‘SYSTEM DYRAMICS:
PORTRAYING BOUNDED RATIONALITY
John D.W. Morecroft
Assistant Professor of Management
System Dynamics Group
Alfred P. Sloan School of Management
Massachusetts Institute of Technology
Cambridge, Massachusetts 02139
To be presented at the
1981 System Dynamics Research Conference
The Institute on Man and Science
Renseelaerville, New York
October 14-17, 1981
273
D-3322 1
SYSTEM DYNAMICS:
PORTRAYING BOUNDED RATIONALITY
John D.W. Morecroft
Massachusetts Institute of Technology
Cambridge, Massachusetts 02139
ABSTRACT
‘Thie paper examines the linkages between system dynémics and the
Carnegie school in their treatment of human decision making. It is argued
that the structure of system dynamics models implicitly assumes bounded
rationality in decision making and that recognition of this assw:
would aid system dynamicists in model construction and in c
other social science disciplines. The paper begins by examiring Simon's
"Principle of Bounded Rationality" which draws attenticn to the cognitive
limitations on the information gathering and processing rowers cf
decision makers... Forrester's “Market Growth Model" is used to illvetrate
the central theme that system dynamics models are portrayals cf tcurded
rationality. Close examination of the model reveals that the information
content of decision functions is limited and that the inforne
processed through simple rules of thumb. In the final pert cf the raper
there is a discussion of the implications of Carnegie rhilosarhy fer syste
dynamics, as it affects communication, model structuring and enelysis, and
future research.
INTRODUCTION
‘The field of system dynamics has long been viewed by its
practitioners as a discipline which is distinct from other major
methodologies dealing with industrial and social systems. In particular, a
distinction is drawn between system dynamics and the dominant concertual
framework offered by economics and operations research: However, to those
outside the field, system dynamics is often regarded as nothing more than a
rather specialized form of simulation modeling which belongs in the general
tool kit available to management scientists.
D-3322 2
Part of the reason for the persistence of these divergent viewpoints
has been the inability of the system dynamics community to express its
4ifferences in modeling philosophy in a language that is understandable to
those outside the field. There are of course exceptions to this general
statenent, for example, in the work of Meadows [1] and Andersen [2]. This
paper adds 2 new dimension to that work. Its purpose is to draw attention
to the fact that there is a widely recognized school of thought with a
philosophy to social systen modeling that has striking parallels to the
vnderlying philosophy of system dynamics. This school of thought offers a
language ani set of concepts that may greatly improve our ability to
coumunicete with other fields, and develop a stronger internal sense of the
contribution that syetem dynamics can make to the analysis of social
systens.
The school of thought is known generically as the Carnegie School,
in reecgnition of the institution where much of the pioneering work was
done in the 1950s and 1960s. A common and powerful theme underlying the
work of the Cernegie School is the notion that’ there are severe limitations
on the infcrmation processing and computational abilities of human decision
nakers. 4s a result, decision making can never achieve the ideal of per-
fect (objective) rationality, but is destined to a lower level of intended
rationality. The Carnegie School contends that the behavior of complex
crgenizations can only be understood by taking into account the psychologi-
cl and cognitive limitations of its human members. Such a viewpoint
focuses attention on the flow of information in a complex system, the
quantity and quality of information that is amenable to human judgmental
274
p-3322 3
processing, and the form of decision rules used to represent judgments. it
is at thie most fundamental level of information flow and processing that
strong parallels with system dyannics can be found. It ie also at exactly
the same fundamental level that the Carnegie School departs radically fron
the traditional views of economics and operations research. [3]
In the main body of the paper, I will develop the ties between the
Carnegie School and system dynamics. In doing so, it is not my intention
to create the impression that the two fields are the seme--they most cer-
tainly are not. Rather, they share some philosophy in common, philosorky
that the Carnegie School has made explicit in its writings and that in
syaten aynimics has alvays been implicit. ‘The development begins vith a
more careful look at the “principle of bounded rationality”, which is the
cornerstone of Carnegie philosophy. Next, the structure and behavior of
Forrester’s “Market Growth Model" [4] ie interpreted in the light of the
principle of bounded rationality, leading to the major conclusion that
system dynamics models are attempts to portray and unravel the consequences
of bounded rationality. Finally, there is a discussion of the implications
of Carnegie pitilosophy for system dynamics, as it affects communication,
model structuring and analysis, and future research.
D-3322 4
‘BOUNDED RATIONALITY
The Principle of Bounded Rationality
‘The principle of bounded rationality was formulated by Simon as the
basis for understanding human behavior in complex systems. The principle
recognizes that there are severe limitations on the thinking and reasoning
power of the human mind. If we wish to predict the behavior of human
decision makere within the context of the systems in which they work and
live, it is first necessary to take account of their psychological
properties.
Simon has defined the principle of bounded rationality in the
following vay: [5]
“The capacity of the human mind for formulating and
solving complex problems is very small compared with the
size of the problems vhose solution is required for
objectively rational behavior in the real world or even
for a reasonable approximation to such objective
rationality."
The principle of bounded rationality provides a basis for the
construction of a theory of organizational behavior. [6] In Simon's words:
{6}
“Organization theory is centrally concerned with
identifying and studying those limits to the achievement
of goals that are, in fact, limitations on the
flexibility and adaptibility of goal striving
individuals and groups of individuals themselves.”
2 275-
D-3322 5
The principle of bounded rationality suggests that the performance
and success of an organization is governed not by the enonynous interplay
of market forces, but rather by the peychological limitations of its
members: the amount of information they can acquire and retain, and their
ability to process that information in a meaningful way. These linitations
in their own tum are not physiological and fixed, they deperi on the
organizational setting within which decision making takes place.
