THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 345
CISD —- A NORMATIVE METHOD FOR LECHNOLOGY ASSESSMENT
Lin Ruiji
Department of Management, Shenzhen University
Shenzhen, Guangdong, People’s Republic of China
ABSTRACT
The formal KSIil (Kane's SIMulation) model is equivalent to a
particular system dynamics (SD) model. On the basis of thia
equivalency, we use the KSIM and cross-impact concepts to
simplify the SD modeling steps, and a new procedure -- CISD
from the abbreviation for Cross Impact System Dynamics which is
technically simpler and more normative, has been introduced.
CISD is well applied in the field of Technology Assessment (TA).
An example for TA of agricultural chemicals with CISD is pre-
sented. A general computer program for CISD which is called
CISD-FORTRAN makes CISD procedure more widely used with faci-
lities even for nonspecialists.
INTRODUCTION
Technology assessment, a relatively new and innovative concept
emerged in the middle of the 1960s, began to change the common
ideas people lad for a long time about the social and economic
functions of science and technology. It promotes the public
policy .and decisionmaking process for programming technological
developments, The term "technology assessment" was first
appeared in a. report submitted to the U.S. Congress by the
former congressman Emilio Q. Daddario in 1966, and it was
described as "a form of policy research which provides a
balanced appraisal to policymaker". "It identifies policy
issues, assesses the impact of alternative courses of action,
and presents findings. It is a method of analysis that. systema-
tically appraises the nature, significance, status, and merit of
the technological pregram".
348 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
Methods used in TA can be divided into three types: (a) qualita-
tive methods, (b) quantitative methods, (c) modeling and simula-
tion methods. system dynamics is one of simulation methods for
TA. the "Limits to Growth" was the earlist and most significant
application of SD to a problem-oriented TA project sponsored by
the Club of Rome, thereafter, SD has been more noticably applied
to different TA problems. However, we are facing difficulties to
puild a general TA model using SD because of its folloeing
deficiencies:
4. A SD model lacks generalization. The model is directed to the
specific problem and it suffers.a lot of changes as the pro-
blem changes a little.
2. SD is technically complex. Modelers are required to be familiar
with the system they are modeling and adept in the 5D methodo-
logy and DYNAMO programming.
3. System struture and policies tested are based on modeler's
intuition, the real decisionmakers are not parcitipative in
the modeling process for some technical reasons.
In generally speaking, subjective and psychological factors play
important roles in TA problems (both technology-oriented and
project-oriented), therefore expertises of wide range are critical
for a TA program. On the other hand, some TA programs are time-~
pressed, i.e., final policies for a technological utilization or
development should be made in a short term because the recent
technologies are speedly evolved and strongly competitive for
commercial and economic purposes. All these call for a method
with simplicity and generalization which not only can congregate
the expertises in a model, but also can reflect the dynamic
behaviour of second and higher order impacts. which the technology
under evaluation makes on natural enviornment, human society,
sovereign economy and technology itself.
This paper will give a method which is called CISD from the abbre-
viation for Cross Impact System Dynamics. CISD uses a Cross
Impact Matrix (CIM) to portray the causal interactions system
elements, and the quantitative relations of system variables are
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 347
expressed in the form of KSIM equations. CIM can be identified
through Delphi, brainstorming and other methods. To some extent,
CISD is a comprehensive combination of SD and KSIM. It is charac-
terized by simplicity, conciseness and its standard format, and
therefore becomes an effective method for TA.
A BRIEF STUDY ON SD
A General Model of SD
Causal diagrams or flow charts essentially indicate the causal
interactions among system variables. These causal interactions
are further divided into two specific types refered to as the
material relation and the information relation.
The material causal relations determine the flow paths in flow
charts. If two quantities are materially related, one of them,
say a> will accumlate the net effect of the other quantity a3?
the quantity q, is called affected variable and q, affecting
variable. More generally, causal relations of this type can be
represented by the following differential equation when there
are n affecting variables:
&y = £4 (aye age veer ay) i (+,) * dio (
In physical meaning, material causal relations reflect the law
of conservation of matter within the universe. Examples of this
type of causality are the relationships of petroleum reserves
to petroleum consumption, of savings to compounded interest, of
population to births and deaths, only to name some. This type of
causal relation can also delineates the dynamic, cumulative,
and memory-possessing characteristics of the system. In flow
diagrams, material flow paths are only through the levels and
rates, the level equation gives the -form of their relations?
