Using Scenario Planning Data in System Dynamics Model Building
Ezzat El Halabi, Matthew Doolan
The Australian National University
Research School of Engineering, Building 32
North Road, Acton, ACT, 2601 Australia
Phone +61 2 6125 5132, Fax +61 2 61252739
ezzat.elhalabi@anu.edu.au, matthew.doolan@anu.edu.au
Abstract
The main contribution of this paper is describing a traceable data analysis procedure for
modelling scenarios using the results of a simplified Scenario Planning workshop. We first
introduce our project where the ultimate goal is to develop a grounded model-based policy
discussion tool for the Australian automotive recycling industry. We review current scenario-
based model building techniques and highlight their practical gaps and then present the
proposed procedure rooted in Qualitative Data Analysis approaches. We show how to update
the Causal Loop Diagrams and Stock and Flow models and how to determine scenario
conditions thus enabling a clear record of model building. Using an example from a real
project, we highlight the main challenges of the procedure that are dealing with data scarcity,
estimating new trends for variables, and deciding on the nature of the changes in the
simulation models. The paper concludes by arguing how the procedure may benefit system
dynamists that require a coherent and structured modelling trail or when it is more feasible to
engage the stakeholders in an abbreviated Scenario Planning workshop instead of Group
Model Building.
Key words
Model Documentation, Model Building, Qualitative Data Analysis, Scenario Planning,
Scenarios
Introduction
One of the main challenges in applying System Dynamics (SD) is documenting the model
building process, especially when grounded models need updating when new information
becomes available. During the past two SD conferences we noticed at several SD model
presentations that the audience inquiries centred on model usefulness almost overlooking
the lack of detail on how the models were developed. While acknowledging the importance
of model usefulness‘, we postulate that maintaining a visible and traceable approach is at
least as important, specifically in grounded applications requiring stakeholder engagement.
In this paper we focus on modelling scenarios, an integral component of SD model
development, determined using the output of a Scenario Planning (SP) workshop involving a
group of stakeholders.
Our project aims to develop a model-based policy discussion tool for the Australian
automotive recycling industry using SD as a guiding framework. When trying to maintain a
transparent modelling approach, we realised that the current model-building procedures are
far from optimal and therefore needed adaptation. Our original intention was to conduct a
series of Group Model Building (GMB) workshops but we soon realised that the proposition
was not logistically or financially feasible. It was too difficult to get a group of stakeholders in
the industry to commit several days of their time to GMB activities as they come from
different enterprises spread across Australia.
Thus, we conducted a series of stakeholder interviews in late 2010/2011 to gather business
process data and capture the underlying decision frameworks. Then, in order to focus the
modelling effort on the most relevant problems, interview data was systemically analysed
and aggregated using an adapted Qualitative Data Analysis (QDA)? approach (El Halabi et al.
2012). Preliminary Causal Loop Diagrams (CLD) were developed for the five emerging focus
areas (El Halabi and Doolan 2012) followed by the Stock and Flow models (SFD). Last year we
+ Within the context of grounded model building, useful models need to be relevant to the stakeholders while
addressing a relevant problem context.
2 QDA is a broad term for a method of systemic inquiry into qualitative data with the purpose to gain insights.
For more information refer to Richards (2009).
designed and facilitated a SP workshop with a group of auto recyclers and representatives
from industry associations (El Halabi et al. 2013) so that we could identify the model-relevant
scenarios and determine the influences of each scenario on each of the five areas of the
model.
The paper is structured as follows: We overview relevant literature while highlighting the
practical gaps. We then present the QDA-based procedure along with an example of its
application focusing on a subset of the SD model ‘Workforce’ in one of the identified scenarios
‘How it should be’. In the discussion section, we contrast the procedure with the one
presented in (El Halabi et al. 2012). We then talk about the pitfall we faced when devising the
procedure that led us to structure in the way presented. We also explore the commonality of
bias observed in three key implementation challenges. We finally emphasise the
transferability of the procedure to other areas of SD model building such as policy modelling.
