JOH 0. BERNARD, M5
B. HENRY PRTE, ED.D.
TECHNICAL EDUCATION
MODELING AMD SIMULATION
TERS
CLEMSON UNIVERSITY - INDUSTRIAL EDUCATION
CLENSOM, SOUTH CAROLINA
THE INSTRUMENT
Abstract
Initial testing is now complete on the TEMS instrument, the
Technical Education Modeling and Simulation system. This research
in the Industrial Education Department at Clemson University is a
three phased project. The phases include developing an instrument
(TEMS) similar to DYNAMO II (Pugh, 1970) for the modeling of
socio-econ-educational systems, reducing world model concepts to
a regional model for the state of South Carolina, and integrating
technical education attributes and effects into the classical
capital sector for this regional model.
The TEMS system is developed in dBASE IV and ’C’. It has all
the model definition building features. and run characteristics of
DYNAMO II. Written for an IBM AT class of equipment, TEMS will
replicate the WORLD2 model (Forrester, 1971) results in 40
minutes for a 100 year run. TEMS supports both the real time
graphic mappings of selected variables and post analysis
graphics. It has both an integrated statistical interface to SPSS
statistics and a reporting system for model runs, definitions,
user created functions, and run time statistics.
Experimentation is in progress to calculate a CHAOS mapping
for the class of level variable equations. Using this Verhulst
equation mapping, TEMS should then dampen any wild ramping and
explosiveness for these selected variables during the simulation.
System Dynamics ‘90 91
Table of Contents
Introduction Pe er 91
Phase I - Instrument Development (TEMS)
Instrument Specifications © we He ee oe we 92
Development of the SmallTalk/V Prototype ... 93
SmallTalk/V Summary Conclusions ew we we we g 96
Development of the dBASE IV/c TEMS System oe 97
TEMS System Description i Wwe Sw ewe wy es 97
TEMS Results and Verification Pe ee ee 102
References ee 104
Introduction
The Industrial Education Department at Clemson University is
conducting research with technical education and its socio-
economic impact. This paper focuses on the first phase of a three
phase research project which is investigating the social and
economic impact of technical education on aggregate households
(people) and on the statewide economy.
The research question which we are investigating is: In the
regional socio-economic model for South Carolina there exists a
significant relationship between technical education and socio-
economic quality of life.
System dynamics modeling and simulation are needed in these
circumstances to study the effects of economic changeovers on the
quality of life for the individual and of the economic health of
the state. States in either growth mode or poor economic
conditions need tools to analyze their predicament, to forecast
their outcomes, and to help design their plans.
In order to successfully conduct this research we needed
both an instrument to describe our model and its intricate
interactions, and a regional socio-economic simulator for South
Carolina. Statistics alone (time series analysis, ANOVA,
ANCOVA . . .) is not sufficient to describe and map complex
variable interactions over time.
The major reasons for selecting system dynamics modeling and
simulation were because there existed proven instruments of
analysis, a bed of literature on the subject, and a track record
of research for this non-linear dynamics science. We are not
exactly creating data and doing experimental research, but
92 System Dynamics '90
rather, applying meta-analysis to extract known relationships and
quantifiable values for these relationships. We then integrate
these relationships and produce (hopefully) new research and new
insights into the technical education system and it relationship
with the industrial community.
The three phases of this research include developing an
instrument (TEMS) similar to DYNAMO II (Pugh, 1970) for the
modeling of socio-econ-educational systems, reducing World model
concepts to a regional model for the state of South Carolina, and
integrating technical education attributes and effects into the
classical capital sector of this regional model. As stated in
Dynamics of Growth in a Finite World, "DYNAMO, the computer
language most often used to express system dynamics models, is
not absolutely essential to the method - other flow diagrams and
computer languages could be used." (Meadows et al, 1974) We feel
that we have achieved an alternative with TEMS.
This paper summarizes our modeling and simulator instrument
development. We conclude this work with both the statement and
evidence to support that we now have a verifiable tool for model
development and research specifically in the area of technical
education.
Instrument Specifications
The instrument or tool for the research is very important to
us for several reasons. A proper instrument (the computer
implementation) provides a pathway for the verification of a
lumped model. The results of the instrument must be repeatable,
and reproducible using other established instruments.
