MicroWorlds and Evolutionary Economics
When | hear, | forget. When | see, | remember. When I do, | understand.1
Michael J. Radzicki
Department of Social Science & Policy Studies
Worcester Polytechnic institute
Worcester, Massachusetts 01609-2280, USA
Abstract
This paper discusses the use of microworlds in evolutionary economics. A macroeconomic
microworld is presented as an example. :
Introduction
Since the time of Veblen, evolutionary economists have viewed the world through a
dynamic and holistic lens and spoken-out against the static and atomistic theory that dominates
economic thinking. In addition, they have championed the idea that economists should spend
their time determining how various policy changes will most likely shape emerging
socioeconomic systems and how the behavior of these systems can be altered in socially
desirable ways (Gruchy 1987). Although these views are appealing and arguably correct, as a
practical matter they are somewhat difficult to implement. More precisely, a large and diverse
body of literature being developed in disciplines such as control theory and psychology shows
that human beings find it very difficult to accurately trace through the evolutionary implications
of proposed policy changes in complex socioeconomic systems.? This difficulty arises
because the behavior of these systems is a direct result of their structures or patterns, which
consist of stock and flow networks embedded in interacting nests of positive and negative
feedback loops, that have themselves been joined by nonlinear couplings. As a
consequence, the circular and cumulative behavior of these systems is often quite different
from what human intuition suggests it will be.3
In recent years Radzicki (1988, 1990a, 1990b) has argued that system dynamics computer
simulation modeling can be combined with evolutionary economic theory to construct dynamic
institutionalist pattern models. The main benefits of these models would be their preciseness,
explicitness, and ability to be simulated. Computer simulation, of course, would make it
possible for the dynamic implications of proposed policy changes to be observed.
Recent,fesearch in the field of system dynamics has shown that turning system dynamics
models into games or “microworlds” can, under certain circumstances, greatly enhance the
insights received by those that play (and develop) them.4 The purpose of this paper,
therefore, is to discuss some of the issues surrounding the creation and use of system
dynamics microworlds in evolutionary economics. A dynamic fiscal policy game is presented as
a means of illustrating the arguments put forth.
MicroWorlds
Although gaming has essentially existed for as long as the computer has, it has only
recently been integrated into microworlds or “learning laboratories” that run on
microcomputers. A microworld is a computerized environment that enables and encourages a
person to explore, and experiment with, a dynamical feedback system. The goals of a
microworld are to help a person discover the ways in which a system's feedback structure or
pattern relates to its behavior and how individual decisions can affect its performance (Diehl!
1988; Simons 1990). Through experimentation, a person working within a microworld is able
experience behaviors that take years to unfold in the real system being mimicked and/or that
have not yet actually occurred because the real system has progressed down a different
evolutionary path. If the microworld has been properly designed, moreover, this freedom to
explore facilitates user self-discovery or “learner directed learning” (Bakken 1989; Peterson
- 533 -
1990; Simons 1990). In terms of evolutionary economics this means that, in principle, an
evolutionary theorist or student could explore the dynamics of a theory or a pattern model
within a microworld and uncover new insights into its structure and behavior.
Broadly stated, there are two types of microworlds that are based on system dynamics
models: computer games and board games. Computer games can be either single player or
multi-player, while board games are almost always multi-player. Multi-player computer games
tend to work best on a computer network, with each player working at a separate machine.
Board games are usually played with all the participants seated around a table.
Computer games use software that simulates a system dynamics model for a finite period
of time (usually one period), pauses, receives and displays feedback from the model, invites a
player to make a decision(s) based on the feedback, and then simulates the model for another
finite period of time. A central feature of this type of microworld is that it keeps track of the
decisions made by a player for later reflection and analysis. One of its main advantages is that it
can usually be played many times by a player, quickly, and without coordination by a game
moderator. The result is that experimentation and exploration are easily undertaken.
