Beruflich Dokumente
Kultur Dokumente
KSIM- A Methodology
for InteractiveResource
PolicySimulation
Jo•xos KA•g, ILA• VERTINSKY,AND WILLIAM THOMSON
Resource Science Centre, University oi British Columbia
Vancouver 8, British Columbia, Canada
A new mathematical language (KSIM) has been developed and designed for interactive
team use. Many features of KSIM make it particularly appropriate for use in formulating
environmental policy. (1) It is easily grasped by the nonmathematical specialist and can
communicatethe workingsof complex,nonlinear feedbacksystemsto such people. (2) It
allowsfor ready entry of such'soft' subjectivevariablesas environmentalquality and national
sovereignty. (3) It emphasizesthe significanceof structural relations rather than numerical
prediction.(4) It is flexible and easily generalized.As an illustration of it• use,we showhow
KSIM can facilitate discussion of the mqltiple impactsof possibleCanadian water salesto
the United States.
with numbers. Indeed there is the major area The KSIM panelsthat we shall refer to have
of geometry, which is concernedwith shapes typically had memberswho were: (1) faculty
and structural relations.Our languagerecognizes and graduate students in disciplinessuch as
this. Think how much is conveyed when we law, communityand regionalplanning,sociology,
say that three peoplehave a triangular relation- civil engineering, and forestry; (2) middle-
ship. No numbers need be given, yet we im- managementexecutivesfrom industry; and (3)
mediately grasp most of the workings of the career officials in the government, particularly
system. The fact that the 'triangle' tells us the Ministry of the Environment.
much, but not all, is also important. Geometric Not all the groupsmentionedabove partici-
concepts permit more slack than arithmetic pated in our water exercisebut in other social
thinking. A geometricstatementis, for example, simulations [Kane et al., 1972; Kane, 1973;
A is bigger than B. This is often all we need W. Thomson, I. Vertinsky, and J. Kane, un-
to know and we expendmuch uselesseffort in publisheddata, 1972]. Even so, we have found
trying to arithmetizethe statementby measur- that much of the group interaction and ques-
ing A and B. Subjectiveevaluationsgenerally tioning has characteristicsthat are independent
correlatewell with geometricunderstanding.If of the particular compositionof the group or
not too much (precision) is asked for it is its focus of interest. From our experienceswe
possibleto get more (understanding);i.e., it have come to emphasize that what we call
,
to extend their understandingof a problem and equations either because the factors are un-
the processof communicatingthese results to known or becauseto incorporate them would
others. pose insurmountableproblemsin mathematical
Our KSIM impact calculus is designedto formulation. To reduce the distortions intro-
impart a feeling for the linkages that cross- duced by such difficulties,our calculusassumes
connectpolicy variables.As a prime hypothesis at the outset that all variables are bounded,
we assumethat in actual policy implementation scaled for convenienceto range from 0 to 1.
more insight is needed in geometric concepts
FORMULATION
(the connections.
betweenvariables,the direction
of forces,and threshold and saturation of varia- The algorithm operates as follows' For the
bles). To us, such considerationshave far more state variables x,(t), (i -- 1, 2, ." , n) we have
importancethan arithmetic specificationof pa-
rameters.Consequently,KSIM is designedpri- 0 < x•(t) < i i = 1, 2, ... , n t >_ 0
marily to impart an appreciationof the geom-
etry and dynamics of the system rather than In addition we may include several variables
an appreciationof numericalbookkeeping. that act on the system, but are not themselves
SIMULATION MODEL acted on (i.e., interventions). These additional
state variables will be denoted x, (t), (i -- n q-
Embedded in the simulation model are five
1, --' , m). We calculate the updated state
basicconditionsthat underliethe impact calcu- vector x,(t q- At) by the transformation
lus. Unlike postulates,which are inflexible,these
basicconditionsare zero-orderassumptionsthat x,(t-Jr-At) -- x,(t)p'(t) i -- 1, 2, ... ,n
can subsequentlybe perturbed. They are:
1. All system variables are bounded. where the exponent pt(t) is chosenexplicitly
2. Variables change according to the net as
p,(t) --
-[
.= xi dt l -- oti
(3)
1-1--
•At•ß= Olii
+•i._•i
xi d•i
dt] + i xi dt IJ
xi
impact of all the other variables. The matrix elementsa, give the impact of xj
3. The responseof a variable to a given on x,, and the entries /•, are matrix elements
impact goes to 0 as the variable approaches giving the impact of the percentagechangein
either bound, threshold, or saturation. xj, d/dt(ln x•) on xj, and At is the time period
4. All else being equal, a variable produces of one iteration. Whereas the a, and /•,• are
greater effectson the systemwhen it is larger. generally constant, they may be functions of
5. Complexinteractionsare describedby an the variables x, and time.
array of binary interactions.(Synergesticeffects Equation 3 implies p{(t) > 0. Hence the
are modeled by more advanced versions of transformation (2) maps the open interval (0,
KSIM.) 1) onto itself, and preservesthe boundedness
We have chosento adopt mathematical algo- of the state variables (1). Equation 3 can be
rithms that embody these characteristicsfor made somewhat clearer if thought of in the
a variety of reasons.Foremost amongstthese is following form'
the realization that no element in any human
p,(t)
system can grow without limit [Holling, 1969].