A Review of Current Techniques
We searched the literature for a detailed and clear procedure we could follow for using
scenarios workshop outcomes to further develop our SD models. To our disappointment we
could only find references to overall approaches where the specific procedures are either
overlooked or simply assumed. In this section we provide an overview of relevant literature
to highlight the technical gaps in current approaches.
Within the realms of SD literature, Maani and Cavana (2007) adopt Schoemaker's (1993) list
for building scenarios in their Systems Thinking and Modelling method but do not
demonstrate the technical aspects of integrating the scenarios elements into the SD models.
Heijden (2011), while using SD within the context of quantifying the scenarios and gaining a
better understanding of the scenarios, does not detail the technical aspects either. It is a
similar story with Belt (2004) who employs SD, scenarios, and other approaches into her
holistic Mediated Modeling paradigm. Still on the same path, Alcamo (2008) in the
Environmental Scenario Analysis approach borrows from SD to help quantify the scenarios
but falls short of demonstrating the procedure.
Most recently, Morecroft (2007) showcases how SD models can be run through different
scenarios by changing variable values to challenge existing mental models of the users, but
does not touch upon revisiting and revising the structures that underpin the models. Stowell
and Welch (2012) refer to the importance of modelling scenarios further reiterating
Forrester's emphasis on the usefulness of SD in decision making (Forrester 1968) but do not
provide the required how-to detail. Most relevant to our work, Olabisi et al. (2010) present a
real world example of using scenarios insights in SD modelling in a participatory setting.
Although the authors discuss interesting scenarios/models, they do not share sufficient detail
on how the models were updated.
In broader non-SD contexts, still within the systems thinking literature and relevant to our
work, Williams and Hummelbrunner (2010) talk about a generic scenario technique adopted
from (Schwartz 1996; Heijden 1996; et al.). They give an example of a real world application
that goes to the level of systemically assessing influences on factors and estimating trends.
While we acknowledge its practical implications, applying or transferring this technique into
a SD context is not immediately obvious.
Furthermore, observing SD literature from a lateral view point, we notice how the term
‘scenario’ is employed broadly without common adherence to a norm. The term which
appears under sensitivity analysis (Sterman 2000; Morecroft 2007; et al.) describes the
process of varying the values of the variables to gauge model response under different
circumstances. Forrest (1998) makes a theoretical distinction between the terms sensitivity
analysis and scenario analysis in SD and indicates that the latter should only be used when
different futures? are modelled through different structures. Similarly we refer to the futures
determined through the SP workshop as scenarios. And while different futures may be
modelled by changing the values of the variables, we believe a more consistent and thorough
grounded approach to modelling would be to analyse and update the underlying models
(CLDs and SFDs) where appropriate.
3 Futures refers to plausible future scenarios that the model can be subjected to.
Building on the above, we can safely deduce that while most SD literature does a good job
presenting the overall guidelines and emphasizing the importance of scenarios in modelling,
there exists a practical gap for developing scenario-based models using SP data. The QDA-
based procedure we propose below aims to address this gap. We would like to emphasize its
work-in-progress nature. By presenting it we hope to generate discussion among SD
practitioners and interested researchers for feedback and improvement.
Proposed Procedure
It is important to highlight the two required input elements for our procedure (Figure 1). The
first one is the set of grounded? CLDs and SFDs (along with the list of the underlying variables
and their causal links) used to define the focus areas of the model. The second element is the
SP workshop? data containing the identified scenarios and the impacts of each scenario on
each sector of the model as noted by the participants (Appendix 1 for a template of the
scenarios matrix).
Scenario
CLDs and SFDs Planning
Workshop Data
,
Z
Proposed Procedure
7
\
& y
Scenario Based
CLDs and SFDs
Figure 1: Inputs and Output of our Proposed Procedure
The proposed procedure is as follows:
4 Developed from stakeholder interviews data like in (El Halabi et al. 2012) but could come from other
processes like GMB.
