflow Loss

v Growth Fraction

and collapse’ and is drawn from Richmond et al. (1990).

Behaviour

Finite

Consumption

Consumptlon per unit RP

Renewable Populatlon

Finite Resource

Structure The system structure is shown here without the accompanying equations as a stock and flow diagram. The loops have been labelled ‘R’ and ‘B‘ to indicate reinforcing and balancing effects. The stock of Finite Resource influences the outflow from the Renewable Population stock, therefore acting as an implicit goal. In turn, the level of the Renewable Population determines the rate of consumption of the Finite Resource. Behaviour The characteristic behaviour of both stocks is shown. The value of Renewable Population initiallyrises but this increases the rate of consumption of the Finite Resource, depleting the contents of this stock. The subsequent increase in the Loss Fraction decreases the value of Renewable Population. The value of Renewable Population collapses to a steady state that is much lower than the values experienced earlier.

1961), product launch (Forrester 1968a), urban development (Forrester 1969), were all previously regarded as generic models. The second viewpoint, “that there are structures that can be transferred across fields” (p. 126), is identified with approaches similar to that of Andersen and Richardson described above. Paich is not attempting a precise definition of generic structures, but it is worth mentioning that the aspect of his definition which is not discussed is


the nature of the “insights” which generic structures are held to support. However, it is clear that the differences between the two viewpoints hinge on whether the application domain or the abstract dynamic structure is the primary focus of interest.

Co un ter-in t uitive system archetypes

This interpretation of generic structure may well constitute the most popular form of system dynamics practice today, its high profile resulting from the work of Peter Senge. In essence, these structures attempt to characterise common dysfunctions in high order dynamic systems and through an understanding of the dynamic structure suggest antidotes. The nature of the dysfunction and that of the antidote are not obvious, i.e. they are both counter-intuitive. This is because “[tlhe complex system presents an apparent cause that is close in time and space to the observed symptoms. But the relationship is usually not one of cause and effect. Instead both are coincident symptoms arising from the dynamics of the system structure.” (Forrester 1969, 110). Structures that develop these ideas to capture this counter-intuitive aspect of dynamic systems we have termed “counter-intuitive system archetypes” (see Fig. 4).

The roots of this interpretation are to be found in Forrester’s characteristics of complex systems which he described in Urban Dynamics (pp. 107-114). Forrester describes high order dynamic systems as being “devious and diabol- ical” (p. 110) and identifies a number of principles or important behaviour characteristics of complex systems, e.g. “complex social systems tend toward a condition of poor performance” (p. 112). These principles encapsulate “generic insights” (Richardson 1991), which help people to understand the relationships between structure, behaviour and policy.

In one of the key works in the area of generic structures, Meadows (1982) represented these insights, and advanced her own, illustrated by simplified causal loop diagrams, e.g. Addiction (p. 105). She neatly characterises them as “computer-free systems insights that any adult can carry in his or her head to deal with the persistent, system-dependent malfunctions of a complicated society” (p. 98). This paper is the source of the actual form of half, and the style of all, of the archetypes in The Fifth Discipline (Senge 1990). While For- rester is primarily concerned with deviousness of dynamic systems, Meadows goes further to propose antidotes. For example, in discussing the tendency of dynamic systems to drift to low performance, she suggests “An obvious antidote . . , is to keep standards absolute. Another is to make goals sensitive to overperformance as well as underperformance.” (Meadows 1982, 105).


Efforts to embody counter-intuitive system behaviour in generic structures were continued at MIT. The “New Management Style Program” explored the role of generic structures in improving managers’ understanding of dynamic behaviour in organisations (Senge 1985). In this paper Senge identifies two levels of structure, generic structures and “archetypal structures”.

Senge defined generic structures as “relatively simple models of dynamic processes that recur in diverse settings and that embody important management principles “ (Senge 1985, 791). Such generic structures are a part or parts of a generic model, a generic model being the final form of a series of generic structures of increasing complexity. The management principles are then lessons concerning types of dynamic behaviour that result as a series of generic structures becomes more complex. Senge gives an example, based on the market growth model (Forrester 1968a). A two-loop model representing market growth with fixed capacity reveals the management principle, “no firm can sell what it doesn’t produce”. The addition of variable production capacity reveals “imbalances between market growth and capacity expansion result in fluctua- tions in demand and sales” (Senge 1985, 791-792) . Although the form of these structures would resemble canonical situation models, their purpose and method of employment is quite different. Within this definition, generic structures are restricted to a particular application domain but-unlike generic models and Paich’s “field dependent” generic structures which seek to be applicable to a class of systems-the primary focus is now on the lesson to be learnt, the management principle that the structure reveals. Implicitly, a structure is identified as generic because there is a transferable lesson to be found within it. Perhaps a little perversely, in this interpretation the structure is virtually defined by the dynamic behaviour insight with which it is associated.

From defining generic structures in this way, Senge moves on in this same paper to identifying what he argues is a deeper level of insightful structure. These he terms “archetypal structures”. Generic structures deal with “management principles” applicable to a particular application domain. In contrast, Senge claims that archetypal structures reveal general system principles. The examples he cites are the very same structures described by Meadows such as “shifting the burden to the intervenor” (Meadows 1982). Senge claims that these archetypal structures are abstract dynamic structures which are “the mathematical (sic) building blocks of all systems” (Senge 1985, 794).3 Forrester had presented his original behavioural characteristics as empirical observations. Meadows elaborates these behavioural characteristics as dynamic structures and discusses their implications. Senge, however, is claiming that these characteristics are caused by fundamental feedback structures which are the building blocks (N.B.) of all systems. This startling claim, for which he cannot be said to have given a complete justification, is aban-


doned in The Fifth Discipline. Instead, it is the concept of dynamic structures revealing general system principles which is advanced in this book.

These same ideas reappear in The Fifth Discipline in the guise of ‘systems archetypes’ (Senge 1990) and are developed further in Senge et al. (1994). This first book has done much to popularise counter-intuitive system archetypes and this third interpretation of generic structure finds its strongest articulation there. Senge’s presentation of counter-intuitive system archetypes emphasises their role as templates or patterns of structure and he does not stress his earlier claim that they are the “building blocks” of systems. Here, counter-intuitive system archtr;-pes are simple causal loop diagrams, though the causal links have no +/- or s/o labels and the variable names are be- quently difficult to imagine quantifying. Each of these “structures” comes with a “business story” and a “management principle”, which the counter- intuitive system archetype neatly brings to life (Fig. 4). Indeed, it is this linkage between the diagram and the principle which it illustrates that is the dis- tinctive characteristic of counter-intuitive system archetypes.