Bounded Rationality and Organizational Decision Making
‘The principle of bounded rationality leads one to expect that
organigations will undertake decision making in such a way as to greatly
simplify the information processing and computational load placed on the
human decision makers it contains. ‘The pioneering work of Cyert and March
[7] indicated that decision making in real business firms is indeed ruck
simpler than one would anticipate based on classical models that assure
objectively rational behavior. In the following section, ve will drew on
the work of Oyert and March to identify a number of empirical features of
organizational decision making that can be interpreted as consequences of
the principle of bounded rationality. We will consider such things es
organizational at weoeetiee and the information collection and processing
habits of human decision makers. Our ultimate purpose is to show that rany
of these features are implicit in the structure and policy formulations of
a system dynamics model.
1.3322 6
1. Factored Decision Making
Common experience with human organizations will reveal that decision
making responsibility is factored or parceled out among a variety of
cutunits. For exenple, many business firms adopt @ functional structure
that divides decision making between marketing, production, pricing,
finance, labor management, etc. Cyert and March point to a “division of
lator" in decision making. The decision problems that an organization must
solve are so conplex that they cannot be handled by an individual.
Seperabie fieces of decision making are assigned to organizational subunits
in the form of subgcals. Each subunit is charged with the responsibility
of necting its own subgoal and thereby contributing to the broader
objectives of the organization. Of course this scheme will work perfectly
if there is no inherent conflict in goals - something which cannot be
on!
guaranteed shen subunits are interdependent. Nevertheless, factored
decision mekizg is a necessary feature of a complex organization, and it
goes hand in rand with a multigoal structure. Factoring simplifies
decision mekirg, but at the cost of focusing the attention of subunits
narrowly on rerforzance relative to the subunit goal.
2. Partial and Certain Information
Decisicns are made on the basis of relatively few sources of
infornetion that are readily aveilable, and low in uncertainty. While the
above statement is not a direct quote from the Carnegie School, it can be
inferred from comments that are made about the way that information is
obtained and yrocessed by an organization. Enpirical observations of
decision making in organizations indicate that decision makers seek only a
276
D-3322 7
small proportion of the information that might be considered relevant to
full consideration of a given situation, ‘Their search for information
tends to be conditioned by a focus upon problem aynptons and by a desire to
avoid the use of information that is high in uncertainty. Both these
tendencies in information selection favor the use of local feedback
information reflecting current conditions in the inmediate operating
environment of a subunit, rather than information gathered more widely
whose impact upon the subunit can be only vaguely conjectured. Both cyert
and March, and Simon, coment on the frequency with which euch locai
feedback information ie used rather than more global information required
for “optimal” decision making.
3. Rules of Thunb
The organization uses standard operating procedures or rules of
thumb to make-and implement choices. In the short run, these procedures, do
not change, and represent’ the accumulated learning embodied in the factored
decision making of the organization. Rules of thumb need enploy only suell
amounts of information of the kind that would be made available through
local feedback channels. Rules of thumb process information in a
straightforward manner, recognizing the computational limits of normal
human decision makers under pressure of tine.
Consider, for example, the pricing decision of a business firm.
Microeconomic theory would suggest that pricing decisions result fron a
sophisticated profit maximizing computation which equates marginal cost and
marginal revenue. In fact, there is evidence to euggest that computation-
D-3322 8
ally simpler markup pricing is common, Under this method, average variable
cost is taken as a base and is increased by a fractional markup to obtain
the selling price. The markup is a rule of thumb which is heavily
influenced by past tradition and by feedback information on profit, return
on investment, market share, etc. (For an example of rule of thumb
pricing, see Mass [8], pp. 31-36.)
BOUNDED RATIONALITY IN A SYSTEM DYNAMICS MODEL
In this section we will take an existing system dynamics model and
interpret its structure and behavior in the light of the principle of
bounded rationality. The model selected is based on Forrester [9] and
describes the policies governing the growth of sales and production
capacity in a new product market. Forrester’s original model resulted from
4 case study of an electronics manufacturer, and represents the opinions of
senior management of the company about the vay that corporate growth is
maneged. We will first discuss the atructure of the model and show how it
embodies the organizational features of factored decision making, partial
information and miles of thumb. We will then show, using simulation rns
of the model, how the bounded rationality of organizational subunits can
cause problems in market growth.
2277
D-3322 9
Factored De on Making in the Market Growth Model
In common with the Carnegie School view, the market growth model can
be broken down into a number of organizational subunits each of which is
responsible for a part of the decision making that produces growth in the
system as a whole.
Figure 1 depicts an organization with decision making factored into
four subunits. In subunit 1 on the right of the figure, customers make
their ordering decisions. Ordering is influenced by the runter of customer
contacts made by the marketing department and by customers’ rerceptions of
the delivery delay in obtaining the product. In subunit 2 the marketing
department makes decisions on the hiring of marketing personnel. fn upper
Limit on marketing personnel is set by a marketing budget which coves in
proportion to sales volume. Hiring adjusts personnel to this budgetary
limit. In subunit 3 the firm makes decision on order filling. The rate of
ty and its
order filling depends on the available production capac
intensity of utilization. Finally, in subunit 4, the firm rekes decisions
on capacity management. Additional capacity is ordered whenever high
delivery delay indicates there is a capacity shortage.
Partial Information and Rules of Thumb
In this section, we will consider in more detail the decision rules
for capacity management, marketing, and customer ordering to illustrate
examples of rules of thumb and the use of partial and certein information.