5 Assuming the workshop uses pre-identified focus areas to guide the SP activities and discussions. The SP
results and transcribed participant notes are summarised in a table, presenting the supporting trends and the
effects of each scenario on each area of the model.
For each scenario, create a tabulated memo of all areas of the model® in rows, and the
following headings in columns (Observation, Identified Variables, Causal Links Updates, CLD
Updates, SFD Updates, Justifying the SFD Updates, Main Variables Behaviour, Threads for
Future Investigation). Then address’ the headings as per the instructions in Table 1.
Summarize the transcribed notes by reiterating the following
steps:
1- Review and analyse the notes given by participants and
existing CLDs/SFDs.
Observation 2- Articulate the effect of the scenario on this area of the
model.
3- Analyse the ramifications of changes within this area to
other areas of the model.
Identified Variables Add newly identified factors (with units) that can capture the
changes.
Causal Links Updates
From the list of variables identified in this area and the
current CLDs indicate the updated relationships using a
simple one way causality notation. Use the approach ®
presented in (El Halabi and Doolan 2012).
CLD Updates
Shortlist the changes in the CLD and verify whether loop
polarity is affected.
SFD Updates
Shortlist the changes in the SFD (adding
convertors/flows/stocks, altering connectors, modifying
values of transfers/flows, and/or updating the
equations/values).
Justifying the SFD Updates
Justify the choice of these particular SFD updates over
others. Indicate other possible updates and the reasons for
not choosing them.
Main Variables Behaviour
Estimate’ the trend (increase, decrease, level/no change) for
each updated/new variable in the SFD variable. If needed
specify the characteristic of the trend (rebound, oscillation,
exponential).
Threads for Future Investigation
Note any observation or theory that could be tested when
modelling.
Table 1 Analysis Instructions for Headings
Once the data analysis is complete, update the CLDs and SFDs with the new information by
creating a new version for each scenario.
© Refer to Appendix 1 for the memo template.
7 Based on the coding process from Richards (2009).
8 Adapted from Vennix (1996).
° Borrowed from the ‘Trend Projections’ step in the scenario analysis technique presented by Williams and
Hummelbrunner (2010).
An Applied Example
To help put steps 2 and 3 into perspective, an example taken from our project is presented:
the ‘Workforce’ sector under scenario B ‘How it should be’. The reason for focusing on this
subset is because it was used as an example in (El Halabi et al. 2012) to demonstrate the
approach presented in that paper. It is worthwhile to note that the CLD and SFD shown here
as inputs are updated and simplified versions of the ones that feature in (El Halabi et al. 2012).
Automotive Recycling Industry
Model/Focus Areas
ELV Parts/
Sirsa Materials
Industry Demand
Image
Premises Workfroce
Figure 2: The Five Focus Areas Identified from Interview Data
National enforced
licensing
Scenario A Scenario B
Smart Auto Waste How it should be
Throw-away Fully
: ~ resalable
vehicles .
vehicles
Scenario D Scenario C
Big Drama Wild West
Loose fragmented
licensing
Figure 3: Four Scenarios Identified during the SP Workshop
Furthermore and for the purpose of keeping this example clear, this sector is treated as
standalone while ignoring some relationships that are linked in other sectors of the overall
model (Figure 2). We designed the SP workshop to meet a strict timing (three hours) and
participant involvement constraints. We facilitated° the SP workshop in October 2012 in
Sydney with a group of eleven stakeholders comprised of auto recyclers and industry
1° Refer to (El Halabi et al. 2013) for more details about the workshop and processes followed.
associations’ representatives from around Australia. Four scenarios were identified during the
workshop (Figure 3). The participants, working in groups/pairs, discussed and noted the
influences of each scenario on each of the five focus areas"! identified earlier (Figure 2).
Required inputs
The first set of input is the CLD and SFD for the Workforce sector developed from stakeholder
interview data as per the method presented in (El Halabi et al. 2012). The input data’? is
shown in Table 2 on the left while the output of the procedure is on the right.