Rather confusingly, Senge now equates these counter-intuitive system archetypes with generic structures (Senge 1990, 94). In fact, counter-intuitive system archetypes are presented in this book as patterns of structure that recur again and again in all aspects of life, so that understanding and master- ing these archetypes can change our perspective in a fundamental way, “[flor . . . only when managers start thinking in terms of systems archetypes, does systems thinking become an active daily agent, continually revealing how we create our reality” (p. 95). Here generic structures-in the guise of counter- intuitive system archetypes-have broadened their primary focus of interest from being the means to develop dynamic insights to the somewhat grandiose ‘ways of seeing’. The key notion seems to be that recognising the counter- intuitive system archetype that best explains a problematic situation can aid the formulation of successful policies. This contrast strikingly with Meadows’ more modest view (Meadows 1982). The all-embracing nature of Senge’s “learning organization” vision actually makes it difficult to engage with him at the same level of debate as with other views on generic structures. His vision, its premises, language, implications and consequences must either be accepted or rejected. This problem is exacerbated by the degree of uncertainty concerning his precise intentions for counter-intuitive system archetypes.

To summarise, counter-intuitive system archetypes, which find their strongest expression as Senge’s “systems archetypes”, attempt to embody system and management principles in simple dynamic structures (see Fig. 4). The common characteristic of counter-intuitive system archetypes is the counter-intuitive relationship between policy and behaviour which these structures exhibit.

loo System Dynamics Review Volume 12 Number 2 Summer 1996

Fig. 4. A “counter- intuitive system archetype”. These are recurrent patterns of structure which attempt to explain commonly occurring dysfunctional behaviours in complex dynamic systems. Each involves a behaviour considered problematic and, by explaining that behaviour in terms of a feedback structure, suggests a counter-intuitive management principle for altering the behaviour. This example concerns “drift to low performance” or “eroding goals” and is drawn from Meadows (1982), as adapted by Senge (1990).

Counter-intuitive System Archety e: Drift to Low Performance / Eroding e oals

Dvsfunctional Be haviour

- - - - Goal

Condition

Structure

P R E S S ~ E S TO B ADJUST GOAL

GOAL \

ACTIONS TO CONDITION 6 IMPROVE

Nanaaeme nt Principle

Goals should not be adjusted in response - > aSt under-performance; “hold the vision” by maintaining standards even though their attainment appears distant.

Dysfunctional Behaviour An obstinate gap between actual and desired conditions increases pressure to lower the goal. Perversely, easing the goal does not close the gap, rather, the actual condition worsens. Structure The system is represented in causal loop diagram form but, in the style of Senge, the links have not been signed. Two balancingprocesses exist to bring Condition and Goal into line. However, the delay between actions and results in the lower one makes the course of response represented in the upper loop more attractive. However, lowering the Goal slightlyreduces the Gap and hence the Actions, thus eroding the condition further. Management Principle see figure.


Generic structures: three views

By tracing the evolution of the generic structure concept we have identified three different views of what they are concerned with (Table I).

In the first view, generic structures yield significantly different modes of dynamic behaviour in an application domain. For these we have employed the term ‘canonical situation models’. These generic structures are closely related to general or generic models, being case studies reduced to their essentials in order to make explicit the causal explanation (or theory) of the dynamic behaviours that the structure generates; see Figure 2.

In the second view, generic structures are relatively abstract combinations of system dynamics components which generate commonly observed behaviour modes. We have called these “abstracted micro-structures”; see Figure 3.

In the third view, generic structures are recurrent patterns of structure which illustrate common but counter-intuitive relationships between policy and behaviour in complex dynamic systems. These we term “counter- intuitive system archetypes”; see Figure 4.

We will now examine the applications of generic structures in system

Table 1. Proposed sub-division of the term ‘generic structure’ with illustrative examples. Note that a mix of past and more recent examples has been chosen.

Generic structures

Counter-intuitative system archetypes

Drift to low performance- Do not adjust goals in response to past underperformance; fix standards (Meadows 1982)

Tragedy of the commons- do not exploit a common resource for relative gain; establish a shared resource management mechanism (Senge 1990)

Organisational addiction- do not reward crises management when more effort produces no gains; reward fundamental solutions to problems (Kim 1992)

Canonical situation models

Abstracted micro-structures

Product Launch (Forrester 1968a)

Urban development (Forrester 1969)

Commodity production cycles (Meadows 1970)

Ambitious product development (Graham 1988)

Economic growth and income distributions in a developing country (Saeed 1994)

Exponential growth as a stock reinforces its own inflow (Andersen and Richardson 1980)

Oscillations as two stocks influence each other’s flows (Forrester 1968)

Overshoot and collapse as a stock’s growth destroys a second, goal-setting stock (Richmond et al. 1990)


Table 2. The three interpretations of generic structure and the contribution that they make when applied to different activities within the field of system dynamics. The dominant contributions of each of the three is indicated by italics. See text for full description.

Counter-intuitive Canonical Abstracted system archetypes situation models micro-structures

System Shortcuts from Constitute a Can suggest dominant conceptualisation dysfunctional library of general structures necessary

behaviour straight models to explain behaviour to management (= hypotheses) to principles be drawn from

Formal model Moving to stocks & Precursor to specific Pre-validated construction flows is problematic application model; building blocks for

[though beyond tailored by complex models intended use) parametrisation

Domain Provide qualitative Serve as general u understanding insights + Sensitise theories of structure (only via completion

perceptions to the and behaviours of of the whole effects of feedback a domain structure of a model)

of using the experimentation structure/behaviour feedback perspective with generalised pairs, rigorously

management demonstrating case studies feedback effects

Teaching Illustrate the value Allow meaningful Introduce simple

dynamics, in order to be able to discuss whether the theoretical distinctions made here hold up in practice.

The application of generic structures

Although generic structures now play an important role in system dynamics’ claim to be an integrative theory of dynamic behaviour, they were generally conceived of for practical purposes. In this study of the application of generic structure we have considered four main application areas: system conceptualisation, model construction, domain understanding and teaching. The role of generic structures in each of these areas is examined in turn (Table 2). Note that we now try to use the three terms proposed in the previous section in order to clarify the interpretation of generic structure that is under discussion at any point.

Generic structures for system con ceptu alisa tion

System conceptualisation refers to the processes by which ideas and theories about the structure of a system and its consequent behaviour begin to be proposed and debated.