DYNAMO II (Pugh, 1973) is an excellent continuous simulation
language for model building and simulation. We wanted feedback
characteristics of a continuous simulation language, the
programmatic mechanics of DYNAMO II, the functions found in large
scale social systems, and these additional modeling and
simulation system characteristics:
e@ Chaos analysis and interaction with the simulator.
e@ On-line integration with graphics and statistics for both
real-time and post processing.
e@ Interactive and available on multiple desktops with an
ease of use requiring minimal training.
e@ User interface which would stimulate usage by researchers
in education, sociology, and labor economics (not only
system dynamics or operations research).
e A system tool customized for general purpose socio-
economic modeling.
System Dynamics '90 93
It is essential that our modeling system deal with
disequilibrium and chaotic behavior. As Sterman points out: most
economic models are in a state of disequilibrium as a rule.
(Sterman, 1986) The science of chaos has advanced to the point
that new understandings of chaotic processes often give rise to
order. We want our modeling system to help identify and employ
dampening techniques to our non-linear social systems in various
states of turbulence. We must adapt these processes for the
"higher rule of order." (Killian, 1989).
After our initial Smalltalk/V development we realized that
real-time interaction rather than post processing interaction was
a time saver and gave us a more dynamic thinking process. Real-
time graphics with a built-in statistical interface can magnify
research productivity and validation.
We decided to develop our own continuous simulation system
to meet our requirements for a research tool. We started with a
Smalltalk/V (Digitalk Inc., 1988) development, abandoned it, and
redeveloped a dBASE IV (Ashton-Tate, 1988) system. The
description of the Smalltalk/V system follows.
Development of the SmallTalk/V Prototype
In this phase we had a goal to design and build a dynamic
interactive tool to build models and simulate the model running.
Object-Oriented Language Systems (OOPS) seemed like an excellent
place to start for doing model building and simulation. The
inherent concepts that OOPS performed operations and functions in
a fashion more closely to the human mind convinced us that this
was the design tool for the future.
The basic premise is that objects are created which have
attributes, operations (methods), and communicative messages
which they may pass to other objects or classes. The human mind
is said to cluster similarly. As system dynamics researchers are
aware, models resemble networks or webs with nodes linked
together, so that a programming language designed around this
concept has to be "natural."
As an example the equation to drive the integration of
quality of life (qol) through the first forward difference was:
| dial |
dial := ((dials at: #qol) reading).
(dials at: #qol) changeReading:
(dial + (dial * capInvToQualOfLife) +
(dial * pollToQualofLife) )!
In this code reading and changeReading are methods found in
subclasses created for this application. They read the screen
dial and update the screen dial for qol. The values of the
individual equation for gol are actually stored in the screen
94 System Dynamics '90
image. The needles are read from the screen image and converted
to current values. From a dynamics point of view, this concept
added new meaning to the phrase "what you see is what you get."
For a simple experimental prototype we used this simple
network relationship to tie some of the basic level variables of
WORLD2 (Forrester, 1971) together for testing purposes:
[_Tndustriai Education - Network Prototype! sd
Natural Resources Pollution
Quality of lite a
Population Capitalization
Figure 1
For prototypes we used this network with simple equations to
create a real system of data relationships which gave a
characterization of a lumped model (a somewhat arbitrary test
model which was not validated, only logically linked).
Using a Smalltalk/V sample program developed by Digitalk
Inc., we adapted it for our use. The original program was the
real-time interaction of someone with the controls and dashboard
gauges of an automobile. After modification, if the capital
investment rate was a little sluggish then we could do things
like step on the accelerator or throw the capital transmission
into a lower gear for more torque. If population was growing out.
of hand, we simply could step on the population brakes and slow
it down. If we wanted to play "what-if" games with the pollution
gauge, we could just change the pollution needle. The entire
model could even be thrown into reverse.
System Dynamics '90 95
We designed a simple simulator with the prototype model hard
wired into Smalltalk/V code so that we could get the feel for the
user interface and the speed/space performance of the computing
aspects of the system. Our prototype looked like this:
Quality of Lite Pollution
‘
fe[R Te Efe T2
S50. Transmission
Investnent
Figure 2
We included warning indicators for when natural resources
were too low or pollution was too high. Even here we began to set
the conditions for monitoring serious turbulence or chaos. The
researcher, in the next figure as an example, could reduce the
overload or force the model to continue in its turbulent state,
similar to what Ed Lorenz, Peter Scott, and Robert Shaw do with
meteorology and fluid models. (WGBH Educational Foundations,
1989)
The out of balance conditions which were hard-wired into the
Smalltalk/v simulator would be brought back to their initial
state values if the researcher selected "reduce" in the pop-up
window, as seen in the next figure. Alternatively, one could
select "continue" and then pick up the needle on the screen and
place it at any value.