Depending on their design, board games based on system dynamics models can be
played either entirely without the aid of a computer, or with the aid of a computer that runs off to
the side and out of the players’ central stream of consciousness. In the former case, the
players themselves simulate the model through time with the aid of the game board and other
accounting devices such as game pieces and paper and pencils. In the later case, a computer
simulates the model and keeps track of the accounting calculations while a game moderator
relays information between it, the players, and the game board. In this type of Microworld, the
game board functions primarily as a conceptualization device. Meadows (1989) argues
convincingly that board games are superior to computer games when multiple players are
involved because the round-table atmosphere promotes a higher level of discussion and
interaction. The drawback to this type of microworld is that its format makes exploration and
experimentation somewhat difficult.
Design and Implementation
Currently, a literature is emerging that describes how best to turn system dynamics models
into computerized microworlds (Morecroft 1988; Graham et al. 1989; Kim 1989; Meadows
1989; Andersen et al. 1990; Peterson 1990; Saeed 1990; Simons 1990). Although there is
still much to be learned, the keys to the creation of a successful microworid -- that is, one that
facilitates learner directed learning -- appear to be proper design and proper implementation.
The former involves the physical construction of the microworld and the determination of the
rules of play and experimentation, while the latter involves the determination of the information
a person will receive before, during, and after play. In each case, the overriding goal is to
minimize the chance that a player will treat the microworld as a video game or toy, and maximize
the chance that it will stimulate his or her curiosity and motivate him or her to think critically.
In designing a microworld, it is helpful to remember that human beings make virtually all
decisions via bounded rational rules (Kleinmuntz 1985; Morecroft 1988; Sterman 1989a,
1989b). As a result, care must be taken to ensure that a player is not bombarded with
information and asked to make too many decisions during a round of play. Both Andersen et al.
(1990) and Peterson (1990) present guidelines involving topics such as screen design that
are aimed at preventing this from occurring. It is also useful to remember that a successful
microworld can only be created from a good system dynamics model and that not all good
system dynamics models can be tumed into successful microworids. In particular, studies (and
common sense) indicate that the perspective from which the underlying system dynamics
model is built (e.g., the level of aggregation chosen; the selection of the particular decision
processes portrayed) will have a substantial effect on the success of the microworld built
around it (Andersen et al. 1990).
In terms of the proper implementation of a microworld, Bakken (1989) has accumulated
evidence indicating that the ability of a person to control a system being simulated within a
microworld is strongly related to their understanding of its structure (feedback loops, delays,
nonlinear couplings, et cetera -- i.e., the very things that cause evolutionary behavior).5
Moreover, both system dynamics studies (Diehl 1988; Bakken 1989; Sterman 1989a, 1989b)
and traditional psychological studies (e.g., Brehmer 1987) show that outcome feedback, or
information on the dynamic implications of decisions made in the game and received during
play, is not effective in helping people learn about system structure. All of this implies that a
person should receive training on the relationship between a system's structure and behavior
before beginning to use a microworld; should have access to information on system structure
and behavior during play; and should be debriefed on the subject vis-d-vis their experience
after play. Kim (1989) describes the improvement in employee learning that took place in a
corporate setting, only after a program of briefing and debriefing was added to the gaming
experience. Graham et al. (1989) provide a comprehensive overview of successful and
unsuccessful strategies for using microworlds in teaching and research settings.
Fiscal Policy Game
In order to illustrate many of the concepts discussed above, an actual microworld created
by the author will be described. It is an example of how gaming can be used to explore the
dynamics of macroeconomic theory and fiscal policy. A similar microworid, based on a case
study of the rise and fall of People’s Express Airlines (Sterman 1988), is an example of how
gaming can be used to explore the dynamics of what is essentially an institutionalist pattern
model.
The goals of the fiscal policy microworld are to: 1) provide players with an environment for
exploring the dynamics of Keynesian macroeconomic theory and the multiplier and accelerator
processes; 2) emphasize the interactions in the world economy; 3) reiterate the necessity of
making choices in economics; and 4) drive home the point that, due to feedback effects, the
decisions people make both shape their emerging world and (often) come back to haunt them.
Design
The system dynamics model that underlies the fiscal policy microworld is based on a
synthesis of Keynesian macroeconomic theory, Milton Friedman’s permanent income
hypothesis, and the Samuelson-Hicks multiplier-accelerator model. More specifically, it follows
the directions originally taken by Low (1980) and Forrester (1982) by recasting the Samuelson-
Hicks muttiplier-accelerator model from a system dynamics perspective and extending it with an
international sector, an endogenous aggregate demand forecasting sector, an endogenous
permanent income sector, and a relatively sophisticated investment sector. In addition, it has
been replicated and linked to its copy via international trade so that secondary feedback
effects in each economy are possible, and it corrects some of the well-known problems
associated with traditional Keynesian theory.”