The first suggestion of using generic structures for system conceptualisation was made by Forrester when he noted that system dynamicists approach


problem situations by drawing on a mental library of previously studied systems (Forrester 1968b). These “system examples” are our canonical situation models; situation-based generic structures, used as an initial “hypothesis” about the important features of dynamic structure in a specific problem situation. Morecroft et al. (1991) describe using Forrester’s market growth model to generate “in-depth questions for fuelling discussion and structuring information gathering” (p. 115). In using these canonical situation models to focus system conceptualisation, there are a two different views which can be taken on the role of the “hypothesis”. In one view, the “hypothesis” is a scientific theory to be tested and verified. An alternative view, based on a proposal by Lane (1994a), is that the “hypothesis” is similar to Max Weber’s ideal types. These may be viewed as thinking aids, drawn from real phenomena, with which a situation is compared in order to understand its significant components and so generate explanatory value. In either case, verification of the “hypothesis” proceeds by testing the various assumptions in the generic structure for their equivalents in the problem situation. Although the philosophical assumptims behind the two views are quite different, the practical limitations of using canonical situation models for system conceptualisation are rather similar. Both views assume that a sufficient number of suitable “hypotheses” can be found to cover any specific situation that will be encountered. The risk is that using an inappropriate structure will misdirect the investigation of the problem.

The role played by comrm-intuitive system archetypes has evolved somewhat since their appearance in Senge (1990), owing to the work of a number of contributors. Although system behaviours-variables changing over time- were implied by Senge’s “description” of each archetype, actual graphs are now employed in the relevant section of Senge el aJ. (1994, pp. 99-158).4 There is also an element in the new ideas on how to choose which counter- intuitive system archetypes to employ, illustrated in “the archetype family tree” by Goodman and Kleiner (Senge et d. 1994, 149-150). The goal, nevertheless, is to use an approach which then “might suggest an archetype” (Senge 1990,117). Counter-intuitive system archetypes move the user from an idea of problematic behaviour, through a causally-based diagnosis of the reason for the dysfunction and then straight to a surprising management principle inbi- cating ways of alleviating the problem. Put bluntly, this is the startlirg idea that lurks throughout the writings in this area and it has its roots in The Fifth Dis~ ip l ine .~ We can therefore see that counter-intuitive system archetypes seek to leap-frog the whole process of model buildmg, moving the user straight from conceptualisation to insights about r, system.

Abstracted micro-structures have also becii suggested as a tool for system conceptualisation. Andersen and Richmison (1986) advocated that students


of system dynamics build their own catalogue of abstracted micro-structures and situation-based examples as an aid to understanding “general characteristics of structure and behaviour which transfer from one system to another, whatever their surface differences may be” (p. 101). A central premise of system dynamics is that structure generates behaviour, and consequently behaviour can go some way in helping to infer structure. Andersen and Richardson note, however, that the identification of structure from behaviour for real-world systems is suggestive rather than certain and that skill is required to recognise these elementary structures in more complex systems6 Transferability of understanding using abstracted micro-structures comes from being able to make a hypothesis about the dominant elements of system structure from the reference mode. The main limitation to using abstracted micro-structures for system conceptualisation in this way is that only low order structures are of practical use. High order abstracted micro-structures are capable of generating so many different behaviour modes that inferences, beyond identifying the one or two most dominant feedback loops, become impossible.

Generic structures and formal model construction

Formal model construction refers to a rigorous process by which a conceptu- alised system structure is carefully represented in logico-mathematical terms (Lane 1995a). This will probably be in the form of some computer language, thus facilitating simulation experiments.

The most direct use of generic structures for formal model construction is to employ abstracted micro-structures as components of the model. System dynamics software user guides are emphatic in advocating their use. For example, the iThink user guide (Richmond et al. 1990) identifies three levels of generic micro-structure, excluding elementary components, as generic flow processes, generic infrastructures and generic sub-systems. Many of the generic flow processes and the generic infrastructures are identical to the abstracted micro-structures discussed earlier, being simple abstract structures which are known to generate particular behaviour modes. In addition, func- tional structures are described, i.e. generic infrastructures and generic sub- systems, which represent functions such as stock ordering with a delay or the finance sector of a typical company. The rationale for using all of these structures is purely pragmatic; it is claimed that significantly better models are built in significantly less time. These structures were identified by analysing a large number of constructed models and extracting the stock/flow arrange- ments w-3ich recurred most frequently. Some of the larger sub-systems, e.g. a marketing st.c?or, have a superficial similarity with the canonical situation


models discussed earlier. The focus here, however, is on the correct arrange- ment of stocks and flows so as to be internally consistent, rather than the relationship between structure and behaviour in an application domain. Structures used in this way for formal model construction, whether abstracted micro-structures or larger components, can almost be thought of as reliable, de-bugged chunks of code that fulfil a particular purpose. Additionally, the link between their structure and their behaviour has been established rigorously by computer simulation.

Considering the role of canonical situation models in model construction returns us to the early applications of system dynamics, in which a general model was seen as being the precursor to building a specific model of the application problem (Forrester 1961). However, it is not clear whether this general model is operational, with representative parameters, or a logical model used to “home in” on the fundamental relationships in the application domain. The urban growth model (Forrester 1969) is an operational model representing a general theory of urban development, a model which could be taken intact and then specialised to a particular urban situation. Forrester and Senge (1980) suggest this course of action when discussing confidence tests for system dynamics models:

The general t,?enry is embodied in the structure of the model. The special cases are embodied in the pxameters. To make the [family member] test, one uses the particular member of the genera: family for picking parameter values. Then one examines the newly parameterised model in terms of the various [other] model tests to see if the model has withstood transphtation to the special case (p. 221)

Probably the most detailed example of such an analysis is offered iii Yeadows (1970) .7

In the case of counter-intuitive system archetypes, formal model construction is arguably not a relevant issue. This interpretation of generic structure aims to be a complete experience of “systems thinking” and is not particularly intended as an aid to model cmstruction. Nevertheless, John Sterman has argued that “as a first approach to understanding systems, they are valuable and revealing” (Senge et ~ l . 1994, 177). However, it is crucially important to note that he goes OR to observe that counter-intuitive system archetypes constitute merely a hypothesis concerning the shcturehehaviour link and that he argues strongly far the use of computer simuIation for the rigorous testing of such hypotheses. Such accommodation of differing views might seem com- forting but it is necessary to add that Dowling et d. (1995) have reported in detail the difficulties that they experienced in attempting to represent two counter-intuitive system archetypes as stock and flow diagrams in order to simulate their behaviour.’ This adds weight to the idea that many counter-


intuitive system archetypes are too poorly posed to be recreated as formal models.

Generic structures for domain understanding

Domain understanding refers to the learning and insights relating to a particular appIication area that result from applying to it the ideas of system dynamics.