CLD and SFD Input Output Under Scenario B
Business ae ees + Marginal
Turnover Workforce es Workforce Costs
“, Sie Turnover Workforce —
yt 7, Size \
. + / %
Business 4B? Workforce ( ws
Profits ” Costs Business = ee
costs
a , Profits
See NS
wo Workforce 8 FE) Wieritorce 6
Workforce Size Workloree Sie
_ = = Fe ( rx 1
ve | ne ammeeenc G va . 1
| «
3 ? Workistoe Costs
{ © Ware cot Licdiisoe] C5
Industry Tumover a es abi
é SS ee Industry Prot Werte Costs
Industry Profit Workiroce Efficiency ne
Worktrose Efficiency
Table 2 CLD/SFD Input and Output of our Procedure
The second set of input is the SP workshop data. These include details of the scenarios
identified through the workshop along with the transcribed stakeholders’ notes delimiting
the effects of the scenarios on each area of the model. For the sake of simplicity we select
Scenario B for the labour related (Workforce) focus area:
o Scenario Title (chosen by the participants): “How it should be”.
o Dimensions: “National enforced licensing + Fully resalable vehicles”.
+1 Refer to (El Halabi and Doolan 2012) and (EI Halabi and Doolan 2013a) for more details about these areas.
22 Refer to Appendix 2For data descriptors including causal links, model equations, and model response.
o Stakeholder notes for the effects of this scenario on the Workforce sector:
More staff required; higher wages attracting people to industry.
Applying the procedure
The resulting analysis for each heading is shown in Table 3.
Participants indicated that due to higher demand for used
parts and the availability of ELVs on the market, the industry
will need to grow its workforce to cope with the perceived
Observation increased demand and might have to increase wages in order
to attract more people to the industry. The effect of
significantly increasing wage costs may impact on industry
profits in the long run.
Identified Variables Marginal Workforce Costs (percentage).
Causal Links Updates Add Marginal Workforce Costs, all other polarities
unaffected.
Shortlist the changes in the CLD and verify whether loop
CLD Updates polarity is affected.
Add Marginal Workforce Costs; Link into Workforce Costs;
Change Workforce Costs equation by adding Marginal
Workforce Costs to the Workforce Size; Modify ‘Hire’ Flow
formula to match the expected behaviour.
SFD Updates
The participants indicated that stronger demand for used
parts will drive the industry to grow its workforce by hiring
more staff. A similar Workforce Size response can be made
through changing the equations in the ‘Dismiss’ flow and/or
the ‘Workforce Efficiency’ transfer but these changes cannot
be explicitly grounded in the supplied data.
Justifying the SFD Updates
Hire: Increase, Dismiss: Decrease, Workforce Size: Increase
Main Variables Behaviour (rebound), Workforce Costs: Increase (rebound).
Effect of Marginal Workforce Costs on industry profits and
Threads for Future Investigation sustainability.
Table 3 Resulting Analysis for Workforce Sector under Scenario B
Discussion
In this section the key challenges faced when devising and implementing the current
procedure are highlighted before discussing the similarities and differences between this
procedure and the one presented at last year’s conference (El Halabi et al. 2012) and finally
the transferability.
Devising and Implementation Issues
When devising this process we initially attempted to address each area of the model
separately across all scenarios. The analysis was spread over five memos, one for each focus
area with all four scenarios addressed within each memo as rows within the table. While in
theory this may provide a coherent snapshot on the impact of different scenarios on one area
of the model, in practice it proved otherwise: it was difficult to update variables in other areas
of the model under the same scenario as they were spread across five separate tables. The
end result for overcoming this pitfall was the proposed procedure (i.e. each scenario on a
memo covering all model areas).
Moving onto the implementation of the proposed procedure, three key issues were faced.
