Rohrbaugh, John, "Evaluating Objective Function Trajectories: What is in the Eyes of the Beholder", 1981

Evaluating Objective Function Trajectories:

What is in the Eye of the Beholder

John Rohrbaugh
Institute for Government and Policy Studies
Rockefeller College of Public Affairs and Policy
State University of New York at Albany
Introduction

Although the system dynamics literature covers issues of how to
construct, analyze, test, validate, and implement dynamic models, surprisingly
little attention has been paid to how managers react to and interpret the
output from system dynamics models (see Gardiner and Ford, 1980; Rohrbaugh and
Andersen, 1979). That is, system dynamicists construct feedback models that
are simplifications of a complex reality and then conduct policy tests on
these abridged representations. However, decision’ makers not trained in
system dynamics may find that even these allegedly simplified models may be
quite complex and difficult to evaluate, since model output typically consists
of scores of variables interwoven over time.

From a psychological point of view, evaluating the output from a system
dynamics model raises several interesting questions. First, how do decision
makers integrate information about numerous variables that change over time
and how do managers evaluate changing patterns of system states over time
(some such patterns have even been nicknamed as “worse before better," “better
before worse," and "trade-off" patterns of behavior)?

A second question is whether or not important differences exist between
individuals with respect to how they evaluate a system's output. If so,
policies preferred by one decision maker may not be preferred by another.

‘This question quite obviously has important implications for implementing

134

model results. For example, do some decision makers emphasize steady atate

responses to policy innovations while other decision makers emphasize the
transient response? If so, these two classes of decision makers would tend to
evaluate "worse before better" policies in consistently different ways.

‘Thie paper sumarizes some recent work on these two questions concerning
how managers evaluate system trajectories over time and how researchers can
attempt to measure individual differences in decision maker's preference
structures. In particular, the paper presents an empirically derived taxonomy
for classifying patterns of individual differences in decision makers'
evaluation of the output from system dynamics models. Before delving into the
research methods and conclusions, a brief overview of some of the existing
problems with evaluating objective functions is presented along with a sketch
of the psychological theory that is used to underpin much of the research

presented below.

Existing Patterns with Traditional Objective Functions

The creation of workable and reliable objective functions has been an

elusive goal of economists, policy analysts, and dynamic modelers since the

beginning of quantitative analyses of social policy. The development of such
functions would produce several dramatic benefits for the formation of social
policy. Not only would objective functions provide a precise index of system
performance, but they would also clarify and explicate the criteria being used
in the process of policy formation. Perhaps more importantly, objective
functions would allow analysts to rank order preferred sets of policy

alternative:

From a technical perspective, as well, the creation of workable and

reliable objective functions would be extremely valuable in dealing with
questions related to parameter sensitivity. The testing of sensitivity would
be greatly simplified if analysts could evaluate the reaction of the overall
system performance to changes in parameters, rather than focusing on
trajectories, characteristic modes, or recommended policies. Furthermore, the
field of automatic control is immediately available with a host of
optimization techniques which would be applicable to dynamic models upon the
development, of workable and reliable objective functions. Unfortunately,
severe conceptual and technical problems currently appear to inhibit their
possible use.

Typically, engineers and economists have been able to side~step many of
the difficult problems involved in the construction of dynamic objective
functions by evoking notions of minimum energy (or cost), maximum efficiency,
or minimum total energy. In general, the problem has been solved by
specifying a quadratic objective of the form

oft) = x(t) Q x(t)T
where O(t} is the objective function at some specified time t, X(t) is the
vector of system states at time t, and 9 is a matrix of weights applied to the
various quadratic terms. The evaluation of dynamic trajectories typically has
been handled by taking the integral of the quadratic objective function
defined above with an added term for the evaluation of the system's end-state

with the form
tes
Ofte) = | X(s) QX(s) ae + £(x(t_))
to

where O(t¢) is the overall evaluation of the objective function at the final
time te and £(X(tp)) is the relative weight given to the system's final state
(assuming that the system is stable). The justification for the use of such a

simplified dynamic objective function has typically rested on a priori

135

deductions concerning what the proper objective should be rather than upon
detailed empirical investigation of what is the actual preference structure
of an individual decision maker.

