Making System Dynamics Cool III:
New Hot Teaching & Testing Cases
Erik Pruyt* — E-mail: e.pruyt@tudelft .nl
August 14, 2011
Abstract
This follow-up paper presents seven actual cases for testing and teaching System Dynamics
developed and used between January 2010 and January 2011 for one of the largest System Dy-
namics courses (250+ students per year) at Delft University of Technology in the Netherlands.
The cases presented in this paper range from short to long and can be used for teaching and
ing introductory/intermediate System Dynamics courses at university level as well as for
If study. Additionally, the use of Multiple Choice questions for testing System Dynamics is
presented and discussed.
Keywords: System Dynamics Education, Case-Based Teaching and Testing, Fish-
eries, Radicalization, Energy Transition, Real Estate Boom and Bust, Mineral/Metal
Scarcity, Slow Students, Ecosystem Management
1 Introduction
Many ‘hot’ ~i.e. current real-world- teaching and testing cases are used each year in the Introduc-
tory System Dynamics courses taught at Delft University of Technology’s Faculty of Technology,
Policy and Management. Hot cases were found to be excellent tools for motivating students, for
illustrating the relevance of SD modeling for real world problem solving, and for showing the way
SD could be applied to real world cases. For more (information on) ‘hot’ teaching/testing «
see (Pruyt 2009c) and (Pruyt 2010a). Most of the publicly available cases focus on basic to inter-
mediate SD model building skills, and —to a lesser extent~ on basic model use and communication
skills. Students need some modeling experience/practice in order to be able to deal with these
s but not too much in order to be challenged by the pre-specified nature of the cases. Two
¢ enough for bringing students to the level
S.
ses,
cas
computer s
required for dealing with the first (the easiest) hot
ons with tutorials and simple exerc
Making a good hot SD case for a particular level is time consuming and difficult. Preparing
a large number of new hot cases is even more challenging and time consuming: 7 to 8 new cases
need to be developed every year for the Introductory SD courses at Delft University of Technolog;
Sharing new cases with colleagues and at a later stage~ students may partly solve this problem.
This paper therefore promotes the exchange of teaching/testing cases by sharing several hot cases
developed over the last year. Thes s can be used for self-study and for teaching and testing
purposes at other universities as long as the authorship is respected and the cases are correctly
referred to (e.g. ‘developed by’ or ‘based on a case developed by’) even for exam cases.
ing the overall quality of dif.
ferent SD courses/curricula, and may help to improve them. publishing our cases leads in any
Openness about ‘exam grade’ cases may also be useful for ass
“dr. Erik Pruyt is Assistant Professor System Dynamics at the Faculty of Technology, Policy and Managementof
Delft. University of Technologywhere he manages and teaches the Introductory System Dynamics courses.
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society 2
way to more reflection about the:
curricula~ too: in this paper we reflect about our use of MC ques sting purposes, both
for testing of general SD skills/knowledge and for testing SD modeling skills, as well as about new
(resource poor as opposed to the current resource rich) designs of our SD curriculum. Finally,
this paper aims at sharing some of our most recent experiences in teaching SD to large groups of
-both bachelor and master~ students.
Section 2 provides a quick overview of our SD curriculum, the SD skills focused on in the intro-
ductory SD cou! s currently used in the introductory SD course, and an overview of publicly
ailable SD cases developed for and used in Delft University of Technology's Introductory SD
courses, The new cases are presented in more detail in section 3. New evaluation modes by means
of quick scanning and model-related MC questions are discussed in section 4. New experiences,
opportunities for our SD curriculum and conclusions are discussed in section 5. And, last but not
least, the appendices contain refer to~ the new c ented in this paper: appendix A contains
the ‘Oostvaardersplassen Natural Reserve’ case, appendix B contains the ‘North-Eastern Bluefin
Tuna’ case, appendix C contains the ‘Boom and Bust in Dubai’ case, appendix D contains the
‘De/Radicalization’ case, appendix E contains the ‘Energy Transition’ case, appendix F contains
the ‘Slow Students’ case, and appendix G contains the ‘Rare Earths Scarcity’ case.
2 Case-based SD Teaching & Testing
2.1. The Current SD Curriculum
System Dynamics (SD) is an integral part of two of the study programmes offered at our faculty.
Meyers, Slinger, Pruyt, Yucel, and van Daalen (2010) describe the different SD courses in the cur-
riculum and their learning goals, and an explanation of the way in which real world complexity is
introduced in a quadruple jump approach over the whole curriculum.
The introductory SD course for bachelor and master students fo on introducing the SD
methodology and on conveying basic and intermediate SD modeling skills. Although most stu-
dents enter the course without prior SD knowledge, all students have followed an introductory
course in Differential Equations and an introductory course in Policy Analysis. Pre-testing shows
that a large s is able to read graphs properly and solve elementary stock-flow
problems. ‘Hot? cases of 1-2.5 pages are used during the 7-week 6-hours-per-weck course to demon-
strate the use of SD in addressing current issues. The explicit learning goals of the introductory
SD course are (i) to have basic knowledge of the SD field/philosophy/method, (ii) to be able to
apply the SD method using SD software packages, and (iii) to have a basic understanding of SD
model use and to have gained basic experience related to the SD modeling process. Cases
for teaching range from small and simple, over technical, to intermediate/difficult.. Exam cases
are always intermediate/difficult of about 2 pages. During the exam, students have 3 hours to
answer 20 MC questions related to SD methodology/insight/... (sce section 4) and for solving a
hot case. Passing rates range from 55% to 95% per year (exam + retake exam) depending on the
student group: passing rates for second year bachelor students are always low (between 55% and
70%), and passing rates for first year master students are always high (between 90% and 100%)
although the exams and the course are the same. Hence, the differen can only
be attributed to time spend on the course, the need to pass the exam!, the level and maturity of
(pre-selected) master students, and most of all, the timing of exam and retake exam?.
are of the students
TDutch BSc students can take the exam over and over again whereas foreign MSc students need to pass their
first or second attempt.
2Master students have 1 week between their exam and their retake exam. Bachelor students formerly had 6
months between their course and exam, and their retake exam — as off last year, they have 3 months between the
exam and retake exam.
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society 3
Case name / theme Approp. | Difficulty] Time | Specifics References (C=case +
(1 to 5) | for $D101_| needed A=analysis)
Dutch Soft Drugs Policy 3 easy T:00 qualitative C&A in (Pruyt 2009b)
Pneumonic Plague 3 T00 small Cim (Pruyt 2010a)
EVs and lithium scarcity 3 staged Cin (Pruyt 2009¢)
Redevelopment of social 3 abstract/highly | based on (Pruyt 2008a)
housing districts aggregated C in (Pruyt 2009e)
Fall of the Fortis Bank 2 200 simplistic Cin (Pruyt 20094)
Oostvaardersplassen q medium 0:5 technical x. Cin this paper
Flu pandemic 5 medium 2:30 staged, builds | Cin (Pruyt 2010a); A in
up from simple_| (Pruyt and Hamarat 2010b)
Cholera in Zimbabwe q medium C& Ain (Pruyt 2009a)
Overfishing of NBF Tuna q medium Staged Cin this paper
Concerted run on DSB 7 difficult, z better than Cin (Prayt 2010a); Aim
Bank Fortis case (Pruyt and Hamarat 2010a)
De/Radicalisation 7 difficult, 230 Tot staged ‘Cin this paper
counterint. A in (Pruyt and Kwakkel 2011)
Bnergy transition 3 difficult, 230 bridge to Cin this paper
SD project A in (Pruyt, Kwakkel et al. 2011)
Boom & bust in Dubai 3 difficult, Tight /wrong Cin this paper
The ‘slow students’ fine 3 difficult, pulse train, ete _| C im this paper
Mineral/metal scarcity IT q difficult, many specific | C in this paper
functions A in (Pruyt 2010b)
Mineral/metal scarcity T T very difficult, Timajor loop Cin (Pruyt 2010a)
Energy versus Food 2 difficult, bridge to Cin (Pruyt 2010a)
Security and long SD project A in (Pruyt 2008b)
Table 1: Publicly available in order of difficul nent of their ap-
propriateness for testing intermediate modeling skills from 1 (lowest) to 5 (highest), their difficulty,
minimal time needed, specifics, and references
less pre-specified and structured case of about 20-25 pages (
in (Meyers, Slinger, Pruyt, Yucel, and van Daalen 2010).
fterwards, students can do a BSc thesis in SD, follow the Advanced SD course, take Simulation
Classes, and do an MSc thesis in SD. Students who follow all the SD courses on offer have
almost the equivalent of a one-year, fulltime master programme in SD (Pruyt et al. 2009). In their
full curriculum, however, students learn a range of problem exploration and structuring methods
and study other modeling methods, such as Agent-Based Modeling and Discrete Modeling.
2.2 Publicly Available Hot Teaching & Testing Cases Developed for the
Introductory SD Course
Table 1 lists most of the publicly available teaching/testing cases developed and used for this partic-
ular Introductory SD course. These fully pre-specified cases typically focus on basic/intermediate
SD modeling and simulation skills: mainly model specification, some model testing, some sensitiv-
ity testing and scenario analysis, some Causal Loop Diagramming (detailed and aggregated), some
focus on the link between structure and behavior, and some policy testing. From the end of we
three on, only hot cases are used — some smaller exe are used during the first three
weeks of the course. These hot cases require 95-99% of transpiration (applying trained skills),
and only 1-5% of inspiration/insight. Additional MC questions are used during the exams to test
insight, knowledge, e
Following seven c
below:
-developed between January 2010 and January 2011- are presented
a small but rather technical case
e The ‘Oostvaarderspla: with interesting spe
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society 4
cations (pulse train, random, lookup, smooth3I, et cetera) about the mismanagement of the
Dutch natural reserve ‘The Oostvaardersplass
en
The ‘North-Eastern Bluefin Tuna Fisheries Case’ is a rather small and easy case related to
the overfishing of North-Eastern Bluefin tuna and the (in)effectiveness of ICCAT polic
¢ The ‘Boom and Bust in Dubai Case’ is an intermediate c
bust of the real estate sector in Dubai.
related to the boom and potential
The ‘De/Radicalisation Case?
towards harmle: sm (de:
an intermediate case related to the development of activism
adicalization) or extremist activism (radicalization).
© The ‘Transition towards Sustainable Energy Systems Case’ is a staged but rather lengthly
and difficult case about the competition of new energy technologies against old energy tech-
nologies and against other new energy technologies for the same subsidi
The ‘Slow Students Fine Case’ is a very hot case about a Dutch cabinet policy proposal to
fine universities based on the number of (relatively) slow students enrolled. Rather specific
functions and structures need to be used in this intermediate case.
The ‘Mineral scarcity II Case’ case is a second case about potential mineral /metal scarcity,
easier than the first mineral scarcity case, but still relatively lengthy.