Generic models - in our nomenclature, canonical situation models- “attempt to provide a general theory of the behaviour of a class of systems” (Bell and Senge 1980, 66). It follows that a good general theory should be useful and ultimately be accepted in the application domain for which it is intended. The canonical situation models that have gained widespread atten- tion have also generated widespread controversy, e.g. the urban model and the world model. Perhaps because of this resistance, recent work on them has concentrated on “in-house” uses, such as system conceptualisation using existing canonical situation models, rather than the search for new ones (see the earlier sub-section). Such modesty may be misplaced. Richardson has shown that many of the important theories in social sciesce are implicitly theories of dynamic behaviour, which call also be represented as situation- based generic structures, at leae: in causal loop diagram form (Richardson, 1991). Fortunately, w ~ h in the style of Meadows (1970) does still hold an attraction for i’ne system dynamics community, as evidenced by the work on servk:t: quality conducted at MIT (see Senge and Oliva 1993). The development of canonical situation models to address recurring business problems seems to offer an excellent vehicle for non-competing companies to co-operate in the development of general models which can then be parametrised usin.g data from different industries. If work of the same quality as that of Meadows can be done, then powerful theories aboiit business processes may result and transferable insights gained. This xea of research is truly in harmony with the earliest aspirations of t k field.

If we turn to counter-intuitive system archetypes, the hoped-for contribution to domain understanding has two parts. First, as described above, the aim is to provide some qualitative insight into a situation which offers a guide to action. However, there is a more general type of understanding that counter- intuitive system archetypes also aim to produce. The whole thrust of The Fifth Discipline is that systems thinking is a way of conceptualising the world. As such, the primary purpose of counter-intuitive system archetypes is to sensitise individuals to think about the effects of information feedback and hence to improve their understanding of dynamic systems in the real world. This approach has similarities with Morgan’s description of the use of metaphors


for understanding organisational behaviour: “For the use of a metaphor implies a way of thinking and a way of seeing that pervade how we understand our world generally” (Morgan 1986, 12). A metaphor, or any framework for conceptualisation, highlights certain aspects of the world and obscures others, producing a one-sided kind of insight. Counter-intuitive system archetypes, as an integral part of the larger framework systems thinking, are similarly one-sided. Senge attempts to justify this approach by drawing the analogy between archetypes and, “archetypal personalities” such as the eccentric inventor. He notes that they

... lead us to subconsciously categorize unique individual personalities, with the consequence that we often act more appropriately to the archetype than to the person him- or herself. Might archetypal structures lead, in a similar fashion, to new perceptions and actions? (Senge 1985, 795)

Abstracted micro-structures do not, in themselves, lead to domain understanding. However, if such building blocks are used to construct a model which has been validated and which the owners have confidence in, then it would argue against the ruison d’Btre of system dynamics if understanding did not result. These consequences are best seen, though, as deriving from the use of abstracted micro-structures in the model construction phase.

Generic structures for teaching

Teaching here refers primarily to more formal situations of education and training, though the remarks have some extension to the concepts of the learning organisation.

Andersen and Richardson (1980) suggested a pedagogical approach to system dynamics in which generic structures were seen as means of teaching system conceptualisation skills. The use of a catalogue of elementary structures, i.e. abstracted micro-structures, that they suggest students build has already been discussed (see previous sub-section). In addition, they suggest a number of other useful exercises, such as constructing and de-constructing example models (e.g. the market growth model) component by component, loop by loop, and observing how the behaviour changes. Another method is to have students examine isomorphic systems such as the pendulum and inventory/workforce systems (Forrester 1968~). The purpose of these exercises is to teach students of system dynamics about the possibilities and limitations of transferability of structure.

An alternative project was undertaken at MIT, the computer-based case study project, in which part of the educational objectives was to help managers gain a greater understanding of dynamic behaviour in organisations. Generic models, i.e. canonical situation models, were used as the basis for case study


examples as they provided consistent and known generic problem situations. Graham (1988), in a paper which preceded development of the computer- based cases, identifies 17 generic models representing common “problematic syndromes and behaviour modes”. However, while generic models may make reliable and repeatable teaching examples, he notes that “the weaknesses of generic models for management education are, first, requiring a highly-trained modeler to run the model and interpret its structure and behaviour, and second, even with such guidance, substantial difficulties transferring insights from an abstract model to real situations” (p 2). Nevertheless, the ability of such tools to facilitate the conduct of rigorous and hence meaningful policy experiments indicates the potential benefits to be gained (Lane, 1995b).

So far, little has been published on the use of counter-intuitive system archetypes for teaching. However, at least two ideas are in circulation. Both use counter-intuitive system archetypes to illustrate, and to get students involved in using, the feedback perspective. The first is described in More- croft (1992) and concerns the use of counter-intuitive system archetypes as part of the briefing material for a management flight simulator. The aim is to provide users with a protocol that encourages them to think about the feedback structure of the pre-packaged simulation model that is used. In this instance the “balancing loop with delay” counter-intuitive system archetype helps participants to move away from the “video-game syndrome” and to conduct experiments which are related to an understanding of causal link- ages. The second idea is less well developed, but simply involves the use of counter-intuitive system archetypes as convincing stories that help to bring to life a dysfunctional management situation and that then provide an explanation in terms of information feedback which leads to a surprising management principle, or insight. This approach accepts that insights can only be derived from experimentation with simulation models but uses counter-intuitive system archetypes as an exciting representation of the benefits of embarking on a complete modelling course (Morecroft, personal communication).

By examining the ways in which system dynamicists have used generic structures we have tried to show that the three interpretations that we have identified are distinguishable in practice as well as in theory. The different roles that the interpretations play in the various applications of the generic structure concept are summarised in Table 2.

Observations on validity

If the scientist’s first question is “What is going on here?” then the next must surely be “Where is something else like this happening?”. It is a natural urge


in the sciences to try to test for external validity, “the extent to which one can generalise the results of the research to the populations and settings of interest in the hypothesis” (Judd et al. 1991, 25). Forrester’s “20 structures/gO% of cases” comment is an extraordinarily bold claim of external validity for those 20 models (and also a highly audacious claim for system dynamics). How might it be validated? We do not offer here an exhaustive treatment of this issue since we are continuing research in this area. Also, further comment on the validation of canonical situation models is made in the final section. How- ever, as we have discussed above the ways that generic structures are applied, it seems worth commenting briefly on the consequences that our three interpretations have for validation approaches.

Analogies, isomorphisms and external validity

The means by which users can be encouraged to have confidence in a model have been described in terms of a series of validity tests proposed by Forrester and Senge (1980) and re-framed by Richardson and Pugh (1981) and Lane (1995a). These tests centre on the quality of the mapping between the model and the real world. We build on the ideas of Beer in describing these mappings (Beer 1966; 1984).

In the real world we encounter an array of phenomena. To hand we also have an array of ideas and tools from science. To illuminate the former using the latter we must begin to define more clearly the phenomena and ideas of interest. We make a loose selection of the phenomena to consider and of the scientific ideas to bring to bear. These we call “loose conceptualisations” and they are employed when “we find that some circumstance that we understand in one system throws light on a parallel circumstance in another” (Beer 1984, 8). It is then possible to draw an analogy between loose conceptualisations and so gain some insight. The mapping between loose conceptualisations is therefore a weak, analogical one. If we now define things more precisely, choosing the exact scientific ideas to employ as well as the real world phenomena being studied, then we can move to rigorous definitions. The mapping between rigorous definitions is then an ‘isomorphism’ between scientific models and real systems. In other words, the attributes of the model can be tightly compared one-to-one with the attributes of the relevant system.