The first one is dealing with the insufficiency of SP workshop data that, hardly a limitation of
the procedure itself, may be attributed to the workshop process including design and
facilitation. There are two dimensions to this issue. The first one is the amount of information
that the participants communicate back during the workshop using compact sticky notes that
forced them to express their ideas succinctly. The other dimension is data depth. With
moderating the workshop activities under strict timing conditions, participant groups/pairs
were not able to engage in lengthy discussions to fully explore the influences of their assigned
scenario on each focus area. We were left with some cells in the scenarios matrix having too
little (single word) or no data (in one instance). We resorted to the data provided in other
cells/areas and to the video recording of the group discussion to be able to deduce the effects
for the areas lacking the data. In order to maintain a coherent trail of emerging ideas, we also
made explicit the assumptions in the observation cell. This neither eliminated the possibility
of bias nor of reaching inaccurate conclusions due to the reliance on our understanding (i.e.
mental models) and interpretation of other parts of the data.
Another issue when applying our procedure is with estimating the trends for variables. Similar
to the first issue and because not all the required data was available, we had to rely on
deductions made from other focus areas to help determine the expected behaviour of
variables. More importantly, and on a separate level, we found that merely describing the
behaviour of an important variable as increasing or decreasing was insufficient to envisage its
response in the model. Referring to the example in the base scenario, the main stock
‘Workforce Size’ was decreasing. We learnt through the workshop data and resulting analysis
that it would increase under scenario B. But how can a decreasing variable be made to
10
increase? A solution was to introduce a trend descriptor to better communicate the response
of the main stock under this scenario (i.e. Workforce Size will rebound).
The last key implementation issue is dealing with the uncertainties when deciding on the
updates needed in the SFDs. To reiterate, our approach is to develop scenario-based SD
models by systemically analysing the SP workshop data and determining, while documenting,
the required changes to the CLDs/SFDs. In the case of the SFDs it is a triple-edged problem
because the modeller has to figure out, not only the relevant components to change, or the
ranges of values to use, but also the appropriate mix of changes in order to get the desired
model response. In the case of GMB the issue will still be present but dealt with by getting
the participants to reach consensus on the required model changes. In our case, however,
there is no access to the same group of participants that produced the data. The proposed
procedure, while attempting to provide more visibility on the modelling process, circumvents
the problem by having the modeller explain the decision behind the actioned SFD updates
over those not appropriated.
To help illustrate the problem, we refer again to our example where the ‘Hire’ flow equation
was updated to induce the required rebound behaviour in the ‘Workforce Size’ stock. The
equations in the ‘Dismiss’ flow and/or the ‘Workforce Efficiency’ transfer could have been
modified instead and would have resulted in a similar response for the main stock. The
decision to modify the ‘Hire’ variable is justified by the adherence to the workshop data where
the participants indicated the industry will grow its workforce in response to stronger
demand/turnover. A different interpretation, but still valid one, could have been that the
actual dismissal rate would greatly reduce (to zero) as the industry holds off dismissal. This
interpretation ensues an update of the ‘Dismiss’ flow instead of the ‘Hire’ flow. Dealing with
the problem of deciding on and justifying the SFD updates begs the question of how far data
interpretation should go to. An interesting answer, though not optimal, comes from to the
List Extension Method (Coyle 1996), which attempts to coherently identify the influencing
factors of a problem: the bounds of the interpretation and analysis is reached when we start
dealing with exogenous factors. In our example, either interpretations subsume endogenous
factors.
11
From a ‘scientific method’ viewpoint we can see that the problem overarching the three
discussed issues is the introduction of bias. We acknowledge and emphasize, however, that
the purpose of SD modelling is to create useful (Sterman 2000) and relevant models, not
replicable ones. The proposed procedure accords more weight to having a transparent and
well documented SD modelling process. After all, the SP workshop data used as input is based
on the mental models of a select group of stakeholders. Discounting the effects of different
facilitators and a non-standard SP process, another group may have discussed different
scenarios and generated different data even if the overall guiding theme was the same.