The:
of the
cases are briefly presented and discus
. The case descriptions and case que
d below in increasing order of difficulty /length
ions are available in appendic
3 2010-2011 Cases
3.1 The Oostvaardersplassen Case
The Oostvaardersplassen case is a relatively small, but quite technical System Dynamics
about natural reserve management. The Dutch natural reserve the Oostvaardersplass
which hunting was prohibited for over 27 years, was kept clear of willows and bushes by large
herbivores which do not have natural predators in the Netherlands. After the carrying capacity of
the area had been reached, the Dutch population was shocked by movies and pictures of massive
starvation of these large herbivores at the end of winter.
birth season death season
Pad percentage birth mte percentage death mate |
length birth season’ vie % Tength death
season
Pte ae
biths herbivores deaths
ee italnumber of Thee,
randomizer births rge herbivors randomizer deaths
smoothed ind on large smoothed info.on smoothed info on larue
herbivore births large herbivores herbivore deaths
Figure 1: SFD of the small —but technical- OVP simulation model
Students need to make a SD simulation model (see Figure 1) based on the description provided
»e section A on page 25). Doing so, they need to use several special functions (pulse train,
Phe corresponding SD models and answers will be sent upon request to colleagues willing to exchange ‘hot’
cases. All models are available in Vensim and in Powersim formats, all cases are available in Dutch and English.
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society 5
random, lookup, and smooth3I functions), test the model, simulate it and draw the dynamics of
several variables with a particular seed for the random functions (see Figure 2).
2,000 6,000
1,500 4,500
1,000 3,000
500 1,500
" i o
GBs sot 1554 QOOUAOS ODIO GUIS CaM DO TOag USS 1985 1990 1995 2000 2005 2010 2015 2020 2025 2030 2035
Time (year)
Tine (yeas)
snnothed ings onlaige hebivere bins ge herbivores 5
smnothed info on large hetbivore deaths Po BangenganGengungn8 smoothed info on large herbivores Se
Figure 2: Behaviour of the smoothed flow variables (left) and crude and smoothed stock variable
(right)
Then they need to simulate the model again with a different ‘sced’, and again, and again,
and again ~ after which they are asked to gencralize the results/conclusions. Not having been in-
structed about random functions during the course, students need to find out about them during
the exam. They also need to test the model for behavioral sensitivity. They need to make a com-
plete and an aggregated CLD of the model and explain the link between structure and behavior.
Finally, they need to come up with a policy, test it, and conclude.
This case was initially developed as a multiple choice question to test flow-stock behavior skills
(selecting the behavior of a stock variable that corresponds to given inflow and outflow variables),
and used as such during the January 2011 BSc exam. Later, it was reworked into a short —but
ase for ‘slow students’ which always seem to run into time problems.
were not better for this shorter case than for
full- exam
Surprisingly, average performance and grad
longer cases. Especially interpretation and distilling gencral conclusions was rather poor. This
may be an indication of too much focus on model construction and too little focus on model use
and interpretation.
3.2 The North-Eastern Bluefin Tuna Case
The NBF Tuna Fisheries case deals with the overfishing of North-Eastern bluefin tuna and the
(in)effectiveness of ICCAT policies. It should be noted that the current version of the c
is just illustrative/educational: data and policies in the case/model are
partly inspired / based on (Dudley 2008) and the Fishbanks Game.
First, student are asked to make a SFD of a partial model description (green variables in Figure
3(a)) and to write down the corresponding balance equation (as in Figure 3(b)). They also need
to simulate this partial model for different values for the total number of ships (see Figure 4(a)).
After extending the model (to all variables in Figure 3(a)), students need to simulate the model
(Figures 5(a) and (b)), validate it, and test the sensitivity (Figures 5(c) to 5(f)). Their conclusion
should be that the model is mainly numerically sensitive to parameter changes — even to a 10%
change in the ‘effect’ lookup — but not behaviourally sensitive.
After realizing that these cases are insufficient — to say the least — students are asked to test
whether the ‘current policy’ would be suffice if the number of illegal tuna ships would drop —
through strict controlling and sanctioning from 10000 to 0 in the year 2010: the blue curves in
Figures 5(g) and 5(h) show that tuna biomass would take a very long time to recover to about
50% of the initial value and that following the policy— the official flects should remain close to
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society 6
__-——e total number of —
egal ma sips hs faction tuna catch
average efficiency
‘tuna ships
tuna recruitment
[[, netincrease official
nl nite
umber oftuna ships ~~
\ Ny
“
@ toma recruitment — natal deatiss
proposed change in the — meee Z 5
nuiber of tuna ships 3 hier ai
twa wrowihrate
effect delay time ‘al state tra unfshed biomass
s fishery
" 1990
perocived state of. most recent perception change in tua
——
rua fishery concening tm fishery perception lookup
(a) Stock flow diagram of NBF Tuna
total mmber oF Tana catch
una chips (
mT delayed ship Lo)
‘fish loop
5 the
illegal tuna ships 1) free
Oficial Tana Ships :
+ AS Biomass Twa 7
i iil, 1S) io * iG
oe a(rte-e8— Fi) eT daptive number “ ia
at ° eae tural death toop"twal dcaths
on + .
.
ahs
pe ae proposed ret increase of natural deaih
/ the mumber of tuna ships smltipicater loop
— x Ph
r6 ratio of biomass to
perceived state of unfished biomass
tuna fishery
(b) Balance equation (6) Balance equation
Figure 3: Balance equation and SFD (green variables ~ balance equation) of the NBF Tuna case
100,000
75,000
\
25,000 |\\
Biomass: 0
o|® Biomass; 5000
1999 2000 2010 2020 2030 2040 2080 2060 2070 2080 2090 Biomass: 15000
oe Biomass; 25000
(a) Tuna biomass for 0, 5000, 15000, 25000 tuna fishing boats
Figure 4: Partial model behavior for different numbers of tuna ships
Pruyt, 2011. Making System Dynami
es Cool IIL. In: Proc.
100,000 40,000
!
7.000 J} 30,000
\
0.000 |} 20,000
\
5.000 |
254 10,000 \
N titty
° ° \
1990 2010 2030 2050 207) 2090 2110 2130 2150 2170 2190
1999 2010 2030 2080
Tine (Year)
2070 2090 2119 2130
Tine (Vest)
(a) Tuna biomass (base case)
(b) Official tuna fishing boats (base case)
40,000
nao
a °
‘ine Cea Tix (ex)
(c) Tuna biomass (parameter sensitivity) (d) Official tuna fishing boats (parameter sensitivity)
so0.000
75,000 I
50.900
i,
25000
‘
1990 Zo) Toso Bosd Dud au
2170 2200 2230 2260
7290 °
Tane (Year) 1990 2020 2050. 2080 2110 2140 2170 2200 7290
Biomass: Sensitive cnt OPceat Tine (Ves)
(c) Tuna biomass (lookup sensitivity) (£) Official tuna fishing boats (lookup sensitivity)
100,000 40,000
75,000 f 30,000
\
#5000 |} 20,000
\
25,000 ‘ot 10,000
oL NG ° ALU reat
1990 2020 2080 2080 2110 2140 2170 2300 2250 2260 2290 1999 3020 2080 2080 2110 2140 2170 200 2230 2260 3290
‘Time (Year)
Tine (Your)
(g) Tuna Biomass
(h) Official number of Tuna Ships
Figure 5: Base case behavior of the NBF Tuna model (a and b), parameter sensitivit:
‘effect’ lookup
(c and d),
sensitivity (¢ and f), and what-if behavior (g and h) for the base case (orange
the base case without illegal fishing (blue), and the base
(green)
ase without downsizing of official fleets
of the Int. Conf. of the SD Society 7
2150 2170 2190
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society 8
zero for a very long time. A second what-if test what if countries are unwilling to reduce their
fleets leads to even more disastrous consequences (green curves in Figures 5(a) and 5(b)).
Students are asked to make a causal loop diagram that can be used to explain the link between
structure and behavior to fishermen and policy makers alike: Figure 3(c) would be a rather detailed
CLD for doing so. Finally, students are asked to design 2 policy measures that improve the
sustainability of the ICCAT policy modeled before. Two final bonus questions aim at challenging
cellent students.
3.3. The De/Radicalisation Case
The de-radicalization case allows to explore how/why activism may become more extremist /harmful
or moderate/harmless. Students found the case moderately difficult because (i) it is not staged,
(ii) it is somewhat explorative, (iii) the effects of policies are rather counter-intuitive, and (iv) it
is difficult /impossible to make a highly aggregated CLD of this model.
Students need to model this issue (see Figure 6), test the model, simulate the model over a
time horizon of 100 years starting in 1980 and make graphs of key variables (Figure 7(b) and 7(c))-
init vate of
snderyng phenomena SE problem stom ———~m vil ofthe reinbrcement of
pennies a - problem“ the vsbily ofthe
inka readiness to ke a =. robe
action per contact, a amt rte ofincrease of Trt actos
re the tnderbng phenomenon ~
fe ef Teinbreemtoftte
a: - sbi ofthe problem
problematic oodiiin: ‘trough radical and
eet evel em Ac noneradal actions
/
rate of decrease
ataimble mate of through societal change “*”
f noma!
a 1; HS
ease a ee
of the problem a E ad
f 7 / strength of contact ate of presuasiness of
Bs y / / incompatible convinced citizens ‘non radical actions
state carat
pic y \ a.
erg degee of é \
__— conviction of convinced nonmal contact rate of ) \
/ pet soles \ conte ions
of ite a , |
\ er \ radical action level
i 5 average readiness 1 al
y, readiness to tke action per 7" Ly
Ay { gf ‘newly convinced citizen Jake action f
\ {/ ‘ \ ra y
Xs, | / Tpplreadnes —p maieaaion
ely settmrwioainet | PS
station of few se ae oe initial amber of
Sone ‘convinced citizens
initial number of |
caveats
Figure 6: Full Stock-Flow Diagram of the de/radicalisation model
Then they need to test the sensitivity/uncertainty of the model for changes in several param-
eters (Figure 7(d)) and draw conclusions: significant changes in five out of six parameters lead to
a fundamentally different mode of behavior (see Figure 7(d) — the blue curve represents the base
case mode of behavior).
Students are asked to make two interesting and consistent exploratory
results of this sensitivity an the very different scenarios displayed in Figure 7(e) can be
distilled since this model is characterized by a strong bifurcation. Students also need to make
a CLD of this simulation model, and use it to explain the link between structure and behavior.
cenarios based on the
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society 9
underying
7 pbenomenon:
¢
a Og
" a
pereeived problem netincrease
oo ke \
i \
Cb cominced
‘ncominced. contact rate
convinced
\ degree of fi » /
/ average conviction and persuasiveness
yg Wilinaness to act +
frustration of +
convinced \
SX cA ca a
non radical action
= level
if
ae i ndial action evel
(a) CLD of the de/radicalization model
ao ats i aay
1M cl Veer tae
oboe toni Soren
500 contacts!(Vr*citizen) |
it de os |
ee oe Gomi
pie iN
Dim ! |
Salat
O Stes 1980 1995 2010 2025 2040 2055 2070
9 eae 2s 2
1980 1995 2010 2025 204020852070 contactrate of convinced cizens:radicalzation +} contacts (eae*izen)
Bac (ven) et ee
fiustraion of convinced ctizens radicabzaion 3+ Dua
convinced eiizens : radicalization. ——}—+—4—+_+ ++ citizens
ae rae 7 visbillity ofthe problem radicalization +++ ++ Dr
persuasion: radicalization —@—¢—#—@—2 2 —e— citizens/Year “presuasioness of aon radical actions" ‘radicalization G55 ehteasleontacts
(b) Evolution of the stock of convinced citizens (blue -1-) (c) Evolution of main drivers of persuasion flow in case
and persuasion flow (red -2-) in case of deradicalization of deradicalization
frustration of convinced citizens
075
os
028
)
1980 1990 2000 2010 2020 2030 2049 2050 2060 2070 2080 1980 1990 2000 2010 2020 2030 2049 2050 2060 2070 2080
Tine (Year) Tine (Year)
(a) Behavior mode sensitivity (frustration of convinced (e) Two distinct scenarios: deradicalization (red) versus
citizens) for 5 out of 6 parameters further radicalization (blue)
Figure 7: CLD of the de/radicalization model (a), and deradicalization mode (b & c) versus
radicalization mode (d & e) of the de/radicalization model
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society10
However, this question may be rather hard because it is difficult to make an simple/aggregated
CLD of this model ~ a rather detailed CLD is displayed in Figure 7(a).
