Levine, Ralph L., "The Paradigms of Psychology and System Dynamics", 1983

325,

THE PARADIGMS OF PSYCHOLOGY AND SYSTEM DYNAMICS*

Ralph L. Levine
Department of Psychology
Michigan State University

East Lansing, Michigan 48824

ABSTRACT

This paper compares and contrasts the philosophical and
methodological paradigms used by psychologists ‘and system dyna-
micists. Currently, psychologists collect huge amounts of data,
use open loop methods of experimental design, and think that
classical statistical models, such as the analysis of variance
and regression analysis, provide the most useful methods for
studying social phenomina. Behavioral approaches to psychology
differ sharply with the system dynamicists concerning the rela-
tive importance of external vs. internal sources of influence on
behavior. The behaviorists focus on controlling the external
environment, even denying the existance or importance of internal
states. The problems of using external control are illustrate by
contrasting two simple attitude change models; one which modifies
attitudes solely through outside influences and another which
makes the change in attitudes a function of the state variables.
System dynamicists attempt to understand the dynamics of social
processes through the study and analysis of dynamic loop struc—
ture. These techniques would be extremely useful for those
psychologists using correlational analysis and causal modeling
methods, where the implications of dynamic structure are not
always fully understood.

The purpose of this paper is two-fold. The first goal is
to explore areas where the system dynamicist and psychologist can
contribute to each other's knowledge of social systems. The
second goal is to suggest points where the science of psychology
can make strides toward building a more comprehensive theoretical
foundation by adopting paradigms patterned after those used by

system dynamicists.

*

The author wants to thank his colleague, Mark Rilling, for
a helpful discussion of recent trends in modern behaviorism.

Those familiar with system Dynamics know that there is a
distinct set of philosopical assumptions.which place it apart
from many other approaches to understanding social, biological
and phyisical processes. It provides a means of viewing social
systems in a rich manner, using the modern systems approach to
complex problems, Indeed, it is these fundemental philosophical
differences which has already had some profound influences on
social thought. In particular, system dynamic thinking is diame-
trically opposed to much if not all of the philosophical assump-
tions held dear to most social scientists. Frankly, system
dynamics appears to fly in the face of what might be considered

"good" social science.

CHARACTERISICS OF PSYCHOLGICAL INQUIRY

It is difficult to characterize what psychologists do. A
beginning text in the field presents what might be a mixture of
very interesting, but unrelated topics, such as visual percep-

tion, learning in rats, development of motor coordination, cop-

ing, fantasy, learned helplessness, etc. [1]. Indeed, the field

is too complex to address the problem of integrating this area in
this short paper. However, there are major methodological and
philosophical characteristics which transend many of the minor
differences among existing schools. Moreover, with the exception
of the psychoanalytic school, almost every group of psychologists
within the mainstream of academic psychology are patently be~
havioristic, particularly among applied psychologists. This

central focus on behavior, rather than dynamic structure im-
326

3

mediately sets it apart from System Dynamics.

Table 1 indicates some of the major dimensions which char-
acterize modern psychology. First, psychologists stress the
study of information systems, caring less about flow and energy
processes, in counterdistinction to, let us say, economics. Cog-
nitive psychology, in particular, studies perception, learning
and memory, and decision or judgmental processes. Research
takes the cognitive psycholgist into the some exciting realms:

language, brain processes, and computer technology.

To the system dynamicist, the psychologist's preoccupation
with information handling should ring a sympathetic note, for
indeed, following information around the loop structure is a major
task of the model builder. The psychologist, in turn, can ap-
preciate the inclusion of perceptual and informational delays
which are stressed in system dynamic models. in fact,

Forrester's discussion of how the exponential delay functions as
a information and decision process concurs with what psycho-
logists know about how people integrate information [2]. This is
an insightful contribution to the study of decision processes,
one which makes models very realistic from the psychological

standpoint.