10
~t A
“Customer
Contacts
\
sales | \
Volume | \
ORDER
FILLING
ORDERING
Delivery
Delay
| Pelivery
| Petey
Capacity |
\
CAPACITY
‘MANAGEMENT
Pigure 1. Organizational Subunits in the Market Growth Model
1, Capacity Management
Capacity managenent is represented as a two-stage decision making
Frocess involving first the detection of capacity shortage and then the
ordering of capacity to elininate the shortage. During the analysis the
ader should bear in mind that the basic objective of capacity management
is to edjust capacity to a level that will support demand. One could
readily visualize a “rational” decision function in which future
expectations of demand are generated acrosa the lead time of capacity, and
278
D-3322 "
the expectations are used to drive capacity ordering. As we shall see, the
decision making process that is actually used is computationally simpler
and requires far less information. let us first look at the equations for
the detection of capacity shortage:
Dbc(t) = DDRC(t) /DDOG( +) a)
DDRC(t) = DDRC(ty) + \; {5(+)/orn(+)) ~ DDRC(t) dt (a)
to
DooG(t) = DDT(t)*DDW + DDMG*(1-DDW) G3)
por(t) = DDr(t,) + \ ppRc(t) = ppr(t) at @)
ty ‘TOOT
In equation (1), delivery delay condition is an index of capacity
shortage based on the ratio of delivery delay recognized by company DIRC to
delivery delay operating goal DDOG. When DDC is greater than 1, a capacity
shortage exists since it is not possible to fill orders at a rate that will
keep delivery delay equal to the operating goal. Equation (2) states that
DDRC ie an exponential average of the ratio of backlog B to order fill rate
orr. [10] Equations (3) and (4) model the delivery delay operating goal as
an adaptive goal based on a weighted average of a fixed delivery delay
management goal DDMG and delivery delay tradition DD which reflects past
performance. Delivery delay tradition is formulated as an erponential
average of recent delivery delay DDRC.
Consider now the cognitive and information processing assumptions of
equations (1-4). The way the company recognizes the need to expand
capacity is by making judgment on delivery delay condition. The judgment
requires comparison of current delivery delay to the operating goal-
D-3322 12
Current delivery delay is known from information on backlog and order fill
rate which is readily available from concrete operating data, and therefore
low in uncertainty. There is no need to go outside the company to do
elaborate market surveys and project future demand expectations. If demand:
is growing, it will be reflected in a rising backlog. Thus, in equations
(1) and (2), we see a clear use of partial and certain information.
Furthermore, in equation (1), the information is processed in a simple rule
of thurb that compares current: delivery performance to an operating goal.
In equations (3) and (4) the operating goal itself is seen to be a rule of
thumb vhich adjusts to past traditions of performance. In conclusion, we
see, that the entire process of detecting capacity shortage uses only
packlog and order filling information processed on the basis of simple
judenental criteria.
Wow let us consider capacity ordering which is represented by
equations (5-7) below.
con (t) = c(t) *CEP(t) G5)
; t
O(t) = (9) a CAR(t) dt (6)
‘to
cEF(t) = r(pdc(t)) £(1) = 0, £10, £0 (7)
Fquation (5) states that capacity ordering rate COR is the product
of capacity C and capacity expansion fraction CE. Thus capacity ordering
takes place by a fractional expansion of existing capacity. Existing
capacity in equation (5) is simply the integral of capacity arrival rate
CaP (aesuming no capacity discards). Equation (7) states that capacity
expansion fraction CEF is an increasing nonlinear function of delivery 279:
D-3322 13
delay condition,DDC. When DDC is equal to 1, the function takes a value of
aero indicating there is no pressure to expand capacity because delivery
delay is in line with the operating goal. As PDC increases above 1, the
function becomes positive resulting in capacity expansion in equation (5)
The function in equation (7) has a second derivative greater then zero
indicating more aggressive ordering as rising DDC indicates a more serious
capacity shortage.
Again, consider the cognitive and information processing sssunptiors
of equations (5-7). The most striking feature of the equations is that
nowhere is there an explicit attempt to compute the capacity needed to
support demand. Capacity ordering is a rule of thumb that responds to
“pressure” from delivery delay condition signaling capacity shortage.
Delivery delay condition is the only information entering the ordering
decision. There is no information from the market or from the marketing
subunit. A policy of fractional expansion is conputationally sicple--a
judgmental process that causes capacity to change’ in the right direction,
but without the need to compute capacity requirenents.
2. Marketing Expansion
Marketing expansion involves a two-stage decision making process of
budgeting and hiring.
Consider first budgeting as represented by equations (8) and (9)
below.
BM(t) = AOPR(t)*PO*FEM (8)
AOFR(t) = AOFR( ty) a. OFR(t) - AOFR(t) dt (9)
to . TAOFR:
The budget to marketing BN is defined as the product of average
order fill rate AOFR, price of output PO, and fraction of budget to
marketing FEM. The average order fill rate is defined in equation (9) as
an exponential average of current order fill rate OFR. Together equations
{8) and (9) represent a simple budgeting process in which a fixed fraction
M of the total budget AOFR*PO is allocated to marketing.
The hiring of marketing personnel is represented by equations
10-12).
t
(t) = P(t) + N PH(t) at (10)
to
PH(t) = (INP(t) - MP(t)) /TAMP fn
Ct) = BHC t)/MS 12)
in equation (10) marketing personnel MP is defined as the integral
of persone! hiring PH. In equation (11) personnel hiring is formulated as
a goal adjustment process which eliminates the discrepancy between, indica-
ted marketing personnel IMP and marketing personnel MP over a time period
time to edjust marketing personnel. Finally, in equation (12), IMP
is defined 2s the personnel that can be supported at a marketing salary MS
by a budget BY.
Consider now the information processing assumptions of equations
($-12). ng is a simple rule of thumb involving a fixed fractional
allocation to marketing. Such a “frozen” budgetary process is computation-
ally sinpler for an organization than one in which allocation fractions are
derived fron e zero base. Hiring is a goal adjustment process. It uses
280
D-3322 15
only information that is specific to the marketing subunit: the current
level of marketing personnel MP, and the authorized target IMP. Hiring
does not include information about capacity or the delivery delay operating
goal of the organization, both of which could conceivably be of relevance
in a “fully informed" hiring decision.