L LK = Led + DI*(RIN.JK — ROUT.JK)
N Le= oe
348 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
2 aim BK = ET ~ aim (RIN.IK - ROUT.IK)
DbeO DT Dteo
= Tin Tout * x(t.) = X5
The rate can be written in the form of the decision function
r= f(x,y,p) (2)
therefore, we have
ox = f(x,y,p) x(t,) = XQ (3)
where x, y, ©, p are the vector forms of levels, auxiliaries,
rates, and parameters respectively.
Information causal relations, on the other hand, determine the
information path in a flow chart. Information is needed when
decisions are made, therefore, the information path actually
embodies the decisionmaking process. If two quantities are
informationally related, the affected one will be decided by
the affecting one but not accumulate the net effect of it over
time, i.e., the information causal relations are memoryless,
they affect each other instantaneously. This type of causality
can be more generally expressed as a functional equation:
95 = 8644s Go eee » Oy) (4)
In system dynamics, its specific vector form is
y = g(x, ys Pp) (5)
“fo conclude, ali SD equations can be written in general form as
equations (1) and (4), or in the vector form as equations (3)
and (5). Level equations are obtained through integrating the
equation (3).
t
t
KexX + J f(x,y,p) dt = x + j rat (6)
ty ° ty
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 349
The SD Modeling Process
Although many authorities have summarised the stages of the SD
modeling process in their literatures, we here show a standard
five steps of SD modeling:
1, Conceptualization model. Identifying and conceptualizing the
problem under study literally, this may include problem defi-
nition, the purpose, condition and specific requirement of
the model, reference behaviour modes, etc.
2, Causal diagram model. Reflecting different causal interactions
among system variables.
3. Flow chart model. Representing the inner feedback structure
of the system.
4, DYNAMO model. Quantifying the couplings of variables appeared
in the flow chart.
5, Model tests. Sensitivity analysis, policy tests. and analysis,
model validity and development, ete.
KSIM MODEL AND ITS SIMILALITY TO A SD MODEL
A general KSIM model can be written as follows:
dx, n ax
road = -(e, + 3 (a4 55 + day aq!) )x,lnx, (7)
where x; are the state variables of the system, i = 1, 2, ..., n
O€xi sts ay, are elements of the state interaction matrix (nxn)
‘giving the impact of x, on X43 by are elements of rate interac-
tion matrix (nxn) giving the impact of ax, /at on x43 C, indicate
the impact of exogenous intervention on Xe Equation (7) can be
represented in matrix notation as:
&& . D(x) (ax + BSE +c) (8)
where A, B, C can be aggregated in a combined matrix (ABC),
the diagonal weighting matrix D(x) is defined as
350 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
~x,1nx, 0
~Xp nx,
D(x) = .
0 -x,Inx,
Here any element (-x,1nx, ) in matrix D(x) is called Kane's modu-
lation function.
Equation (7) does not meet the requirement of the rule of consis.
tency in system dynamics, however, by defining a5 = dx, /at and
enlargingthe conbined matrix the expression is equivalent to a
consistent equation as
ax ul
St = (ey + a ayy x,) x,lnx, (9)
or in matrix form as
#2 -(¢ + Ax) D(x) (10)
Comparing Eq. (10) with SD equation (3), we obviously find that
the former is a particular form of the latter. Further assuming
r= (C + Ax) D(x), we shall write the equivalent DYNAMO equations
for Eq. (10) as
L u.K = x.J + DD*r.JK
R veKL = -(C + A¥x,K)*x.K*LOGN(x.K) (41)
N x= X,
The equation set (11) represents a minimum submodel of SD flow
diagram (shown in Fig. 1). A complete KSIM model consists of n
submodels of the same form.
Fig.1 A minimum SD model
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 351
CROSS IMPACT SYSTEM DYNAMICS
A Four Step Procedure
As discussed above, the KSIM model is equivalent to a specific
SD model, this model is normatively structured by n minimum SD
models of the same form. If we introduce a CIM (Table 1) to
identify the causal interactions of system variables and the
DYNAMO equations are written after the pattern of the equation
set (11), a simplified four step modeling procedure CISD is
established on the basis of previous SD modeling steps.
Step one -- conceptualization of the problem (the same as SD).
Step two -- structural interpretation. In this step, system
variables are determined and initiated; a CIM which
gives the interaction and its strength aa5 between
any two variables is identified through Delphi.
Step three -- DYNAMO model. The equivalent DYNAMO interpretation
of KSIM relations is utilized.
Step four -- model tests.
