Paul Kailiponi Researcher

1
Paul KailiponiResearcher ERGO
Projectkailipop_at_aston.ac.uk
Analysis of Operational Plans for Precautionary
Evacuation using Decision Theory
www.astoncrisis.com
2
Outline

• Applied Decision Theory
• Prescriptive analysis
• Contextual limitations to Decision Theory
• Model specification
• Value function specification
• Probability specification
• Substantive results for preventative evacuation
• Sample policy comparison
• Improvements to hazard forecasting
• Practitioner feedback
• Conclusions and broadening scope

3
Applied Decision Theory

• Systematic approach to decision-making
• Utilizing Subjective Expected Utility (SEU)
  theories
• Individual or small group decision-making
• Subjective utility and probability elicitation
• State-based preference
• Prescriptive Decision Theory for emergency
  management
• Evacuation decision-making
• Focus on providing insights into the
  decision-making process
• NO attempt at creating a descriptive model
• Theoretical contributions
• Identification of appropriate values for
  evacuation management
• Multiple prescriptive uses through substantial
  practitioner participation
• Identification of risk thresholds policy
  analysis

4
Contextual Limitations

• Emergency management for catastrophic events
• Limited n-size
• Mixture of objective and subjective data
• SEU limitations
• Elicitation process difficulties
• Utility elicitation
• General practitioner aversion to mathematical
  models of decision-making
• Elicitation trade-off toward requisite modelling
• SEU and Multi-attribute utility theory (MAUT)
• Prescriptive Focus

5
Model Specification Utility Function

• Identifying evacuation values
• Evacuation Responsiveness by Government
  Organizations (ERGO)
• Ten participating countries (Belgium, Bulgaria,
  Denmark, Germany, Iceland, Japan, Poland, Spain,
  Sweden, UK)
• 80 interviews, 150 documents
• Various hazards
• Advantages to breadth of data-gathering
• Maximize confidence that all values are
  identified (Bond 2007)
• Verify subsets of values conditional on
  hazard/threat

6
Evacuation Attribute List
7
Objective Function Specification

• For any given emergency context
• Axiomatic verification
• Multi-attribute utility function
• Additive function
• Single utility function

8
Single Utility Function Elicitation

• Elicitation Trade-off
• Theoretical complexity vs. Practitioner
  participation
• Utility assessment process
• Value and measurement identification
• Axiomatic verification
• Utility assessment Certainty Equivalence (CE)
• Software support
• Weight (k) assessment
• Utility (u(x)) assessment

9
Probability Function Specification

• Identification of key uncertainties
• Storm-tide predictions
• (Casualty rates Storm-tide levels)
• (Casualty rates Traffic conditions)
• (Evacuee Compliance Official Order)
• Mixture of subjective and objective probabilities

10
Influence Diagram
11
Risk Threshold Analysis

• The evacuation dilemma
• 12 Hour Storm-tide risk thresholds

12
Risk Threshold Analysis

• 9 Hour Storm-tide risk thresholds

13
Decision Theory Risk Thresholds vs. Existing Risk
Thresholds

• At 12 hour
• MAUT Initial evacuation actions taken at
  forecasts of 640 cm.
• Actual risk thresholds based on experience
  Initial evacuation actions taken at
• Practitioner Verification

14
Policy Analysis

• Policy analysis
• Forecast Improvement versus other policies
• Example of policy analysis
• Given 1 million euro cost of evacuation
• Probability of catastrophic storm surge
• Improvements in forecasting precision can save
  the city 200,000 over 10 years in false
  evacuations
• Sensitivity analysis can provide additional
  insight

15
Conclusion and Broadening Scope

• Theoretical contributions
• Identification of appropriate values for
  evacuation management
• Multiple prescriptive uses through substantial
  practitioner participation
• Risk threshold analysis
• Policy generation and analysis
• Broadening Scope
• A generalized evacuation model for multiple
  hazards
• Risk tolerance of various emergency organizations
  conditional on organizational system
• Scenario building and training applications

16

• Questions?
• Comments?