Lecture 4: Incomplete Information Flashcards
Incomplete Information
So far we have assumed that we can obtain the point estimates for weights by conducting an interview with the DM.
- Tradeoff method: System of equations is under-determined
- Swing method: DM is not willing or capable to allocate exact points to all alternatives
-> Sensitivity Analysis
Sensitivity Analysis (Steps)
- Determine the relationship between the attribute weights (express each weight in terms of the weight you like to alter)
- Rewrite the value of each alternative as a function of the attribute weight that you would like to alter
- Vary the objective weight in order to see how sensitive the decision is
- Draw the graph and determine the pints of intersection
What is the sensitivity analysis good for?
- Does the decision change if the weights are slightly altered?
- How robust is the decision?
Dominance Test
Dominance checks allow for narrowing down the number of relevant alternatives on the basis of a low level information
Dominance
If a dominates b then for any set of weights that is consistent with the incomplete information, V(a) >= V(b) holds and at least for one weight combination V(a) > V(b)
Dominance Check (Steps)
- Pick any set of weights that satisfies the boundary conditions
- Calculate the value of each alternative for the chosen set of weights
- Eliminate on of the possible dominance relationships
- Check for the opposite direction by investigating the sign of max[V(a) - V(b)]
Dominance Check without random initial solution
- Determine min[V(a) - V(b)]
- > for <0: further analysis
- > for >0: a dominates b - Determine max[V(a) - V(b)]
- > for <0: b dominates a
- > for >0: no statement possible
Dominance Check Conclusion
L.4 - Slide 19
Depending on the results for min[V(a) - V(b)] & max[V(a) - V(b)] we get a dominance relationship (a > b or b > a), no statement (min <0 and max >0), impossible (=0 and <0; >0 and <0; >0 and =0) or indifference (=0 and =0)