Table 1 shows the general form of a CIM in CISD. Variables are
Table 1 The CIM in CISD
affecting
affected variable x4 Xo | ove x. c
variable ~ a
*4 314 | 812 | ere | Ain|]
Xo 894 |Ao0 | «++ | onl] Co
—
: : : | 43 : :
Xs any [ano tees | ann | cy
binarily interrelated, the total number of elements in CIM is rn.
The most ostensible difference between CISD procedure and SD
modeling process is the use of CIM to replace the the causal and
352 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
flow diagrams for the purpose of indicating the system structure.
CIM has a very normative form and always keeps the same whatever
changes the system structure. This characteristic of CIM makes
the CISD procedure simple and stardard.
Comparision Between SD and CISD
Some comparisions between the two are shown in Table 2.
Table 2. Comparisions between SD and CISD
~ Generalization-
lacking model
- Delay and table
functions
- Wider range of
Comparisions
in the aspect of? sD cIsD
1- characteristics | -Complex simpler
normative model
narrower range of |
- Causal diagram
- Flow chart
- DYNAMO model
~ Model tests
application application
2. modeling steps | - Problem problem
conceptualization conceptualization
cross impact matrix
DYNAMO model
model tests
3. the system
- Dynamic
~ Non-linear
- System are bounded
- No limit to varias
ble's value
dynamic
linear
open system
Osx <1
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 353
The CISD-FORTRAN Program
In order to dilate the application of CISD to nonspecialists who
are not familiar with DYNAMO, a FORTRAN program for CISD is com-
piled. CISD-FORTRAN also produces graphical outputs. A brief
scheme is shown in Figure 2.
Data input
—-—-~{Do. (0) Ix=1,m
|
Do @0) J=i,n
[
iY (J) =0
ccc Do 110 K=1,n
|
|
|
!
|
I
Y(J) = Y(J) + ACI, K)*xX(I-1,K)
R(T) = -(C(F) + Y(T))*X (1-1 )¥ALOG(X(T~1,5))
I
‘X(1,d) = X(TH1,d) + DI*R(T)
Graphical
output
em)
Pig.2 CISD~FORTRAN Program Scheme
354 THE 1987 INTERNATIONAL: CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
AN, EXAMPLE
Agricultural chemicals have played an important historical role
in agriculture because they prevent and control the plant
diseases, eliminate insect pests. As a result, the agricultural
output remains increasing during the past few decades. However,
they carry pesticide pollution which makes negative impacts on
human health and natural enviornment (e.g., soil, air, water,
etc.) at the same time, TA of agricultural chemicals (Taac)
therefore emerges from neccessity. China is the biggist agri-
cultural country with an agricultural population of eight hun-
dred million, it is imperative that we attach great importance
to TAAC.
As an example, we coordinate with Zhejiang Research Institute of
Chemical Industry to apply CISD'to TA of a specific pesticide
“called Tsumacide, as a result, nine variables are abstracted to
-describe the impact system as follows:
X4.-- Chronic toxicity of the pesticide
XQ -- its acute toxicity
X3 -- the pesticide residue
x4 -~- the efficacy of the pesticide
Xs -- the pesticide production
X¢ 7~ the market share
x -- pésticide user's attitude towards ‘the pesticide
xg -- agricultural product consumer's attitude
Xo increment of crops and grains
A CIM reflecting the binary interactions of the nine impact
variables is obtained through Delphi method and shown in Table 3.
Basic runs (one of them shown in Fig. 3) of CISD-TAAC model
show that x4 and x, will increase in a small range over time
without outside interventions. If outside policies which prevent
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 355
Table 3. CIM of CISD-TAAC
xy Xo Xs x4 5 =e x7 Xg XQ
x; 0 +4] 42 | -1 [4t.5 | 0 o | 05] 0
Xp 0 fo) O |-0.5 | +1 te) (0) i} te)
Xz 0 0 | o | o [+5 |so5 | o 0 |-0.2
x4 0 © [40.5 |-0.5 |-0.3 |-0.3 | +1 ° °
x, | -0.4 | -1 0.3, +2 | -1 | -1 | 42 o | 41
x, | -0.2 | 1 70-2 42 | +1 [90.5 | 44 O |+0.5
Xq | -O.1 | -2 [40.5 | +3 | 40.2 [40.2 | 41 |-0.5 | 42
Xg “2.5 | -0.5 | -2 |-0.5 fe) fo) lo) +1 +1
Xq i) © |-0.4 142.5 ]+0.5 |+0.3 }| 0 0 Cy)
* Note that here a CLIP function is used:
A TIME.K 2 C
CLIP(A,B,TIME.K,C) = {
B TIME.K < c
nw RUN OF CISD FOR TA OF AGRICULTURAL CERICAL
ts
2 ae
1
Fig. 3. Basic run of CISD-TAAC
356 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
the two kinds of toxicity are added to the model (c, = -1 and
Cp = -1), the curves then decline (shown in Fig. 4). Policies
which encourage the pesticide production (eg = +1) and restrict
it (cg = -1) are tested in the same way. The result shows that
the restricting policy is not. sensitive to the production, i.e.,
the demand of the pesticide is decided by the system itself. The
encouraging policy promotes the production markedly.