Figure 1 presents the major feedback loop structure of one of the economies in the fiscal
policy microworld. Inspection of the Figure reveals that six principle feedback loops -- two
positive and four negative -- govern the macroeconomic behavior of the sector. Positive
feedback loops generate self-reinforcing behavior and are responsible for the exponential
growth or decline of systems. The major positive loops in the sector are identified by the large
plus signs placed within parentheses in Figure 1. Negative feedback loops generate
counteracting or goal-seeking behavior and are responsible for stabilizing (or trying to stabilize)
systems. The major negative loops in the sector are identified by the large minus signs placed
within parentheses in Figure 1. The small plus (minus) signs next to the arrows that comprise
the various feedback loops in the Figure indicate places where a change in the variable at the
back of the arrow causes, ceteris paribus, a change in the variable at the front of the arrow in
the same (opposite) direction.8
In terms of the representation of specific macroeconomic processes with feedback loops,
the dynamical behavior of both the multiplier process and the accelerator process is strongly
influenced by positive feedback. In particular, the Keynesian multiplier is represented in Figure
1 by a positive loop connecting Permanent Income, Consumption, National Income, and
Disposable Income, and the accelerator is represented, in part, by a positive loop connecting
National Income, Expected National Income, and investment.
=~ 535 =
The four negative loops shown in Figure 1 depict either leakages from the macroeconomy
or explicit goal-seeking behavior. The first negative loop captures the leakage due to spending
on imports and is a causal chain linking Permanent Income, Imports, National Income, and
Disposable Income.9 The second negative loop portrays the leakage or “fiscal drag” that arises
from the collection of income taxes by the government and is a causal chain linking Permanent
Income, Consumption, National Income, Tax Revenue, and Disposable Income. The third
negative loop represents the leakage due to the wearing-out of the economy's capital stock
and is a causal chain linking Capital and Depreciation. Finally, the fourth negative feedback
loop represents goal-seeking behavior within the accelerator process and is a causal chain
linking Investment and the Capital stock. The goal-seeking behavior takes place when the
model compares the economy’s Capital stock to its Desired Capital stock and adjusts
Investment spending accordingly. Macroeconomic instabilities can occur when Investment falls
as the economy's Capital stock nears its desired level, and rises as it moves away from it.
Figure 2 depicts the game screen for the fiscal policy microworld. Inspection of the Figure
reveals that the screen is divided into four sections: a “Decisions” section, “Reports” section,
“Graphs” section, and a section reserved for the viewing of reports and graphs. The
“Decisions” section is the place where a player enters his or her policy changes in response to
feedback from the model. Four fiscal policy tools are available for player manipulation:
Government Spending, Lump Sum (Non-Income) Taxes, the Marginal (Income) Tax Rate, and
Transter Payments. The “Reports” section is the place where a player can call-up information
on the game (“Created By;” “Instructions’) or its underlying feedback structure (‘Major Causal
Loops”; “Model Overview").1° it is also the place where he or she can call‘up the “Major
Macroeconomic Flows” report shown in Figure 2. This report displays the current values of the
major national income and product accounting flows in the economy and thus changes as the
model is advanced through time. The “Graphs” section is the place where a player can call-up a
time series graph of any of the major variables in the economy he or she is trying to control.
Whereas the “Major Macroeconomic Flows” report displays only the current values of the
selected variables, each of the “Graphs” records and displays the historical values of the
variables produced during the game up to the current point in time. Indeed, it not only does
this for the model’s endogenous variables, but also for its four exogenous ones -- i.e., the
Policy levers trom the “Decisions” section. Thus, a player can go back at any time and examine
his or her policy changes vis-a-vis the behavior of the model.
