1. Systems Modelling and Qualitative Data
Dr Mike Yearworth
Reader in Engineering Systems
12th April 2013
2. ! Purpose
• Presenting an approach to grounding systems
modelling, by…
• Describing system dynamics modelling
• Describing what I mean by qualitative data analysis –
specifically grounded theory
• Bringing the two together
• Practical details – using CAQDAS and some properties of binary
matrices
• Presenting results
• Debating
• Methodology – justifying the approach using arguments of
multimethodology
212th April 2013
3. ! Origins of Dynamic Complexity
• Feedback! A causes B, B causes C, … causes A
• Negative feedback (balancing) – goal seeking, control
system
• Positive feedback (reinforcing) – unlimited growth until
bounded by exogenous factor(s)
• Combined: patterns of +ve or –ve loop dominance,
• Exponential, goal seeking, damped oscillation, limit cycles,
chaotic behaviour
• Delays complicate behaviours e.g. in –ve feedback loops
• This is not detail complexity, system structure can be
quite simple yet still produce complex dynamic behaviour
312th April 2013
5. ! System dynamics modelling
elements
• Causal Loop Diagrams (CLDs)
• Used to surface mental models about the behaviour of elements
(variables) of the system expressed as causal relationships and
feedback loops
• Stocks and Flows (S&F) Maps
• Describe the structure of the system in terms of flows and
accumulations of things
• System Dynamics (SD) Models
• Combined CLD+S&F which describe the dynamic behaviour of a
system
• Model boundary chart
• Endogenous/exogenous/excluded variables
• Sub-system diagrams
• Overall architecture of a model, comprising sub-systems and flows of
things between sub-systems
12th March 2013
6. ! System Dynamics modelling
process
12th March 2013
STERMAN, J. D. (2000) Business dynamics : systems thinking and modeling for a complex world
Real World
Decisions Information
Feedback
Strategy, Structure,
Decision Rules Mental Models of
the Real World
xxx
1. Problem Articulation
(Boundary Selection)
2. Dynamic
Hypothesis
3. Formulation4. Testing
5. Policy
Formulation &
Evaluation X
7. ! Causal Loop Diagrams (CLDs)
• Relationship between variables
represented by arrows linking them
• causal relationship
• can be positive or negative
12th March 2013
Investment Jobs
+
Cost of Fuel Car Journeys
-
Work
Completion
Rate
Project
Completion
Time
Management
Complexity
Number of
Project Staff
+
+
-
+
8. ! Causal Loop Diagrams – textual
analysis
• “Increased government investment in SWRDA
will lead to the number of jobs created in Bristol
going up”
• “Raising the tax on petrol will reduce the overall
number of car journeys”
• Management meeting…
• Director – “Increasing the number of project staff will improve
our work completion rate and improve our project completion
times.”
• Project Manager – “But increasing the number of staff in my
team will make my job way more complex and probably lead
to worse completion times.”
12th March 2013
9. ! Causal links and feedback
loops
12th March 2013
W Z
+ Read as W causes Z with +ve link polarity or mathematically
as (∂Z/∂W>0). If the cause increases, the effect increases
above what it would otherwise have been.
A B
- Read as A causes B with ,ve link polarity or mathematically
as (∂B/∂A<0). If the cause increases, the effect decreases
below what it would otherwise have been.
X Y
+
Read as X causes Y with +ve link polarity but only after some
delay.
B1
Label to indicate a balancing feedback loop.
R1
Label to indicate a reinforcing feedback loop.
10. ! Qualitative Data Analysis
• Summary of Report to the Board of Directors –
“Based on the strength of the order book and the revenues that
were being generated the CEO decided to invest in building up the
sales force to generate more orders and grow revenues in line with
promises to investors. At first this worked well and revenues grew.
However, you started to experience significant operational
difficulties in meeting this growth in orders, a significant backlog
built up and delivery lead times began to get out of hand. This
eventually caused problems for your sales force when word got
around that you were getting later and later in fulfilling your orders.
The sales team started to loose customers at an alarming rate,
revenues fell and the CEO decided to reverse his earlier decision
and cut the sales force to reduce costs.”
1012th April 2013
11. ! CLD – system behaviour
12th March 2013
Size of Sales
Force
Number of
Orders in
Process
Revenues
+
+
+
Order
Backlog+
Sales
Difficulties
-
R B
Delivery
Lead Times
+
+
Expected Order
Fulfillment Time
+
12. ! Stocks and Flows (S&F) Map
12th March 2013
Size of Sales
Force
Number of
Orders in
Process
Revenues Order
Backlog
+
Sales
Difficulties
Hiring Rate Firing Rate
+
Fulfillment
RateOrder Rate
+-
13. ! System Boundary Chart
12th March 2013
Endogenous Exogenous Excluded
Revenues Cost of sales Inventory
Sales difficulties Expected order fulfilment time Products
Delivery lead times Price of product
Order backlog Delay in finding out (about problems)
Work rate
14. ! System Dynamics Model
12th March 2013
Size of Sales
Force
Orders in
Process
Revenues
Order
Backlog
+
Sales
Difficulties
R
B
Delivery
Lead Times
+
+
Expected Order
Fulfillment Time
+
Cost of Sales
Order Rate
+ Fulfillment
Rate
Work Rate
+
Price of Product
+
-
Hiring Rate
+
- Firing Rate
+
Fraction
Invested
+
Delay in
finding out
-
-
+
16. ! Soft
Stocks
• Examples:
confidence,
ability,
drive,
satisfaction,
alignment,
morale,
productivity,
reputation…
• From
the
social/management
sciences
we
have
the
theory
of
scale
types
• Nominal,
order,
interval,
ratio§
• For
quantitative
system
dynamics
models
we
need
to
use
interval
or
ratio
measurements
for
simulations
to
work
1612th April 2013
¶FOWLER, A. (2003) Systems modelling, simulation, and the dynamics of strategy. Journal of Business Research, 56, 135-144.