3. Customer Ordering
An important feature of the market growth model is that customer
ordering is entirely endogenously generated. The initiative for growth
rests with the company. The company must contact customers in the market
and persuade them to buy the product. Customer ordering (here called the
product order rate POR) is represented by equations (13-16) below.
e
PoR(+) = C(t) *=PA(t) (13)
c(t) = MP(t)*NCR (14)
EPA(t) = £(DDRM(t)) —£°<0., (15)
“
DDRC(t) - DDRM(t) dt (16
DPRm(t) = DDRM( ty) + nd TR »
In equation (13) product order rate POR is formulated as the product
of customer contacts CC and effect of product attractiveness EPA (we assune
that each contact can generate no more than one order). Thus a customer
will order only if contacted, and only then if the product seems attrac-
tive. In equation (14), customer contacts are expressed as a fixed
multiple NCR of marketing personnel MP. NCR, the normal contact rate,
represents the average number of contacts made by marketing personnel
during a month. In equation (15), the effect of product attractiveness EPA
is formulated as a decreasing nonlinear function of delivery delay
223322 16
recognized by the market DDRM. Customers are assumed to be sensitive to
delivery delay: they will be discouraged from ordering as delivery delay
grows. In equation (16), DDRM is formulated as an exponential average of
delivery delay recognized by the company DDRC, to represent the customers’
perception of delivery delay.
Consider the information processing assumptions of equations
(13-16). Customers need only two pieces of information to make their
decision: they need to be made avare of the product and they need to judge
ite delivery delay. They need to know nothing at all about the detailed
condition of the company (such as its marketing and capacity plans). Even
their knowledge of delivery delay is local to the market and need not be
the same as actual delivery delay the company ia currently achieving.
We have now completed our review of the major decision functions in
the model. We have shown that the formulations can readily be interpreted
in the light of the principle of bounded rationality. Decision making is
fectored into subunits each striving for separate goals: marketing 1a
striving for a personnel goal dictated by the budget, capacity managenent
is striving to maintain delivery delay at a value dictated by the delivery
delay operating goal. All the decision functions use partial and certain
information which is but @ small fraction of the total availeble to
descrite the state of the system. There are numerous. examples of rule of
thumb decision making.
281
D-3322 17
Bounded Rationality Underlies Problem Behavior
In this section, three simulation experiments will be presented to
show how well-intentioned policies (intendedly rational) can lead to
problem behavior in a complex organizational setting. We start by showing
that our three policies for marketing, capacity expansion, and ordering ere
intendedly rational. In other words, they are capable of producing
reasonable behavior when taken in isolation or in simple combination. A
demonstration of intended rationality is important because it indicates
that there is a rationale to support the existing policies. we then bring
all three policies together in a “complex organizational system" and show
that together they can fail to bring about market growth even though the
marketing policy is striving for growth and there is no inherent Limit to
market size.
Experiment 1 - Interaction of Customer Ordering and Narketing
In Experiment 1, the interaction of customer ordering and rarketing
is examined in isolation from capacity constraints on order filling. With
no capacity constraints, delivery delay is constent so product order rete
POR in equation (13) is directly proportional to marketing pe7sonnel.
Figure 2 shows a simulation run of the simplified system over a time perio¢
of 80 months. Product order rate and marketing personnel both display
unlimited exponential growth. Expansion of marketing personnel leeds to an
increase in product order rate which in turn leads to an increase in the
budget for marketing, thereby justifying further marketing expansion.
Growth is limited only by the delays in personel hiring end by the sales
8/05/81 08:26 RUNTMARKETING AND PRODUCT ORDERING
noc Lat
POR A, ERASE MPH
PAGE FILE:
D-3322 18
Marketing,
Personnel
‘
Figure 2. Marketing and Product Ordering with No Supply Constraints
efficiensy of marketing personnel. We can interpret Figure 2 to mean that
the marketing policy is intendedly rational in the sense that it is able to
bring about market growth under conditions of perfect supply.
Experinent 2 + Capacity Expansion in Isolation
In Experiment 2, we look at the behavior of the capacity expansion
policies in isolation from the market.
We look at the simple question of
282
D-3322 19
whether the capacity expansion policies are able to bring about an increase
in supply in response to an exogenous increase in demand. We are therefore
concerned not with how the demand increase is generated, but merely with
the ability of capacity management to make a “rational response.”
Figure 3 shows the results of a simulation run over a time period of
120 months. ‘The run starts with the system in a state of equilibrium.
Order fill rate is equal to the product order rate of 1000 units per nonth.
Capacity is steady at 2000 unite per month with a utilization of 50°
percent. Delivery delay is equal to the operating goal which is set at two
months. In the 10th month of the simulation, product order rate is
increased by 50 perceht. We trace the adjustment of the manufacturing
system over time.
The simulation run shows that the demand’ change is encothly
accommodated. Shortly after the increase in product order rate, capacity
utilization increases, thereby allowing order fill rate to rise before any
permanent change in the level of capacity has taken place. By month 24,
order fill rate is equal to product order rate, meaning that the demand
increase has been satisfied. Rising delivery delay after week 10 degine to
set in motion a long-term capacity adjustment. As we might expect,
capacity rises only gradually, reflecting long delays in capacity
acquisition and a reluctance to commit to expansion before there is solid
evidence in terms of delivery delay to justify expansion. As capacity
arrives in month 24, utilization and delivery delay gradually return to
their starting values.
28 RUNSCAPACITY MANAGENENT
FILEMGCIS! 6/05/81 09:
PAGE S
D-3322 20
Figure 3. Capacity Management with Exogenous Product Order Rate.
We can interpret Figure 3 to mean that the capacity expansion policy
of equations (1-7) is intendedly rational. It brings about a gradual
expansion of capacity in response to reliable and conclusive delivery delay
information, indicating that expansion is justified.
e 3 ga Ba
i tid iccieetttttete tthe
Orde Rate fT, order Pall. Rate '
. B86
o —
war .
88¢
ess
S38
a &ss
a Seed
: ’ i
3 Delivery + ‘Z
2 cgratnd salar H
iS ee ‘-
gag & wonthe @
5
283°
D~3322 aa
Experiment 3 - Bounded Rationality in a Complex System
In the final-experiment we put all three rolicies together starting
fron the knowledge that the marketing policy can generate growth and that
the capacity expansion policy can follow increases in demand. ‘There. is
good reason to expect a system that will generate continuel grovth in
orders and capacity. In fact, continual growth need not occur, and under
extreme conditions, stagnation and decline can set in.