The validity of these two mappings is quite different. Beer’s ideas empha- sise that loose conceptualisations are only very broad accounts of situations rather than rigorous definitions. 1 He insists that, without clarity, “analogies may be carried too far” (p. 8) and that analogical mappings are far less strong than isomorphisms. As he puts it, “bedevilling disputation . . . attends analogy” (Beer 1966, 113). In contrast, the isomorphism between rigorous


definitions is the most desired form of mapping in science since it tends towards “a mathematical invariance that transcends analogy” (Beer 1984, 17). If there is a rigorous definition of both the model in question and the real world phenomena under study, then neutral scientific language can be used to test the validity of the isomorphism.

In system dynamics, confidence is created when the mapping between model and system is deemed to be a strong one. To achieve this aim, the validity tests proposed by Forrester and Senge (1980) use a variety of methods to compare the structure and behaviour of a model with the structure and behaviour of the system existing or planned in the real world. These tests result from a clear understanding of the part of scientific ideas that is being employed and so allow detailed comparisons to be made.

These tests primarily concern individual system dynamics models (and are described further in the following sub- section). However, the “Family Mem- ber Test” moves beyond this in the direction of external validity. This test addresses the question of whether “a model [is] a general model of the class of system to which belongs the particular member of interest” (Forrester and Senge 1980, 220). The test essentially involves checking whether the structure is appropriate for all members of a designated class of phenomena and also checking whether the application of appropriate parameter values results in behaviour modes that are appropriate to the class. One is now studying a homomorphism, in that many real world systems are being mapped onto a single model. As with isomorphisms, testing the quality of an homomorphism is only possible when rigorous definitions are available.

The idea of generic structures can therefore be seen as involving the creation of a more general hypothesis about the structure and behaviour of a class of “settings of interest” (c.f. Judd et al. 1991). A strong homomorphism is then required between the model and a carefully defined set of real world systems. It is in this context that a clear division appears between canonical situation models and abstracted micro-structures and, alternatively, counter-intuitive system archetypes.

On the external validation of two interpretations

Canonical situation models and abstracted micro-structures can be seen as rigorous definitions with good mapping aspirations. The validity tests result from a clear understanding of the scientific ideas being employed and allow detailed comparisons to be made. Using tests such as “structure verification” and “parameter verifi~ation”~ it is clear that an isomorphism concerning structure is being tested: by creating opportunities for refutation we can form a judgement on the correspondence between the structure of the model and


that of the real world system. Similarly, by rigorously deducing behaviour with a simulation model, tests such as “behaviour reproduction” and “surprise behaviour” allow a judgement to be made regarding a behaviour isomorphism. This may be qualitative or may be more numerically based (Sterman 1984). Finally, the presence of a functioning simulation model also allows experiments to be performed which can add greater weight to the structure and behaviour isomorphism but which also result in policy insights that have been tested using sensitivity analysis. These steps all result in a “good” model of a system, that is, a model in which users have confidence regarding the claimed isomorphism. However, the last step is then the application of the “Family Member Test”, by which one tests the hypothesis that a model of system structure combined with parameter sets and observed past behaviours is related to members of a designated class of systems. The aspiration is to move from isomorphism to homomorphism.

It is because of their rigorous definition that, in the case of canonical situation models and abstracted micro-structures, it is possible to pass through these stages. The isomorphism can be tested and so subsequently can the homomorphism. Of course, there will always be differences regarding the classes that are of interest-in the case of the second these will be drawn in a more abstract manner-but the tests continue to be meaningful. And these remarks do not imply that performing the “Family Member Test” and establishing the external validity of a candidate generic structure is unproblematic. But they do indicate that the rigorous nature of canonical situation models 2nd abstracted micro-structures makes reasonable the hope of establishing such validity by providing a broad empirical basis for a modeI. The situation is very different with counter-intuitive system archetypes.

External validity and counter-intuitive system archetypes

Kim (1992) summarises the use of counter-intuitive system archetypes in the following way: “Often, an event that is seen as a problem symptom can be the starting point. This may lead you to trace out the pattern of behaviour of similar events over a period of time. The archetype can then help you identify the systemic structures that are responsible”. So, rather than building up a structure piece by piece, in this approach the user embarks on what Sterman calls “a multiple choice exercise” (Senge et al. 1994), choosing from a range of counter-intuitive system archetypes and arriving at a revealing management principle. There are two crucial difficulties here, one concerning the system structure, the other concerning the associated behaviour. First, the system structure comes ready made, via the choice of counter-intuitive system archetype. It may be vague and badly posed when compared with other sys-


tem dynamics models (the causal links are not signed and some of the concepts do not readily encourage the application of numerical values). The work of Dowling et al. (1995) supports this view. It therefore has a much weaker mapping to the real world based on criteria such as structure and parameter verification. Second, counter-intuitive system archetypes make a strong, though highly questionable, inference between a particular behaviour pattern and the underlying dynamic structure. This arises both in the dysfunctional behaviour that aids in the choice of counter-intuitive system archetype and in the more acceptable behaviour that is supposed to result from the application of the counter-intuitive management principle. However, with counter-intuitive system archetypes there is no rigorous linkage between system structure and system behaviour: the user is called upon to make an inference rather than a deduction. But this is known to be a highly perilous process. Rich- mond (1994) observes that “using causal-loop diagrams to make inferences about behaviour is a treacherous business” (p. 144) and Richardson (1986) offers fine examples of the difficulties of inferring behaviour patterns from qualitative model. Similarly, Forrester (1994) holds that “causal-loop diagrams do not provide the discipline to thinking imposed by level and rate diagrams” (p. 252). The lack of a simulation model also rules out debate about sensitivity analysis and robustness of the counter-intuitive policy. With counter-intuitive system archetypes, behaviour is merely inferred by mental simulation, a process which is known to be treacherous, and its implied use arguably constitutes a rejection of one of the core assumptions of system dynamics.

Considering these two difficulties, we claim that there can be no question of there being an isomorphism between models and the real world in the case of counter-intuitive system archetypes. In addition, therefore, the homomorphism required for external validity cannot be a realistic aspiration.

