Types of Decision Problem and Applications of Decision Support

Download Report

Transcript Types of Decision Problem and Applications of Decision Support

Types of Decision Problem and
Applications of Decision Support
and Analysis
Simon French
[email protected]
Types of Decisions
Problem Context
Cognitive Factors





calculating power
knowledge & beliefs
risk attitude
preferences & values
…





structured?
uncertainty
time-span
time available
number of alternatives,
states, …
…
Social Context
 single or group decision
making
 stakeholders
 accountability
…
2
Players
Science
Values
Experts
Stakeholders
Accountabilities
and responsibilities
Forecasts of
what might happen
Decision Makers
Process
expertise
Analysts
3
Strategy Pyramid (1)
• Strategic
• Tactical
• Operational
4
Strategy Pyramid (2)
• Strategic
unstructured,
long time spans
of discretion
• Tactical
• Operational
• Instinctive
(recognition primed)
very structured,
short time
spans of
discretion
5
Planned, Orderly Activities
Strategic thinking ….. Tactical thinking …. Implementation
Strategic,
unstructured decision
making
Operational,
structured decision
making
6
Responsive Activities & Emergent
Strategy
Immediate response
……
regain of control
Strategic,
unstructured decision
making
Instinctive,
(rehearsed?) decision
making
7
The interplay between rationalistic
and emergent strategy
Rationalistic decision making
brings coherence to parts of
the strategy
Savage’s
‘small
world’
So decision analysis is usually made against
background of some inconsistency and in
recognition that this will continue
8
Organisational Levels
• Strategic 
Corporate Strategic
• Tactical 
General
• Operational  Operational
• Instinctive 
Hands-on Work
(recognition primed)
9
Levels of Decision Support
Level 0: Acquisition, checking and presentation of
data, directly or with minimal analysis, to DMs
Level 1: Analysis and forecasting of the current and
future environment.
Level 2: Simulation and analysis of the consequences
of potential strategies; determination of their
feasibility and quantification of their benefits
and disadvantages.
Level 3: Evaluation and ranking of alternative
strategies in the face of uncertainty by
balancing their respective benefits and
disadvantages.
10
DSS by levels and domains
Level 3
Level of
Support
Level 2
AI/Expert
OR
Systems models
Decision
Analysis
Forecasting
Soft
modelling
Level 1
Business Intelligence
EIS
Data Mining
Level 0
Hands-on
work
Operational General
Domain of Activity
Corporate
Strategic
11
Cynefin: a Welsh habitat
Complex
The realm of Social Systems
Cause and effect may be
determined after the event
Chaotic
Knowable
The realm of
Scientific Inquiry
Cause and effect can
be determined with
sufficient data
Cause and effect
not discernable
Known
D. Snowden (2002).
"Complex acts of knowing paradox and descriptive selfawareness." Journal of
Knowledge Management 6
pp. 100-11.
The realm of Scientific
Knowledge
Cause and effect understood
and predicable
12
Cynefin and decision making
Complex
The realm of Social Systems
probe,
sense,
respond
Chaotic
act
sense
respond
Knowable
The realm of
Scientific Inquiry
Sense
and
respond
Known
The realm of Scientific
Knowledge
categorise and respond
13
Cynefin and solutions
Complex
The realm of Social Systems
Judgement
collaboration
knowledge mgmt
Chaotic
Explore and
seek insight
Knowable
The realm of
Scientific Inquiry
data assimilation
and fitting
then optimisation
Known
The realm of Scientific
Knowledge
Databases
expert systems, neural nets,
deterministic optimisation
14
Cynefin and statistics
Complex
The realm of Social Systems
Unique
events
Knowable
The realm of
Scientific Inquiry
Chaotic
Events?
Known
The realm of Scientific
Knowledge
15
Cynefin and investigation
Complex
The realm of Social Systems
Knowable
The realm of
Scientific Inquiry
Chaotic
Known
The realm of Scientific
Knowledge
16
Do preferences exist?
• DeFinetti famously said
–
“Probabilities do not exist”
• Do preferences exist?
• or better
– When do preferences come into
existence?
17
Cynefin and Values
Complex
The realm of Social Systems
Unique
events
Knowable
The realm of
Scientific Inquiry
Chaotic
Events?
Known
The realm of Scientific
Knowledge
18
Applications:
Simpler than you think!
Simon French
Decision support means
•