The final simulation run shows the extrere that results in decline.
Jo appreciate this run, the precise conditions under which it wae generated
should first be described. Customer ordering end marketing perform accord-
ing to exactly the same rules used in Experiment 1. Capecity expansion is
slightly modified in two respects in relation to" Experiment 2. First,
delivery delay operating goal DDOG of equation (3) is made entirely a
function of past tradition by setting delivery delay weight ITW equal to 1.
In experiment 2 DDOG was equal to a fixed management goal DING. The change
is a subtle one that in no way alters the apparent "logic" of the capacity
management policy. A condition of excess delivery delay will still elicit
capacity expansion, but continued failure to meet the operating goal will
result in goal deterioration. Second, capacity expansion fraction CEP in
equation (7) is modified to include a delivery delay bias DEB as shown
delow:
CEF(t) = e(ppe(t) - DpB) £(1) = 0, £120, "0 (17)
In the modified equation DDB plays the role of a management attitude toward
capacity expansion. When DDB is greater than 0, management has a congerva-
tive attitude toward capacity expansion, preferring initially to overcone
1:28 RUNSALL SUBUNITS
B-3322 22
* 5
es 8 Hes sess s Gye
= a Saar EWS | BBS
= = Delivery! 7 \ -
Delay “KY © g
144
‘arket}ng Personnel
'
:
“Product Order
Rates
S vonths #
Figure 4. Interaction of Marketing, Product Ordering,
and Capacity Management
suppiy shortages by increasing the utilization of capacity rather than
ordering aii
ional capacity. However, if delivery delay condition rises
sufficiently high, capacity will be expanded in the "normal" way. Again,
the underlying “logic” of the capacity ordering policy has not changed.
ising delivery delay will still elicit capacity expansion, but the
evidence from ielivery delay must be more compelling than in a case when
IDB = 0.
» 284
D-3522 2
Figure 4 shows the behavior of the complete market growth model over
an interval of 120 months. The run starts from a condition in which
delivery delay is 4 months--twice the initiel operating goal, and a clear
sign that capacity expansion is required (even with DDB'= 0.3). Capacity
expands for the first 18 months of the run. In addition, marketing
personnel expand, bringing about a growth in product ordering. The initial
pattern of growth is the one we might anticipate from the previous two
experiments. However, beginning in month 18, growth in the system begins
to falter, and eventually decline in both capacity and marketing personnel
ects in. This behavior can be explained ao a consequence of bounded
rationality in both the capacity management and marketing policies.
Capacity expansion requires convincing and solid evidence of capacity
shortage in the form of high delivery delay--a requirement that is
reasonable in isolation. High delivery delay depresses ordering and
ultimately inhibits growth of the total budget. With e fixed budget
allocation (that worked well in isolation) growth of marketing is
restricted, thereby eliminating the primary driving force behind growth.
The complete set of policies interact in a way that fails to bring forth
the growth potential of the market. Failure arises because the
consequences of a vell-intended (intendedly rational) policy within one
subunit radiates unintended effects elsewhere in the system. It is this
failure due to unintended consequences that is the hallmark of bounded
rationality.
Yo summarize, in this section we have tried to show how a system
dynamics model can be interpreted in the light of the principle of bounded
D-3322 24
rationality. We have argued that the marketing, capacity expansion, and
ordering policies of the “Market Growth Model" are formulated recognizing
implicitly the information processing limitations of human decision makers.
Using simulation runs, we have shown that the policies are intendedly
rational: the marketing policies promote exponential growth and the
capecity policies bring about a conservative adjustment of capacity to
demand changes. We have also demonstrated with a final simulation run that
intendedly rational policies can produce unintended consequences which are
characteristic of bounded rationality. In the market growth model, the
unintended consequences lead to market stagnation and decline.
IMPLICATIONS FOR SYSTEM DYNAMICS
The principle of bounded rationality is a powerful general law
underlying ell socisl systems. Its consequences have been studied and
developed in the widely known literature of the Carnegie School. The
principle is an implicit part of the structure of system dynamics nodels.
‘These three statements strongly suggest that the ideas of the Carnegie
School have sone important implications for the field of system dynamics.
As I see it, there are implications for the communication of the field to
other social science disciplines, for the structuring and testing of systen
tynamics models, and for research on generic structures. Each of these
themes will be developed in more detail below.
285 >
D~3322 25,
Improved Communication to Other Disciplines
System dynamics is not well understood or accepted outside the
system dynamics community. Part of the reason for this lack of acceptance
is that system dynamics models are not clearly differentiated fron the
other mathematical modeling methods of the social sciences such as
economics and operations research. System dynamics often deals with areas
of application similar to economics and operations research. When analysis
yields results conflicting with the more conventional approaches, the
discrepancy is often explained by appeals to the importance of e "feedbeck
approach" or “systems philosophy," neither of which conveys muck mesning to
those outside the field. ‘he Carnegie School offers a lenguege in vhich
the fundamental difference of system dynamics models is explained in te:ra
of the models’ treatment of information flow and information crocessing in
decision making.
System dynamics models are built implicitly on the principle of
bounded rationality. “They portray the bounded rationality of huzen
decision makers and human organizations. They show the distributed
responsibility for decision making that is characteristic of real
organizations. They contain multiple goals. They use local feedback
information in decision making rather than sophisticated future
expectations. Decision functions are portrayed as rules of thunb,
requiring limited information input and limited computation of that
information.
D-3322 26
286
When we realize that system dynamics models are portraying bounded
rationality, we cen understand why they should be different to models of
other dominant social science disciplines. Both operations research and
nicroecononics focus on portrayals of efficient and rational decision
maxing. In contrast, system dynamics focuses on portrayals of bounded
rationality, with the intention of identifying the information structures
that are consistent with bounded rationality. ‘Therefore, as a theory of
decision maxing, system dynamics differs sharply from classical economics
ty assuming that non-rational behavior is both likely to occur and likely
to be eustained over time. Asa tool for normative analysis, systen
dynamics differs from classical optimization by setting out to explain why
inefficiences exist and seeking decision functions that improve on existing
Yehavior, rather than striving for optimal decision functions regardless of
the existing decision making structure of the system.