However, those features that ultimately limit 1 q- At Isumof negativeimpactson x•l
the growth of a variable are frequently ignored 1 -+-At Isumof positiveimpactson x•I
in standard formulations involving differential (4)
68 KANE ET AL.' POLICY SI1ViULATION
When negative impacts outweigh positive ones, evident provided that the c•j and fi•j are
the exponent p• is greater than unity and x• constant.
decreases.When p, -- 1, x, is unchanged.Ii To gain greater insight into this system of
the negative impacts are less than the positive equations, considerthe specialcaseof a single
ones, p• is lessthan unity and x• increases(2). independent variable without any derivative
The remaining properties of this formulation feedback (fi --0). For this situation (5) be-
become clearer if we examine the limiting sys- comes
tem of differential equationsimplied by (2) and
dx/dt = --ax In x (6)
(3) as A/-• 0
The growth characteristicsof the solution of
this differential equation are plotted in Figure
dt - .__ agixinu •i dt/ ' 1 for various valuesof the growth parametera.
The essentialsigmoidalcharacterof the growth
i-- 1, '" ,n (5) shouldbe noted, i.e., limited changenear thres-
An inspection of this system of differential hold (x = 0) or saturation (x = 1). It should
equations (5) indicates that as x• .-• 0 or 1, be observedthat since (6) involved x In x, the
the derivative dx•/dt --) 0 and thus character- behavior of the curves is not symmetric about
istic 3 follows. Therefore the expression --x• x = •. As long as a is restricted to constant
In x• may be said to modulate the responseof values, growth rates near threshold are not
variable x• to the impact given by the summed paralleled by similar positioning near satura-
terms. Characteristics 4 and 5 are similarly tion. This is actually observedin practice in
lo
0.5
0.1
0
time
Fig. 1. The variations of z as given by (6) for differing values of the parameter • 0.1, 02,
0.•, 0.4, and 0.5. In all easesthe initial vMue is 0.5.
KANE ET AL.: POLICY SIMULATION 69
many biological,social, political, and economic terms of plus and minus; e.g.,What impact will
institutions. There is no reason why growth the sale of water to the United States have in
rates near birth shouldbe comparableto growth terms of development of hydropower (a•)?
rates near maturation. Indeed, this is a point Instead of asking for a numericalanswer,we
that yon Bertalan]y [1969] has commentedon ask, Will the impact be positive or negative?
in considerabledetail, and his skewed growth (in this casethe answerwill clearlybe positive)
curves are quite comparable to those given and, Is the impact mild (+), strong (+ +), or
by (6). overwhelming(q- q- q- ) ? Of course,ultimately
At any event the actual shape of the pre- the plus and minus will be translated into
sumed growth curves is of little matter, for if numerical values; e.g., half the group might
one shouldwant symmetriccurves, (6) is easily vote for a strong (++) assessment,whereas
perturbed by choosing• -- •(x -- 1)/ln x, the other half might vote for an overwhelming
and if an arbitrary growth variation is desired (q- q- q- ) assessment.A compromiseis easily
of the form dx/dt -- f(x), all one need do is chosenby setting an = 2.5.
choosea = --f(x)/(x In x). Further mathe- The value of using plus and minus is very
matical details are discussedin a separate important in that it frees the nontechnical
technical paper (J. Kane, unpublished data, members of the group from numerical consid-
1973). erations.When we ask someoneto give a nu-
merical estimate we usually overwhelm his
WORKINGS OF THE PANEL
powers of estimation. Unlike the arithmetic
Preliminaries.In constructingthe water mode] requisition,the geometricchoice(mild, medium,
using KSIM, the panel proceededas follows. or strong) is easily answered and seldom
First, the significantvariableswere introduced. creates psychologicalbarriers. Especially in
In some cases variables have to be relabeled dealing with mathematicalunsophisticates, nu-
or redefined and in others a few distinct con- merical confrontationis a great inhibiting factor
cepts have to be combinedinto one variable, and should be avoided at all costs until the
e.g., environmental quality. Mathematically terminal stagesof the exercise.