Helping the decision makers and the other players
understand
Working at their cognitive level
• Need simple models usually to convey ideas
• Analysts may need complex models
• but more likely they need diagnostics for simple models
•
Paradoxically decision support and analysis drives
to simplicity
• Requisite modelling
• Start simple and build in necessary complexity until there
is sufficient understanding to ‘make the decision’
20
Chernobyl
• The world’s worst nuclear accident
• Complex event at a complex time in
Soviet Union’s history
• Many people affected
• Vast swathes of land contaminated
21
Hierarchy used in
Conference
th
5
Normal Living
Effects
Health
Public
Acceptability
Radiation
Related
Fatal
Cancers
Hereditary
Stress
Related
Affected Region
Rest of
USSR
Resources
22
Decisions based on
Intervention Levels
Measure
of Dose
Above this level, relocation would be
advised and offered
In between these levels, many countermeasures
would be implemented to clean up the area and
protect the population
Below this level, there would be little need to
do anything except reassure the population
23
Details of the Countermeasure
Strategies
Strategy
Number
relocated
(thousands)
Number
protected by
other means
(thousands)
SL2_2
706
0
SL2_10
160
SL2_20
SL2_40
Estimated
number of fatal
cancers averted
Estimated
number of
hereditary
effects averted
Cost (billions of
roubles)
3200
500
28
546
1700
260
17
20
686
650
100
15
3
703
380
60
14
24
Framing Issues
Imagine that you are a public health official and
that an influenza epidemic is expected. Without
any action it is expected to lead to 600 deaths.
However, there are two vaccination programmes
that you may implement:
• Programme A would use an established
vaccine which would save 200 of the
population.
• Programme B would use a new vaccine which
might be effective. There is a 1/3rd chance of
saving 600 and 2/3rds chance of saving none.
25
Framing Issues
Imagine that you are a public health official and
that an influenza epidemic is expected. Without
any action it is expected to lead to 600 deaths.
However, there are two vaccination programmes
that you may implement:
• Programme A would use an established
vaccine which would lead to 400 of the
population dying.
• Programme B would use a new vaccine which
might be effective. There is a 1/3rd chance of
no deaths and 2/3rds chance of 600 deaths.
26
Pareto Plots
27
Sensitivity analysis
28
Chernobyl
• The ‘world’ was a complex as it comes
• The analysis and presentation was
really rather simple
– And hugely effective.
29
Fast and Frugal aids
• Simple heuristics have been shown to
help substantially reduce psychological
biases
• For instance, Gigerenzer has shown that
‘frequency’ presentations can reduce the
issue of ‘forgotten base rates’
30
Probabilities as frequencies
80% correctly ~ 24 cancers
detected
correctly
30 women
detected
0.3% have
cancer
10000
women
99.7% do
not have cancer
20% falsely
cleared
5% falsely
detected
~ 500 false
detections
9970 women
95% correctly
cleared
31
Other fast and frugal ideas
• Consider the opposite
– Challenge your thinking
– Calibrate yourself against past decisions
• Over-define some parts of the model
– Beware of framing effects
32
Other fast and frugal ideas
• Consider the opposite
– Challenge your thinking
– Calibrate yourself against past decisions
• Over-define some parts of the model
– Beware of framing effects
• Positive emotions encourage divergent
thinking
– Brainstorm and formulate issues when you are
happy!
33
Applications of decision
support and analysis is usually
about bringing together
various simple ideas to help
decision makers evolve their
understanding, preferences
and beliefs.
34
The process of decision analysis
Identify the DMs
and stakeholders
Formulate
Identify the
need for a
decision
Evaluate
Decide
Formulate and
structure the
problem and
issues
Select option to
implement
Elicit relevant
judgements from the
DMs
No
Yes
Yes
Identify uncertainties
and gather relevant
data
Formulate
problem:
Review
Requisite?
Clarify and
articulate values
and objectives
No
Are the DMs
comfortable with the
guidance provided by
the analysis?
Review decision
structure:
Evaluate
options:
Combine
information and judgements
in a model and evaluate
options to inform the DMs
Perform sensitivity
analysis in the model
35
DSS by levels and domains
Level 3
Level of
Support
Level 2
AI/Expert
OR
Systems models
Decision
Analysis
Forecasting
Soft
modelling
Level 1
Business Intelligence
EIS
Data Mining
Level 0
Hands-on
work
Operational General
Domain of Activity
Corporate
Strategic
36
Linear programming models
• Huge and complex
• But actually rather simple with respect
to the world
• Algorithms are complex (though idea is
easy)
• But models are simple to explain in
principle
37
Business Intelligence and Analytics
• Is data mining based on simple or
complex models
• Algorithms are complex
• But representation to managers is
usually simple
– Flags and warnings saying ‘check this!’
38