As shown in Figure 3, a person plays the game by intervening in the economy with the
fiscal policy tools available in the “Decisions” section of the microworld. The game is started in
equilibrium with the trade and federal deficits equal to zero. In year three, however, the
government of the second economy cuts its spending by $50 (billion). This shock unleashes
an oscillation with a downward trend in both economies. The player, of course, is instructed to
use the available fiscal policy tools to try and stop this undesirable behavior. As there are no
price levels or monetary sectors in the model, the “down side” to a strategy of continuously
Pumping-up the economy with spending increases and tax cuts is the expansion of the twin
deficits. Thus, the player is also instructed to try to keep the deficits at a “reasonable” level.
One last complication to all of this is the accelerator. Even without the interactions with the
second economy, the first economy will stall and decline whenever its capital stock reaches its
desired level. The.player, therefore, has to battle a second set of forces working to create an
undesirable situation in the economy. Figure 4 presents the graph of National Income for a
typical play of the game, Figure 5 the corresponding Federal Deficit plot, and Figure 6 the
graph of the player's choices for Government Spending during the simulation.
Implementation
As the above description indicates, a player must keep track of many things while playing
the fiscal policy game. In fact, one might even hypothesize that the microworld provides a
player with so many policy options and so much feedback that the assembly of a well thought
out strategy, where effects are systematically related to causes, is impeded. Experience with
the microworld has shown that players do “best,” in the sense of learning about the impact of
various fiscal policies on multiplier-accelerator dyads, if they are first taught the underlying
macroeconomic theory in a traditional manner, 11 and then with causal loop diagrams and/or a
traditional system dynamics model. Moreover, learning appears to be enhanced if players have
access to the model's feedback loop structure during play, and if they are debriefed with a
discussion of the theory vis-4-vis causal loop diagrams after they play. These effects are
identical to those described by Kim (1989). It should also be noted that the “Major Causal
Loops” report is specifically included in the game so that players can refer to it when making
decisions between rounds. A quick glance at Figure 1 will reveal that the places where the four
fiscal policy levers impact the system are clearly marked for the player with rounded rectangles.
One last result that is worth noting at this time is that, generally speaking, it appears that
players do a significantly better job controlling the system if they play the game more than once
or twice -- a luxury actual policy makers do not have. Although this result is probably not
surprising, it does point out the value of exploration and experimentation. It also implicitly
illustrates the evolutionary nature of the simulations as, from play to play, the economy travels
down different dynamical paths.
Future Research
At this time it appears that microworlds have two main uses in evolutionary economics: 1) to
help students learn existing economic theory (traditional or otherwise) in a superior way; and 2)
to aid in the process of developing new theories and pattern models. Although much of the
discussion in this paper was implicitly centered around the first use, the second may ultimately
be more beneficial to evolutionary economists. This is because, in principle, microworlds can
enable evolutionary economists to discover more about their new theories and patter models
than would otherwise be the case. The result will be better models and theories and hence
Policies that are more effective in shaping emerging economic systems in socially desirable
ways.
Endnotes
1. Chinese proverb suggested by Meadows (1989, p. 636) as appropriate for viewing the
contribution gaming can make towards helping people understand the dynamics of
complex systems.
2. See Diehl (1988), Sterman (1989a; 1989b), and Bakken (1989), for overviews of the
literature relating to human errors in dynamic decision making.
3. This argument is implicitly advanced in the field of macroeconomics in the traditional
debate over activist stabilization policies. Those that favor nonactivism cite the difficulty
of detefmining the correct timing and magnitude of fiscal and monetary interventions
due to the lags that exist between problem recognition and policy impact.
4. Other terms that are used more or less interchangeably with “microworld” are “learning
laboratories,” “learning environments,” and “light simulators.”
5. — To be fair, this contradicts the findings of other researchers. See for example Broadbent
and Aston (1978).
6. See Sterman and Meadows (1985) for an example of a.microworld used to explore a
theory of the economic long wave and Saeed (1990) for an example of a microworld
used to explore economic development theory.
7. For example, its incorrect specification of stocks and flows. Excellent discussions of the
problems associated with traditional Keynesian theory are contained in Low (1980) and
Solow (1984).