Sonceptualization--A Focus on the Organization
System dynamics offers a number of structuring principles [11] to
guide model formuletion. These principles are of most value when the
toundary of the model has already been set and the major interacting
elerents already identified. There is very little guidance for the
earliest end sonetines most challenging step of initial conceptualization.
‘what features of a situation make it suitable for analysis with system
dynamics? Are there patterns of structure that one can anticipate in the
construction of a system dynamics model?
p-3322 ar
‘The process of conceptualization would be greatly aided if we
clearly recognized that we are building models of human organizations and
that those organizations are governed by the principle of bounded
rationality. [12] We could then anticipate both the general form and
specific structural features of a model.
In general form, syaten dynamics modele are likely to portray @
depth of organizational structure. They will involve multiple sectors or
subunits with divided decision making responsibility. It is within the
complex structure of a multisector system that bounded rationality is most
likely to proauce najor problens in overall systen behavior. Problens that
are posed within the setting of # single organizational subunit are
unlikely to be suitable for analysis with system dynamics.
Specifically, we would expect decision making within subunits to
reflect the limits of human rationality. Thus, it is extremely unlikely
that a decision function in a given subunit will be gathering large
quantities of information from distant parts of the organization. Tecision
functions are‘likely to employ locally available information, and to
process this information with simple rules of thumb. Where very complex
formulations arise, there are grounds for questioning the complexity and
seeing whether a more compact formulation can achieve the same basic
intention. Decision functions should reflect the multi-goal feature of
large organizations. Different subunits will be responsible for different
goals and their decision making biased toward achieving those goals
independent of their impact on overall system performance.
D-3322 28
‘The observations above are indicative of the structuring aids that
can be obtained from a Carnegie perspective. Much work could undoubtedly
be done in this area building on the fine structure of decision making that
has been described in Carnegie writing, but which is not covered in this
paper. New structuring aids would complement, not contradict, the existing
principles of formulation,
Behavior Analysi: cing Use of Intended Rationality
The analysis of system dynamics models is traditionally broken into
partial and whole model tests. Partial model tests usually perform a
purely technical function, enabling the modeler to eliminate formulation
errors in a small model rather than unravel the same errors in the more
complex setting of the complete model. The explanation of the behavior of
the system is made in terms of whole model tests.
‘The Carnegie School approach to organizations suggests there may be
powerful insights to be derived from contrasting partial and whole model
tests. Pertial model tests can be viewed as demonstrations of the intended
rationality of decision making in organizational subunits. Partial model
tests often reveal that the policies of a subunit make perfect sense when
the subunit is free to act, independent of other organizational constraints.
Whole model tests indicate how intendedly rational policies can break down
and produce problem behavior in a sufficiently complex organizational
setting. Contrasting partial and whole model tests, to show that rational
policies can in fact produce problem behavior, is a powerful method of
287
D-3322 29
generating understanding of complex system behavior. Understanding is
created by building upon the intuitively clear behavior of # subunit or
small group of subunits. As additional subunits are edded, a clear
explanation can be generated of why policies begin to fail in the nore
complex setting.
An example of the contrast of partial and whole model tests vss
presented earlier in the analysis of the market growth model.
Partial model tests of the marketing and capacity management policies
indicated reasonable behavior of the two policies taken in isolation. 4
full model test involving customer ordering, marketing, and capscity
management revealed non-rational behavior in which the firm feiled to grow
when faced with a limitless market for its product.
Research on Generic Structures
‘The ideas of the Carnegie School are likely to be valuable in
providing methodological support for the concept of generic structures. In
common with other disciplines, system dynamics is seeking order and genera?
structure in the social systems with which it is dealing. There is, of
course, already a structure that is common to all system dynanics codels.
‘They all use the same basic building blocks of levels, physical flows,
information flows, and decision functions. However, beyond the conmon rate
level structure, the question remains whether there are larger groupings of
basic building blocks that might occur repeatedly in social end econcmic
systems. These larger groupings are described as generic structures.
B-3322 30 D-3322 bi
288
To adopt a Carnegie School perspective, the question of whether ‘ REFERENCES AND FOOTHOTES
generic structures erist is similar to the question of whether common forms
of organization exist. The principle of bounded rationality tells us that
organization should evolve around the cognitive limitations of its menbera. ty Meadows, D-H. "The Unavoidable A Priori” in Elements of the
Systen des Method, Cambridge, Massachusetts, M.I.T. Press,
It is probable (though by no means certain) that common organizational agp mance Meth
structures have evolved to cope with these common and fundamental limita- fe] Andersen, D.F, “How Differences in Analytic Paradigms Can Lead to
Differences in Policy Conclusions” in Elements of the Systen
tions of human decision makers. It is also probable that common problems Dynanics Method.
are generated by these organizational structures. {3] For an interesting analysis of the difference between rational and
. organizational paradigms af decision making see Allison, G.T.,
Essence of Decision, Boston, Little, Brown and Company, 1971.
The ideas of the Carnegie School lead us to a research method for [4] Forrester, J.W. “Market Growth as Influenced by Capital
Investment." Sloan Management Review 9, no. 2, pp. 83-105, Winter
findimg generic structures end testing empirically whether they are indeed a 1968. ‘
meric. Generic structures should be defined by the breakdown of (5) Simon, H.A.\ "Rationality and Decision Making", in Models of Man.