§STEVENS, S. S. (1946) On the Theory of Scales of Measurement. Science, 103(2684), pp. 677-680.
17. ! Soft
stock
example
–
NASA’s
safety
culture¶
1712th April 2013
¶http://cpmr.usra.edu/Leveson-Year1-Review.ppt
18. ! Qualitative/Quantitative
Debate
• Models can be quantitative or qualitative
depending on purpose
• Quantitative models : the normal way of using
System Dynamics as per method described
• Qualitative models (CLDs – only) : emphasis
on identifying feedback paths that produce
either balancing or reinforcing feedback which
can be used in a learning process
Coyle, G. (2000). Qualitative and quantitative modelling in system dynamics:
some research questions. System Dynamics Review, 16(3), 225-244.
12th March 2013
19. ! Grounded
Theory
1912th April 2013
An abductive approach to theory generation
• Glaser and Strauss (1967)
• “the discovery of theory from data”
• Strauss and Corbin (1998)
• theory that is “derived from data, systematically gathered and
analyzed through the research process”
• Methodology
1. Data collection: for example interviews, transcripts, and documents
2. Procedures for interpretation and organizing data
a) conceptualizing, reducing, elaborating and relating data; which
collectively are referred to as coding
b) analytical procedures, such as non statistical sampling, writing
of memos and diagramming
3. Output: Written and verbal reports
GLASER, B.G., STRAUSS, A.L., (1967). The Discovery of Grounded Theory: Strategies for Qualitative Research. Hawthorne: Aldine.
STRAUSS, A. L. & CORBIN, J. (1998) Basics of qualitative research : techniques and procedures for developing grounded theory,
Thousand Oaks ; London ; New Delhi, Sage
20. ! Grounding
Systems
Modelling
• Background:
open
ended,
inductive
system
dynamics
modelling
projects
where
understanding
behaviour
of
complex
organisational
systems
was
a
requirement
• Repenning
and
Sterman
(1997)
–
inductive
approach
to
modelling
dynamics
of
process
improvement
• Morrison
(2003)
–
inductive
modelling
of
organizational
change
• Leading
to
the
idea
of
Grounded
Theories
expressed,
or
encoded,
as
Causal
Loop
Diagrams
or
System
Dynamics
models
• Hypotheses
about
dynamic
behaviour
• A
modeller,
skilled
in
the
art
of
CLDs/System
Dynamics,
would
probably
argue
that
they
do
this
naturally,
models
do
not
appear
out
of
thin
air.
• However,
by
using
the
Grounded
Theory
approach
and
using
CAQDAS
tools
provide
an
explicit
audit
trail
an
explicit
linkage
from
data
to
models
2012th April 2013
MORRISON, J. B. (2003) Co‐evolution of process and content in organizational change : explaining the dynamics of start and fizzle.
REPENNING, N. P. & STERMAN, J. D. (2002) Capability traps and self-confirming attribution errors in the dynamics of process improvement. Administrative
Science Quarterly, 47, 265-295
21. ! Morrison, on his approach…
2112th April 2013
“Data analysis included listening to the recorded interviews and reading the
transcriptions, coupled with a review of field notes. I identified patterns of
interest and recurring themes in the data, bounding the analysis with a focus
on efforts to implement change in the first production cell. As is typical in
developing grounded theory, I organized the data into categories, which I
represented with variables and causal relationships between them (Glaser et
al., 1967). I combined variables and causal relationships to begin identifying
causal loops as a description of the feedback processes gradually emerging
from this analysis. During the data analysis, I occasionally translated
portions of the emerging feedback structure into formal mathematical models
and simulated their behavior in order to gain a richer understanding of the
relationship between the feedback structure and the dynamic behavior. The
iteration between the grounded data, causal loop diagrams, and formal
mathematical models led to additional insights and generated new questions
that I could explore in the available data or pursue with my respondents.”
22. 2212th April 2013
Morrison, J. B. (2003) Co-evolution of process and content in organizational change:
explaining the dynamics of start and fizzle. PhD Thesis Sloan School of Management.
Massachusetts Institute of Technology.
23. 2312th April 2013
Repenning, N. P. & Sterman, J. D. (2002) Capability traps and self-confirming attribution
errors in the dynamics of process improvement. Administrative Science Quarterly,
47(2), pp. 265-295.