The next logical development would have been to analyze the
change in variable levels or rates rather than using critical set
points for chaos. We incorporated this in the dBASE IV simulator,
described later.
96 System Dynamics '90
: G8
ate vie FIRE 12)
a Daen Pa Dad Transmission
Birth Rate °°? = investnent Rate.
Figure 3
SmallTalk/V Summary Conclusions
We abandoned this prototype for three primary reasons.
The small simulator ran very slowly, and as time went on its
computation for each year (with DT = 1) increased 20 percent. The
model would compute Year 1 in five seconds and Year 7 was taking
20 seconds. Even after adding floating point capability and
optimizing the space/speed options the simulation speed did not
improve very much.
Smalltalk/V is a more complex programming language than most
programmers care to admit. This language comes with over two
thousand methods and one hundred classes. Thirty five pages of
cross referencing these methods and classes only compounds
Digitalk’s observation that ". . . interim versions of methods
are an excellent way to ‘divide and conquer’ .. ." and that
programming tips are "survival tips." (Digitalk Inc., 1988).
Finally, the language may be too powerful. The concept of
networking helps us understand the complex, but the logic may not
be intuitive. Each time we made a simple change to the model,
like stepping on the capital investment accelerator, could be a
complete dissertation within itself. Simulators with this kind of
flexibility may only cause outside critics of this process to say
that the simulator is only an out-of-control results producer.
System Dynamics '90 97
These reasons should not be conceived, however, as
criticisms of Smalltalk/V because in large research projects with
many researchers and with INTEL 386/486 microcomputers with
megabytes of memory, this flexible language could be an excellent
research and development tool.
Development of the dBASE IV/C TEMS System
We settled on developing our second prototype in dBASE IV
and ‘C’ for several reasons. These software languages had
important ancillary support packages such as graphics and
statistics. When the computations were intense (floating point or
graphic computations) the load and call to an external program
was there to expedite processing. These software products both
have automatic application generators which significantly reduced
instrument development time. Finally, the simulation produces
voluminous data which needs to be cataloged, managed, and
evaluated. A data management system with a built in procedural
programming language was very desirable.
The prototype network illustrated in Figure 1 was again used
for initial testing, and after our successful testing, we then
built WORLD2 and simulated it producing statistically identical
results (described later).
TEMS System Description
Our simulator is classically constructed: the software
evaluates each node of a linked set of equations and pulls in
each input value from either other nodes, constants, or tables.
The equations are evaluated in such an order so as to avoid
simultaneous equation solving. The clock is then incremented and
the process repeated. (Aburdene, 1988)
To support the simulator we have a module to build the
equations and to set their initial values. Another module
establishes the criteria for the individual run (times, DT,
etc.). Because these modules integrate with a data management
system, the researcher can conveniently track different runs,
models, and results. Secondly, an integrated interface to both a
graphics and a statistics subsystem (SPSS) permit the results to
be analyzed immediately without any intermediate steps.
Our system is interactive with selection windows that move
the researcher from major screen to minor screen. By simply using
the arrow keys on the keyboard, a selection could be highlighted
and then picked by pressing the enter key. The Control Menu
carries the researcher to the other master menus for building the
model, setting up the runs, actually running the model, and
reviewing the results. In the review of the results, bar chart
and line graphics may be selected for four variables at a time to
be displayed on the screen, printer, or plotter. Also from this
submenu the statistics package is automatically set on the stats
98 System Dynamics ‘90
database. The researcher may start selecting data samples, the
type of statistics, and the output specifications.
This figure gives an overall insight into our model builder
and simulator:
B=. 84/88/98 Technical Education Simulation Control Menu
Build/vEdit Model Run Paraneters
Build/Edit Model Definition Equations
Simulation
Model Run Statistics
From InEdsin
Work with a model run.
Figure 4
Here one can see the Control Menu which permits the user to
select one of five activities: building the model, setting up a
run of the model, running the model, reviewing the results of the
run, and exiting the system.
The next submenu for building the model and setting up the
run for the model are similar because they permit the researcher
to add new elements, change elements, browse through all the
elements, reorganize the elements, and exit the submenu.
The Definition submenu, Figure 5, seen on the next page
permits the researcher to develop a set of equations, similar to
DYNAMO II, to describe the model. Because this is a specialized
simulator, most of the mathematical equations typically found in
a social simulator are predefined as functions. The researcher
only has to name the equation ( also specifying an abbreviation)
and give it a relative number and equation type: level,
auxiliary, or rate. Further, the other nodes of the model which
input to the current node are specified as well as any constants
or table values to feed the predefined mathematical equations.