New York, Wiley, 1957.
different crganizational subunits, by the channels of communication between
[6] Simon, H.A. Administrative Behavior, 3rd Edition, New York, Free
subunits, and by the mental shortcuts embodied in rules of thumb for Press, 1976.
decision making. If in a particuler application we observe a piece of (7] Cyert, R.M. and J.. March. A Behavioral Theory of the Firm. Kew
Jersey, Prentice Hall, 1963-
structure that is responsible for problem behavior, we might then dissect
: {8} Mass, N.J. “Introduction to the Production Sector of the National
the strusture to ask whether it can be explained as a consequence of Model." Syatem Dynamics Group Working Paper D-2737-1, July 1977,
Sloan School of Management, M.I.T., Cambridge, Massachusetts.
bounded rationality. What is it about the structure, and in particular the
fo]. Forrester, J.¥. "Market Growth as Influenced by Capital
assumed complexity of the information network, that limits the rationality Investment.” Sloan Management Review.
of decision making? What changes in the information network. would be [10] For an example of how (2) reduces to an exponential average see
. : Forrester, J.W. Industrial Dynamics, Appendix E, Cambridge,
compatible with more rational decision making, and why do they not Massachusetts, M.I.T. Press, 1 +
currently exist? Answers to questions like these could form the basis of a (11) Forrester, J.W. Principles of Systems. Cambridge, Mass., M.I.T.
Press, 1968.
refutable empirical study of other organizetions similar to the one that
[12] Models of biological and ecological systems have also been
yielded the generic structure. constructed. I am suggesting these be distinguished from social
systema on the grounds that their principles of organization are
not necessarily the same as social systems, and the principle of
bounded rationality may not apply so strongly to them.
D-3322 32
APPENDIX A
DYWAMO Listing of Market Growth Mode)
TOF:
00010 NOTE MARK TNELUENCED BY CAPITAL IN
09009 NOTE aCiey GReATeD Sy SOHN Doky RU yay teat
0093¢ NOTE FROM KGCI FOR PAPER ENTITLED SYSTEM Bit PORTAYALS:
0080 NOTE OF EOUNUED RATIONALITY PRESENTED AT 1861 CONFERENCE
0089 NOTE IN SYSTEN DYNA
ROES) NOTE TeTMODEC TST EAGED ON I¢He FORRESTER’S MARKET GROWTH MODEL
00070 NOTE IN SLOAN MANAGEMENT REVIEW, VOLS, NO. 2. WINTER 1968, PP2Q-105
mi
ne
2000, _UNITS
KLEPR. JK
ALSC.KAUC.K¥SHOCE((B.K/DDMG)#(1-SHEC))
RY DELAY CONDITION
ASIRE
BORG. J+ (DT/TDDRC) (DDI. J~DDRC. J)
engetcn
gowanze z
Reece Ss
ta
tate
TE S
Mae Hi
2G 3
SSB!
SSSSSSSsSSSSSE'
eeese2sseees:
8:
ESSE:
Sit
3:
eg
at
>ZDOzrD;
STANDARDS
60231 10G..K= (DDT.K) (DDW)+(DDMG) (DDKE)
00240 N DDWC=1~D0W
09989 € bnGe2 now
99270 & DUT Ke att NSIBT/TODT)(DDRC. J-DDT. JD
00280 N DOT
caze0 C pots 12 AONTHS
00300 NOTE
00310 NOTE CAPACITY MANAGEMENT AND SUPPLY
90920 Ls KaC. J+ (DT) (CAR SK
2380 f Tet2000_untrs oF ouTPUT
Sire CGR AL SDeL AFSL CR Ak com
470 4 LOE. K=Uot., J+<DTD (COR. JK-CAR. JKD
Ooae0 HU
90380 R COR (L=C.KACEF .K¥SHC
=TABHL TCEE «(DDC K-DDB) 0124515)
07/=.02/0/ .02/.07/.15
99 6 DDB
00440 NOTE UTILIZATION OF CAPAGITY
90450 A UC ARs TASH (TUC,DDM. A Or5
0460 T TUG=0/ 25/.9/.67/.| 1) 897-887 95/.97/ 90/1
00470 DOM. K=B.K/E.K
00480 NOTE
00480 NOTE | MARKETING
00550 L BP,K=HP. J4(DT) (PH.JK)
00510 N HP=I0
00520 R PH.KL=(IMP.K-MP.K)/TAHP
530 C TAAP=20 PaNTHS
00560 A BH.K=AOFR.K#POSFEM
06570 4 ACen AOFR. J+(DT/TADFR) (OFR. JK~AOFR. J)
00520 N AGFR=OFR
93 530 TRORRST MONTH
94600 C POsS80 S/UNIT
G0810 C FEHS.1
00620 NOTE
289 t
D~3322 33
NOTE. PRODUCT oRDER:
8288 Foe HEPA ae ##(198TEP(SPOR SPOR) )
oes E Gosi"Gepenscontacy
Goseo € Sran-0/ fePoR= ay
00870 A CC.K=(MP.KASKMAMPC® (1-SHM) )#NCR
BoE ab ane
00900 € NER [00 CONTACTS/eE90700 © NCR=100 CONTACTS/PERSOR-HONTH
90710 A NEPA. K#SHPAYEEPA.K( {~SHPA
00720 ¢
oro 6 N AgtE (TERA, DDRM KO, 10;
900740 T
CRETE 08/ .09/ 02
00780 LDR ALAR. J (BabA CBDR. eeebanaedy
00890 NOTE CONTROLS
99840 PLOT PO HE RReE sve Syont-oywe-n/cerer
150 PRINT oR EA: BHP /OFRsDDRM, PHC, DDOG.CEF
Soseo spec prs S/LEN /PLTPER=2/PRTPERSO
00870 RUN COMPII
00880 C LENGTH=80
00880 & SuM=1
e800 E Suce+O
goaie © IEPA 75
890 PLOT heen E/MPH
140 RUN ING AND PRODUCT ORDERING
eNenatns120
BLIPER:
gasses
S28
iit
es
‘3!