8. More precisely, a small plus sign next to an arrow indicates:
2 Variable at Head of Arrow 0
2 Variable at Tail of Arrow >
and a small minus sign next to an arrow indicates:
2 Variable at Head of Arrow 0
3 Variable at Tail of Arrow <
9. Unlike traditional Keynesian theory, imports in the fiscal policy microworld are a function
of permanent income, rather than national income. This implies that, in each economy, all
capital goods are produced domestically.
10. The “Major Causal Loops” report is shown in Figure 1. The “Model Overview” report is
shown in Figure 3.
11. For example, by having them read the description of the interaction between the
multiplier and accelerator provided by Samuelson and Nordhaus (1989, Chapter 10).
Experience has also shown that having players examine the original multiplier-
accelerator model is not at all useful in helping them to intuitively understand its structure
and behavior, let alone in helping them to play the game well.
References
Andersen, David F., Ik Jae Chung, George P. Richardson, and Thomas R. Stewart. 1990.
“Issues in Designing Interactive Games based on System Dynamics Models.” pp. 31-45.
In: David. F. Andersen, oolge P. nicherdson, and John D. Sterman, as 7 - Syslem
Volume . Available trom the System Dynamics Society. Massachusetts institute of
Technology.
Bakken, Bent E. 1989. “Learning in Dynamic Simulation Games; Using Performance as a
Measure.” pp. 309-316. lin: Peter M. Milling and Erich O. K. Zahn, eds. Computer-Based
Management of Complex Systems. Berlin: Springer-Verlag. g
Brehmer, Berndt. 1987. “Systems Design and the Psychology of Complex Systems.” pp. 21-
32. In: Empirical Foundations of information and Software Science Ill. Jens Rasmussen
and Pranas Zunde, eds. New York: Plenum Publishing Company.
Broadbent, Donald E. and Benjamin Aston. 1978. “Human Control of a Simulated Economic
System.” Ergonomics 21: 1035-1043.
Diehl, Ernst W. 1988. “Participatory Simulations as Training Tools -- A Study Based on the
Market Growth Model.” pp. 52-65. In: Jack B. Homer and Andrew Ford, eds. Proceedings
of the 1988 International System Dynamics Conference. Available from the System
Dynamics Society. Massachusetts Institute of Technology.
Forrester, Nathan. B. 1982. A Dynamic Synthesis of Basic Macroeconomic Theory:
Implications for Stabilization Policy Analysis. Unpublished Ph.D. Dissertation.
Massachusetts Institute of Technology.
Graham, Alan K., Peter M. Senge, John D. Sterman, and John D. W. Morecroft. 1989.
“Computer-Based Case Studies in Management Education and Research.” pp. 317-326.
In: Peter M. Milling and Erich O. K. Zahn, eds. Computer-Based Management of Complex
Systems. Berlin: Springer-Verlag.
Gruchy, Allan G. 1987.
Institutional Economics. Westport, CT: Greenwood Press.
Kim, Daniel H. 1989. “Leaming Laboratories: Designing a Reflective Learning Environment.”
pp. 327-334. In: Peter M. Milling and Erich O. K. Zahn, eds. Computer-Based
Management of Complex Systems. Berlin: Springer-Verlag.
Kleinmuntz, Donald N. 1985. “Cognitive Heuristics and Feedback in a Dynamic Decision
Environment.” Management Science 31: 680-702.
Low, Gilbert W. 1980. “The Multiplier-Accelerator Model of Business Cycles Interpreted from a
System Dynamics Perspective.” pp. 107-124, In: Augusto A. Legasto, Jay W. Forrester,
and James M. Lyneis, eds. System Dynamics. Amsterdam: North Holland.
Meadows, Dennis L. 1989. “Gaming to Implement System Dynamics Models.” pp. 635-640. In:
Peter M. Milling and Erich O. K. Zahn, eds. Computer-Based Management of Complex
Systems. Berlin: Springer-Verlag.
Morecroft, John D. W. 1988. “System Dynamics and Microworlds for Policymakers.” European
Journal of Operations Research 35: 301-320.
Peterson, Steven O. 1990. “Designing Learning Environments.” pp. 852-862. In: David F.
- 538 -
Andersen, Geome: P. Richardson, ar dohn D. Sterman, eds. System Dynamics ‘90:
ence. Volume III. Available
from the System ‘Dynamics Sociely. ‘Massachusetts Institute of Technology.