Finally, students are asked to formulate (counter-intuitive) policy advice based on the link
between structure and behavior, and to give advice related to future refinements and extensions
of the model.
Pruyt, 2011. Making System Dynamics Cool IIL In: Proc. of the Int. Conf. of the SD Society11
3.4 The Transition towards Sustainable Energy Systems Case
This case deals with the competition of new energy technologies with existing technologies and
other new technologies, and could be extended to spreading /concentrating subsidies for innovative
renewable energy technologies. The Energy Transition towards Sustainability case description can
be found in appendix E on page 31. Students found this staged case difficult and lengthy.
First students need to make a SDF (Figure 8(a)) and a detailed CLD (Figure 8(b)) of a small
part of the model description. Second, students need to add a learning curve structure 8(c)) and
plot the marginal c
" average construction
desired fraction new se ones 7
capacity Tl initial capacity under initial capacity T1 lifetime
\ construction T1 | techmology T1
a capacity under = pe installed capacity | —
newly planned — construction 1) Cotmssioning 11 deconsissioning
capacity T1 capacity T1 7 capacity TL
expected new capacity Phuning period
tobe installed ~~ total installed
capacity
expected capacity
required = “Wp
(a) SED of the first partial Energy Transition model
xy ey fy
commissioning decommissioning
cguetneton 1) capacity TL ‘pac a AD capacity T1
op
vee renee Ay cata
eapacity TH ‘elose the capacity installed
Rye gap’ loop .
expected new capacity “. expected capacity to
tobe installed "be replaced.
(b) Full CLD of the first partial Energy Transition model
marginal cost marginal cost aj ————__ ind
capacity previous year TL journey esta
os ‘ ie aed capacity previous year T1
initial marginal cost
desired fraction new new capacity T1
capacity Tl experience curve
parameter T1 cumulatively installed
capacity T1
progress ratio T1
(c) SFD of the learning curve added to the first SFD
ion model
Figure 8: Partial SFDs and CLD of the Energy Transi
Then they need to extend the model with a sustainable alternative (Figures 9(a) and 9(b)),
simulate it, and make graphs. Students also need to explain how this structure generates this
behavior. They need to test the sensitivity of the model for changes in the parameters of the
in a function. Finally students need to add another sustainable
t the influence of spreading investments over two alternatives.
learning curve and for change
technology (Figure 9(c)) and t
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society12
‘marginal cost
new capacity :
smarginal cost new ‘marginal cost a ——___
capacity previous year 11 |__Tl capacity TL cumulatively installed
capacity previous year TL
intial marginal cost
sew capacity TL
<desied faction <desired faction gtr
new capacity T3> new capacity T2 omer cunmibatively installed
\ ‘capacity T1
— ratio TL
desired faction new average construction i
capacity TI time TL initial cumulatively
initial capacity under intial capacity TL deconmissioned
construction TL | eanee TL capacity T1
it camilativel
w|i] — Sm |
newh planed — construction T1| commissioning Gecomtisinig | capacity TL
capacity TL capacity TL capacity TL
expected new capacity Planning period
tobe installed = total installed
.<
Se Fy sista faction of
“rene CNPected capacity Pe Z Pe total installed capacity
aured capacity T3> capacity T2>
marginal Cost new marginal cost 4 _
capaciy previo year T2 a
earindticock intial marginal cost s capacity previous year T2
marginal cost new ee .comaersae
cxpacy TS F panier 12 N
desired faction new __ curmiatvely installed
capacity T2 progress ratio T2 ‘capacity T2
ii Sa Br a ees ita ema
capacity TL P techmology T2 jecommissio
ae intial capacity T2 capacity T2
zi salled | cumiaively
expected new pe capacty under Smet | decameisioang? | decommissioned
GHG poe panewbyplamed ~ [constuction T2) commissions jecommissionng | Capacity T2
ale capacity T2 capacity T2 capacity T2
4
2 Tiara cost
Teninlcocew new capacity |‘margial cost
capacity previo year T3- = capacity T3 Cummtatvely installed
<tuansnal cost new Yt argizal cost =o
capacity T2 ew capacty Ts sperms cue —
= "1S" parameter 73 ~
.. ~
Aesied faction new . cunmulitivey installed
capaci progress ratio TS ancy T3
as
average constuction
A teins | ‘ttl cmiatvely
<suarginal cost 2 ae tine TS \ decommissioned
capacity pa op inialcapaciy T3_~ elmeew TSN. capacity TS
——>a! ? = cumin
espana pe! capacity under aa a decommissioned
Gihcn (C5. seo pammed” [commaon 3] commésonhe cnensioina | “copay 13
‘stad copaety T2 capacty T3 capacity T3
Figure 9
: SFD of three competing Energy Technologies
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society13
3.5 The Boom and Bust in Dubai Case
The ‘Boom and Bust in Dubai’ case was developed specifically for testing purposes. It was inspired
by a news paper article about ‘The miraculous recovery of Dubai’ (NRC Handelsblad 2010) (see
Figure 12(a) on page 16). The case description can be found in appendix C on page 29.
First, students need to make a Stock Flow Diagram (green variables in Figure 10(a)) and a
detailed CLD (green variables in Figure 10(b)) of the first part of the case description. Second,
they need to extend the simulation model to the full description (Figure 10). They need to be able
to specify the right max(0,...), min, delay, and lookup/graph functions. Their models generate
nonsensical behaviours if one or more of the crucial functions are poorly specified.
They need to verify, validate, simulate, and plot graphs of the model, first without crisis
settings (see Figures 11(a) and 11(b)). Later they need to add crisis settings to the model and
test whether that leads to the unfolding of a real estate bust after month 10 (see Figure 11(c) and
11(d)).
These changes not being enough to generate a true collapse
of the uncertainty related to the average immigration time (1 ~ 3 months) (Figure 11(e)) and
the REU construction time (1 — 4 months) with crisis settings (Figure 11(f)) on the number
of immigrants, and their combined effects without crisis settings (Figure 11(g)). After these
simulations, students should recognize that two different modes of behavior can be simulated
with combinations of different values for these two parameters (exponential growth and a partial
collapse followed by exponential growth), that total collapses without redress are not experienced
without crisis settings, contrary to simulation with crisis settings.
Students should also be able to deduct that the exponential growth is caused by: more im-
migrants, more REU needed, more REU under construction, more immigrants, etc. The partial
collapse is caused by an initial surplus of immigrants and REU under construction for runs with
small values of the immigration time and construction time in which the construction time is
smaller than the immigration time. Hence, the REU under construction initially in the pipeline
are completed before new immigrants can be attracted.
However, only a very small fraction of our students was able to distill these conclusions during
the time-constrained exam: only those few students were able to make an appropriate aggregated
causal loop diagram of the model (see Figure 12(b)) for explaining the link between system struc-
ture and behavior and providing appropriate policy advice derived from the model.
students need to test the influence
This case served as a testing case on 25 October 2010 for a group of 45 MSc students and
on 18 Jamuary 2010 for a group of 70 BSc students during their time-constrained exam of the
introductory SD course. The passing rates were rather low because (i) passing rate at the exam
are always low compared to passing rates at the retake exam, and (ii) because the case cannot be
solved (at all) without the correct min-max specifications.
Building blocks addressed in this case include stock-flow modelling and detailed and aggregated
causal loop diagramming of aging chains, formulation of special functions (lookup functions and
time series), and exploring plausible model behaviour. In teaching, this case may be used in the
of the introductory SD course (sce (Pruyt et al. 2009).
last weel
of the Int. Conf. of the SD Society14
Cool IIL. In: Proc.
Pruyt, 2011. Making System Dynamic
expected REU Al Maktoum investment avg REU lifetime
shortage ratio of current REU supply, as,
REU under
construction! meetin S23
se pate bt REU demotion
approved commissioning
aveREU ee _™ 4
approval time workers per REU /
under construction of
| REU construction
a | , == REU shortage
Bis der ve REU shortage
expected new REU normal cost REU_ eee
due to immigration “MK
¢ REU price
i ase ‘a REU demand
immigrants
™ _
\ relative REU demand
inunigration \ attractiveness 2 per
sultipication Letor normal salary ‘2 immigrate —
ge \
4 NE on 4ermnigrant ‘Time:
joned sana ‘integration rate
shortage aa
immigrants cals
workibree | maga i
‘immigration integration
avg emigration
worktitee gg time
itrmnigrants a <i SS ————__ exogenous
previous perio eae
a our demand:
(a) SED of preliminary model (in green) embedded in the SED of the full model
Al Maktoum
‘REUpush
REU demand _-eXPeted new REU
due to anmigration
ne average REU
Tifctime
N ~
expected REU.
shortages a aes a
op
current REU supply
+ fa REU supply cA REU demolition
i Pits oo immigration ae ~
average REU approved ee staat 5: adjustment IZ cs
approval time
ie, workers per REU
# under construction
‘onstruction
~ tay ke
construction
tal REU construction time
(b) Complete CLD of preliminary SFD — in other words, of the green variables
Figure 10: Full SFD and partial detailed CLD of the Boom and Bust model
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society15
Selected Variables.
40M person 40
40M REV
30
20M person eo Lert |
20M REU 4% al
10
© person i
0 RED °
0 om 48 72 96 20 144 168 102 216 240 0 4 48 72 06 12D 14 168 i92 216 240
Time (Mouth) Tine (Mont)
Exam Question! person EU shortage: ExamM2Questnnl, ++} 4
REV supply :EsanM2Questionl 2222 a 29s 2 REL Ibbow shortage -EvamM2Quesionl 222 ee —p a
(a) Immigrant and REBU
supply without crisis settings (b) Labour and REU Shortage without crisis settings
ais 6
ie a8
ow a 3 eRe
hee
° °
om 4 72 96 120 1 168 192 216 240 0 2 48 72 96 12D 144 168 192 216 240
(Month) Tine (Mont)
Exam Questions EU shortage : ExamM2Questn t+ +} $4
REU supply: ExamM2Question’ -@—e—e—2—e—a aa — Inbour shortage ;EuamM2Quesion3. —¢—2—2—e—2 a2 2 —
(c) Immigrant and REU supply with crisis settings (d) Labour and REU Shortage with crisis settings
immigrants immigrants
20M 4.003.
rt 3.002 M —
=
: ea
P 200m
tom a
son
- °
0
o 24 4g 72" 86 120 «144 ies "192 “216 240
Tie ont) =
Eamon Ecsite Conte ines ttt
(c) changes in immigration time with crisis settings (f) changes in construction time with crisis settings —
detail
6.004 Mt
perma
3.000. M
0 2 2 36 48 60 72 84 96 108 120 182 IM 156 168 190 192 204 216 228 240
‘Tine (Month)
inmigemtsEemS0QS_ C13 115} }——inmigants temags cry + ~-
feemgamte EemanQe crs rts ee Regs mrQs CTD
femicaus EmasDQs CTS 172
2
(g) changes in construction times without crisis settings
Figure 11: Plots with crisis settings (a and b), plots without crisis settings (¢ and d), sensitiv-
ity/uncertainty analysis related to the immigration time (ce), construction time (f), and combina-
tions of immigration and construction times (g)
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society16
ch REU under average REU approval
Faneuil RE - iilsctcknn and construction time
uurpis dita per arante ater oa aconen expected REU + y
1 Goniddete nape par arkame mater pervorgeerl kta shortage EES
‘econ —— 26 i shortage
‘00 +
S Ap) oo
immigration loop +
ie expected REU expected net
eo demand inmigration 5
oe a ae acute workforce
i é
ar ap ae oF oF af oF oF oop go ot of or erperd B= emiraton
“(1 dirmam= 18 eurccent , Koes T8710) ‘NHC WINN Br: Ch Es population
(a) ‘Business market Dubai collapsed’ — Or- (b) Aggregated CLD
ange bars: rental price per m?