Cognitive psychology is making strides toward understanding
memory, decision processes, perception and learning. However,
although this area has been fruitful, nevertheless progress has

been limited somewhat in at least three ways. First cybernetics

2,

Table 1

A Comparison of Psychological and System Dynamic Methods

Psychology
Empirical Orientation

Use of Computers
for Data Analysis

skinnarians- Stress
Control Through Changes
In Outside Environment

Experimentalists- use
Open Loop Thinking

Experimentalists- Study
Interactions with
Analysis of Variance

Correlationists- Stress
Measurement and
Reliability Through
First Order Correlations
& Multivariate Analysis

Take an Empirical Approach
Time Series Analysis

Make No Hypothesis
Concerning the Direction
O£ Loops in Causal
Models Before Statistical
Estimation Of Parameters

lL
2.

6

System Dynamics
Model orientation

Use of Computers
for Simulation

Stress Control
Through Changes in
Internal Structure

System Dynamicists-
use Closed Loop
Thinking

Study Interactions
Through Analysis of
Feedback Loops

Stress Dynamic Process
Modeling & Qualitative
Fit of Models to Data

Stress Dynamic Loop
Structure Underlying
Time Series

Hypothesize Direction
And Purpose of Loops in
Simulation Models
327

and feedback concepts are not well known to most cognitive psych-
ologists. Secondly, psychologists define a decision much more
narrowly than is necessary. Borrowing the perspectives of sta-
tistical decision theory and economic choice theory, psycholgists
are interested in short-term, discrete situational choices which
may or may not lead to dynamic consequences [3]. This approach
follows from much of modern economic decision making models,
which deal with extremely static situations, without regard to
dynamic effects [4]. Finally cognitive psychology is limited by
their reliance on the classical experimental paradigm and their
insistence on statistical analysis of their data, a topic which

will be discussed shortly.

It is interesting to contrast this approach with that used
in system dynamics. The decision is a key concept in system
dynamics, However, decision making takes on a much a much
broader context. Decisions deal with continous processes which
affect rates of flow into and out of storages. Although it is
possible to consider individual decisions, system dynamicists
focus on long-term effects of policies at decision points, which

are represented by rate equations [5].

At the moment, academic psychology appears to be roughly
divided into two general approaches, namely the experimental and
correlational approach. This is only a convenient simplificaton,
and indeed there are exceptions to this generalization. The cen-
tral methodolgical theme cutting across both approaches is the

stress on empiricism. At the end of the 20th century, psychology

has few general theories to fall back upon, yet has a rich ac-
cumulation of empirical literature that has been growing rapidly

throughout the last 190 years.

EXPERIMENTAL PSYCHOLOGY

As the name implies, experimental psychology deals with
the study of responses to manipulating one or more independent
variables. Two school have emerged in recent years. The smal-
lest in number are the operant conditioners, who use a set of
methods pioneered by B.F. Skinner [6,7]. In applied areas, these
Skinnarian techniques have been extended to cover many practical
situations, Today behavioral modification is somewhat popular as
a tool in mental health, education, and in industrial settings

(8).

The idea behind behavioral modification is that, by using
modern reinforcement techniques, one can control and change beha-

vior. This is certainly not new or different. However it is the

site of change which does differ from the central assumptions of

the system dynamicist. To a behavioral modifier, the site of
change comes from the outside, not from a change in inner
structure, As a matter of fact, the idea of inner state vari-
ables appears to be avoided entirely. The individual frequently
is considered as an unknown black box, potentially controllable
from the outside environment. A skillful behaviorist can rid
the individual of bad habits through the control of the immediate

environment, through reinforcement techniques, etc., not by
328
7

changing the dynamics of what is inside the black box.

It is ironical that the extreme behaviorist pays so much
attention to the individual, yet completely ignores internal
structure, which may determine much of the behavior that is the
central content of the science of psychology. In contrast, the
system dynamicist focusses upon internal structure. According to
this group, most of the problems arise when the structure cannot
cope with stress from the outside. one of the major activities
of S/D is to discover the dynamic structure, locate points where
problems arise, and to suggest changes in structure which may

buffer the system from external control.

The two perspectives are indeed 180 degrees out of phase.
Historically, Skinner has brought into experimental psychology
the notion of control, and the ability to change individuals. 1t
certainly is very close to being an empirically oriented control
theory. There are three major requirements for controlling be-
havior. First one must have complete control of the environment.
Second, one must know the past history of reinforcement sche-
dules, and finally, third, control of behavior can be a function

of the genetic makeup of the individual.