After the model has been defined then TEMS sorts the equations to
ferret out simultaneous equation relationships and reassigns the
relative equation numbers.
Other functions may be built by the researcher to define
relationships which are not routinely provided in the simulator.
If the equation under development is a level equation then the
System Dynamics ‘90 99
Chaos Function may be set to "Y" to indicate to the simulator
that the results of this equation should be monitored for
dampening in the event that this level is in a wild collapse or
ramping mode.
Records GoTo Exit font
Model Definition Equations
DEF_NO q TYPE VAR_NO q
UAR_NAME ZUR ABBY ij FUNC_NO
INIT_VAL 1658008088 288880]
NLIML1 By wiINz 7 HIND N_IN_4
N_INS 8 wine
CHAOS
CHAOS_FN
CONSTANTS AND TABLE
Heat [csNus\nodeI\ROD DERN Tec 0G Fite
ENTER THE MODEL DEFINITION NUMBER.
Figure 5
The model building and run options have a common attribute:
Verhulst equation tracking. We have incorporated an experimental
mechanism to control and suppress model collapse modes.
As seen in the Model Definition and Run submenus (Figures 5
and 6), Chaos can be selected at both menus. When the model is
running and Chaos = ’Y’ for both the run and the specific
equation then the simulator checks the five previous values to
see if there exists a rapid increase or decrease in the change of
the level, and if the values are moving away from equilibriun.
The Verhulst equation:
Your = Ye(2 + R) - RY,2,
where ¥ is the level variable under consideration and R is
the rate of change. The first portion of the right hand side of
the equation is a classic first forward difference, but the
second portion of the equation, - RY,“, dampens large changes.
On any level equation (population as an example) created, —
the researcher may elect to automatically have the calculated
values for that variable tracked against the known equilibrium
paths of the Verhulst equation.
The Run submenu on the next page, Figure 6, permits the
researcher to specify which model to run, the purpose of the run,
100 System Dynamics '90
and comments about the run. The beginning and ending times are
chosen, as well as the DT interval, the date, time, and the type
of graphics (line or bar charts) dynamically displayed during the
run. Four variables can be graphically monitored.
Records Go To Exit
Model Run Information
RUN_NO
DEF_NO
DESC This run is with iv al values not e zero and big DT = 8.
DATE
TIME
CLOCK_NO
LENGTH
DT
DATA_PUB BEE
Graph Graph Variables 1 2 4
CHAOS
RESULTS Stat data shows Model within P=4H44 OF MIT WORLDZ model.
ICNuswnodeINMOD MUN Tec 374 Tite
¥ OR N TO CHECK AGAINST CHAOS FUNCTIONS IN EACH CLOCK TICK...
Figure 6
We can see from the next figure the Verhulst equation
plotted without transients paths removed.
Chaos Map - POP
tes 1 vith Transients
ow
°
7
acunanrncem
o.0 0.4 o8 1.28 1.60 2.0 2.4 2.8 3.2
RATE VARIABLE
Figure 7
System Dynamics '90 101
Transients are defined as an arithmetic series of points
which are plotted at every R. At any R one can have an infinite
number of points which represent the current value of the
population. The population Y¥ in this plot goes from .1 to a
stable equilibrium value of 1. As seen on the previous figure
some points may actually map above 1. (Dewdney, 1987)
To paraphrase Pierre-Francois Verhulst, an eighteenth
century Belgian researcher in population growth, "An absolute
shortage of food when there are lots of rabbits enables different
control mechanisms in nature than when there are just a few
rabbits." It is interesting to note from the transient equation
graph (Figure 7) that along several transient paths both R
decline and R incline can produce an extinction of a species.
The equation further describes the effect of a large
population on equilibrium: the brakes are applied more
significantly at higher population levels. As seen in Figure 8,
an R less than 2.0 and a small initial population evolves to a
stable population of 1. What follows are oscillation paths
(attractors) where a population may map into an equilibrium path.
Figure 8 shows the plotting of the equation with the first
6000 transients removed. Now we can clearly see the equilibrium
path and the attractors (alternate equilibrium paths). These
temporary equilibrium values, have been logged into an internal
table in the simulator. The sea of chaos is evident at R > 2.6,
and once the level variable collapses in this area, there
presently is no recovery in TEMS.