For many engineering problems and gome economic problems, a priori .
specification appears to be justifiable. Most policy problems, however, seem
to demand a more complex integration of the variety of social and political
variables contained in a dynamic model. Early developments in the field of
Gestalt psychology coupled with several recent advances in peychologist's
ability to measure and quantify decision makers’ preference structures provide
the tools needed to develop more workable, empirically derived objective

functions.

An Organizing Psychological Framework

The framework needed to integrate and evaluate complex patterns of many
variables changing through time is contained in the arguments of Max
Wertheimer and other Gestalt psychologists in the second decade of this
century, arguments that ultimately redirected structural and behavioristic
psychology, They insisted that perception, for example, is more than the sum
of individual sensations or that behavior ie more than a “bundle of reflexes."
These radical configurationists rejected molecular psychological models and
Proposed that one should begin with the complex, holistic system as the basis
for scientific progress. As Wolfgang Kohler suggested in his definitive
statement of the Gestalt theory:

The stimulus-response formula, which sounds at first
so attractive, is actually quite misleading . . . . When the
term is taken in its strict sense, it is not generally “a
stimulus” which elicits a response. In vision, for
instance, the organism tends to respond to millions of
stimali at oncer and the first stage of this response is

organization within a correspondingly large field... +
The right psychological formula is therefore: pattern of

stimulation-~organization~-response to products of
organization. ‘er, ete Bp eee ~

Kohler proceeded to outline a type of “system dynamics" theory of
psychological functioning as a keystone for the Gestalt approach:

Everything in this field (sensory experience] points
toward a theory in which the main emphasis lies on dynamic
factors rather than on anatomically prescribed conditions.
Moreover, in many observations field dynamics is almost
directly revealed to the subject. This is the case for
instance, when sudden stimulation, or a change of
stimulation, is followed by sensory events rather than
states . . . . Without the great historical prestige which
machine theory still enjoys, nobody would hesitate to take
these observations as evidence of dynamic interaction... «
There is no question that so long as dynamics remains
undisturbed by accidental impacts from without, it tends
to establish orderly distributions . . . . Dynamic
self-distribution in this sense is the kind of function
which Gestalt Psychology believes to be essential in
neurological and psychological theory. (Kohler, 1947, pp.
12232)

Due to Kohler's belief that the dynanic distributions of sensory
organization and sensory fields are functional wholes, he stressed the need
for psychologists to investigate the overarching forms of environmental
stimilation, as well as the aystematic integrations that underlie the
organism's responses. This molar rather than molecular approach is

"Pindanestall tovoue praca sntavese 4 evaluating objective function
trajectories.

In the work reported below, social judgment analysis, a set of empirical
techniques grounded in the theory of experimental cognitive psychology is
Proposed as a basis for the development of objective functions that could be
used to sumarize the multivariate performance of a variety of social systens
(see Fanmond, McClelland, and Mumpower, 1980; Hammond, Mumpover, and smith,
1977). To A1lustrate the use of social judgment analysis in the development
of workable and reliable objective functions, examples have been drawn from
Forrester's (1969) Urban Dynamics, a complex, nonlinear, dynamic, feedback

model designed to capture many of the interactions in a generic urban area.

136

‘The focus of the research reviewed in this section is on the construction of
such dynamic objective functions that swmarize the patterns of stimulation
provided by the output of the Urban Dynamicg model in response to a variety of

policy tests.

Cross-sectional Summary of the Sensory Field

In order to form an objective function, each n-tuple in an n-state model
should map into a single value. The resulting unidimensional measure allows
for the preferential ranking of all possible states of the system. Of the 124
system variables that Forrester tabled in Urban Dynamics, 36 were selected as
potentially useful to the summary of the urban system sensory field. These 36

key system variables were combined and organized to form the hierarchical

judguent ‘model shown in Figure’ 1. More specifically, the 36 key systen.
variables were variously combined to derive 31 explicit criteria at the bottom
of the hierarchy which were subsequently clustered to form 6 separate system
goals: job availability, housing quality, population distribution, industrial
conditions, density composition, and tax structure. These 6 system goals,
when integrated by a decision maker, provide the basis for constructing an
objective function of overall system performance.