These three assumptions concerning the conditions for con-
trolling behavior have become the basis for a new technology.
Let us examine them in light of the perspective of system dynam-
ics. First, in terms of controlling the environment, in many

respects the purpose of system dynamics is to buffer oneself from

a hostile environment, so that the site of change, as we have
noted is internal, i.e., within the boundary of the individual.
Behavioral modification has been successful, but what are the
costs of controlling the environment? The general question re-
mains whether it is more efficient to control from the outside or
make changes from inside the system boundary. This point will be

brought out a later discussion of attitude change.

The second assumption deals with a knowledge of past rein-
forcement histories of the individual. From a systems point of
view, knowledge of the past history really is a substitute for
knowledge of the state of the system. The modern notion of state
plays an extremely important role in system dynamics modeling
[9]. Skinner refuses to recognize the usefullness of internal
state concepts, yet in actually, if one individual acts differ-
ently from another because it has had a different history of

reinforcement, then each is in a different state, State repre-

sentation is not the only way to describe system behavior. The

problem with Skinner's requirement of knowing the past history of
the individual is primarily practical, for knowing every moment
of the individual's past history of reinforcement is totally

impossible. Moreover, in a deterministic system, two different

‘past histories can lead to the same values of the state vari-

ables. Once this happens, the two trajectories will be identical
from that point on. Spending time seeking a knowledge of past

histories is wasteful of time and effort in some cases.
329

The third requirement for controlling behavior deals with
the individual's genetic nature. Historically genetic components
appeared late in the evolution of behavioristic thinking.
Origin-ally, the founder of behaviorism, John B. Watson, felt
that one can train any animal to do almost anything [16]. Indeed
Skinner's original worked stressed this point. For example, one
could see films of pigeons learning to play ping pong. Later
people began to question these assumptions by indicating situa-
tions where genetic capacities introduced constraints on the
learning process [11]. Indeed today there is a growing active -

field o£ investigation looking for limitations to learning [12].

The third condition appears to be a modification of the
classical environmental assumption of behaviorism. ‘This, how-
ever, indicates that another type of dynamic internal structure,
namely enzyme pathways, helps to determine behavior. Thus, in
reality, in two out of three proposed major factors which deter-
mine behavior, modern behaviorists are not far from a state
variable position. in terms of the site of changing behavior,

however, system dynamics and behaviorism totally conflict.

DYNAMIC VS. STATIC ANALYSIS OF A PROBLEM, Not every psychologist
subscribes to all of the practices of the Skinarians. For exam-
ple Skinner concluded, after finding that the mean learning curve
of a group of rats did not resemble individual learning curves,
that the science of psychology should drop all statistical
analysis and just follow individual cases over time. This issue

separates the Skinnarians from other experimental psychologists.

10

In general, most researchers think that statistical analysis and
design is the major framework with which to develop a science.
Graduate students spend their first year learning the intricacies
of repeated measures designs, Latin squares, orthogonal poly-
nomials, and eta squares. A good graduate program in psychology
brings the student far along the road of statistical sophistica-
tion. It is the hallmark of a department if their students know
the latest form of the analysis of variance or perhaps are

acquainted with a new time series model.

the major problem with emphasis on statistical analysis of
data is that the statistical model underlying the analysis be-
comes a substitute for a process model, rather than being a
device for organizing the data, exploring response surfaces, and
Suggesting functional relations among variables when nothing is
known about the phenominon. Analysis of variance and regression
models are very poor representations of social processes. In

particular, dynamics are not well handled using analysis of

variance, even when time becomes an explicit variable.

Another problem the statistical techniques used by experi-
mental psychologists is that in some sense, methods like the
complex analysis of variance become almost uninterpretable in
situations where the experiment is a function of many independent
variables. It is almost impossible to interpret seventh order
interactions, yet we know that behavior is a function of many

variables. System dynamics models attempt to capture the nature
330
1

of the system, The interactions among variables are carefully
studied as the structure of the model is developed. There is
little problem in interpreting models which handle many vari-
ables, and indeed system dynamic models, such as the M.I.T.
National Model, may involve hundreds of interacting variables
[13]. The analysis of variance can only handle a restrictive
number of variables before becoming totally useless to the re-

searcher,

What are some of the problems with not perceiving diffe-
rences between dynamic and static designs, situations, and
processes? Although there might be many exceptions, especially
among learning theorists, in general the differences between
dynamic and static proceses are not always clearly distinguished
in psychological studies. ‘Take for example, the classical dosage
response curve. One might want to vary the concentration of
sugar as a reinforcement to see the effects on learning perfor-
mance. In the independent groups design, each rat is assigned to
one and only one concentration group, while in the repeated mea-
sures design, each rat would get all concentrations in perhaps
random order [14]. The data from both types of studies would be
presented the same way, namely mean performance as a function of
increasing concentration of sugar. Indeed, a person reading a
report of the study could not tell which design was used without

reading the methods section.