Furthermore the SP workshop process, being a group activity, is not immune to the pitfalls of
group facilitation observed in GMB such as groupthink, social loafing, and competition
(Vennix 1996). In short, while the introduction of bias may be of concern, it is not specific to
the presented procedure but rather to both SD and QDA paradigms that procedure is based
on.
Comparison with Procedure from Last Year
Both procedures serve the goal of building transparent grounded SD models. They are also
similar in terms of reiterating between steps, addressing areas sequentially, and in the use of
memos and tables. From a scenarios perspective, the previous approach could be seen as
developing SD models for the base scenario, hence the similarities. There are several key
differences however, summarised in Table 4.
| Procedure presented last year Procedure presented in this paper |
| | Extracting causal links from interview | Updating SD models using SP data |
Purpose
data
| Input | Interview transcripts | CLDs/SFDs and SP data |
| Analysis | Organised by theme of questioning Grouped by scenario and organised by |
focus area
| Output | CLDs for focus areas | Scenario-based CLDs/SFDs for focus areas |
Transferability
Finally, in terms of transferability of the procedure, it must be noted that the instructions are
generic enough and therefore are applicable using different input data sets. The assumption
here is that the SP data should result from a workshop where the pre-identified focus areas
are used to guide the discussions and activities. To help explain this further, let us assume
12
that we want to apply our procedure to the same example but using data from a differently?
designed SP workshop where the ‘Workforce’ sector does not feature as an area guiding the
discussions. Executing our procedure would prove troublesome as more data interpretation
become needed to mesh the influence areas, identified in the SP workshop, with the
‘Workforce’ area. Questions will arise about whether this sector is worth modelling in the first
place, whether the interview data was misinterpreted to give it so much importance, and
whether the workshop participants may have simply overlooked it as a result of the
aforementioned group facilitation pitfalls. Still on the topic of transferability and more
substantially, our procedure can also be used to study the impact of policies/strategies on
different areas of the SD model. Using the same CLDs/SFDs as the first input, the second input
can be a set of policies/strategies identified either through stakeholder engagement or from
anecdotes in literature (Figure 4). The effects on the SD model can then be analysed by
following the same instructions“* of our procedure.
Policy/Strategy
CLDs and SFDs Data
/
Z
Proposed Procedure
7
\
% Z
Scenario Based
CLDs and SFDs
Figure 4: Another Potential Use for Our Procedure
Conclusion
There is an increasing need for more transparent and structured procedures enabling the use
of SP in SD modelling. Similarly, there is a drive to document the SD model building process
whether motivated by client requirements or by system dynamists wanting to develop
3 The SP workshop activities would rely purely on the participants to structure the areas of influence instead
of using the areas of influence identified from interview data.
4 In the instructions of our procedure, replace the term ‘scenario’ with ‘policy/strategy’ where applicable.
13
credible grounded models. The proposed procedure in this paper aims to address a practical
gap in the literature and to provide practitioners with a useful tool for developing SD models
using SP workshop data while enabling a coherent and structured modelling trail.
The real world application presented in the paper indicates that the procedure can be used
in contexts where it is more practical to rely on the data of an abbreviated SP workshop than
having to conduct a series of costly GMB workshops. We argue that despite its shortcomings,
the procedure is generic enough to transfer to other applications such as policy analysis.
Future work will focus on improving the procedure, specifically when dealing with SP data
scarcity, and handling uncertainties when updating the SD models.
Acknowledgements
This original research was proudly supported by the Commonwealth of Australia through the
Cooperative Research Centre for Advanced Automotive Technology (AutoCRC) and the
Australian National University (ANU). We thank the Victorian Automotive Chamber of
Commerce (VACC), the Auto Parts Recyclers Association of Australia (APRAA), and the Auto
Recyclers Association of Australia (ARAA) who helped putting us in touch with most of the
stakeholders. We also thank Charles Featherston for contributing his time and expertise to
help us design and facilitate the SP workshop.