In order to develop a data base for the current research, the effects of
11 different urban policies on the 36 key system variables were extracted frow
the tabled values in Urban Dynamics for the two time periods cited: 10 years
following implementation of a policy and 50 years following implementation.
The 10-year cross section captured the short-term effects of the policy being

implemented and the 50-year cross section reflected the long-run equilibrium

t of conditions defined at

effects of the policy. By including one more
initial equilibrium, a total of 23 alternative observations of the 36 system

variables and, therefore, 23 alternative profiles of the 31 derived criteria,

Managerial Job Avattability
4
Laber Job Availabitity
aber Una Raby I seh avait abtty
Undereroloyed Job AvatTaitity
w/v

Undereraloyed Unward Mobility
((uTL-LT4)/u)

Proportion of Preniun Housing
(Pa /807,

Adecuacy of Premiun Housing
(a

Rate of Change of Preniun Housing
((PHC-PHO)/PH)
Dragertion of Yorker SN

“nay Of Workar Housing me

of Charge of Worker Housing =n
"alt cene Seale ———F
Prapartoy of Udereralayed Housing
Asgsaty of Undererpoyes Hous 9
te .
Rate of Capae of unarecloyed Yusing
Catortee?=sud) 08)

tropertion of Managers:
werein) ~ Overal?

Chance of Marsgertal Pepytetfon System
_— Performance

(Conese:
reten or Labo

Population Distributton

51?)
fe of Change of Underevoloyed Pooulatfon
((usesmeL TU-utL-u0)70)

Proportion of New Enterprise
(4E/PUT)

ceperetlon of to Enterovise
eee at Rate of ae
aise ‘ton of Mature Enterprise as
Masure Taleroctsig ‘Net Rate of Change —_s
((5£0-¥30) 8]
Prozcrticn of Geclining Industry
(br/PuT),
DecTinina Indust Net. Rate of Change
(430-010) /01)

PLT oo )
we ‘of Fe st at Density Composition

(C2pur7.

TCS orRAN PD
|. OT RxAY
Tay fate ic Tas Structure

(50.0TR)

Industriat Conditions

FIGURE 1, Hierarchical Judgment Model of the Urban System.

137

were constructed. Thus the'23 profiles included one base equilibrium run plus
‘one short-run (10-year) and one long-run (50-year) set of effects for each of
the 11 policies investigated by Forrester. .
Bach complete profile was split into sections corresponding to the 6
separate system goals. For example, Figure 2 illustrates two of the 23
profile sections pertaining to job availability, ae well as two of the 23
profile sections pertaining to population distribution. It should be noted
that, although the profile sections for each system goal contain the sane
criteria, the criteria take on different values in each profile section.
The evaluation of system states on the basis of the 31 criteria is a
complex cognitive problem that can only be understood in the context of ~
individual judgment. According to social judoment theory (Hammond, Brehner,

Stewart, and Steinmann, 1975), such an evaluation process demands the

integration of the sensory field containing any or all of the 31 criter:
Social judgment theory proposés that the sumary of such multiple stimuli in
the judgment process can be represented by (a) the particular degree of
importance placed on each criterion--referred to as weight; (b) the specific
form of the functional relation between each criterion and the final
judgment--referred to as function form and (c) the particular method for

1

integrating all of the criteria~-referred to as the organizing principl
repeated judgments are made about a variety of eystem states, the covert
cognitive process of an individual's judguents can be mathematically modeled
using multiple regression statistics, as well as converted to pictorial
representation by means of interactive computer graphics (Hammond, Rohrbaugh,
Mumpower, and Adelman, 1977).

Four students of system dynamics in the Graduate School of Public Affairs

at the State University of New York at Albany evaluated the 23 alternative
*ROFILES OF JOB AVAILABILITY

fanagerial Job Availability
(managerial jobs/manager) 72

.abor Job Availability
(labor jobs/laborer) 1.03

abor Upward Mobility
(% laborers to managers/year) %

Inderenployed Job Availability
{underemployed jobs /underemp loyed) +55

‘ndereaployed Upward Mobility
(net % underemployed to laborers/year) 1.5%

anagerial Availability
(managerial jobs /manager) 78

abor Job Availability
(labor jobs/laborer) 1.10

abor Upward Mobility

(% laborers to managers/year) 11t
aderenployed Job Availability
(usderenployed jobs /underemployed) Bi

aderemployed Upward Mobility
(net % underemployed to laborers/year) 3.0%

FIGURE 2.