A fundemental difficulty arises when carryover effects,

Such as learning, adaptation, etc. occur, which would violate the

standard analysis of variance models used to analyze the data.
Now if there were carryover effects, then the results of a study
using independent groups design would differ from those generated
from a repeated measures design. Carryover effects, which are
quite common, imply the influence of other state variables which
are not picked up by measuring a single dependent variable. The
number of levels or state variables is greater than one. The
carryover effects themselves may be part of the true dynamics.

In one case, the analysis of variance model could be extended to
at least pick up the presence of carryover effects. In the
independent groups design, carry over effects would be completely

missed.

OPEN SYSTEMS THINKING. There are reasons for the popularity of
statistics in psychology. Historically, the paradigm of the
experimental method and statistical analysis came from work in
agricultural and biological research. In these areas of expe-
rimentation, one literally goes to an experimental field, applies
varying concentrations of materials, such as nitrogen, phosphor-
ous, and other nutriants, and then measures the response to
changes in the concentration of the nutriatnt. ‘The classical
experimental paradigm is an example of open loop thinking as
opposed to a closed loop approach [15]. A stimulus variable,
varying in intensity, elicits one or more responses, without

thought to possible feedback effects.

The lack of interest in feed back effects appears to be
331

13
ubiquitous throughout much of experimental psychology. The
Skinnarians literally place a box between the experimenter and
the subject, and treat the system in a black box fashion. The
rest of the experimental field spends time performing open loop
experimentation. The problem, of course, is that one cannot avoid
feedback effects between the subject and the environment. This
can lead to faulty thinking, and in the context of applications,

it may lead to faulty policy formation.

Here is an example of some of the problems with the experi-
mental psychologist's version of open loop thinking. One of the
most exciting new areas of applied psychology is the evaluation
of public programs and policies [16]. For example, investigators
might want to evaluate the effects of a particular type of work-
shop on energy conservation behavior by giving different experi-
mental groups (with a control) varying types of materials, les-
sons, or hands-on experience. This can be performed in the tradi-~
tion experimental manner, randomly assigning people to workshop
conditions, and then observing changes in conservation behavior

over a specified time period.

The major problem with this procedure is the assumption that
a single dose or pulse into the system has little connection with
other components of behavior, and that the effects first noted in
the experiment are permanent. In this area of experimental
program evaluation, researchers use designs which only assess the
impacts of discrete events, such as a single workshop. The

design does not take into consideration the strong likelihood of

14

feedback effects. Indeed, Forrester has pointed out that in
social systems, major changes will at first go in the desired
direction, After a time, inhibiting forces, represented fre-
quently by negative loops, dominate the system, so that eventual-

ly behavioral patterns drift back to the original level [17].

This is precisely what may happen to the experimental
groups, if one waits long enough. The open loop approach assumes
that a single pulse is sufficient to cause long-term changes in
organizations and individuals, an alternative method method
would be to (1) develop a hypothesized set of dynamic loops to
account for the potential effects of workshops, (2) generate a
model of the system, and (3) assess strategies for maintaining
the system on target. The model can then be tested experimenally
by varying the number and or timing of workshops to keep this

behavior at a given level.