If it is not possible to test the external validity of counter-intuitive system archetypes then what is their status? Beer provides the key diagnosis. The scientific ideas employed in counter-intuitive system archetypes are a weakly outlined sub-set of the field of system dynamics. Similarly, the models used are imprecise. Both ideas and models can be seen as loose conceptualisations. In consequence, counter-intuitive system archetypes have merely an analogical relationship with real-world phenomena. By offering a shortcut to understanding dysfunctional system behaviour, counter-intuitive system archetypes simply ask us to see the world through these structures because we may find it helpful. It is not possible to question the appropriateness of the structures. Similarly, it is not really possible to establish clearly when to apply a particular counter-intuitive system archetype, partly because they are so general and imprecise. Their definition is merely analogical, therefore making validation


impossible. A comparison with metaphors has already been made. But they can also be likened to the “issue-based models” employed in soft systems methodology (Checkland and Scholes 1990), where the sole measure of validity is whether the model gives rise to insights which are deemed to be helpful (Checkland 1995). A case can be made for this approach: in weakening or los- ing the possibility of structure and behaviour isomorphisms, “[this] validation approach would seem to presage the triumph of usefulness over representa- tiveness” (Lane 1995a, 113).1° However, this stance devastates the external validity aspirations of this interpretation of generic structure. As Beer puts it, “the validity of a particular analogy must always be open to disputation, and its relevance in a particular instance cannot be formally demonstrated” (Beer 1966, 112).”

Conclusions

In tracing the evolution of the generic structure concept we have tried to elu- cidate some of the subtleties of meaning which have been implicit in different views of that concept. Generic structures in all their forms express the basic ambition to transfer experience and understanding from one dynamic situation to another. We have identified three different interpretations of what a generic structure is and what transferability means (Table 1). These different views are distinguishable in practice as well as in theory (Table 2).

In the first view, generic structures are theories of a number of significant modes of dynamic behaviour in an application domain; canonical situation models (Fig. 2). They offer transferability of both dynamic structure and particular causal assumptions about dynamic behaviour in that application domain. These generic structures are, in a technical sense, indistinguishable from any general or generic model which has been reduced to its dynamic essentials. The rigorous definition of such models, if combined with an equally rigorous statement of the application domain to which they are relevant, means that validation of each model is possible since one is dealing with isomorphisms between models and phenomena. This is not to say that establishing external validity is unproblematic. One of the difficulties for this form of generic structure is that there is no obvious way of validating or refut- ing a generic structure as being one of the 20. As Barlas and Carpenter point out, “model validation is a gradual process of building confidence in the usefulness of a model; validity cannot reveal itself mechanically as a result of some formal algorithms. Validation is a matter of social conversation, because establishing model usefulness is a conversational matter” (Barlas and Carpen- ter 1990, 157). This comment is particularly relevant since Forrester’s asser-


tion that 20 generic structures would cover 90 per cent of the policy issues that most managers encounter can never be proven. Greater confidence in using a canonical situation model for a particular problem situation can really only come about when the structure has been socially accepted into the application domain as a valid theory for interpreting a particular class of problems. This suggests that more effort should be put into engaging in debate about the use and validation of canonical situation models in application domains. Within the system dynamics community, such general models have been suggested as being useful primarily for system conceptualisation and for revealing general domain understanding. In addition to the MIT case study project, more needs to be reported on the success or failure of using these structures in practice. As stated above, this area of research is consistent with the earliest aspirations of the field. Research difficulties and previous controversy should therefore not discourage work in this area since canonical situation models may prove to be the single most profound and influential interpretation of the ‘generic structure’ concept.

In the second view, generic structures are combinations of abstract system dynamics components which generate single, commonly observed behaviour modes; these are described as abstracted micro-structures (Fig. 3). They offer transferability of structure, pure and simple. Abstracted micro-structures are more easily definable than canonical situation models, as almost any abstract dynamic structure can be a candidate for consideration as one. For system conceptualisation, however, only low order structures are of practical use because of the difficulties of inferring structure from behaviour in complex systems. The main contribution of abstracted micro-structures is to model construction, where their value lies in their reliability and predictability as model sub-systems with a rigorously validated link between structure and behaviour. As with canonical situation models, it is possible to make sensible tests of the external validity claims of this interpretation of generic structure. Clearly delimited though this area of applicability may be, the contribution made by abstracted micro-structures has been very important to the field and there are no reasons to believe that this will alter.

In the third view, generic structures are recurrent patterns of structure which illustrate common but counter-intuitive relationships between policy and behaviour. These we have called “counter-intuitive system archetypes” (Fig. 4). As a result of the influence of The Fifth Discipline (Senge 1990), their primary application is to act as shortcuts from identifying dysfunctional behaviour in dynamic systems to suggesting managerial solutions. Again, the validation of these structures is conversational within the particular community which uses them, as they are seen as equally applicable to all types of dynamic system. As an integral part of a larger conceptual framework, belief


in their usefulness depends on acceptance of that framework. Regarding the external validity aspirations of counter-intuitive system archetypes, the loose link between structure and behaviour is a concern to us. But perhaps the more disturbing aspect of them is that they constitute a retreat from the rigours of scientific, isomorphic comparison and instead encourage analogical thinking. Analogical thinking normally precedes scientific thinking, but system dynamics already possesses a detailed set of scientific ideas. In this respect, counter- intuitive system archetypes appear to be a significant retreat for the field. We certainly feel that the nature and choice of counter-intuitive system archetypes cannot be validated in a meaningful scientific sense. So what may be concluded? Counter-intuitive system archetypes may reduce the perceived need for model construction and the carefully validated application of the feedback perspective. Alternatively they may generate further interest in the whole process of system dynamics modelling. It is simply too early to say what effect this most recent interpretation of ‘generic structure’ will have on the field.

By tracing the evolution of the ‘generic structure’ concept and examining the way that generic structures have been used in practice we have tried to show that at least three distinct interpretations of the concept can be identified. As interest in dynamic systems increases and the scope of systems dynamics thinking grows to encompass different variations and even paradigms (Lane 1994b), we feel that it is no longer desirable for different notions of what is generalisable and transferable about dynamic systems to be concealed by such an ill-defined and all embracing term as ‘generic structure’. As stated previously, this term has become overburdened with meaning. For research in this area to advance in a rigorous and constructive way we believe that it is time to unbundle the concept of ‘generic structure’ so that three distinct interpretations can be debated. Following on from our opening Chalmers (1992) quotation, we hope that we have begun to make the underlying theory of the ‘generic structure’ concept more precise and coherent and that the three interpretations of this important concept that flow from this are more clear. We further hope that this paper may allow system dynamicists to evaluate more critically the theoretical and practical uses of generic structures with respect to the range of assumptions inherent in their problem solving and modelling approaches.

Notes

1. This paper presents a more developed version of the ideas which first appeared as Lane & Smart (1994).


2. One example of the re-statement of this idea may be found in Forrester (1993).

3. Note that the assumptions behind these archetypal structures have quite different roots to what we have termed abstracted micro-structures. It is not clear in Senge (1985) whether archetypal structures are simple causal loop diagrams, stock and flow diagrams or functioning simulation models. In contrast, the form of abstracted micro-structures is clearly defined.