14
Bibliography
Belt, M. van den. 2004. Mediated Modeling: A System Dynamics Approach To Environmental
Consensus Building. \sland Press.
Coyle, R. G. 1996. System Dynamics Modelling: A Practical Approach. Chapman & Hall.
El Halabi, E., and M. Doolan. 2012. “Causal Loops in Automotive Recycling.” In Proceedings of the
56th Annual Meeting of the ISSS.
http://journals.isss.org/index.php/proceedings56th/article/view/1968.
———. 2013. “Operational Challenges in Automotive Recycling: A System Dynamics Perspective.” In
R gil ing Manufacturing for inability, edited by A. Y. C. Nee, B. Song, and S. Ong.
Springer.
El Halabi, E., M. Doolan, and M. Cardew-Hall. 2012. “Extracting Variables and Causal Links from
Interview Data.” In Proceedings of the 30th International Conference of the System Dynamics
Society. http://www.systemdynamics.org/conferences/2012/proceed/papers/P1293.pdf.
El Halabi, E., C. Featherston, and M. Doolan. 2013. “System Dynamics and Scenario Planning:
Implementation Challenges.” In Proceedings of the the 2013 Systems Engineering and Test
and Evaluation Conference. In Press.
Forrest, J. 1998. “System Dynamics, Alternative Futures, and Scenarios.” In Proceedings of the 16th
International Conference of the System Dynamics Society.
http://www.systemdynamics.org/conferences/1998/PROCEED/00095.PDF.
Forrester, J. W. 1968. Principles of Systems. Pegasus Communications.
Heijden, K. van der. 2005. Scenarios: The Art of Strategic Conversation. John Wiley & Sons.
———. 2011. Scenarios: The Art of Strategic Conversation. John Wiley & Sons.
IBISWorld. 2011. Motor Vehicle Dismantling and Used Part Dealing in Australia Industry Report,
F4624. Retrieved from IBISWorld database.
Maani, K., and R. Cavana. 2007. Systems Thinking, System Dynamics: Managing Change and
Complexity. Pearson Education New Zealand.
Morecroft, J. 2007. Strategic Modelling and Business Dynamics: A Feedback Systems Approach. John
Wiley & Sons.
Richards, L. (2009). Handling Qualitative Data: A Practical Guide - Second Edition. London: Sage.
Schmitt Olabisi, L., A. R. Kapuscinski, K. A. Johnson, P. Reich, B. Stenquist, and K. J. Draeger. 2010.
“Using Scenario Visioning and Participatory System Dynamics Modeling to Investigate the
Future: Lessons from Minnesota 2050.” Sustainability 2 (8): 2686-2706.
doi:10.3390/su2082686.
Schoemaker, P. J. H. 1993. “Multiple Scenario Development: Its Conceptual and Behavioral
Foundation.” Strategic Management Journal 14 (3): 193-213.
Schwartz, P. 1996. The Art of the Long View: Planning for the Future in an Uncertain World. Reprint.
Currency Doubleday.
Sterman, J. D. 2001. Business Dynamics: Systems Thinking and Modeling for a Complex World.
McGraw-Hill Education.
Stowell, F., and C. Welch. 2012. The Manager’s Guide to Systems Practice: Making Sense of Complex
Problems. John Wiley & Sons.
Vennix, J. 1996. Group Model Building: Facilitating Team Learning Using System Dynamics. 1st ed.
Wiley.
Williams, B., and R. Hummelbrunner. 2010. Systems Concepts in Action: A Practitioner’s Toolkit.
Stanford University Press.
15
Appendix 1 - Scenarios Matrix Template
Model Sectors /Focus Areas/ Areas of Interest
Scenarios Scenario Areal Area 2 Area 3 Area 4 Area 5
Dimensions
Scenario A Cetaers
Scenario B +
Scenario C Rretaers
Scenario D +
Appendix 2 - Scenario Memo Template
Scenario X: “Title”
Dimensions: ... +...