Green curve: percentage change to previous
quarter
in dirham;
Figure 12: Graph from the newspaper article (Source: NRC Handelsblad 16/10/2010) and aggre-
gated CLD of the model partly explaining the behavior
3.6 The Slow Students Fine Case
One of the most recent cases —the ‘Slow Students Fine’ case~ was too hot too last: the case was
developed one day before the exam of 18 January 2011 (precisely because this hot issue was on
every student’s mind), but was outdated less than two weeks later (when the strongly opposed
policy proposal was partly abolished after mass demonstrations).
In 2010, the new Dutch cabinet Rutte-I launched plans to fine ‘slow students’ -students in
higher education with more than one year delay- €3000 per extra year on top of their normal
annual college fee (for Dutch students) of about €1900. The plans specified that the institutes
for higher education needed to pass on the additional college fee of their slow students to the
government, but also that these institutes for higher education needed to pay an additional annual
fine of €3000 per slow student. These plans aroused serious opposition and protests, also from
the institutes for higher education because these proposed policies were expected to seriously hurt
those institutes for higher education —especially the three technical universities and other difficult
studies— and to lead to undesired side-effects (eroding goa ve lay-offs, et cetera). These
plans were also argued to be damaging to the (future) Dutch knowledge economy, and running
counter to the ambition to return in the top of the world of dmowledge and innovation countries’
by 2020 (Kennis and Innovatie Agenda 18/01/2011).
More than 80000 out of 550000 students in the Netherlands had —at that time- accumulated
more than one year of study delay. Especially the technical studies have because of the level of
difficulty- a large fraction of slow students: 22.5% of the students need 4 years instead of 3 to
finish their Bachelor of Science (BSc). At Delft University of Technology, the fraction was even
higher because of the difficulty of the studies as well as the dazzling student life.
But such fines for slow students imposed upon the institutes of higher education would hurt
all students since they would reduce the overall educational amount of moncy available (there is
no subsidy for slow students) without providing the tools to speed them up or kick them out (no
numerus clausus, no binding targets, ct cetera). In the proposed form, it was simply a distributive
code for the largest budget cuts in Dutch higher education in decades (€360+ million).
In the corresponding staged case, students first of all need to model the through-put of BSc
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society17
nieuwe BSc
instromers.
‘minimal BSe
bikomende jaarijkse
“y %, vertraging BSc
BSc uitstroom mdien enkel i™
vaste vertaging door ninimak 2a BSc utstoomna vaste
studieduur Ps bijkomende vertraging
fiactie
Jatt BSc
jafvallers ee
<TIME STEP>
‘ao = _
instoomaomeat Patt BSe_T stroom BSe ‘ibtoom BSe
Pies studenten studenten
<Time> pp evolutic wan dé nieuwe
BSc instromess
(a) SFD of the ‘BSc’ submodel
minimale BSc ‘bikomende jaarlijkse
studieduur.
<a BSc.
BSc uitstroom indien ene! f he
vaste verte door rinale ” {| pS ital vaste en.
Staab oan of ae vertraging.
s J eg afvallers
T Bsa lL, Yai —
Semen, Se studenten, ucte avalkrs BSe
instroom BSc eg = MOBS ‘ . ee
studenten ut aes ee
studenten,
(b) Detailed CLD of the ‘BSc’ submodel (c) Aggregated CLD of the ‘BSc’ submodel
owt»
sae
minmale BSc bjkomende jaarlikse bijkomende jaarlese
ia nici MSc komende jal
“we vertaghyg BSe “— sncaheg vermaging MSc
[BSc utstroom inlien enkel if! — \ Be =
‘vaste ver door minimale “ge: BS* wistmom na vaste (MSc utstroom indien enkel
“one en bond serena vaste venga door iueale ——p- MSCutsroom vaste
va ‘ eee eabjioneni erage
ee | Sect van BSe ""teain
tale B80 mare So
stats m0 7 \
smitci> |_|
arljlce BS
ae Tih bse ret Bante se coro Ne
See studenten studenten eel studenten smdenten
a Tine pe bt van de Meme m4 /
ie ed /
set y,
<Tme>
~ \
ste
\ .
boete per AG
sok
eye
i wetoleys ie
cies TE =e
celia
se porte
tromer
ide per MSc
itn sen —
“Seen “in
docesten
wereheafd
(a) SED of the entire ‘slow students’ model
Figure 13: SFDs and CLDs of the BSc submodel and the entire ‘slow students’ model (in Dutch)
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society18
students and make corresponding detailed and aggregated CLDs (see Figure 13). They need to
extend the description with a submodel about the through-put of MSc students (copy-past with
), and a simplified submodel about the functioning of the faculty (finances and person-
nel) (see Figure 13(d) for the full SFD). Then they need to simulate the evolution of the faculty
without the fine system (Figures 14(a) and 14(b)) and with the fine system (Figures 14(c) and
14(d)), perform several what-if analyses, make/discuss two proposed corrections and propose other
adaptations/changes.
800 shaders 40
—_ ‘Nan
"| NARI Hi NNR
IK i,
i ra |
0 stent
0 exo 2 sl td a a ° #|
199) 1996 2002 2008-2014 2020 ~—=2008 199) 1994 1998 2002 2006 2010 2014 2018 2022 2026 2030
‘Time (Year) ‘Time (Year)
(a) Without fines for teaching to slow students: BSc stu- (b) Without fines for teaching to slow students: out
dents (blue —1-), MSc students (red -2-), and amount of flow of BSc students (blue — 1-) and outflow of MSc
money available for education (purple ~3-) students (red —2-)
ae oo {| i lak
1s) 1996 go02 2008 20182020 1965 1364 1968 3003 F005 Doxd sori bis “foxk” 836 “Zos0
Tine (Year) Tine (Yeu)
(c) With fines for teaching to slow students: BSc students (d) With fines for teaching to slow students: outflow of
(blue -1-), MSc students (red -2-), and amount of money BSc students (blue — 1-) and outflow of MSc students
available for education (purple ~3-) (red -2-)
Figure 14: Left hand side: BSc students (blue -1-), MSc students (red -2-), and amount of money
available for education (purple -3-); Right hand side: outflow of BSc students (blue — 1-) and
outflow of MSc students (red -2-)
Building blocks addressed in this case include stock-flow modelling and detailed and aggre-
gated causal loop diagramming of aging chains, formulating special functions (lookup functions
and time series), and exploring plausible model behaviour. In teaching, this case could be used in
week 6 or 7 of the curriculum (sce (Pruyt et al. 2009).
This case may have been too hot. Several things happened in the days and weeks after
using this exam question which made the question obsolete. Three days after the exam, mass
demonstrations took place in the Netherlands against these proposals. However, the cabinet
refused to make any changes. But on 1 February, the ‘Raad of State’ (the legal advisor of the state
and highest administrative/legal court) published its negative advice against the fines imposed
upon the universities based on their number of slow students. After this negative advice, the
cabinet turned the fines for slow students into a lower contribution to the universities of exactly
jected to generate... The introduction of fines imposed upon
ed.
the amount those fines were pr
students was postponed, not aboli
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society19
Joss of RAM in,
goods
use of RAM in
production of goods
A
initial amount of
RAM in goc
supply of RAM
processing of
RAM
(a) First submodel
gemiidede levenschur
in goedern.
gebruik van ZAM inde
pproduktis van goederen.
(b) Full CLD of first submodel
gemiddelde levensdinr
in gocderen.
a 4
verles aan ZAM in
goederen.
‘gebruik van ZAM in de
produkiic van gocderen..
processing van
ZAM.
(c) Aggregated CLD
production wit
‘intrinsic demand
decommissioning
ander) commissioning
rewh-planed — |_Construction rape
extraction capacity a
itil extmction capac ‘ptecise construction time
der constructic “ofextraction capacity
a
(d) Full SFD of the REM model
extmaction capacity at end
of liétime wa
A
Figure 15: Partial SFD, partial CLDs, and full SFD of the REM model
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society20
3.7 The Rare Earths Scarcity II Case
Expected or plausible mineral/metal scarcity issues were, still are, and will continue to be hot
issues. Several SD simulation models about particular mincral/metal scarcity issues have been
and still are being developed by our research team. Pruyt (2010a) already presented a gencric
teaching and testing case about mineral /metal sc Most students found that case too difficult
for the (exam of the) introductory SD course. Moreover, better SD models about mineral/metal
scarcity have been developed by our team, such as the generic model presented in (Pruyt 2010b).
Hence, the author further simplified the model presented in (Pruyt 2010b) and turned it into the
‘Rare Earths Scarcity II Cas
In this new scarcity case, students first need to make a small SFD, a detailed CLD and an
aggregated CLD from the first description (see Figures 15(a)(b)(c)). This very simple submodel
needs to be extended in two further steps: with the demand for REM (variables in yellow in Figure
15(d)) and mining/processing industries (variables in green in Figure 15(d)).
Students need to perform the necessary verification and validation tests, extend the model with
an ‘intrinsic demand’ structure and related scarcity output indicator, and simulate the behavior
of key variables (sce Figures 16(a) and 16(b)). Then students need to investigate what would
happen if the ‘initial extraction capacity under construction’ would be zero (sce Figures 16(c) and
16(d)), and what would happen if the ‘initial extraction capacity under construction’ would be
zero and the ‘economic growth rate’ would amount to 3% instead of 5% from 2011 on (see Figures
16(c) and 16(f)).
Following up on the what-if analyses, students need to perform a sensitivity analysis. Finally,
they need to make an aggregated CLD of the model (see Figure 17) and explain the link between
structure and behavior.
4 From Detailed Evaluation to MC and Quick Scan Evalu-
ation
Not only is it time consuming to make appropriate exam models, it is also time consuming to
them properly. At least, it was. Until the end of 2010, exam models were evaluated
in detail by lecturer and assistants — on average 30 minutes per exam... Significant cuts in the
number of student assistants available for the introductory courses forced the author to test new
ways of evaluating exam models without eroding the goals and quality of the course and exams.