PROFILES OF POPULATION DISTRIBUTION

Proportion of Managers
{managerial persons/total population)

Rate of Change of Managerial Population
(% growth of managerial population/year)

Proportion of Laborers
(labor persons/total population)

Rate of Change of Labor Population
(% growth of labor population/year)

Proportion of Underemployed
{underemployed persons/total population)

Rate of Change of Underemployed Population.
(% growth of underemployed population/year)

Proportion of Managers
(managerial persons/total population)

Rate of Change of Managerial Population
(% growth of managerial population/year)

Proportion of Laborers
(labor persons/total populatfon)

Rate of Change of Labor Population
(% growth of labor population/year)

Proportion of Underemployed
(underemployed persons/total population)

Rate of Change of Underemployed Population

249

(% growth of underemployed population/year) -1.3%

Two of the 23 Alternative Profile Segments Depicting Criteria for the

Evaluation of Job Availability and of Population Distribution.

profiles of system states (divided into the 6 separate system goals) described
above. Hach individual's 6 sets of judgments were made on a 20-point rating

scale from 1 (a completely unacceptable system state) to 20 (a very acceptable
system state). Figure 2 illustrates the exact nature of the judgment tasks
that were given to the four participants. The participants would consider the

Profile sections (such as those shown in Figure 2) and express their relative

preferences for the profile sections by assigning to them ratings on the
judgment scale (1-20).

Once judgments had been made concerning each of the 6 profile sections at
the more detailed side of the hierarchical model shown in Pigure 1,
participants were then asked to specify how to combine information from all 6
system goalg into a single, overall evaluation of system performance. Figure
3 depicts two of the possible 23 alternative profiles that participants
evaluated in assessing overall system performance. As in Figure 2, each
participant responded to the profiles using a 20-point rating scale. It
should be noted that these profiles’ (as shown in Figure 3) were constructed
by using the prior judgment scales that ranged from 1 to 20. ‘Thus, the 6 sets
of judgments for each individual became a final set of profiles themselves,
about which a last set of judgments were made concerning the overall ,
acceptability of the 23 system states.

Stepwise multiple regression analyses were used'to develop models of the
judgment process of the four participants. criteria vere entered into the
regression equations only if they were found to be statistically significant
predictors of the participants’ judgments (p < .05). The resulting multiple
Rs ranging from .74 to .99 (an average of .94) indicated that a major
proportion of the variation in judgments could be reliably predicted by the

Linear, additive models; nonlinear and nonmetric models requiring additional
PROFILES OF OVERALL SYSTEM PERFORMANCE

Job Availability 6
Housing Quality 15
Population Distribution 7
Industrial Conditions 5
Density Composition 9
Tax Structure i
Job Availability 18
Housing Quality 6
Population Distribution 15
Industrial Conditions 18
Density Composition W
Tax Structure 16

FIGURE 3. Two of the 23 Alternative Profile Segments Depicting Individual
Ratings as a Basis for the Evaluation of Overall System Performance.

139

predictive terms were not tested due to the limited number of profiles
available for the data base (details of this analysis have been presented in

Rohrbaugh and Andersen, 1979, and Andersen and Rohrbaugh, 1979)-

nsory Field.

c Summary of ¢!

The construction of an empirically derived cross-sectional objective
function as discussed above represents a fairly straightforward application
of extating techniques of judgment analysia to the field of dynamic syateas.
However, extending the evaluation of system performance over an extended tine
frame poses several conceptual and technical problems. The system dynamics
Literature is filled with suggestions that longitudinal evaluation of systex
performance might prove difficult. For example, many systems exhibit
unexpected or even counter-intuitive behaviors over time (Forrester, 1971). A
system that initially shows relative improvement may soon reverse itself and
show deterioration. 7