An advantage of the system dynamic approach to this situa-

tion is that it can explain why programs and policies succeed or

fail. Moreover, one sees common mechanisms spanning many simi-
lar phenomina, so that in other applications of experimental
methods, one is aware of the communality of many of these dynamic
processes, In this case, new habits die out quickly,. especially
when the circumstances for their performance depends totally upon
external events, Fig. 1 shows several examples of such a hy-
pothesized loop structure. A person with a skin disease has

notable symptoms, which may be disturbing. In fact, the person
15

ENERGY ar

CONSERVATION
BEHAVIOR
+ +
SENSITIVITY TO

PROBLEM
ae KNOWING TECHNIQUES

OF CONSERVATION

Td I.
RATE OF FORGETTING
SKIN PROBLEM

USE OF
MEDICINE

+
SENSITIVITY
TO PROBLEM
. TAKING PILLS

(b) A +

BLOOD PRESSURE
PROBLEM

un ©
SENSITIVITY

TO PROBLEM

FEAR OF

(co) CONSEQUENCES
TOL,
RATE OF
FORGETTING

Fig.! Three Examples of a Similar Structual Process

16

may go to to a dermatologist, and begin using medicine to relieve
the symptoms.. The disapearance of symptoms lowers the person's
sensitivty to the problem which in turn, after a time, alows the

skin problem to reappear once more.

The same mechanisms (Fig. 1c) are at work in the area of
high blood pressure, where physicians frequently worry about
their patients neglecting to take medicine to control hyperten-
sion. Unfortunately, people with high blood pressure dis-con-
tinue taking their medicine after several months of monitoring
pressure, They become somewhat sensitive to the dangers of -high
blood pressure, upon a routine visit to the doctor's office, and
begin taking pills to keep the blood pressure level within
bounds. The pills have the immediate effect of keeping blood
pressure in line, but after several. months of getting very posi-
tive reports, the patients begin avoid taking the medicine on a
regular basis. It is interesting that a reinforcement theorist
would say that being successful in lowering blood pressure by the
act of taking the pills is reinforcing. This in turn should lead
to the continuation of taking pills, because it is reinforcing,
and not to decreasing the behavior, A simple reinforcement

hypothesis leads to the wrong conclusions.

Consider again the example of the workshop experiment.
First, almost any new activity has negative inhibitory processes
attached to it. These indicate a behavioral counterpart to an
entropy effect, so that higher the level of commitment to a new

set of habits, the faster the rate of decrease in commitment
333
7

occurs. Although the variables are different, this negative loop
is common in physics and chemistry, The behavioral approach
stresses control from the outside, and one is forced to contin-
uously "pump" up habits by periodically retraining the subjects.
The retraining process both serves to give them facts concerning
how to save energy and heighten their awareness of the costs of

wasting energy.

This may not be the only method for making habits more
permanent, There may be ways to change the structural loops
associated with the subjects behavior. In any event, a model can
tell one what would be the impacts of a single pulse into the
system, or indeed many pulses, Knowing that a single pulse will
bring the system back to baseline leads to a much more sophisti-

cated experimental verification of the model.

CORRELATIONAL ANALYSIS ;
The second paradigm in psychology is characterized by the
use of paper and pencil tests or questionaires to study the
implications of relationships among so-called "dependent varia
bles." No attempt is made to use experimental methods, con-
sidering them impractical outside of the laboratory situation.
Correlational analysis is based on the theory of meaurement,
derived from psychometrics and multivariate analysis. In par-
ticular, applications of multivariate techniques are frequently
made to the study of personality and social processes. Moreover,

applicatiions of these principles have been found quite useful in

18

clinical and industrial areas of psychology.

Perhaps the most impressive contribution to this area of
research in psychology deals with the theory of measurement and
reliability [18]. Psychologists study variables which are
thought to be crucial in understanding social and personality
processes, Frequently these concepts are difficult to measure.
The system dynamicist also has to deal with some concepts which
are "fuzzy" in nature, such as quality of life. Neither group
shies away from using those variables when they are thought to be

important.

There are two major differences between the correlationists
and the system dynamicist, beside from content area. First the
correlationist, as the name implies, relies quite heavily on the
notion of correlation to assess both static and dynamic relation-
ships. This strategy has paid off dramtically in the static
case. The theories of measurement and reliability have enabled
‘the psychologist to obtain stable and relevant measures of fuzzy
concepts, such as anxiety, cognitive complexity, and locus of
control. Those measures, once operationally defined, can be used

in a variety of situations.