4. This book contains contributions by very many authors. However, these are not always discrete chapters but rather appear sometimes as small sections or even just comments. For this reason we have chosen not to attempt a citation for each idea that is referenced but have given page numbers for the whole volume.

5 . To give one example; “deeper systemic causes . . . are not recognized. With the aid of the systems archetypes, these causes often can be understood and, in many cases, successful policies can be formulated” (Senge 1990, 117).

6. Regrettably, the implications of this crucial proviso would appear to be less attractive than the perceived promise of the more superficial readings which have become popular of late.

7. In this book a gneral model of production, inventory, consumption and price of a commodity is established. This model is then applied to the case of pork production and, after a very careful process of adaptation, is shown to provide a theory for “the hog cycle”. A similar, though truncated, argument is used to apply the model to the case of beef and hence the cattle cycle and finally an appendix offers the parametrisation necessary to adapt the model to the production of chickens.

8. In November 1995, the independent consultant Gene Bellinger made available via the system dynamics bulletin board on the Internet a suite of iThink simulation models based on counter-intuitave system archetypes. From his address (CrbnBluBaol.com) he produced 11 files. These are based on the list of 10 in Senge (1990) but with the omission of “Shifting the Burden to the Inter- venor” and the addition of “Reinforcing Loop” and “Accidental Adversaries”. (The first addition will be clear to all readers, but further details of the second may be found in a piece by Kemeny in Senge et al. 1994, 145-148). The 11 models are presented in the most recent version of iThink, having integrated explanatory text, graphs and control panels. These last feature sliders which enable the models to be run in “flight simulator” mode, changing parameters mid-simulation.

Although “Fixes That Fail” involves a curiously specific-and unconvinc- ing-example, most of the models attempt to preserve very general variable labels. However, this leads to some very odd system structures, the lack of realismof which would cause them to fail validity tests. Also, the use of the flight simulator interface draws the user into a regrettable discrete event approach. Reliance on the flight simulator approach may additionally be the reason why some of the key feedback loops are not represented in the actual model structures, being left instead to assumed user intervention for implementation. For example, “Growth and Underinvestment” is a strange distortion of the original (Forrester 1968a), the structure of which does not contain the second, capacity-regulating, balancing loop. This process must be implemented


by input in the flight simulator mode but, since this is set up to allow only “bursts” of investment which are reset to zero after one time unit, discrete rather than continuous output results and the potential insights are hard to deduce. Similarly, “Tragedy of the Commons” has a bizarre structure which allows continued activity of the players even when the Common Resource has fallen to zero. Although the reinforcing loops are mapped out, the balancing loops are implemented solely as an arbitary “Non-negative” setting for the Common Resource stock. The model supports the expected behaviour mode at first but the unrealistic structure leads eventually to modes that are incompre- hensible. On a minor note, the “Balancing Loop” model has no delay and so no oscillations result, which would seem to separate the model from the delay-management insights described in Senge’s original. Finally, although over-ridden in the flight simulator mode, two of the models contain algebraic errors which reverse the polarity of important loops.

We have examined these models with care since considerable work has clearly been put into them. However, the description above does nothing to diminish our view that creating rigourous models of counter-intuitive system archetypes is fraught with a range of difficulties.

9. These are the terms used in Forrester and Senge (1980). In more general simulation terms they may be thought of, respectively, as “face validity” and a combination of conceptual and numerical correspondence (see Lane 1995a).

10. In this paper Lane argues that archetypes-and other forms of “qualitative system dynamics-sacrifice “formulational validity”, “experimental validity” and “data validity” in order to increase “process effectiveness”. He comments that, whilst the resulting “operational validity” may still be high, this is pur- chased at a severe cost to “analytical quality”, (Lane 1995a).

11. It is perhaps worth indicating a parallel between our argument here and one employed by Morecroft et al. (1995). They develop a model of water sharing in a multiple shower system and then propose this as a “metaphor” for the co- ordination of subsidiaries in multinational corporations. We are in disagree- ment with these authors regarding their use of language; in describing types of links between real-world phenomena and scientific ideas/models they are implicitly using the terms “metaphorical” and “analogical”, whereas above we have employed “analogical” and “isomorphic”. However, such linguistic squabbles are strictly secondary. The important point is that they too are con- trasting loose and precise (or rigorous) links. They are clear that, although their model is a rigorous representation of the shower system, it is merely a metaphor for the corporate phenomena. Their careful statements about the difficulties and limits of employing such loose connections between ideas are therefore in harmony with our concerns regarding counter-intuitive system archetypes. Although the authors are clearly intrigued by the effectiveness during class teaching of the metaphorical usage, their caution regarding the validity of any consequent deductions is laudable. Users of counter-intuitive system archetypes would benefit from following their example.


References

Note: Books originally published by MIT Press have been republished by Productivity Press, Portland, OR.

Andersen, D. F., and G. P.Richardson. 1980. Toward a Pedagogy of System

Barlas, Y . and S. Carpenter. 1990. Philosophical Roots of Model Validation: Two

Beer, S. 1966. Decision and Control: The Meaning of Operational Research and

, 1984. The Viable System Model: Its Provenance, Development and Path-

Bell, J. A., and P. M. Senge. 1980. Methods for Enhancing Refutability in System

Chalmers, A. F. 1982. What is This Thing Called Science? (2nd ed.). Milton

Checkland, P. B. 1995. Model Validation in Soft Systems Practice. Systems

Checkland, P. B. and J. Scholes. 1990. Soft Systems Methodology in Action.

Coyle, R. G. 1977. Management System Dynamics. London: John Wiley & Sons. Dowling, A. M., R. H. MacDonald, and G. P. Richardson. 1995. Simulation of

Systems Archetypes. In Proceedings of the 1995 International System Dynamics Conference: Volume 11-Parallel Program eds T. Shimada and K. Saeed, 454463. Tokyo: Gakushuin University and the International System Dynamics Society.

Forrester, J.W. 1958. Industrial Dynamics: A Major Breakthrough for Decision Makers. Harvard Business Review 36 (4): 37-66.

Dynamics. In Legasto et al. 1980: 91-106.

Paradigms. System Dynamics Review 6 (2): 148-166.

Management Cybernetics. Chichester: John Wiley & Sons.

ology. Journal of the Operational Research Society. 35 (1): 7-125.

Dynamics Modeling. In Legasto et al. 1980: 61-73.

Keynes: Open University Press.

Research 1 2 (1): 47-54.

Chichester: John Wiley & Sons.

. 1961. Zndustrial Dynamics. Cambridge, MA: MIT Press.