speaking, the states of such multidimensional In filling in the entries of the interaction
variables cannot be unambiguouslyordered. matrix there is another hiddenadvantage.Each
However, we have no mental difficulty in de- entry must be given a value, which ultimately
cidingthat the environmentalquality in British might be chosenas 0. However, settingan entry
Columbia is. higher than that in say Harlem. to zero is a consciousact. It is one thing to say
The value framesand perceptionsof the KSIM there is no interaction and another to forget
panel establishan ordering suitable for further to specify the interaction (setting it to zero by
operations. default and in so doing, losing an important
In specifying the variables they are all re- feedbackloop). There is a great benefit in going
scaled to the open interval (0, 1). It proves through the full bookkeepingof all pairings
very suggestiveto think of such values on the of variables. Many significantcouplingsare so
unit interval as percentages.For example,water introduced that would otherwise be overlooked.
sales to the United States are measured as a In all our workings with KSIM groups we
fraction of the Canadian capacity. Thus the ex- have found one point of confusionto be almost
pressionx• -- 0.27 does.not refer directly to universal. Nearly all panel members do not at
so much acre-feet of water but rather it says first properly understandthe differencebetween
that Canada is selling27% of its water to the the a,• and the fi,• entries.The former repre-
United States. Clearly such relative measures sentsconstantcouplingand the latter derivative
are much more suggestivein putting together coupling. The a• entry describesthe impact
mental pictures of the workings of complex that A has on B simply becauseof the existence
systems. of A; e.g., the sun has a positive effect on
The water model is defined once initial values chlorophyll production. On the other hand, the
and a,• and •,• are specified.In choosinga,• fi• entry describesthe effect of a changein A
and • we have found it of great pedagogical on B. The interaction of environmentalquality
value to have •he panel assessthe entries in and an individual's emotional state is of this
70 K•Ns ET AL..' POLICY SI1ViULATION
O.5
0.0
0 i i i i ß - 25i i t I i ' dO
(YEARS)
analysiswasconducted concernedpopulation
that especially and industrialization,
environmental
such soft variables as Canadian sovereignty. qualityin the regiondrops.However,increased
awareness of pollutiondangersleadsto rapid
RESULTS
growthin abatementactivities.Thus environ-
The models were run for 50 simulated years. mental quality declinesonly slowly. In this
The results of the basic simulation (no water simulation we see that Canadian sovereignty
export)andthreesimulationswithintervention increasesslightly as economicand industrial
arepresented
in Figures com- growthlead to lessCanadiandependence
2-5. To facilitate on
parisons,
summaries of the levelsof selected importsfrom the United States.
variablesare presentedfor these same four When we intervene in this model by commit-
simulations and two additional ones in Tables ing a portionof the British Columbianwater
3--8. supply for export, we find severaldifferent
trends.Populationgrowthis retardedsomewhat
DISCUSSION
to 2.4% peryearwhen25% of the watersupply
Lookingfirst at the basicsimulationwe find is exportedand to 2.0% per year when50%
that the population slightly of the water supply is exported.
growthrate declines
over the 50-yearsimulatedperiod,the average Underlyingthis observationis the effect of
growthratebeing2.7%peryear.Currenttrends storagecapacityrequirements neededto export
are continuedin industrializationas economic substantialquantitiesof water. A substantial
and populationgrowthprovidethe basisfor storagecapacityrequiresa significantflooding
industrialdevelopment.
With theseincreases
in of valleysthat are neededfor the expansionof
1.0
Canadian
g --. . ....-
........ ,-OVereight
v •',,•".
.........................
................
. •.• •-.-- ._ . ....•.•..
,• ,• ......................
•.•.,•.: . '..............
• .• ..'
- / •'• ."
•' • - • . i•o
•.-'/"
.••. • • ..
" i I ' m i
ø-øo ....
TIME (YEARS)
Fig. 3. Projected
system
behavior
withwaterexports
committed
at a 25%levelof supply.
74 KANE ET AL.' POLICY 8Ii•ULATION
'• o.5-
TIME (YEARS)
market forces prevailing within Canada, since in environmental quality during the sales pe-
it was assumedthat no contractual or political riod. This decline reduced the attractiveness of
commitment had been made to sell water to the the region,thus inhibiting normally high immi-
United States. We observed a continuous decline gration to the region. When substantialwater
in salesleadingto a terminationof saleswithin sales were committed for a period of 30 years
20 years for initial salesof up to 50% of the or longer, results were similar except for the
available water supply of British Columbia. appearanceof a severe threat to Canadian
This experimentdid not attempt to model a sovereignty.Whereas the water sales initially
realistic situation, but served to illustrate the enhancedthe economicstrength of the region,
potential forces in the region opposingsales. contributing
to Canadiansovereignty,
the effect
The salesinitially led to a sharp increasein of continued sales led to a sharp decline in
industrialization in the region, the increase sovereigntyafter 15 years.