To address this difficulty of deriving a full longitudinal objective

function, the following approach waa developed. ‘The 7 regression equations
constituting the full judgment model for each of the participants (6 equations
for each of the initial sets of judgment tasks, e-g., housing quality and job
availability, and one equation for the overall evaluation) were attached to
the Urban Dynamics model in order to create a new objective function sector.
When the model subsequently was run, plotted output from the objective
function sector showed how the individuals’ preferences varied dynamically as
the performance of the system fluctuated over time, as illustrated in Figure

4
b auniyg

S842} LOg [RAFAH URGAN 4NOY 4oJ SUOZZDUNy anLaoa~qU aueuUNS

wes6oid Bujupes3 pexolduasapun

wesod uojzeaua qof pedo, duauepun

bed function

Objective

Buty

os

Eres

os

Objective

° function

uLSMOY wNLS

ueuBoud u0}34 Lowep

S-

Suh}

os

Objective 140

function

02

Objective
function

s

3

rt

wesGoad u0139nJqsu09 asjd4aqua MaK

os

Sensory Organization-~Objective Function Trajectories

How does one respond to the patterned stimuli that comprise an objective
function varying through time? What sensory organization is brought to bear
on the shifting characteristics of the curves such as those show in Pigure 4?
For example, consider two systems, A and B, that begin at the same point (as
measured by the objective function) and improve to the same equilibrium point
that represents a 20% improvement in overall system performance. System A,
however, rapidly improves by 508 in the first fifteen years, deteriorates to
its initial conditions by the thirtieth year, and finally settles into
equilibrium at the end of a fifty year period. in contrast to this
oscillating improvement pattern, system B rises slowly but steadily to its
final equilivrium, in the beginning showing much less drastic improvement than

Whether a decision maker

system A but never deteriorating as does syst
prefers the pattern of oscillating improvement or the pattern of gradual
Amprovement is an additional problem requiring individual judgment. It would
appear to be quite difficult to quantify and measure exactly what it is about
these two dynamic objective functions that is preferable. The problem is
further exacerbated when one considers the broad range of behavior that can
emanate from a dynamic system.

‘A fundamental difficulty to be overcome in order to evaluate alternative
trajectories is determining a set of dimensions that can be used to classify
the various curves generated by the objective function sector of the Urban
Dynamics model. Furthermore, do decision makers evaluate system performance
through time by consistently using a fixed set of curve characteristics? To
begin to investigate these problems, a class of analytically “well-behaved”
curves that closely paralleled the trajectories generated by the objective

function sector of the Urban Dynamics model were explored further. The curves
is

analyzed in this portion of the study were generated by orthogonally varying

: ey 8 ; S four characteristics that almost fully defined their trajectories: number of
| | j years from initial equilibrium to maximum, number of years from maximum to
; 2 second minimum, maximum point, and final equilibrium point. The systematic
i a variation of the four characteristics at five levels (e.g-, 5, 10, 15, 20, and
| & ¢ “ 25 years from initial equilibrium to maximum) produced a set of 25 alternative
i * $ . curves, 4 ofjwhich are illustrated in Figure 5.
—_— = ‘The set of 25 curves was presented to the 15 research participants
‘ 2 (advanced graduate students in the field of management) who vere instructed
‘ é that these curves represented hypothetical objective functions tracing the
a a7 g ae = 2 longitudinal acceptability of urban system states. the participants were
@atqoerqo aay3oafqo = sked to evaluate the curves on a scale of 1 (a completely undesirable
: 2 trajectory) to 20 (a very desirable trajectory). Again, through the use of
2 stepwise multiple regression analyses, judgment models were derived which
3 could be used to predict consistently the desirability of a wide range of
i 8 foesguinee 8 7 curves. Characteristics of the curves were entered into the regression
i = equations only if they were found to be statistically significant predictors
i 2 of the participants" judgments (p < .05). ‘The resulting multiple Rs ranging
| 4 $ from .89 to .97 indicated that a major proportion of the variation in
2 £ & judgments could be reliably predicted by the linear, additive models based on
r or ra the four characteristics identified.
jurwyxey £ Table 1 presents the standardized regression coefficients in relative
i i H = form (i.e+, constrained to sum to 1.00) for each participant. The judgment
| ; sj models shown in Table 1 represent diverse cognitive approaches to the
Ss woryouny ar) worzouny e' evaluation of the 25 alternative objective function trajectories. Six of the
s 9arz9afqo @atqoefa0