The system dynamicist, as described earlier, also uses
“soft" variables in modeling social problems. These variables
are frequently the factors suggested and used by the clients and
decision makers who are touched by the problem. At first little

or no attention is made to obtaining estimates of the
334
19

reliabillity of the measures used to fit the model, especially if
the purpose of the model is to assess general trends and impacts
of proposed policy and not for exact forecasts. The recent
literature in system dynamics does address measurement error and
using S/D models for quantitative forecasts [19]. For example,
GPSIE, a program to the estimate parameters of S/D models, has
been developed to address the problem of both measurement errors
in the empirical time series and errors due to misspecification

of the model,

A second difference between correlationists and system
dynamicists is in the order of performing tasks. The predominant
operating mode in psychology is to spend the most time on the
empirical aspects of psychological research, waiting for later
stages of their research or for others to put the pieces together
into a coherent theory. The regression or correlation model
becomes one's temporary theoretical framework until a substantive
model can be formulated. ‘This distribution of effort is also
reflected in graduate training. Psychology graduate students are
well versed in multivarite analysis and psychometric theory, but
have little formal training in building models of substantive

processes,

Waiting for theories to emerge can lead to some relatively
difficult problems for the correlationist. First, one must rely
heavily upon the size of correlations in the data set. Problems
of the use of first order correlation coeffiencts have been

pointed out before by system dynamicists [20]. Basically, cor-

20

relations are indices of what is happening at the surface, and
first order correlations neither hold other factors constant nor
effectively measure non-linear relationshps. Moreover, a new
criticism has emerged, which reflects the fact that many correla~
tion coefficients are based upon extremely small samples, at
least in psychology. The size of the sampling error is so great,
that most relationships cannot be detected by sample correlations
[21]. Currently psychologists have been developing meta~analytic
methods to combine correlation studies to increase sample size,

and statistical power [22].

CAUSAL MODELING. A second problem with a predominately empirical
approach to correlation analysis has appeared in using a new
methodology, which psycholgists have borrowed from quantitative
genetics, econometrics, sociological methodology, namely causal
or structural modeling. First, there is still some confusion
between static and dynamic processes. In experimental psycholo-
gy, this was manifested by using between group designs when
repeated measures designs were called for. In the case of cor-
relational analysis, the confusion between dynamic and static
processes is particularly evident in situations where investiga-
tors use these causal modeling techniques [23]. Frequently, this
occurs when fitting data so-called non-recursive linear models,
in which feedback loops have been hypothesized. Feedback, of
course, implies a time process. However, often the data have
been obtained at only one point in time, i.e., measurements have

been taken simultaneously, through a single questionaire or paper
335
a1

or pencil test to cut the costs of research. In this case,

change in a variable is represented as variability among sub-
jects, not change within subjects. The assumption that static
correlations are equivalent to dynamic serial correlations is
risky at best. Computer programs which estimate path coeffi-
cients are completely insensitive how these correlations were

obtained.

Another problem with the use of causal modeling is that
invariably, in the process of hypothesizing loop structure, the
correlationist may not make a commitment concerning whether the
loop is positive or negative. The action of the loop is con-
sidered an empirical matter to be determined by the sample data
set. Statistical programs are available to estimate loop para-
meters, i.e., path coefficients. Frequently, the investigator
will wait to see the results of the computer run to tell whether

or not a given loop was positive or negative.

The empirical “wait and see" approach to causal
modeling has problems. For example, psychologists use stan-
dardized scores, i.e., scales which have a mean of zero and a
Standard deviation of 1.6, instead of the set of original
measurements, raw scores. Unfortunately, statistically it is
possible to find that a given loop is positive, i.e., the product
of.the path coefficients is positive, when using standardized
Scores, and the same loop is negative when using the original raw
Scores (or vice versa). Which is correct? There are no empiri-

cal guidelines to indicate to the researcher the appropriate

22

direction of the loop. Unfortunately the direction of the loop

is an extremely important characteristic of any dynamic system.

The system dynamic approach to modeling hypothesizes a loop
structure, as does the correlational analyst. However, this is
where the methods diverge. First, in building an S/D model, the
purpose of each loop is clearly delineated from a dynamic point
of view. Every loop corresponds to a hypothezed feedback
mechanism which accounts for growth, collapse, and reversal of
direction. Thus the system dynamicist hypothesizes the direction
of the loop and justifies the assumptions underlying those hy-
potheses. One should know whether the loop is positive or nega~
tive. In many instances one can check the hypothesized action

of the loop with actual time series.