stemmingfrom the injection of capital ex- Another experiment introduced an extreme
pendituresandrevenues in the water high price on water use. This intervention
originating
sales.This industrialgrowthwas.reducedwhen inhibited industrial growth and changed the
sales were terminated. However, the total composition of the economyin the region,stimu-
growth over the 50-year simulationperiod lating recreationrelated industries.Environ-
reached a similar level to that obtained in the mental quality declinedonly slightly, and the
basic simulation. The price to the region of attractivenessof the regionwas maintainedin
water salesmanifesteditself in a sharp decline spiteof slowereconomic growth,leadingto the
1.0 '
0.5-
0.0
o 5k)
TIME (YEARS)
same population growth as that of the basic sonal communication,1966) has observed,they
model. In this experiment no water was sold to are 'counterintuitive,' often behaving in an en-
the United States. When the experiment was tirely unanticipated fashion. For example, the
modified by a short-term commitment to sell apparently contradictory result that to decrease
half of available water suppliesto the United the attractiveness
of the environment
of Britis•
States, industrial and agricultural development Columbia will increase environmental quality
was severely retarded. Additionally, although becauseit will tend to reducemigration. (2) Un-
water sales were not committed, the lack of like the classicaltwo or three-body problem,
industrial and agricultural developmentforced which can be continuouslyperturbed, complex,
the region to rely on water sales to maintain multidimensional systems are often extraordi-
the regionaleconomy.Thus salesof water to the narily resistantto change.They seemto absorb
United States never dropped below 30% of small intrusions without a significantmodifica-
available supply in this experiment. tion in behavior. Even if more and more intru-
And what is the conclusion? Have the de- sions are introduced there still seems to be little
scribed exercisesmade a significantimpact, if effect until one intrusion too many is added.
any, on Canadian water policy? No, not yet. Then the system 'flips,' adjusting itself to an
It was not our intention at this point to play a entirely different configuration; i.e., high-order
key role in the decision-makingprocess.Ulti- complexsystemsdo not respondcontinuouslyto
mately and hopefully yes, but for the moment small changesbut rather tend to absorb them
suchaspirationsare premature. until a critical threshold is reached when it
For the moment our prime objectiveis educa- jumps suddenly into an entirely new state.
tion. In particular we seekto impart an experi- These are two of the mathematicalpoints we
ence into the workings of intricately linked wished to communicate. On the whole our dem-
feedback systems.The behavior of such com- onstrations have been successfulin achieving
plex systemsis entirely different than the sim- this aim. As a secondgoal we wishedto impart
plified, dimensionallyreducedsystemscurrently a new pattern of holistic thinking. Have we
emphasizedin the academiccurriculum.Systems achievedthis? Perhaps. At least one point was
involving many linked variables exhibit the conveyed.At no time was control on any one
following differences:(1) As J. Forrester (per- variable successful even if the intrusion were so
TABLE 6. Selected Indices for the Simulation of Zero Population Growth and No Water Exports
these have significantinterest in themselves,our systems, Brookhaven Syrup. Biol. 22, 128-141,
1969.
goal has not been to produce a working model
Kane, J., A primer for a new cross-impact lan-
bugrather to presenta technique,KSIM, which guage (with examplesdrawn from transportation
has the potential of being a valuable learning planning), Technol. Forecasting Soc. Change,
tool. Ultimately, we feel what is essentialis that 4 (1), in press, 1973.
a sufficient decision-makingbody be educated Kane, J.. W. Thomson, and I. Vertinsky, Health
care delivery: A policy simulator, Socio-
in the processesof mathematical articulation,
Economic Plann. Sci., 6(3), 283-293, 1972.
and that at least they be appreciative and un- Martino, J.P., An experiment with the Delphi
derstandingof holisticconceptsand the dynam- procedure for long range forecasting, IEEE
ics of complex feedbackstructures.If not ...? Trans. Eng. Manage., EM 15(3), 138-144, 1968.
von Bertalanfy, L., General Systems Theory, G.
]•EFERENCES Braziler, New York, 1969.
Williston, R., A Canadian water viewpoint on
Dalkey, N. C., and O. Helmet, An experimental water export, Proc. Annu. Amer. Water R eso•tr.
application of the Delphi method to the use of Con. 3rd, 44, 1967.
experts, Manage. $ci., 9(3), 458-467, 1963.
Ho]ling, C. S., Stability in ecological and social (Received February 21, 1972;
systems, in Diversity and stability in ecological revised June 1, 1972.)