participants appear to be particularly concerned about the final equilibrium

point, placing over 80% of their relative weight on that one characteristic of

141
Table 1

Relative Weights Comprising. Judgment Models for Objective Function Trajectories

v

“142

Cluster type 1A 1B Ic 110 HE LIF
Participant 123 4 5 6 7 8 9 WN 12 13 415
Final

equilibrium 1.01.01.01.0 .82 81 77 475 68 .65 .62 .57 56 235.30
Maximum

point ses set see see see - 23.25 .32 .35 .38 .43 30 +40 249
Years to

Micime: eevee ig eee | Gas ake ck eames ee 18 ll
Years to

maximum aes see nee eee ase 19 ee wee wee eee wee eee lene 412 .09

Multiple R +96 .97 .97 .96 90 89 +92 .95 .94 .95 .92 .91 395 9

97

the trajectories. The remaining nine participants appear to be concerned both
about the final equilibrium point and the maximum point of the trajectories,
some placing almost 50% of their relative weight on the latter characteristic.
Concern about the number of years from initial equilibrium to maximum and the
number of years from maximum to second minimum also differentiate the sensory
organization of the participants but to a lesser degree.

By grouping participants into clusters of individuals with similar
methods of responding to the patterned stimuli provided by the objective
function trajectories, it is possible to describe at least six discrete
judgment policies reflected by the decision makers in the present study:

TA: concern for maximizing final equilibrium

IB: capcern for maximizing final equilibrium and years to
second minimum

Ic: concern for maximizing final equilibrium and years to
maximum point

IID: concern for maximizing final equilibrium and maximum
point z

ITE: concern for maximizing final equilibrium, maximum pointy
and years to second minimum

IIF: concern for maximizing all four characteristics but
particularly maximum point

A series of two choices between objective function trajectories will determine
for any decision maker which of the six methods of sensory organization is
being used. The first choice, shown at the top of Figure 6, indicates whether
the decision maker is in cluster type I or cluster type IT. If the decision
maker prefers the trajectory associated with cluster type T, the second
choice, shown at the lower left of Figure 6, indicates whether the cluster
type is IA, IB, or IC. If the decision maker prefers the trajectory
associated with cluster type II, the second choice, shown at the lower right

of Figure 6, indicates whether the cluster type is IID, IIE or ITF.
9 oun yy

Saanqon4rs a2u94a4a4g 2494 Luly

40 S497SN {9 X}s UaaMeg YsinvU;3S}0 03 pasp sa;u0r2IPeAL

ig

Because a regression equation exists to mathematically model the sensory

Objective organization that each cluster of decision makers brings to the patterned

function
= stimuli created by the objective function trajectory, any run of the Urban

Dynamics model can produce a single index of overall performance of the urban

system with xespect to a particular urban revival policy. For example,

because the decision maker whose objective function trajectories shown in

Figure 4 produced a judgment policy indicating a concern only for maximizing

uy

final equilibrium (cluster type IA), it is clear that he would prefer the

Objective
é poact i 8 effects of an underemployed training program (shown at the bottom left of

Pigure 4) to the effects of an underemployed job creation program, slum

housing demolition, or new enterprise construction program.

Put in terms more familiar to system dynamicists, the reaults presented

|

in Table 1 suggest that, when presented with output such as that found in

Urban Dynamics, decision makera tend to fall into two broadly defined

Subd

clusters. The first cluster consists of those who tend to place considerable

Objective

8 . <
° function 8 emphasis on the steady state behavior of a system in response to a policy

-4- change. The second cluster consists of decision makers who place relatively

more emphasis on the transient response of the system. The most important

transient characteristic is the maximum point of the transient peak followed

OS

by the speed of decline (measured as time to minimum) and the speed of

increase in overall system performance (measured by time to maximum).

These results suggest that, when faced with worse before better" or

"better before worse" types of system performance, these two clusters of

decision makers would tend to split in their evaluation of preferred policies.