In January 2011, the author tested the use of MC questions about the modeling and simulation
in combination with quick scanning of the model (about 5 minutes per model) and compared it
to detailed grading of the same models (about 30 minutes per model). Almost all students ob-
tained lower grades via the MC questions and higher grades via the quick model scan, resulting
nilar to the detailed grades. Although it takes more time to prepare exams with
MC questions about the modeling and simulation, there is an enormous time gain with this new
approach: for 200 exams, the difference amounts to about 70 hours (200 exams x 25 minutes
difference per exam - 180 additional mimutes to develop a MC version times 3 versions minus 210
minutes extra consistency chec i
in averag
Introducing MC questions to evaluate students’ modeling and simulation of the exam case also
led to the realization that students were over-tested in terms of specification (special functions and
delays), model testing and sensitivity analysis, unit/dimension analysis, CLD and SFD modeling,
and linking structure and behavior. Apart from the case, students also have to answer 20 MC
questions deal with:
e SD Philosophy, SD Methodology, or ‘SD speak’
© Formulation (special functions and delays)
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society21
600,000 Year
1
. [.
300,000 t'Year
Daal
e 8
ae
0 veer
o ot o LL tee be
200070122024 2036 2048 2060-2072 2000 S008 2016 2024 2052" 2040 2010 05S Doon “20732080
Tine (Yea)
Tine (Year)
(a) Base case behavior: installed extraction capacity (blue ~ (b) Base case behavior: the relative price (blue —1-)
1), demand for REM (red —2-), the output indicator (pur- and scarcity price effect (red —2-)
ple -3-)
2M Year 200
cea ar Ae
Nal NA « /\ /
° Year lanaae@r rer olen Ly
2000 2008 2016 2024 2032 2040 2088 2056 2064 2072 2080 2000 2008 2016 2034 2032 2040 2048 2058 2064 “2072 2080
Tine (Yeas) Tine Wea)
(c) What-if 1: installed extraction capacity (blue -1-), de- (d) What-if 1: the relative price (blue —1-) and
mand for REM (red ~2-), the output indicator (purple ~3-) scarcity price effect (red —2-)
600,000 t'Year f\ 8
150
PY
2000 2012-2024
300,000 t'Year
1
0 Year
°
2036 —-2048—«2080 «2072 2000 2005 2016 2024 2032 2080 2048 2056 206% 2072 2080
‘Tine (Yeas) ‘Tine (Yeas)
(c) What-if 2: installed extraction capacity (blue -1-), de- (f) What-if 2: the relative price (blue —1-) and
mand for REM (red ~2-), the output indicator (purple ~3-) scarcity price effect (red —2-)
Figure 16: Left hand side: Base case and what-if behaviors of the installed extraction capacity
(blue -1-), demand for REM (red -2-), the output indicator (purple -3-); Right hand side: Base
case and what-if behaviors of the relative price (blue —1-) and scarcity price effect (red -2-)
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society22
demand for REM
ror REM use in
decrease of demand prodocton:
through substitution 7
dpe, SC AS2
relative price
dental desired
of substitute EES ceancaon
relative price capacity
‘Te scarcity price wa“ 4 she
%, effect 7 y
---qereal extraction
average extraction bee:
costs t* extracted ae
Gp a. 6 oo
installed ge ge
extraction:
ae’
Figure 17: Aggregated CLD of the REM scarcity model
e Link between structure and behaviour
e From CLD to SFD and from SFD to CLD
© Calculation/assessment (of behaviour)
Archetype to fit the description
¢ Model testing and sensitivity analysis
Units/dimensions
e Policies
Hence, the number of traditional MC questions could be reduced and could be oriented towards
SD Philosophy /Methodology, ‘SD speak’, calculation/assessment (of behaviour), fitting archetypes
to descriptions, and policy analysis. This will save time — time most students lack during the exam.
5 Future Changes to the SD Curriculum
The high level attained by students after the introductory SD course —mainly because of the use
of these hot teaching and testing c: also makes that the curriculum could —and may need to-
be changed.
The well-specified cases used during the SD project may now be replaced by less structured
cases with less supervision. The project used in the SD project course taught to about 45 master
students per year will therefore change from the academic year 2011-2012 on into an almost
entirely unstructured hot case. From then on, pairs of students will be able to choose between
two topics proposed and mainly supervised by two senior lecturers. The relatively large number
of junior supervisors could then be reduced to a few junior super of students
with technical problems. The case topics will be chosen from current i
lecturers. That way, models developed by the students may at a later stage be used for re
purposes, for example to explore model uncertainty and explore the robustness of polic
different model spec
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society23
The Advanced SD course could ~and may need to~ be changed too: even more time could
be spent on really advanced issues (cigenvalue elasticity analysis, multi-player serious gaming,
Exploratory System Dynamics Modelling and Analysis, et cetera).
6 Conclusions, Lessons Learned, Proposal
All new testing/teaching
have been based on ‘hot’
developed over the last two years for this introductory SD course
may well be the main cause of a significant improvement of the SD
modelling skills: although it is difficult to prove, it seems that the use of these testing/teaching
es has accomplished more than the other measures discussed in (Pruyt et al. 2009). Moreover,
using ‘hot? 2 good way to enthuse students and to arouse their interest in applying SD in
case of real-world issues. Applying SD to ‘hot’ issues illustrates the relevance of SD for dealing
with real-world complex s, which takes SD testing/teaching models one step further than
being didactically responsible exercises. Although actual real-world testing/teaching cases are
often more motivating, they are also more difficult than exercises developed in the first (and only)
place to test/teach, b ¢ they need to be sufficiently close to the complex issue at hand to be
relevant and credible as a ‘real-world case’.
The main goal for introducing such cases to bridge the gap between the introductory SD
course and the SD project course by raising the level of difficulty of the introductory SD course—
has been reached. Students now learn all basic SD modelling skills (and more) where they ought to
learn these basic skills: in the introductory course. Hence, the SD modelling skills of those passing
the Introductory SD course are high enough to allow the SD project to be organised in a different
s resource intensive- mode. This allows us to change the SD project course into what a SD
course should be: an almost real-world project —with little but high quality supervision—
in which an unstructured and complex issue is structured, modeled, tested, simulated, analysed,
used for policy testing, and communicated in time to problem owners / stakeholders.
Two previous papers presenting ‘hot teaching and testing cases’ have also lead to the start up
a small (informal) network of university-level lecturers interested in sharing ‘hot’ testing/teaching
s. However, it may be desirable to set up a ‘central case depository’ managed by the System
Dynamics Society and to make agreements on a set of criteria and a specific standard/format.
A depository should set minimal quality standards —but more importantly— should be open to
different type c for example the cases in (Ford 1999), (Sterman 2000), (Martin Garcia 2006)
the D-series, (Meyers, Slinger, Pruyt, Yucel, and van Daalen 2010), etc), and should make a dis-
tinction between lecturers and students for levels of access. The Proceedings of the International
stem Dynamics Conference may —in the absence of a depository— be useful for sharing cas
Enjoy! But use with care...
The use of ‘hot’ cases
cas
References
Dudley, R. (2008). A basis for understanding fishery management dynamics. System Dynamics
Review 24, 129. doi: 10.1002/sdr.392. 5
Ford, A. (1999). Modeling the environment:
environmental systems. Washington (D.C.
Martin Garcia, J. (2006). Theory and praci . 23
Meyers, W., J. Slinger, E. Pruyt, G. Yucel, and C. van Daalen (2010, July). Essential Skills for
System Dynamics Practitioners: A Delft University of Technology Perspective. In Proceed-
ings of the 28th International Conference of the System Dynamics Society, Scoul, Korea.
http: //systemdynamics . org/conferences/2010/proceed/papers/P1096.pdf. 2, 3, 23
NRC Handelsblad (2010, 16 October). Miraculeus herstel van Dubai. NRC Handelsblad. 13
an. introdu
: Island Pre
ion to system dynamics models of
. 23
cal exercises of System Dynami
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society24
Pruyt, E. (2008a, July). Dealing with multiple perspectives: Using (cultural) pro-
files in System Dynamics. In Proceedings of the 26th International Conference
of the System Dynamics Society, Athens, Greece. System Dynamics Society.
http: //systemdynamics . org/conferences/2008/proceed/papers/PRUYT424. pdf. 3
Pruyt, E. (2008b, July). Food or energy? Is that the question? In Proceedings of the 26th Inter-
national Conference of the System Dynamics Society, Athens, Greece. System Dynamics So-
ciety. http://systemdynamics. org/conferences/2008/proceed/papers/PRUYT372. pdf.
3
Pruyt, E. (2009a, July). Cholera in Zimbabwe. In Proceedings of the 27th International Con-
ference of the System Dynamics Society, Albuquerque, USA. System Dynamics Society.
http: //www.systemdynamics.org/conferences/2009/proceed/papers/P1357 . pdf. 3
Pruyt, E. (2009b, July). The Dutch soft drugs debate: A qualitative System
Dynamics anal In Proceedings of the 27th International Conference of
the System Dynamics Society, Albuquerque, USA. System Dynamics Society.
http: //www.systemdynamics . org/conferences/2009/proceed/papers/P1356. pdf.
3
Pruyt, E. (2009¢, July).
ing & Teaching Cas
Making System Dynamics Cool? Using Hot Test-
In Proceedings of the 27th International Conference of
the System Dynamics Society, Albuquerque, USA. System Dynamics Society.
http: //www.systemdynamics. org/conferences/2009/proceed/papers/P1167 . pdf.
1,3
Pruyt, E. (2009d, July). Saving a Bank? The Case of the Fortis
Bank. In Proceedings of the 27th International Conference of the — Sys-
tem Dynamics Society, Albuquerque, — USA. em Dynamics Society.
http: //www.systemdynamics . org/conferences/2009/proceed/papers/P1273. pdf.
3
Pruyt, E. (2010a, July). Making Sy:
stem Dynamics Cool Il: New hot testing and teach-
ing cases of increasing complexity. In Proceedings of the 28th International Con-
ference of the System Dynamics Society, Scoul, Korea. System Dynamics Society.
http: //systemdynamics . org/conferences/2010/proceed/papers/P1026.pdf. 1, 3, 20
Pruyt, E. (2010b). Scarcity of minerals and metals: A generic exploratory sys-
tem dynamics model. In Proceedings of the 28th International Conference
of the System Dynamics Society, Seoul, Korea. System Dynamics Society.
ntp://systendynanics.org/conferences/2010/proceed/papers/1268, pat. 3, 20
Pruyt, EB. ot al. step and jump towards real-world complex-
Tn Proceedings of the 27th International Con-
ference of the System Dynamics Society, Albuquerque, USA. System Dynamics Soci-
cty. http://www.systemdynamics .org/conferences/2009/proceed/papers/P1140. pdf.
3, 13, 18, 23
Pruyt, E. and C. Hamarat (2010a). The concerted run on the DSB Bank: An Ex-
ploratory System Dynamics Approach. In Proceedings of the 28th International Con-
ference of the System Dynamics Society, Scoul, Korea. System Dynamics Society.
http: //systemdynamics . org/conferences/2010/proceed/papers/P1027 .pdf. 3
Pruyt, E. and C. Hamarat (2010b). The Influenza A(HIN1)v Pandemic: An Ex-
ploratory System Dynamics Approach. In Proceedings of the 28th International Con-
ference of the System Dynamics Society, Scoul, Korea. System Dynamics Society.
http: //systemdynamics . org/conferences/2010/proceed/papers/P1389.pdf. 3
Pruyt, E. and J. Kwakkel (2011, July). De/radicalization: Analysis of an exploratory SD model.