Chaos Map - POP
ma 9 Transients Surpressed - Zoomed=in
»
zo
rs
senwmenrncem
a
1.9 2.0 2.14 2.2 2.3 2.4 2.5 2.6 8.7 2.8 2.9 3.0
RATE VARIABLE
Figure 8
102 System Dynamics '90
Again, if there has been a sudden change, the simulator
looks for the nearest equilibrium or attractor point in the
Verhulst table and forces the new current value in that
direction. The current value is "dampened" or "clipped" to
emulate exogenous control found in physical and social phenomena.
We have a screen display of the simulator while running
WORLD2. Back in the Run submenu, Figure 6, there is a factor
called the Data Publish factor which permits one to specify the
interval that the results are to be put in the stats database.
When they are published, the screen graphics update to reflect
the last five data values for the four selected variables.
- TENS Graphs-
PoP cr
Figure 9
Finally in addition to the actual running of the model there
is the review of the results by examining individual variables or
by browsing all variables or by printing the entire set of run
data or by graphing it. The researcher may look at four different
variables at a time on the screen or pass them all to the printer
or plotter.
TEMS Results and Verification
Statistically our simulator produces repeatable and
reproducible results when compared to another verifiable
simulator using a validated model, WORLD2. Additionally, we have
System Dynamics '90 103
conducted the necessary correctness proofs for the program
modules, reprogrammed critical modules for a comparison of
results, and traced input-output relationships as suggested in
the literature. (Sargent, 1984)
During our WORLD2 simulation run, using TEMS we produced a
data set of results which we used to compare to the DYNAMO II
WORLD2 results. A matching of the two data sets revealed:
sen. 84/88/98 Review of Model Run Statistics
Paired samples t-test: INEDCIAF
WLDZCIAF
Variable Nunber Standard Standard
of Cases Nean Deviation Error
INEDCIAF 11 2582 043 813
WLD2C IAF 14 2508 043 813
@ifference) Standard Standard
Mean - Deviation Error
2-Tail t Degrees of 2-Tail
Corr. Prob. Value Freedom Prob.
1.608 808 1.49 18 +167
Figure 10
The five levels and the QL auxiliary values from WORLD2 were
t tested using pairing of the values in DYNAMO II and TEMS. As
seen in Figure 11 above, a high correlation value indicated a
near perfect degree of association and the t value was 1.49. All
equation variables were equally matched between the two data
sets. We can accept the null hypothesis that there exists no
significant difference between the values of each data set.
(Best, 1988)
The very small differences were attributed to differences
in the floating point chip and floating point routines of the two
different computers. We are able to maintain an 18 significant
digit precision with dBASE IV so we did not attribute any slight
differences to numeric precision.
Authors’ Note
Program listings of the Smalltalk/V prototype described in
this paper will be made available at the conference for those who
wish to have a copy for further development.
104
System Dynamics '90
1.
11.
12.
13.
14.
References
Aburdene, Maurice F. Computer Simulation of Dynamic
Systems. Dubuque, Iowa: Wm. C. Brown Publishers, 1988.
Ashton-Tate Corporation. dBASE IV Reference Manual.
Torrance, California, 1988.
Best, John W. and Kahn, James K. Research in Education.
Englewood Cliffs: Prentice-Hall, Sixth Edition, 1988.
Bronson, Richard. "Computer Simulation: What it is and
how it is done." Byte Magazine, (March, 1984): 95-102.
Dewdney, A. K. "Probing the strange attractions of
chaos." Scientific American 257: 108-111, July, 1987.
Digitalk Inc. Smalltalk/V Programmers Handbook. Los
Angeles, May, 1988.
Engle, Charles B. "Chaos on computers." Science (D 1,
1986): 246:1172. :
Forrester, Jay W. World Dynamics. Cambridge: Wright-
Allen Press, Inc., 1971.
Killian, Anita M. "Playing Dice with the Solar System."
(an MIT treatise on chaos), Sky & Telescope, Vol. 78,
No. 2, Belmont, Massachusetts: Sky Publishing
Corporation, August, 1989.
Meadows, Dennis. L. et al. Dynamics of Growth in a
Finite World. Cambridge: Wright-Allen Press, Inc., 1974.
Pugh, Alexander L. DYNAMO II User’s Manual. Cambridge:
The MIT Press, Fourth Edition, 1973.
Sargent, Robert G. "Verification and Validation of
Simulation Models." 1984 Winter Simulation Conference
Proceedings (1984): 115-121.
Sterman, John D. "The Economic Long Wave: Theory and
Evidence." Systems Dynamics Review Vol. 2, No. 2: 88 -
125, Summer, 1986.
WGBH Educational Foundation. "The Strange New Science of
Chaos." #1603, New York: Journal Graphics Inc., 1989.
CC BY-NC-SA 4.0