. 1968a. Market Growth as Influenced by Capital Investment. Zndustrial

. 1968b. Industrial Dynamics-After the First Decade. Management Science

. 1968c. Industrial Dynamics-A Response to Ansoff and Slevin. Manage-

. 1968d. Principles of Systems. Cambridge, MA: MIT Press.

. 1969. Urban Dynamics. Cambridge, MA: MIT Press.

. 1971. World Dynamics. Cambridge, MA: MIT Press.

. 1980. System Dynamics-Future Opportunities. In Legasto et al. 1980:

. 1983. Future Developments of the System Dynamics Paradigm. System Dynamics Working Group Paper D-3715. Sloan School of Management, MIT, Cambridge, MA.

. 1993. System Dynamics and the Lessons of 35 years. In A Systems- based Approach to Policymaking ed. K. B. De Greene, 199-240. London: Kluwer.

. 1994. System Dynamics, Systems Thinking, and Soft OR. System Dynam- ics Review 10 (2-3): 245-256.

Management Review (now the Sloan Management Review) 9 (2): 83-105.

14 (7) 399415.

ment Science 14 (7) 601-618.

7-2 1.


~ ~ ~~

Forrester, J. W. and P. M. Senge. 1980. Tests for Building Confidence in System Dynamics Models. In Legasto et al. 1980: 209-228.

Goodman, M. R. 1974. Study Notes in System Dynamics. Cambridge, MA: MIT Press.

Graham, A. K. 1988. Generic Models as a Basis for Computer-Based Case Studies. System Dynamics Working Group Paper D-3947. Sloan School of Management, MIT, Cambridge, MA.

Judd, C. M., E. R. Smith and L. H. Kidder. 1991. Research Methods in Social Relations (6th ed.). London: Holt, Rinehart and Winston.

Kim, D. 1992. Systems Archetypes. Toolbox Reprint Series 6 (2). Cambridge, MA: Pegasus Communications.

Lane, D. C. 1994a. With A Little Help from Our Friends: How System Dynamics and Soft OR can Learn from Each Other. System Dynamics Review 10 (2-3):

. 1994b. Social Theory and System Dynamics Practice. Proceedings of the 1994 International System Dynamics Conference-System Dynamics: Method- ological and Technical Issues, eds. E. Wolstenholme and C. Monaghan, 53-66. Stirling, Scotland: University of Stirling. Expanded as LSE Working paper in operational research No. LSEOR.95.20.

. 1995a. The Folding Star: A Comparative Reframing and Extension of Validity Concepts in System Dynamics. In Proceedings of the 2995 International System Dynamics Conference: Volume I-Plenary Program, ed. T. Shimada and K. Saeed 111-130. Tokyo: Gakushuin University and the International System Dynamics Society.

. 1995b. On a Resurgence of Management Simulations and Games. Journal of the Operational Research Society 46 (5): 604-625.

Lane, D. C., and C. Smart. 1994. Mad, Bad and Dangerous to Know? The Evolution and Limitations of the ‘Generic Structure’ concept in S.D. In Proceedings of the 1994 International System Dynamics Conference-System dynamics: method- ological and technical issues eds E. Wolstenholme and C. Monaghan, 67-77. Stirling, Scotland: University of Stirling.

Legasto, A. A., J. W. Forrester, and J. M. Lyneis eds. 1980. System Dynamics: TIMS Studies in Management Science. Volume 14. Amsterdam: North-Holland.

Meadows, D. H. 1982. Whole Earth Models and Systems. CoEvolution Quarterly. Summer 1982: 98-108.

Meadows, D. H., D. L. Meadows, J. Randers, and W. W. Behrens 111. 1972. The Limits to Growth. a Potomac Associates Book New York: University Books.

Meadows, D. H., D. L. Meadows, and J. Randers. 1992. Beyond the Limits. London: Earthscan Publications.

Meadows, D. L. 1970. Dynamics of Commodity Production Cycles. Boston: Wright- Allen Press.

Morecroft, J. D. W. 1988. System Dynamics and Microworlds for Policymakers. European Journal of Operational Research 35 (3): 301-320.

. 1992. Design of a Learning Environment. In System Dynamics 1992: Pro- ceedings of the International System Dynamics Conference, 1992, Utrecht, The Netherlands eds. J. A. M. Vennix, J. Faber, W. J. Scheper and C. A.Th. Takken- berg, 465-474. Boston, MA: System Dynamics Society.

101-134.


Morecroft, J. D. W., D. C. Lane, and P. S. Viita. 1991. Modeling Growth Strategy in a Biotechnology Startup Firm. System Dynamics Review 7 (2): 93-116.

Morecroft, J. D. W., E. R. Larsen, A. Lomi, and A. Ginsberg. 1995. The Dynamics of Resource Sharing: A Metaphorical Model. System Dynamics Review 11 (41: 289-309.

Morgan, G. 1986. Images of Organisations. Newbury Park, CA: Sage Publications. Paich, M. 1985. Generic Structures. System Dynamics Review 1 (1): 126-132. Richardson, G. P. 1986. Problems with causal-loop diagrams. System Dynamics

Review 2 (2): 158-170. . 1991. Feedback Thought in Social Science and Systems Theory. Philadel-

phia: University of Pennsylvania Press. Richardson, G. P. and A. L. Pugh 111. 1981. Introduction to System Dynamics Mod-

eling with DYNAMO. Cambridge, MA: MIT Press. Richmond, B. 1994. Systems Thinking/System Dynamics: Let’s Just Get On With

It. System Dynamics Review 10 (2-3): 135-157. Richmond, B. M., P. Vescuso, and S. Peterson. 1990. iThinkTM, Distributed by

High Performance Systems, 145 Lyme Road, Suite 300, Hanover, NH 03755, USA.

Saeed, K. 1994. Development Planning and Policy Design: A System Dynamics Approach. Hong Kong: Avebury.

Senge, P. M. 1985. System Dynamics, Mental Models, and the Development of Management Intuition. In Proceedings of the 1985 International Conference of the System Dynamics Society: Volume II, ed. D. F. Andersen, N. B. Forrester and M.E. Warkentin, 718-798. Boston, MA: System Dynamics Society.

Senge, P. M. 1990. The Fifth Discipline: The Art and Practice of the Learning Organization. New York: Doubleday.

Senge, P. M. and R. Oliva. 1993. Developing a Theory of Service Quality/Service Capacity Interaction. In System Dynamics 1993: The Role of Strategic Modelling in International Competitiveness, ed. E. Zepeda and J. A. D. Machuca, 476485. Boston, MA: System Dynamics Society.

Senge, P. M., C. Roberts, R. Ross, B. Smith and A. Kliener. 1994. The Fifth Disci- pline Fieldbook. London: Nicholas Brealey Publishing.

Sterman, J. 1984. Appropriate Summary Statistics for Evaluating the Historical Fit of System Dynamics Models. Dynamica 10 (winter): 51-66.