In Proceedings of the 29th International Conference of the System Dynamics Society, Wash-
ington, USA. System Dynamics Society. 3, 31
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society25
Pruyt, E., J. Kwakkel, G. Yucel, and C. Hamarat (2011, July). Energy transitions towards
sustainability: A staged exploration of complexity and deep uncertainty. In Proceedings
of the 29th International Conference of the System Dynamics Society, Washington, USA.
System Dynamics Society. 31
Sterman, J. (2000). Business dynamics: systems thinking and modeling for a complex world.
Irwin/McGraw-Hill: Boston. 23
APPENDIX — APPENDIX — APPENDIX — APPENDIX
A Mass Starvation in the Oostvaardersplassen ( /25)
The swampy natural reserve the ‘Oostvaardersplassen’ (OVP) ~approximately the area between
Almere and Lelystad in South Flevoland, the Netherlands~ was created when it was d
time after the impoldering- to create a large area where geese could pass the molting season. Large
herbivores were needed in order to keep the area free of willows and other brushwood. Hence, a
small group of Heck cows were introduced in 1983, followed by Konik horses in 1984 and red deer
in 1992. The area was supposed to bring back Dutch primal nature: the area would be left to
nature — no hunting would be allowed.
The three populations of herbivore
as could be expected with herbivores
However, the last couple of y
winter. Movies of many of the animals dying of starvation caused major public outrage and
discussions whether the area should be managed or not and thus whether animals should be shot
or not, and if so, when (just before the winter season or just before dying).
Make a System Dynamics simulation model about the large herbivores in the Oostvaarder-
splassen based on following information.
ided -some
increased prosperous, at the beginning even exponentially,
without natural enemies on such an large pasture area.
a large fraction of the large herbivores did not survive the
Suppose that there were 75 large herbivores in the Oostvaardersplassen in 1985. One could
model the births as the product of the number of large herbivores, the percentage birth rate, the
birth season and the birth randomizer divided by the length of the birth
model the deaths as the product of the number of large herbivores, the percentage death rate, the
death season and the death randomizer divided by the length of the death :
The birth season could be modeled with a pulse train starting in 1982.5 with the length of the
birth season and a frequency of 1 year until the final simulation time. The death season could be
modeled with a pulse train starting in 1982 with the length of the death season and a frequency of
1 year until the final simulation time. Make your model such that the length of the death season
and the length of the death season equal the time step.
eason. And one could
ason.
Suppose that you obtain the information in Figure 18(a) regarding past assessments of two
aspects of the carrying capacity of the area for Heck cows and that you generalize the information
regarding Heck cows to all large grazers in the Oostvaardersplassen as in graph 18(b). Use the
ssments displayed in Figure 18(b) to model the percentage birth rate of the large herbivores
and the percentage death rate of the large herbivores.
ass
so add a birth randomizer di
deviation 0.5 and seed 2 (the seed is a number from which a (pseudo-)random number is gencrated).
And add a death randomizer distributed uniformly between 0.5 and 3 with seed 3.
The information related to the three output indicators of interest (number of herbivores, births
and deaths) still needs to be smoothed for at least two reasons:
ributed normally between 0 and 2 with average 1 and standard
ible to monitor
e Since the OVP reserve is a rather large OERwildlife reserve, it is imp
the exact numbers of births, deaths and large herbivores over time: assessment of these
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society26
Draagkrachtberekening 08
@ Procentuele sterfte o7
50% @ Procentuele aanwas 06
‘i e 05 =
° Pedi
40% |=—* 04
, bee e ? 03 ==
30% er eh 7 02 =<
20%|—# ots 2 04 —
55 : ee|,° © o
10% o e t i e 0 2000 4000 6000 8000}
8 E totaal aantal grote grazers
a 100200300 pulatinn 1 jancor| [ =#—procentuel sterfie _ —=— procerfuele aanw as
(a) Carrying capacity for Heck cows in the OVP ~ Source: (b) Generalized carrying capacity curves for all large
NRC Handelsblad 11/12/2010 herbivores in the OVP
Figure 18: Carrying capacity and gencralized carrying capacity for the OVP
evolutions could, at best, be made based on periodic assessments (right before and right
after the birth season and death season) and some additional calculations/smoothing.
e Rather discrete modelling constructs used in the description above may lead to discrete flow
behaviours that need to be tured back to continuous evolutions.
Add therefore a variable smoothed info on large herbivore births to smooth the birth
to third order exponential smoothing with a delay of 1 year. Do the same for the variables
info on large herbivores and the smoothed info on large herbivore deaths.
according
smoothed
1. (/10) Make a SD simulation model based on the description and information provided above.
2. (/2) Test the model: list two useful validation tests (except sensitivity anal
perform them, and briefly describe results/conclusions.
see below),
3. ( /4) Simulate the model and draw the evolution of the births and deaths (in the same
graph), as well as the number of large herbivores. Simulate the model again with a different
‘seed’ and draw the results. Do this again, and again, and again. Generalise and conclude.
4. (/2) Perform the necessary sensitivity analyses. To which parameters and assumptions/functions
is the model behaviourally sensitive? Briefly describe the analyses performed and draw only
the interesting outcomes.
5. (/4) Make a complete and an aggregated CLD of this model.
a
. (/1) Explain the link between structure and behaviour.
x
. (/2) Policy? Implement it in the model and test it. What is your conclusion and why?
B_ The Bluefin Tuna Files (_ /25)
Tuna experts fear that the Atlantic bluefin tuna may be extinct in few years from today. According
to environmental organisations, the collapse of the East-Atlantic blucfin tuna is imminent as a
consequence of systematic overfishing and illegal catches in the Mediterranean. Hence, their call
for a moratorium in the eastern part of the Atlantic Ocean and the Mediterrancan.
The East-Atlantic bluefin tuna is a migratory predator that commutes between the Atlantic
Ocean and the Mediterrancan Sea. Almost the entire catch is exported to Japan for its thriving
sushi and sashimi markets. The tuna population has been in sharp decline in recent years and
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society27
ion for the
many tuna experts considered the most recent meeting of the ‘International Commi
rvation of Atlantic Tunas’, ICCAT* for short, as the last chance to remedy the situation. ..
inning of the eighties, drastic catch restrictions were agreed upon for the Western part of
the Atlantic Ocean. But those drastic measures were too little too late. The population only
stabili ent of the 1975 level.
ised in the nineties at only 20 perc
It is clear that action needs to be taken now to save the East-Atlantic bluefin tuna. That is
why you are asked to make a SD simulation model concerning this threatened tuna species (from
now simply called tuna).
Fish in the Sea ( /8)
The current tuna fish biomass, estimated to amount to the un-fished tuna biomass® of 100000
tonnes in the year 1990, increases through growth of the current tuna biomass and delayed tuna
recruitment®, decreases through natural tuna fish deaths and tuna fish catches.
The growth of the current tuna biomass is equal to the current tuna biomass times the rate of
tuna growth of ‘adult tuna fish’ of 4% per year. The delayed tuna recruitment equals the current
, but is delayed (exactly) 4
tuna biomass multiplied by the tuna recruitment rate of 1% per ye
years”.
The natural tuna fish deaths is estimated to amount to the current tuna biomass multiplied by
the ratio of current biomass to un-fished biomass over the normal tuna lifetime of 20 years. The
ratio of current biomass to un-fished biomass is simply the current tuna biomass divided by the
un-fished tuna biomass.
‘And tuna fish catches depend on the current tuna biomass and the tuna catch fraction. The
tuna catch fraction depends in turn on the total number of tuna fishing boats and the tuna boat
efficiency of 0.0004% per boat per year. [Assume for now that:] The total number of tuna fishing
boats is composed of 15000 official tuna fishing boats [this constant will be turned into a variable
in the next section] and 10000 illegal tuna fishing boats.
1. ( /6) Make a first System Dynamics simulation model of this issue.
2. (/1) Simulate the model. What happens with a fixed total number of fishing boats of 25000,
15000, and 5000? What would be the total number of fishing boats that would keep the tuna
biomass in equilibrium at the current level? Draw the four results in terms of the current
tuna biomass, both on your computer and on your exam copy.
3. (/1) Write this system as a balance equation (make sure to choose appropriate symbols and
explain their meaning).
Fishery and Fleet Management ( /17)
The International Commission for the Conservation of Atlantic Tunas (ICCAT) is an inter-
governmental fishery organization responsible for the conservation of tunas and tuna-like species
in the Atlantic Ocean and its a , more specifically the Mediterranean. ICCAT fleet size
regulations are among the most important measures for preventing the extinction of the bluefin
tuna homing in the East Atlantic-Mediterrancan. ICCAT’s functioning and policy-making can be
seen as a high-level policy loop. This high-level ICCAT-policy loop may be summarised as follows:
The ratio of current biomass to un-fished biomass is estimated and interpreted with a ‘change
in tuna fishery perception’ function in order to form the latest perception of the tuna fishery
4The ICCAT is an intergovernmental fishery-organisation responsible for the preservation of tuna and tuna-like
in the Atlantic Ocean and bord s, like the Mediterranean.
5 The un-fished tuna biomass is imated tuna biomass without any (pr rent) tuna fishing.
®Recruitment means reaching a certain size or reproductive stage. With fisheries, recruitment usually refers to
the age a fish can be caught and counted in nets.
7Tuna sumed to mature at four years of age in the Mediterranean. It is also assumed here that harvesting
ious 01
tuna under the age of 4 is not interesting from economic and ecological points of view and/or feasible altogether
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society28
status. The ‘change in tuna fishery perception’ function connects following points: (0,-10), (0.25,-
2), (0.5,0), (0.75,1.5), (1,10). This latest perception of tuna fishery status is smoothed into the
ICCAT perceived state of tuna fishery using a time needed to change the tuna fishery perception
of 2 years, starting from the initial ICCAT state of tuna fishery in the year 1990 of 10 (which
corresponds to ‘excellent’).
The official number of tuna fishing boats, initially 15000, increases (and decreases) by means of
the net increase of the official number of tuna fishing boats equal to the proposed change in tuna
fishing boats divided by the time to implement the tuna boat policy of about 2 years.
The proposed change in tuna fishing boats equals the effect of ICCAT’s perceived state of
tuna fishery on the number of tuna fishing boats times the official number of tuna fishing boats.
Given past and expected ICCAT states and decisions, you can assume that this effect of ICCAT’s
perceived state of tuna fishery on the number of tuna fishing boats amounts to -0.9 for ICCAT’s
perceived state of tuna fishery of -10, to -0.5 for ICCAT’s perceived state of tuna fishery of -7.5,
to -0.25 for ICCAT’s perceived state of tuna fishery of -5, to -0.1 for ICCAT’s perceived state of
tuna fishery of -2.5, to 0 for ICCAT’s perceived state of tuna fishery of 0, to 0.075 for ICCAT’s
perceived state of tuna fishery of 2.5, to 0.15 for ICCAT’s perceived state of tuna fishery of 5, to
0.21 for ICCAT’s perceived state of tuna fishery of 7.5, and to 0.25 for ICCAT’s perceived state of
tuna fishery of 10 or more.
1. (/5) Extend the simulation model with the information provided above and save it. Verify
the model briefly. Simulate the model and make graphs of the official number of tuna fishing
boats and current tuna biomass.