Of course, these results strictly hold only for behaviors similar to those

emanating from the Urban Dynamics study. Separate analyses would have to be

conducted to identify similar clusters of preference structures for other

classes of aystem performance.

143
An interesting speculation centers on what causes these differences in
preference structures for various clusters of decision makers. One unexplored
question in this study is the degree to which the participants tended to view
the sample trajectories as definitive statements about what the future would
actually be like versus a belief that the trajectories are fallible
predictions about the future originating from a model or other predictive
device. One hypothesis that might explain the differences between the two
major cluster of decision makers is that those who favor steady state may
believe that the trajectories are actually a statement of what will happen.
They may be pessimistic about their ability to intervene and change the
outcome in the long run, favoring a policy that leads to a presently
established sure gain. On the other hand, those who favor the transient
response tay be skeptical of the accuracy of long term predictions and
optimistic about their ability to intervene and reverse a bad or deteriorating
situation in the long run. Hence these decision makers may favor short-term
improvenents at the expense of the longer run consequences of the policy being

examined.

Conclusions

‘The work presented above has developed an empirically derived method,
based oR social judgment analysis, for quantitatively deriving and modeling
the implicit preference structure for individual decision makers. When
coupled with a system dynamics simulation model, these preference structures
can be turned into dynamically varying objective functions that summarize at
any point in time how an individual decision maker ranks the overall system
performance. This result is especially useful because it allows researchers

to determine to what degree individual decision makers differ in the criteria

2

144

that they employ for evaluating overall system performance over time. For a
sample of fifteen, important individual differences in the criteria

used to evaluate system performance were found. Further, these individual
differences were found to cluster into several distinct types. The major
difference between the various cluster centered on whether individuals more
heavily weighted the system's transient response versus the system's steady
state response. Important characteristics of the system's transient response
were found to be (in order of importance) the maximum point for the transient
curve, the speed of decline, and the epeed of improvement in the overall
system performance. The paper concludes with some speculation on what might
be the psychological determinants of the empirically observed clusters of

individual @ifferences.
23

REFERENCES:

Andersen, D. F. and Rohrbaugh, J. (1979). “objective Function Dynamics:
Evaluating Urban Systems Through Time," Proceedings of the International
Gonference on Cybernetics and Society, Denver, Colorado.

Porrester, J. W. (1970). "Counterintuitive Behavior of Social Systems,"
Technology Review 73.

Forrester, J. W. (1969). Urban Dynamics. Cambridge, Ma!

at MeI-T. Press.

Gardiner, P. C. and Ford, A. (1980). “Which Policy Run is Best, and Who Says
So?" in Legasto, AeA., Forrester, J-W. and Lyneis, J.W., (eds),
TIMS Studies in the Management Sciences Vol. 14: Systems Dynamics,
‘Amsterdam: North Holland Publishing Company.

Wamnond, K. R., McClelland, G. yi. and Mumpower, J. (1980). Human Judgment
and Decision Making: Thedries, Methods, and Procedures. New York:
Praeger Publishers.

Hammond, K. R., Mumpower, J. Le, and Smith, T. H. (1977). “Linking
Environmental Models with Models of Human Judgment: A Symmetrical
Decision Aid." IEEE Transactions, Systems, Man, and Cybernetics SMC-7:
358-367.

Hammond, K. R., Rohrhaugh, J., Mumpower, J., and Adelman, L. (1977). “Social
Judgnént Theory: Applications in Policy Formation," in M. F. Kaplan and

8. Schwartz, (eds.), Human Judgment and Decision Processes: Applications
in Applied Settings. New York: Academic Press

Ramona, K. R., Stewart, T. Re, Brehmer, B., and Steinmann, D. 0. (1975).
“Social Judgment Theory," in M. F. Kaplan and S. Schwartz, (eds.), Human
Judgment _and Decision Processes: Formal and Mathematical Approaches.
ew York: Academic Press. ~

Kohler, W. (1947), Gestalt Psychology. New York: Liveright+

Rohrbaugh, J. and Andersen, D. F. (1979). "Specifying Dynamic Objective
Punctions: Problems and Possibilities,” Proceedings of the 1979 Summer
Computer Simulation Conference, Toronto, Ontario, Canada.

145