‘YOWD RUM OF CISD FOR TA OF ACRICULTIRA. CEENICAL
i
Fig. 4. Changes for rerun of CISD-Taac
7 Reta
Based on the policy tests, some conclusions and suggestions are
reached and submitted in a report to the institute. The brief
main points are as follows:
1. Tsumacide is proved to be a pesticide of little toxicity, it
can prevent and control plant diseases and pests effectively.
2. Although Tsumacide is slightly poisonous, any toxicity
prevented steps can not be neglected.
3. Tsumacide clicks, we suggest that encouraging policies be
made for these kinds of pesticides with slight toxicity (e.g.,
investment: prevailing policy).
THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA 357
CONCLUSIONS
System dynamics seems to be more rapidly developed and more
widely used in many fields of research in the past one or two
decades. But what is its future? Is there any prospective direc-
tion that will lead the way for it?
On one hand, system dynamics, as a methodology, should be
established on a very firm basis, it has its own logic and rules,
theoretical research on SD itself is therefore emphazisea. On
the other hand, it is an applied tool for modeling socioeconomic
system, the simpler it is, the more often it can be used.
This paper not merely introduces the CISD procedure, it is more
meaningful that it combines two methods together, and a new
simplified procedure is born as a result. Could this "marriage"
bring us some revelation for our further SD study?
REFERENCES
Burns, J.R. (1977) "Converting Signed Digraphs to Forrester
Schematics and Converting Forrester Schematics to
Defferential Equations", IEEE Transactions on Systems,
Man, and Cybernetics, Vol. SMC~7 No.8, pp695-707
Burns, J.R. and Marcy W.M. (1979) "Causality: Its Characteri-
gation in System Dynamics and KSIM Models of Socioecono~
mic System", Technological Forecasting and Social Change
14, pp387-398
Soates, J.P. (1976) "The Role of Formal Model in Technology
Assessment", Technological Forecazting and Social Change
9, ppt39~190
Forrester, J.W. (1968) Principles of Systems MIT Press, Cam~
bridge, Mass.
?orrester, J.W. (1961) Ind
358 THE 1987 INTERNATIONAL CONFERENCE OF THE SYSTEM DYNAMICS SOCITY. CHINA
Mass.
Forrester, J.W. (1975) Collected Papers of J.W. Forrester,
Wright-Allen Press, Cambridge, Mass.
Kane, J. (1972) " A Primer for a New Cross-Impact Language --
KSIM", Technological Forecasting and Social Change 4,
pp129-142
Lin, R.J. (1986) "Cross Impact System Dynamics and Its Applica-
tions in Technology Assessment", Master Dissertation,
Zhejiang University.
Lin, R.J. (1986) "Some Rules and Characteristics of System
Dynamics", Journal of Shenzhen University, Vol. 3, No. 4,
PP21-30
Linstone, H.A., et al. "The Use of Structural Modeling for Tech-
nology Assessment", Technological Forecasting and Social
Change 14, pp291-327, 1979
Porter, A.L. and Rossini, F.A. (1980) "Technology Assessment
Enviornmental Impact Assessment: Toward: Integrated Impact
Assessment", IEEE Transactions on Systems, Man, and
Cybernetics, Vol. SMC-10, No.8, pp417-424
Richardson, G.P, and Pugh III, A.L. (1981) Introduction to
System Dynamics Modeling with DYNAMO, MIT Press, Cambridge,
Mass.
Roberts, E.D. (1978) Managerial Applications of System Dynamics,
MIT Press, Cambridge, Mass.
White, Jr., K.P. (1981) "A Constructive Approach to KSIM Model
Analysis", IEEE Transactions on Systems, Man, and Cyber-
netics, Vol. SMC-11, No.8, pp571~574
CC BY-NC-SA 4.0