Analysis Headings
Model Observation Identified Causal Links | CLD Updates | SFD Updates
Sectors / Variables Updates
Focus Areas (Codes)
Justifying the
SFD Updates
Main
Variables
Behaviour
Threads for
Further
Investigation
Areal
Area2
Area3
Area 4
Appendix 2 - CLD/SFD Data Descriptors
Model equations and response for the
inputs
Model equations and response for
Scenario B
Business Turnover -> (+) Workforce Size
Workforce Size -> (+) Business Costs
Business Turnover -> (+) Workforce Size
Workforce Size -> (+) Business Costs
Causal Links | Business Costs -> (-) Business Profits Business Costs -> (-) Business Profits
forthe CLD | Business Profits -> (+) Workforce Size Business Profits -> (+) Workforce Size
Marginal Workforce Costs -> (+) Workforce Costs
Workforce_Size(t) = Workforce_Size(t - dt) + (Hire - Workforce_Size(t) = Workforce_Size(t - dt) + (Hire -
Dismiss) * dt Dismiss) * dt
INIT Workforce_Size = 4080 INIT Workforce_Size = 4080
INFLOWS: INFLOWS:
Hire = - Hire = -50*LN(50*Industry_Profit/Industry_Turnover)
100*LOGN(100* Industry_Profit/Industry_Turnover) +200
+200 OUTFLOWS:
OUTFLOWS: Dismiss = if (Workfroce_Efficiency <7) Then 150 Else 0
Dismiss = if (Workforce_Efficiency <7) Then 150 Else 0 | Industry_Profit = GRAPH(TIME)(0.00, 251), (1.00, 265),
Workforce_Costs = Workforce_Size*1000 (2.00, 265), (3.00, 263), (4.00, 254), (5.00, 256), (6.00,
Model Workforce _Efficiency = 250), (7.00, 250), (8.00, 240), (9.00, 245), (10.0, 250),
(industry_f . Size)*100 (11.0, 254), (12.0, 257), (13.0, 257), (14.0, 254)
for the Industry _Profit = GRAPH(TIME)(0.00, 251), (1.00, 265), | Industry_Turnover = GRAPH(TIME)
SFD* (2.00, 265), (3.00, 263), (4.00, 254), (5.00, 256), (6.00, | (0.00, 937), (1.00, 982), (2.00, 1003), (3.00, 1023), (4.00,
250), (7.00, 250), (8.00, 240), (9.00, 245), (10.0, 250), | 1032), (5.00, 1045), (6.00, 1054), (7.00, 1068), (8.00,
(11.0, 254), (12.0, 257), (13.0, 257), (14.0, 254) 1052), (9.00, 1062), (10.0, 1082), (11.0, 1098), (12.0,
Industry Turnover = GRAPH(TIME)(0.00, 937), (1.00, | 1118), (13.0, 1123), (14.0, 1119)
982), (2.00, 1003), (3.00, 1023), (4.00, 1032), (5.00, Marginal__Workforce_Costs = 200
1045), (6.00, 1054), (7.00, 1068), (8.00, 1052), (9.00, | Workforce_Costs =
1062), (10.0, 1082), (11.0, 1098), (12.0, 1118), (13.0, (Workforce_Size+Marginal_Workforce_Costs)*1000
1123), (14.0, 1119) Workfroce_Efficiency =
(industry_Profit/Workforce_size)*100
400 —— Baseline 4200 ——aaseline
4,000 «sss Baseline Projected 4000 «sees Baseline Projected
3,900 Model Output 3,900 —— Scenario B
3,800 3,800
Model 3,700 3,700
Response* 3,600 3,600
(Workforce | | 3500 3,500
Size) 3,400 3,400 eeeeeee,
3,300 3,300 :
Ssessesassasaas Sees oe Sea easaae
Bashan gagcaanmaas a BE gegoudasga
SESESSESSESESES ERREESERSSaRESS
* Data sourced from (IBISWolrd 2011).