3:
$4
20
Bes:
ef
#
DB =O
or en
Or PoRs# t=0/ PASE. UE*U/DDI=D/HP=H/CEF*F
LENGTH =120 :
BEB8E88
azo
Zannonan:
2gngener 3:
FE Ses a
3
gescesosooces:
EEESSss:
Bs8ssss'
3322 4 D-3322 35
APPENDIX B
DYNAMO List of Variable Names
macros DYnawe 8705/81 wcctet DYNANO 2/05/81
LIST OF VARIABLES PLTPER © 30 PLOT PERIOD (MONTHS) <30>
PO c 21.3 PRICE OF OUTPUT (DOLLARS/UNIT) <21>
SYMBOL T WHR-CMP DEFINITION POR R 22° PRODUCT ORDER RATE (UNITS/NONTH) <22>
R 4 PRODUCTION RATE (UNITS/MONTH) <4>
ACER & 21 AVERAGE ORDER FILL RATE (UNITS/MONTH) <2t> PRIPER: c 30 PRINT PERIOD (MONTHS) <30>
Nn Qt” ‘SPA c 27.2 tat IN PROOUCT ATTRACTIVENESS (OIMENSIONLESS)
a & "2 eacKtos (unsTs) <a>
Noa ‘spor © 22:2. STEP-IN PRODUCT ORDER RATE (DIMENSIONLESS) <22>
eM A 29 GUDGET TO MARKETING (DOLLARS/MONTH <20> ‘swe c 13.1 SWITCH FOR CAPACITY (DIMENSIONLESS) <13>
c L 10 CAPACITY (UNITS OF QUTPUT/MONTH) <10> ‘swee c 4,1 SWITCH FOR CAPACITY CONSTRAINT (DIMENSIONLESS)
N 10.8 <a>
caR ® 1 CAPASTTY ARRIVAL RATE (UNITS OF OUTPUT/MONTH/ ‘Sw © 23.1 SWITCH FOR MAPKETING (DIMENSIONLESS) <23>
BOTH) <tt> SwPA © 2414 SuiTcH FOR PRODUCT ATTRACTIVENESS
A 23 CUSTOMER CONTACTS (CONTACTS) <23> (DIMENSIONLESS) <24>
a 14 CAPCITY EXPANSION FRACTION (FRACTION/MONTH) <14> TAM: c 18,1 TIME TO ADJUST MARKETING PERSONNEL (MONTHS) <16>
Rg 13 CAPACITY UPDER RATE (UNITS OF OUTPUT/MONTH/ TAOFR c 21.2 ‘TIME TO AVERAGE OROER FILL RATE (MONTHS) <2t>
<ia~ ‘TCEF T 14.1 TABLE FOR CAPACITY EXPANSION FRACTION
c DELAY BIAS (DIMENSIONLESS) <5> {OIMENSIONLESS) <14>
A GELAY CONDITION (DIMENSIONLESS) <5> TODRC c 6.2 TIME FOR DORC (MONTHS) <6>
a7 DELAY INDICATED (MONTHS) <7> Toor © 26.2 TINE FOR DORM (KONTHS) <26>
A 18 DELAY BIR IHU<(MONTHS) <16> toot S (9.2 TINE FOR ODT (WONTHS) <9>
c oa DELAY MANAGENENT GOAL (MONTHS) <@> teen T 25.1 TABLE FOR EFFECT OF PRODUCT ATTRACTIVENESS
a) DELAY OPERATING GOAL (MONTHS) <a> (OIHENSTONLESS) <25>
L686 DELAY RECOGNIZED BY COMPANY (MONTHS) <6> SPA © 27.3 TINE FOR STEP IN PRODUCT ATTRACTIVENESS <27>
y8. SPOR © 2212 TINE FOR STEP IN PRODUCT ORDER RATE (MONTHS) <22>
L 26_—DELIVERY DELAY RECOSNIZED BY MARKET (MONTHS) <26> Tuc T 15-1 TABLE FOR UTELIZATION OF CAPACITY
N 26.1 (DIENSTONLESS) <15>
£9" DELIVERY DELAY TRADITION (MONTHS) <9> uc A 4S “UTILIZATION OF CAPACITY (DIMENSIONLESS) <15>
“ a. voc L 12 UNFILLED ORDERS FOR CAPACITY (UNITS OF OUTPUT/
pow © 8.2 DELIVERY DELAY WEIGHT (DIMENSIONLESS) <@> No 42.1 MONTH) <12>
eoes N G.1. DFLIVERY DELAY WEIGHTING COMPLEMENT
(GISENSTOULESS) <6>
or © 30
EEDA A 27 enzenivenrAL effect OF PRODUCT ATTRACTIVENESS
(DEHEN SIGNLESS) "<2 -
PA A 24 errcr oF PRODUCT ATTRACTIVENESS (DIMENSIONLESS)
<2e>
rom © 24.4 FRECTION CF BLOGET TO MARKETING (DIMENSIONLESS)
<21>
18 © 2.2 IWITTEL BacaLOs (UNITS) <2>
16 $10.2 INITEAL ceDAcITY (UNITS OF OUTPUT/MONTH) <10>
ft S 27L1 INITIAL E*FEECT OF PRODUCT ATTRACTIVENESS
(OMENS ONCE SSO <27>
te a 49 3IZATED MARKETING PERSONNEL (PERSONS) <tg>
Lenora © 30 QF SIMULATION RUN (WONTHS) <30>
ure cane 7 FAR CAPACITY (WONTHS) <tt>
> Ctr’! abekerivs pemsonec (rensons) 17>
£ “
wee c PEASCNNEL CONSTANT (PERSONS) <23>
45 € SALARY (OOLLARS/PERSON-iONTH) <19>
eR c SGUTACT RATE (CONTACTS/®ERSON-HONTH) <23>
NEPA A 25 iL EFFECT OF PRODUCT ATTRACTIVENESS
(GUIENSIONLESS) <25>
oc c 22.3 ORDZ2S PER CONTACT (ORDERS/CONTACT) <22> ~
ore R 3) ORDER FiCL RATE (UNIS/NONTH) <3>
Pa R 18 PERSONNEL HIRING (PERSONS/MONTH) <18>