Finally, the empirical approach can lead to rather am-

biguous conclusions, under circumstances where a model fits the

data well, but appedrs to make fail many of the criteria for

realism [24]. Several articles have appeared in which
empirically all the loops have turned out to be positive, with
no negative loops to inhibit growth, Statistically chi square
tests of significance appeared to show excellent fit to cross=
sectional data. However, without negative inhibitory processes,
the actual behavior of the system should be quite unstable,
growing infinitely large, or collapsing. In terms of validation,
the model is completely misspecified, yet from a statistical

point of view, it the model appears to fit the data quite well.
336
23

APPLICATIONS TO ATTITUDE CHANGE. Many of these princples can be
illustrated by a simple example. an area of interest to psycho-
logists is changing attitudes, and as a matter of fact, much of
the social psychological literature during the past 15 years has
been devoted to this area. The behavioristic approach to atti-
tude change would place much emphasis upon the role of outside
influences upon attitudes, which in turn might be defined as
social behaviors [25]. The key mechanism is in the form of a
message from the outside world to the individual, Attitude
change is generated by controlling messages. Let us take then a
simple behavioral model found in Hunter and Cohen [26]. Assume
that the message given to a person can be measured on the same
scale as the attitude itself, The rate of change in attitude, A,
to can be represented by the derivative of A and might take the

following form:
.
A = (Pers) *M, (i)

where
Pers = the persuasiblity of the person receiving the
message, and

M = the intensity of the message in attitude units.

As one can see, the rate of change in A is not a function
of A itself, but only a function of the intensity of the message.
Stronger messages generate more change, a typical behavioristic

assumption. From a systems theory point of view, the behav-

2h

iorist's insistence upon control from the outside leads to a very
difficult time managing the attitude change process. For exam-
ple, at least in this model, attitudes do not change without
outside influences. Moreover, suppose a person's original atti-
tude was -4 on a scale running from -19 to +19. If another
individual attempts to get the person's attitude to +5, for
example, giving too intense a message may shove the attitude
beyond the target level of 5, so that one would have to reverse
the sign of the attitude message to bring lower the attitude.
Thus, for example, suppose someone was negative about the Presi-
dent of the United States, and his wife wanted to make him moder-
ately positive about the President. If by accident she gave the
President too strong an endorsement, according to the model, she
would have to reverse herself and tell her husband something
negative about he President to bring down her husband's attitude.

There is no evidence for this type of behavior in the literature.

A simple alternative to the behavioristic approach to con-
trol of attitudes is encapsulated in the following simple linear

differential equation:
°
A = {Pers)*(M-A) (2)

This equation represents a feedback approach to attitude
change. Hunter, Levine, and Sayres [27] have developed a more
complex model which includes the dynamic effects of both external

messages and internal messages that determine attitude change.
337
25

In the present simple version, the intensity of the attitude
itself becomes part of the general negative loop. The larger the
discrepency between the message and attitude, the more the atti-
tude will change toward the target represented by the message.

In this case, all the wife has to do to get her husband to a
target level of +5 on the scale is to continue to give +5 mes-
sages. She does not keep changing the intensity and direction of
the message, according to the model. Also her husband does not
have of keep oscillating back and forth in finding a stable
equilibrium point, while he tries to figure out why his wife is

so hot and cold about the President of the United states.

This, then in summary is an example of two different para-
digms concerning social behavior, The first attempts deal with
the complexities of behavior by outside influences, external
control, etc. The second paradign is much more theoretically
oriented, and, while not denying outside influences, stresses the
nature of internal structure determining the dynanmics of be-
havior. Psychology has gone far during the last 25 years, but it
is constrained by the limitations of its methods and viewpoints
concerning the dynamics of behavior. With interest in applied
areas increasing, psychologists may be ready to evaluate other
approaches to social problems besides extreme empiricism. An-

other, more balanced paradigm awaits review.

REFERENCES

1. Houston, J.P, Bee, H., and Rimm, D.C. Invitation to Psy-
chology, New York: Academic Press (1982).

2s.

12.

13.

14,

15,

16.

17.