24. 2412th April 2013
Dunford, C. N., Yearworth, M., York, D. M. & Godfrey, P. (2012) A View of Systems
Practice: Enabling Quality in Design. Systems Engineering.
Use of Taught SE
Techniques
Training in
System
Engineering
Natural forgetting to
use tools over time
Overall Quality of
Systems Practice
Knowledge of
Systems Practice
R
Evidence
Number of projects
where SE applied at
quality per month
Improvement
+
-
Application
+
Quality decay
Learning
+
Knowledge
decay
Coupling
Communicate
the Value
+
Awareness of
Systems
Engineering
+
+
Engineer's
Appreciation of
Systems Practice
Ease of Tailoring
the Systems
Approach
+
+
+
+
Logistical
Complexity
-
-
Cross
Lifecycle
Working
+
Frequency of
training events
+
Enthusiasm
Application
Appreciation
Expertise
R
B
R
-
-
25. 2512th April 2013
Yearworth, M. (2010) Inductive Modelling of an Entrepreneurial System. 28th International Conference of the
System Dynamics Society. Seoul, Korea.
Yearworth, M. & White, L. (201x) The Uses of Qualitative Data in Multimethodology: Developing Causal Loop
Diagrams During the Coding Process. European Journal of Operational Research - In Review.
Confidence in
management
Methods to
ensure success
+
Entrepreneurial
drive
+
Cooperation
between
investors
+
Parallel
investment
+
Equity funding
success
+
Financial outcome -
Return on Equity
(RoE)
Idea
generation
+
Intellectual property,
creating and
defending
Proof of concepts
and prototypesSources of early
funding
Entrepreneurs'
equity stake
+
Entrepreneurs'
risk appetite
+
Investors' risk
appetite+
Equity
funded
+
++
Meeting
customer
needs
Evidence of
revenue and
projections
+
Persuasiveness of
business model
+
+
+
+
+
+
Portfolio of
funds
+
+
R1
R4
R2
R3
-
B1
+
R0
+
"Spotting Opportunities, Testing, and Validation
(SOTV)"
"Realistic Equity Position
(REP)"
"Scale-Up and Exit
(SUE)"
26. ! Coding and an axiom
• In addition to axial coding…
• A possible relationship exists between two
codes (concepts, categories) if the two
categories code data within the same
scope of the source
2612th April 2013
27. ! Causality analysis
2712th April 2013
Rabinovich, M. & Kacen, L. (2010) Advanced Relationships Between Categories Analysis
as a Qualitative Research Tool. Journal of Clinical Psychology, 66(7), pp. 698-708.
Maxwell, J. A. (2004) Using Qualitative Methods for Causal Explanation. Field Methods, 16(3), pp. 243-264.
28. ! Proposition
• The value of the method is in the potential
to
1. Introduce dynamic sensibility to qualitative
data analysis§
2. Provide a more rigorous approach to the
formation stage of system dynamics
modelling¶
2812th April 2013
§LANE, D. C. & OLIVA, R. (1998) The greater whole: Towards a synthesis of system dynamics and soft systems methodology.
European Journal of Operational Research, 107, 214-235
¶Kopainsky, B. & Luna-Reyes, L. F. (2008) Closing the Loop: Promoting Synergies with Other Theory Building Approaches to
Improve System Dynamics Practice. Systems Research and Behavioral Science, 25pp. 471-486.
Luna-Reyes, L. F. & Andersen, D. L. (2003) Collecting and analyzing qualitative data for system dynamics: methods and models.
System Dynamics Review, 19(4), pp. 271-296.
29. ! Multimethodology – theoretical
underpinnings
• “…it seems apparent that the question is not if to
use qualitative data, but when and how to use them
appropriately?”[6]
• approaches such as grounded theory constitute a
toolset that helps build “…relevant system dynamics
models, grounded in data, and with higher potential
to provide rigorous and relevant generic
structures”[7]
2912th April 2013
[6] Luna-Reyes, L. F. & Andersen, D. L. (2003) Collecting and analyzing qualitative data for system dynamics:
methods and models. System Dynamics Review, 19(4), pp. 271-296.
[7] Kopainsky, B. & Luna-Reyes, L. F. (2008) Closing the Loop: Promoting Synergies with Other Theory Building
Approaches to Improve System Dynamics Practice. Systems Research and Behavioral Science, 25pp. 471-486.
30. ! Conclusions
• Value…
• Theoretical underpinning in multimethodology
• Tool support, both mathematical and software
• Practical examples of application
è rigorous grounding of SD modelling
è adding dynamic sensibility to grounded theory
• …and?
3012th April 2013
31. ! Conclusions
• The codeàconceptàcategoryàtheory grouping and
free node/tree node (axial coding) in NVivo leads
thinking towards hierarchical structuring (arborisation)
• Matrix structures lead towards network thinking
(reticulation), but much harder to do and less support for
it
è work close to the data using a parallel/bridging strategy
where questions of causality are posed constantly and
dynamic hypotheses generated/tested abductively
è requires better tool support to make this an easier
process
3112th April 2013