decision analysis your logo here jane hagstrom university of illinois jane hagstrom university of...
TRANSCRIPT
Decision Analysis
Your Logo Here
Jane HagstromUniversity of Illinois
Jane HagstromUniversity of Illinois
What is Decision Analysis? In the broadest sense, an organized
approach to solving a decision problem.
In the narrow sense, a method of formulating a decision problem in terms of decision trees.
Where Is Decision Analysis Used? Cost/benefit/risk analysis of capital
investments Cost/benefit/risk analysis of new
product introduction Cost/benefit/risk analysis of power
plant operation Cost/benefit/risk analysis of medical
procedures
When Should We Use Decision Analysis? When we’re not sure of our objective When we’re not sure what factors
should influence our decision When we’re uncertain about the
values of factors affecting our objective
When method of determining a good decision is not obvious
Why Should We Use Decision Analysis? Organizes our thinking about hard
decisions Allows us to model major factors that
might affect our decision Output gives us a good
(optimal/reasonable) decision policy Allows us to perform sensitivity analysis
on all components of our model.
Organizing Our Thinking about Hard Decisions1. Problem context2. Fundamental and means objectives (How
will we evaluate our decision?)3. Alternatives4. Problem structure5. Assessments of chance, utilities/values6. Decision analysis algorithm gives
“optimal” policy7. Sensitivity analysis produces requisite
model
Modeling Factors that Affect Our Decision Low-level fundamental objectives provide
value objectives for an influence diagram Means objectives and problem context
suggest decisions that should be made Problem context and objectives suggest
values (risky and known) that will affect objective values
Sensitivity analysis allows us to arrive at a requisite model
Optimal Policy Mathematical formulation allows the
computation of an optimal policy. A requisite model with good numerical
quantities produces a truly optimal policy. In the absence of good numerical
quantities, the computation of an “optimal” policy either provides a good policy or provokes us to think about what aspects of our model are not satisfactory.
•Sensitivity Analysis Helps Us Understand the Problem Context Allows us to start with a simple model and
then helps us develop a requisite model Identifies values which might vary enough
to affect the goodness of our decision Allows a decision-maker to identify the
important factors to consider in making the decision
Sensitivity Analysis Tells Us What Information Is Important Should we consider the possibility of
deciding to change the value of a controllable quantity?
Should we gather more information about uncontrollable quantities?
Should we assess probabilities for the values of uncontrollable quantities?
Should we assess utility for the objective?
Sensitivity Analysis Helps Us Choose Our Analysis Method Simple requisite model allows use of
decision analysis Need for many chance events suggests
use of Monte Carlo simulation or special computations of queuing model values
Advantage in having large number of decision events suggests use of linear, nonlinear, or integer optimization
Limitations in Using Decision Analysis Methodology
Difficult to create large model Only handles small number of alternatives
for one decision event Only handles small number of states for
one chance event Computation time requirements grow
large as the number of events grow large Even utility may not completely model
decision-maker’s attitude toward risk
Important Skills in Decision Analysis Creating a Good Problem Structure Associating Values and Probabilities
with Events Interpreting a Policy Tree Interpreting Sensitivity Analysis Creating a Requisite Model Computing Value of Information
Creating a Good Problem Structure Identifying objectives Determining how low-level
fundamental objectives should be measured
Identifying alternatives Influence diagram Time sequence of events
Associating Values and Probabilities with Events Gathering information Assessing probabilities Assessing utilities
Interpreting a Policy Tree Intermediate values and probabilities are
conditioned on the events higher up/to the left in the tree.
Policy recommendation is based on maximizing expected utility/monetary value/..
Optimal policy may include recommendation as to what to do based on which outcome of some chance event occurs.
Sensitivity Analysis Any approach should be used
iteratively Try different values Tornado diagram Rainbow diagram Risk profiles
Tornado Diagram Shows a comparison of how the
policy and objective value varies when each of several input values in the model is changed
Is used to identify the most important events affecting the decision policy and objective value
Rainbow Diagram Shows how the policy and objective
value vary as a single input value is varied.
Identifies importance of gathering more information and/or suggests an appropriate approach to modeling the value using a chance or decision event
Risk Profiles Give a more accurate picture of how
risky a policy is Allow comparison of the riskiness of
different policies Allow elimination of a policy from
consideration when it is stochastically dominated by some other policy.
If You’re Going to Be a Decision-Maker ... Keep the Clemen book and the software Use the early chapters in Clemen to help
you organize your thinking about the decision
Unless the decision policy is obvious, use the software to do a rough analysis
Either continue to use the software until you feel you can make a good decision, or at some point switch to some other method of analysis.
If You’re Going to Assist in Making a Decision … Same as if you’re the decision-maker,
plus Create a good prototype model and solve
it Be prepared with sensitivity reports as
well as a policy recommendation If more decision analysis seems
appropriate, keep improving the model Use all of Clemen to help you assess
probabilities and utilities
What next? Simulation Utility Theory Algebraic Methods of Optimization Bayesian Statistics Stochastic Models Group Decision-Making