2. ( /1) Validate the model. List 2 validation tests (with the exception of sensitivity testing),
perform them and describe the results/conclusions (briefly).
3. ( /2) Test: the sensitivity of the model (more specifically of the official number of tuna fishing
boats and the current tuna biomass) for changes in 3 parameters of your own choice (choose
them well!!!) as well as the effect of ICCAT’s perceived state of tuna fishery on the number
of tuna fishing boats. Briefly describe the tests you performed, as well as your results and
conclusions.
4. (/1.5) What happens to the current tuna biomass and official number of tuna fishing boats
if the number of illegal fishing boats falls due to strict controls and severe punishments-
from 10000 down to 0 in 2010? Apply, rename the model, and draw and briefly ibe the
results. [Preserve this drastic reduction of the number of illegal fishing boats described in
this what-if question for the remainder of the questions.]
5. (/1.5) What happens if countries studiously refuse to scale down their tuna fishing fleets, in
other words, if the net increase of the official number of tuna fishing boats does not become
negative? Apply, rename the model, and draw and briefly describe the results.
6. ( /4) Make an extremely aggregated causal loop diagram of the model to explain the main
feedback loops. Use it to explain the link between structure and behaviour.
7. ( /2) ICCAT policy making is heavily criticized for its unsustainability. Devise 2 feasible
policies to improve the sustainability of the current high-level ICCAT policy-loop. Describe
them, implement them (rename your model), test them separately and (if possible) together,
and briefly describe your conclusions: is this ICCAT+ policy more sustainable?
Fishery and Fleet Management in a Changing World? ( / 4 Bonus)
1. (/2) It is well known that the ecosystem capacity for Tuna keeps on deteriorating (pollution,
overfishing of species predated on by tuna, etc) and that the efficiency of tuna boats keeps on
increasing. Model both evolutions and test the appropriateness of the ‘ICCAT policy’ and
the ‘ICCAT+ policy’ (the ICCAT policy plus the policies devised in the previous question)
given these evolutions. Draw and describe your outcomes and conclusions.
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society29
2. ( /2) From an individual fishery perspective, it makes
nse to catch the fattest (and con-
sequently the oldest) fish. But the older age classes actually contribute more to reproduc-
tion... Adapt the model accordingly. What does this information mean for the effectiveness
of the ‘ICCAT policy’ and your ‘ICCAT+ policy’?
C Case: Real Estate Boom and Bust in Dubai ( /25)
Two weeks ago, Dubai announced that it overcame the crisis which started after Dubai World's
announcement that it had to default on its debt. However, it seems a bit premature to a
sume
that all problems have been solved. Following the real estate bubble burst 11 months ago, the real
estate market is threatened today by permanent lack of occupancy (especially many buildings of
poor quality in the desert).
C.1 Real Estate Sector ( /6)
The phrase ‘Real Estate Unit’ (or REU) is used in the remainder of the text to refer to one
house/appartment or one 1-person office space. Suppose that the RBU supply initially consists
of 1.800.000 of these REUs. The REU supply decreases by means of REU demolition after an
average REU lifetime of almost 42 years (or 500 months).
The REU supply increases through REU commissioning of REU under construction. REU
commissioning normally equals the number of immigrants divided by the product of the REU
construction time and the number of workers per REU under construction of 25 persons per
REU. Note that REU commissioning can never be greater than the REU under construction over
the REU construction time of 3 months. Set the initial value of REU under construction to the
number of immigrants times the REU construction time divided by the workers per REU under
construction.
REU under construction increases by means of new REU plans approved. New REU plans are
of expected REU shortages over an average REU
approval time of 1 month as well as in response to investment desires of the ruling Al Maktoum
family. Suppose that the Al Maktoum family invests an investment ratio of current REU supply
of 1% of the REU supply. Suppose that the official calculation of the expected REU shortage does
not take into account demolition and therefore equals the REU demand minus the REU supply
approved in response to non-negative estimate
minus the REU under construction plus the expected new RBU due to immigration.
1. (/2) Make a $D model of this description.
2. ( /1) How do we call such stock-flow structures?
3. (/3) Make a complete causal loop diagram of this (partial) simulation model.
C.2 Population: locals and immigrants ( /0)
Suppose for the sake of simplicity that locals initially 220.000- do not work as workers (at least
not in the real estate construction business), that all immigrants —initially 2.000.000- ar
the labor market (in other words, immigrants come to Dubai without famili
members are / are not counted as immigrants in your model),
and that all immigrants work in the real estate construction sector.
‘tive on
or inactive family
imply not entered into the statis
The number of immigrants increases through workforce immigration, and decreases through
workforce emigration and through integration. Workforce immigration —which should always be
positive- can be modelled as the relative attractiveness to immigrate times the number of existing
immigrants over the average immigration time of 1 month. The normal workforce emigration
-which cannot become negative can be modelled as the number of immigrants minus the labor
demand, divided by the average emigration time of 1 month. Immigrants can become locals
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society30
if/when they integrate and find a self-sustaining job outside the REU business: this integration
flow amounts to the immigrant integration rate of 0.001 per month times the number of immigrants.
Both immigrants and locals need REUs: their total REU demand is the product of the sum of
these populations and the REU demand per person. Suppose that the REU demand per person
increases linearly from 1 REU per person at the start of the simulation time to 2 REUs per person
at the end of a 20 year time horizon.
C.3 Linking population to real estate to population to
-(/19)
Define labor shortage as the labor demand over the available number of immigrants. Labor demand
is the product of workers per REU under construction and the REU under construction.
Suppose that the average immigrant salary amounts to the labor shortage times a normal
immigrant salary of 1000 dollar per person per month. The relative attractiveness to immigrate is
ly proportional to the average immigrant salary divided by the normal immigrant salary and
sely proportional to the REU price divided by 960. The proportionality coefficient is equal to
1. Dividing by 960 is motivated by the assumption that 75% of the housing cost is subsidized by
the companies and/or the Emirate, and a mortgage with a duration of 20 years can be obtained
for the remaining amount.
The REU price equals the normal RBU cost times the REU shortage price effect applied to
the REU shortage. The normal REU cost amounts to $50.000 per REU (material costs) plus
the product of the average immigrant salary, the REU construction time, and the number of
workers per REU under construction. The REU shortage price effect consists of a curve connecting
following couples (0,0.6), (10,4), (50,7.5), (100,10). REU shortage can be defined as the REU
demand over the REU supply.
The expected new REU due to immigration equals the product of the immigration multiplication
factor of 1 and the difference between the number of immigrants and the number of immigrants
in the previous period. ‘Immigrants in the previous period’ refers of course to the number of
immigrants in the previous time period.
1. ( /5) Extend the simulation model with the information provided above. Verify the model
briefly. Simulate the model and make graphs of the immigrants, the REU supply, the REU
shortage and labor shortage.
2. (/1) Validate the model. List 2 validation tests (with the exception of uncertainty testing)
perform them and briefly describe the results/conclusions.
3. (/3) Use the model to try to simulate the unfolding of the real estate bust after month 10:
Let the Al Maktoum family’s investment ratio of current REU fall instantly from 1%
to 0% at the beginning of month 10 .
¢ Add following non-negative term to the formula of workforce emigration: exogenous em-
igration/ average emigration time that allows you to simulate an exogenous emigration
of 200.000 immigrants in month 10.
Save your model using a new name. Simulate the model and make graphs of the immigrants,
the REU supply, and a combined graph of the REU shortage and labor shortage. Are these
changes enough to generate a real estate bust (collapse)?
4. ( /3) Keep the crisis settings from the previous question. Now, test the influence of the
uncertainty related to the average immigration time —test for instance average immigration
s of 1 month, 2 months, and 3 months~ on the number of immigrants. Make a graph
of the effects in terms of immigrants. Do the same for the uncertainty related to the REU
construction time —test for instance REU construction times of 1 month, 2 months, 3 months,
and 4 months.
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society31
5. (/2) Remove the crisis settings, and test the combined effect of different average immigration
times and REU construction times on the number of immigrants without crisis settings.
Briefly discuss your results and explain these effects and what causes them.
6. (/3) Make an extremely aggregated causal loop diagram of the model without crisis settings
which allows you to explain the main feedback effects in case of an average immigration time
that does not lead to a collapse. Use the CLD to briefly explain the link between system
structure and behavior (in other words: why does the system not collapse?). Now, make a
new aggregated causal loop diagram or adapt the previous one (use a different color) to the
case of an average immigration time of 1 month. Again, use the CLD to briefly explain the
link between system structure and behavior.
7. (/1) Suppose that the ruling family still wants to turn Dubai into the regional capital. The
boom needs to be sustained in order to do so: what do you advice the ruling family to do
-without spending/losing too much money- in order to sustain a continued boom? Limit
your policy advice to two 2 sentences.
8. (/1) This model is just a preliminary model. What would you add/change/...to improve
the model and make it really useful for real-world policy analysis? Don’t do it.
D_ De/Radicalization ( /25)
For this case description, readers are referred to (Pruyt and Kwakkel 2011).
E_ Energy Transition towards Sustainability ( /25)
For this case description, readers are referred to (Pruyt, Kwakkel, Yucel, and Hamarat 2011).
F The ‘Slow Students Fine’ Case ( /25)
A mass demonstration was organised on 21 January 2011 ~a few days after the SD exam- in order
to demonstrate against proposed legislation to fine ‘slow students’ and univers hing to
‘slow students’. Students were asked to model the potential consequences for our faculty based on
the description below.
The BSc Student ( /10)
First, model the inflow of BSc students. Annually, there is an annual inflow in the BSc studies
at the BSc inflow moment. Suppose for reasons of simplicity that this inflow moment happens
once a year ~ use a PULSE TRAIN(start, width, tbetween, end) with a width equal to the time
step. Model the annual BSc inflow as the evolution of the new BSc inflow divided by the time
step times the BSc inflow moment. Suppose that the evolution of the new BSc inflow gradually
increased from 20 new BSc students in 1990 to 90 new BSc students in 1995 to 120 new BSc
students in 2000 to 130 new BSc students in 2008 to 200 new BSc students in 2010 and that it
stabilizes at 200 until the year 2030. The real inflow of BSc students is then the product of the
annual BSc inflow and the quality (the lower the quality, the lower the inflow will be). For now,
set the quality equal to 100%.
The inflow of BSc students is added to the group of BSc students. The group of B.
decreases through the outflow BSc students when/if students obtain their BSc or as BSc quitters.
Model the outflow of BSc quitters simplistically (but not entirely correctly) as the fraction of BSc
quitters times the BSc outflow after fixed and additional delay. Suppose that 30% of the studen
quits the first year, 10% the second year, and 5% the third year. The fraction of BSc quitters
s between 0 and 1 is then the sum of these quit fractions divided by the quality (the lower
students
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society32
the quality of the studies, the more quitters). Those who do not drop, obtain their BSc diploma,
eventually: the outflow of BSc students then equals the BSc outflow after fixed and additional
delay multiplied by the complement of the fraction of BSc quitters. Model the BSc outflow after
fived and additional delay as the first order delay of the ‘BSc outflow if only minimal fixed delay’
with a total delay time equal to the product of the minimal BSc study time of 3 years and the
additional yearly delay BSc of (on average) 50% divided by the quality. And model the variable
‘BSc outflow if only minimal fixed delay’ as the delay of the inflow of BSc students with a fixed
minimal BSc study time of exactly 3 years.