Radzicki, Michael J. 1990a. “Methodologia Oeconomiae et Systematis Dynamis.” System
i 6(2): 123-147. 1990.
Radzicki, Michael J. 1990b. “Institutional Dynamics, Deterministic Chaos, and Self-Organizing
Systems.” Journal of Economic Issues 24(1): 57-102.
Radzicki, Michael J. 1988. “Institutional Dynamics: An Extension of the Institutionalist
Approach to Socioeconomic Analysis.” Journal of Economic Issues 22(3): 633-666.
Saeed, Khalid. 1990. “Bringing Practicum to Theory-Based Social Science Disciplines: An
illustration with a User-Friendly Simulation Laboratory on Issues of Economic
Benen ene: Ppp. 963+ 971. In: David F. Andersen, Georne F Bietarcson, and John D.
sterman, eds.
Dynamics Conference. Volume Il. Available from the System Dynamics Society.
Massachusetts Institute of Technology.
Samuelson, Paul A. and William D. Nordhaus. 1989. Macroeconomics. 13th edition. New York:
McGraw-Hill.
Simons, Kenneth L. 1990. “New Technologies in Simulation Games.” pp. 1047-1059. In:
David F. Andersen, Sore f Headly and dohn D. Sterman, eds.
ence. Volume Ill.
Available from the System Dyan Society. Massachusetts Institute of Technology.
Solow, Robert M. 1984. “Mr. Hicks and the Classics.” Oxford Economic Papers 36: 13-25.
Sterman, ohn Dd, 13888 “Misperceptions of Feedback in Dynamic Decision Making.”
na sesses 43(3): 301-335.
Sterman, John 0. 1989. “Modeling Managerial Behavior: Misperceptions of Feedback in a
Dynamic Decision Making Experiment.” Management Science 35(3): 321-339.
Sterman, John D. 1988. “People’s Express Management Flight Simulator.” Software and
os available from the System Dynamics Group. Massachusetts Institute of
‘echnology
Sterman, John D. and Dennis L. Meadows. 1985. “STRATEGEM-2: A Microcomputer
Simulation Game of the Kondratiev Cycle.” Simulation and Games 16(2): 174-202.
Consumption +
Imports
(@) Income
ee f
Desired +
Capital
+
Capital-Output
Ratio
Figure 1: Major Feedback Loop Structure of One Economy in the Fiscal Policy Game
ne
= === Major Macroeconomic Flows === E)|
Fiscal Palicy Ge
Ep] Deciciene
Gevernment6per 200)
National rrerne = Consumption + [Ell] pagonragen =e
500 a8 ‘TrancforPaymer 100]
:
Investment Gov't6pending
a + ona + i
Beets . rnports |
« 48 :
i
Trade Deficit
0
Dynamic Dynamic
Keynesian Keynesian
Economy 1 Economy 2
The Beonmy
You Goniiol
Figure 3: Overview of the Fiscal Policy Game
— =
National_Incame CE Fiscal Policy 6¢
1257.21 .
1792.13 7
1627.53 4
1462.45
1297.27 4
1182.11 4
966.95 4
Ol 791 4
636 631 4
471 472+
Nattanatincamo |
arginal_Tax Rate
aD.
° ae
Figure 4: National Income Plot After a Typical Play of the Fiscal Policy Game
ys
ca. GE
“SEE
SSS Federal_Deficit Fiscal Policy Ge
DD FederalDefiit Decisions
|®GoverreneniSper 4n0|
*LumpsumTax 100)
PMarghallaxRats a
|e TrancforPaymer 300
4105164
364.903 Step) Exploration End
Reporte
319.29 cried ba
272.4774 structions
Major Causal Loops
220.064 4 Major Maorceaanomio Flaws
Model Overview
102.4524
138.8394
912256 4
4551294
o4
a 24
Figure 5: Federal Deficit Plot After a Typical Play of the Fiscal Policy Game
ELD === Government_Spending
© GavarnmmontSpending
4004
317.778
a554564
3353334
a1 Ail4
200 409 7
266 467 +
244 4444
222.222
200 -B
- 542 -