26
Forrester, J.W. Principles of Systems, Cambridge, Mass.:
M.I.T. Press, (1968).

Hammond, K.R., McClelland, G.H., and Mumpower, J. Human
dudgment and Decision Making. New York: Praeger (1980).

Anderson, D.R., Sweeney, D.J., and Williams, T. An Intro:
duction to Management Science: Quantitative Approach to Deci-
sion Making, (2nd Ed.), St. Paul, Minn.: West Publishing

Co. .

Roberts, N., Andersen, D., Deal, R., Garet, M., and Shaffer,
W., Introduction’ to Computer Simulation: A System Dynamics
Modeling Approach, Reading. ng, Mass: Addison-Wesley (1983).

Skinner, B.F. Are Theories of Learning Necessary, In
Cumulative Record, B.F. Skinner ( Ed.), New York: Appleton-
Century—Crofts (1959).

Skinner, B.F. Why I am not a Cognitive Psychologist. In
Reflections on Behaviorism and Society. B.F. Skinner (Ed.),
Englewood Cliffs, N.J.: Prentice-Hall (1978).

Tharp, R.G., and Wetzel, R.G. Behavior Modification in the
Natural Environment. New York: Academic Press (1969).

Luenberger, D.G. Introduction of Dynamic Systems. New York:
John Wiley (1979)7

Watson, J.B. Behaviorism. New York: Norton (1925).

Breland, K. and Breland, M. misbehavior of Organisms,
American Psychologist, 16, 681-684 (1961).

Seligman, M., and Hager, J.L. Biological Boundaries of
Learning. New York: Appleton-Century-Crof >

Forrester, J.W., Mass, N.J., and Ryan, C.J, The System
Dynamics National Model: Understanding Socio-Economic Beha~

vior and Policy Alternatives, Technological Forecasting
and Social Change. 9, 51-68 (1976).

Wood, G. Fundementals of Psychological Research. (3rd Ed.)
Boston: Little Brown (1981).

Richardson, G. and Pugh, A. Introduction to System Dynamic
Modeling with DYNAMO. Cambridge, MasS.: M.I.T. Press, (1981).

Struening, E.L., and Guttentag, M. Handbook of Evaluation
Research, Vol. 1, Beverly Hills: Sage Publications (1975).

Forrester, J.W. Urban Dynamics, Cambridge, Mass. M.I.T.
Press (1969).
338

27

18. Nunnally, 3. Psychometric Theory. (2nd. Ed) New York:
necraw-aill Tie7}

19. Peterson, D.W. Statistical Tools for System Dynamics. In
Elements of System Dynamics Methods. J. Randers (Ed.)
Cambridge, Mass.: M.I.T. Press (1980).

20. Mass, N.J. and Senge, P.M. Alternative Tests of Selecting
Model Variables. In Elements of System Dynamics Methods. J.
Randers (Ed.) Cambridge, Mass.: M.I.T. Press (1980).

21. Levine, R.L. and Hunter, J.E, Regression Methodology: Cor-
relation, Meta-analysis, Confidence Intervals, and
Reliabillity. Journal of Leisure Research (In Press).

22. Hunter, J,E., Schmidt, F.L,, and Jackson, G.B. Meta-analysis:
Cumululating Research Findings Across Studies. Beverly atts?

Sage Publications (1982).

23. Blalock, H.M. Theory Construction. Englewood Cliffs, New
Jersey: Prentice-Hall (1969

24. Schmitt, N. and Bedeian, A.G. A Comparison of LISREL and
Two-Stage Least Squares Analysis of a Hypothesized Life-Job
Satisfaction Reciprocal Relationship. Journal of Applied
Psychology, 67, 886-816 (1982).

25. Kiesler, C.A., Collins, B.E., and Miller, N. Attitude
Change. New York: John Wiley (1969).

26. Hunter, J.E, and Cohen, S.H. Mathematical Models of Attitude
Change in the Passive Communication Context. Unpublished
Manuscript, Michigan State University (1972).

27. Hunter, J.E., Levine, R.L. and Sayers, S.E. Attitude Change
in Concept Hierarchies. In Mathematical Models of Attitude

Change and Grou} mamics, J.E. Hunter, J.E. Danes, and
S.H. Cohen (Eds.) New York: Academic Press (In Press).