1. (/7) Make a SD model of the description above.
2. ( /3) Make a complete Causal Loop Diagram (CLD) and a strongly aggregated Causal Loop
Diagram of this partial simulation model.
The MSc Student ( /3)
Model now the throughput of MSc students: almost the same applies to MSc students as to BSc
students. Following details are different:
The inflow of MSc students equals the quality of the education times the sum of the annual
MSc inflow of new students (not flowing semi-automatically from the BSc studies) and the product
of the outflow of BSc students and the fraction of BSc students that flow semi-automatically from
the BSc to the MSc. The evolution of the new MSc inflow was 0 students per year until 2007,
started with 2 students per year in 2008, rose to 5 students per year in 2010, and is assumed to grow
to 15 students per year in 2015 and 20 in 2020 after which it is assumed to remain constant. The
fraction of students that flows semi-automatically internally from BSc to MSc was about 100%
before the year 2008 — suppose that it fell to 80% of the students in 2008 and afterwards. The
minimal MSc study time is equal to 2 years. And the fraction of MSc quitters ~always between 0
and 1- is lower too: 10% in the first year and 10% in the second year. In summary: the structure
of the MSc students submodel is the same as the BSc students submodel — a hand full of new MSc
students is absorbed by a larger —maar decreasing~ group of students flowing semi-automatically
from the BSc to the MSc.
1. ( /3) Extend the SD model with the description above.
The Faculty ( /6)
The quality is a function of the professor hours per student: if the number of professor hours per
student is 0 then the quality is 10%, if it is 50 then the quality is 60%, if it is 100 then the quality
is 90%, if it is 150 or more then the quality is 100%.
Model the professor hours per student as a third order delay of one year of the product of 1000
hours per professor and the number of professors divided by the total number of students. Make
sure in the previous formula that the denominator cannot become 0.
Model the number of professors ~initially 5 in 1990- and the increase and decrease of the
number of professors in a rather simplistic way: suppose that the net hiring of professors equals
the difference between the maximum number of professors and the number of professors, divided
by the average hiring time. The hiring time (and here firing time) for (good) prof
long — on average 2 years from the moment a new professor is actually needed. The maximum
number of professors then equals the amount of money available for education divided by the
average cost of a professor of €100000 per professor per year.
The amount of money available for education —initially 0- is increased by the inflow of money
available for education and decreased by the outflow of money available for education. Without a
fine for slow students, the outflow of money available for education approximately amounts to the
number of professors times the average cost of a professor.
The fraction of slow students seems to be —at least partl:
education: if the quality is 0% then the fraction of slow studen'
is rather
a function of the quality of the
is 90%, if the quality is 25%
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society33
then the fraction of slow students is equal to 85%, if the quality is 50% then the fraction of slow
students is equal to 66%, if the quality is 75% then the fraction of slow students is equal to 40%,
if the quality is equal to 100% then the fraction of slow students is equal to 25%.
The inflow of money available for education equals the subsidy per new BSc student times
theinflow of BSc students plus the subsidy per BSc graduate times theoutflow BSc students plus
the subsidy per new MSc student times theinflow of MSc students plus the subsidy per MSc
graduate times the outflow of MSc students plus the annual lump sum and other subsidies. The
subsidy per new BSc student amounts to €15000, the subsidy per BSc graduate €5000, the subsidy
per new MSc student €5000, and the subsidy per MSc graduate €5000. Suppose that the annual
lump sum and other subsidies for educational purposes amount to an additional €1 million.
1. ( /3) Extend the SD model based on the description above.
2. (/3) Simulate the model without fines for slow students from the year 1990 until the year
2030. Make graphs of following variables: BSc students and MSc students, outflow of BSc
students and outflow of MSc students, professors, and amount of money available for educa-
tion. Is the faculty healthy without the proposed system of fines?
And now with fines for slow students...( /9)
But what does the proposed system of fines for slow students mean for the faculty? Model therefore
the system of fines as follows. With the system of fines the outflow of money available for education
becomes fraction of slow students times the total number of students times thefine per slow student
plus the number of professors times the average cost of a professor. Anticipating (at least) one
year delay caused by opposition and demonstrations, you can assume that fines for slow students
will only be introduced from 2012 on. From then on, the fine per slow student would amount to
€3000 per year. This is of course only part of the picture: increased tuition fees to be paid by
slow students are not taken into account here.
1. (/3) Extend the SD model with the description provided above.
2. ( /2) Simulate the model with the system of fines. Make graphs of following variables:
students and MS¢ students, outflow of BSc students and outflow of MSc students, professors,
and amount of money available for education. Is the faculty healthy with the proposed system
of fines?
oo
. (/1) What if the number of new students increases until 2020 to 300 BSc students per year
and to 60 new MSc students per year?
4. (/1) The cabinet is furious: they claim your model is totally wrong because the outflow of
money available for education still needs to be divided by a factor 2.5 (the average number
of study years of BSc and MSc). Is that correct? What would be the consequence?
5. (/1) After the previous proposed correction has been made, the cabinet now argues that the
LOOKUP function of the fraction of slow students needs to be adapted too — the cabinet
assumes after all that students will study faster in the new system. Change the lookup,
discuss the new function and the consequences of this change for the faculty.
6. (/1) It should be clear that this
more realistic and better? De
changes/corrections.
model requires further adaptatio
ibe what and how, do it, and de
: how would you make it
be the results of these
G Case: Scarce ‘Rare Earths’ ( /25)
From Extraction and Processing to Production of Goods ( /4)
Rare Earths Metals (REM) are often used in very small quantities in modern appliances or appli-
cations. Assume that the REM in goods initially amounts to 2000000 ton. REM in goods are lost
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society34
after an average lifetime in goods of some 15 years (recycling of REM is (currently) not feasible).
The use of REM in the production of goods depends on the demand for REM or the available
supply of REM ~initially equal to 250000 ton- if the available supply of REM is smaller than the
demand for REM. The available supply of REM only increases through processing of REM which
in turn follows the real REM extraction.
1. (/1) Make a SD model of the description above.
2. ( /3) Make a complete Causal Loop Diagram (CLD) and a strongly aggregated Causal Loop
Diagram of this partial simulation model.
Demand and Supply of REM ( /4)
The demand for REM —initially equal to 100000 tons in 2000- increases in principle by means of
an increase of the demand for REM and decreases through a decrease of demand through price
elasticity of demand of door substitution losses
The increase of the demand for REM is simply the product of the economic growth rate and
the demand for REM. Suppose for reasons of simplicity that the economic growth rate was equal
to 3% until the year 2009, that it fell to -10% in 2009, and that it jumped to 8% in 2010. Assume
that the economic growth rate remains constant at 5% from 2011 on.
The decrease of demand through price elasticity of demand could be modeled as:
1 1
ia of demand xdemand for REM + ee — relative pie of the previous year
Felative price of the previous year
—price elas
With the relative price equal to the product of the average REM extraction costs, the scarcity
price effect and (1 plus the normal profit margin of 15%). Suppose that the price elasticity of
demand is 10%.
If the relative pr
the substitution los
greater than the relative price of the cheapest price substitute then
crease to the product of the demand for REM, —ar-
difference between the relative price and the relative price of the cheapest pric . The
substitution losses due to price substitution effects are in other words non-negative. Suppose that
the relative price of the cheapest price substitute is constant at 100 (in other words 100 times the
normal price of REM).
Suppose also that the scarcity price effect amounts to 100 if the supply demand ratio is 0, 10
if the supply demand ratio is 0.55, 1 if the supply demand ratio is 1.1, 0.75 if the supply demand
ratio is 2.2, 0.5 if the supply demand ratio is 11, 0.2 if the supply demand ratio is 22. The supply
demand ratio is equal to the available supply of REM divided by the demand for REM.
and the
i
1. ( /4) Extend the SD model following the description provided above.
Commissioning and Decommissioning of Extraction Capacity ( /17)
Suppose that the mining industry is myopic and has limited foresight: the desired extraction
capacity is then equal to the demand for REM. The newly planned extraction capacity then equals
the product of the profitability of REM extraction between 0 and 1 and the difference between
the desired extraction capacity and the installed extraction capacity (Note: the value of the latter
ily lies between 0 and the value of the installed extraction capacity).
The newly planned extraction capacity increases the extraction capacity under construction
“initially equal to 60000. The e.
sioning of extraction capacity which exactly delays the newly planned extraction capacity with the
difference nec
raction capacity under construction decre
s through commis-
precise construction time of extraction capacity of 8 years. The commissioning of extraction capac-
ity initially equals the extraction capacity under construction divided by the precise construction
time of extraction capacity.
Pruyt, 2011. Making System Dynamics Cool III. In: Proc. of the Int. Conf. of the SD Society35
The commissioning of extraction capacity leads of course to an increase of the installed extrac-
tion capacity, initially equal to 100000. The installed extraction capacity decreases on the one hand
through decommissioning of extraction capacity and on the other hand through decommissioning
of unprofitable extraction capacity. Model the decommissioning of unprofitable extraction capacity
as follows: installed extraction capacity * (-MIN(profitability of REM extraction,0)).
The decommissioning of extraction capacity then equals the installed extraction capacity di-
vided by the average lifetime of extraction capacity minus the decommissioning of unprofitable
extraction capacity. Note that formula of the decommissioning of extraction capacity needs to be
. Set the average lifetime of extraction capacity at 30 year.
The maximum REM extra squal to the installed extraction capacity. The real REM
extraction normally equals the installed extraction capacity, wn scarcity price effect is
smaller than 1, then it equals the installed extraction capacity times the scarcity price effect.
Cumulation of the real REM extraction gives the cumulatively extracted REM —initially equal
to 4000000 ton. The difference between the cumulatively extracted REM and the initial cumula-
tively extracted amount is needed to calculate the average REM extraction costs. To do so, use
a function with the difference between the cumulatively extracted REM and the initial cumula-
tively extracted amount as argument and following couples: (0, 1), (2.000.000, 2), (4.000.000, 4),
(6.000.000, 8), (8.000.000, 16), (10.000.000, 32), (12.000.000, 64), (14.000.000, 128), (16.000.000,
256), (18.000.000, 512).
Finally, the profitability of REM extraction is equal to the difference of the relative price and
the average REM extraction costs, divided by the average REM extraction costs.
non-negat
tion is
1. (/7) Extend the SD model with the description provided above.
2. ( /3) Perform the necessary verification and validation.
3. (/1) Extend the model: model the ‘intrinsic demand’ ~ in other words, the demand in the
absence of decrease of demand through price elasticity of demand and substitution losses —
and make an output indicator ‘fraction produced of intrinsic demand’ that allows to visualize
the use of REM in the production of goods in function of the intrinsic demand over time.
4. ( /1) Simulate and draw the behavior of the output indicator, of the demand for REM, of
the installed extraction capacity, the scarcity price effect and the relative price.
5. (/1) What happens if the ‘initial extraction capacity under construction’ is 0? Compare,
draw and conclude.
6. (/1) What happens if -on top of the previous what-if- the ‘economic growth rate’ amounts
to 3% from 2011 on? Compare, draw and conclude.
For which parameters and functions is the model
e? Draw (only) interesting outcomes.
ity anal:
7. ( /2) Perform a s
behaviorally sensi
8. (/1) Make an aggregated CLD and explain the link between structure and behavior.
