Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 78 views

    Engineering Risk Benefit Analysis

    Uploaded by

    Riki Mandol
    by George E Apostolakis, Massachusetts Institute of Technology.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd

    Engineering Risk Benefit Analysis

    Uploaded by

    Riki Mandol
    78 views14 pages
    by George E Apostolakis, Massachusetts Institute of Technology.

    Original Title

    Engineering risk benefit analysis

    Copyright

    © Attribution Non-Commercial (BY-NC)

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    by George E Apostolakis, Massachusetts Institute of Technology.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    78 views14 pages

    Engineering Risk Benefit Analysis

    Uploaded by

    Riki Mandol
    by George E Apostolakis, Massachusetts Institute of Technology.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    You are on page 1of 14

    Engineering Risk Benefit Analysis

    1.155, 2.943, 3.577, 6.938, 10.816, 13.621, 16.862, 22.82, ESD.72, ESD.721

    DA 2. The Value of Perfect Information


    George E. Apostolakis Massachusetts Institute of Technology

    Spring 2007

    DA 2. The Value of Perfect Information

    Recall the evaluation of the survey results (Slide 14, DA 1):


    Strong P(s/L2) = 0.8 Mild P(m/L2) = 0.2 Weak P(w/L2) = 0.0 1.0 P(s/L3) = 0.2 P(s/L4) = 0.0

    P(m/L3) = 0.6

    P(m/L4) = 0.3

    P(w/L3) = 0.2 1.0

    P(w/L4) = 0.7 1.0


    2

    DA 2. The Value of Perfect Information

    Perfect Information (Clairvoyant)

    A clairvoyant, CV, is always correct, i.e., P[CV says L2 /L2 materializes] = 1.0 = P[s/L2] P[CV says L3 /L2 materializes] = 0.0 = P[m/L2] P[CV says L4 /L2 materializes] = 0.0 = P[w/L2] Receiving the CVs report removes all uncertainty.
    DA 2. The Value of Perfect Information 3

    Calculations for survey result is s or survey says L2 (Slide 18, DA 1)


    Payoff Prior Likelihood Prob. L2 L3 L4 0.3 0.5 0.2 1.0 P(s/ L2)=0.8 P(s/ L3)=0.2 P(s/ L4)=0.0 Product Posterior Probability P(L2/s)=0.706 P(L3/s)=0.294 P(L4/s)=0.000 1.000

    0.24 0.10 0.00 0.34

    P[L2 materializes/survey says L2] = 0.706, because the survey is not perfect.
    DA 2. The Value of Perfect Information 4

    Bayes Theorem for the Clairvoyant


    P[L2 materializes/ CV says L2] =
    =
    4

    P(CVsaysL 2 / L 2materializes)xP(L 2materializes)

    P(CVsaysL 2 / Limaterializes)xP(Limaterializes)
    2

    1xP(L 2materializes) = =1 1xP(L 2materializes) + 0 + 0


    P[L2 materializes/ CV says L2] = 1 regardless of the prior probability, because the CV is perfect.
    DA 2. The Value of Perfect Information 5

    The original decision tree


    Decisions States of Nature Payoffs

    s, 0.3 m, 0.5 N w, 0.2

    L2 $300K L3 $100K L4 -$100K

    1.0
    DA 2. The Value of Perfect Information

    L1 $150K
    6

    Modifications
    In a decision tree, the order of the nodes is chronological. With perfect information, the uncertainty is resolved before the decision is made (a chance node is followed by a decision node). The evaluation is done a priori (before the CV is hired). Therefore, the DM believes that the CV will predict L2 with probability 0.3, L3 with probability 0.5, and L4 with probability 0.2.
    DA 2. The Value of Perfect Information 7

    Decision tree with a clairvoyant


    s, 0.3 m, 0.5 N
    =$150K

    L2 $300K L3 $100K

    w, 0.2

    L4 -$100K

    CV

    O
    =$195K

    1.0

    L1 $150K

    DA 2. The Value of Perfect Information

    The value of alpha

    EMV, if the terminal decision is to be made with perfect information at no cost.

    = 0.3x300 + 0.5x150 + 0.2x150 = $195K

    DA 2. The Value of Perfect Information

    The value of beta


    What is the EMV without any information? We solved this problem in DA 1 (original decision tree). EMV[no information] = $150K : EMV, if the terminal decision is to be made without any opportunity to obtain additional information. Note:The chance node follows the decision node.
    DA 2. The Value of Perfect Information 10

    Expected Value of Perfect Information (EVPI)


    EVPI - = $195 - $150 = $45K The EVPI is an upper bound on the amount the DM would be willing to pay for additional information. The expected value of any information source must be between zero and the EVPI. In DA 1, the cost of the survey was $20K < EVPI.
    DA 2. The Value of Perfect Information 11

    Decision tree with a clairvoyant


    s, 0.3 m, 0.5 N
    =$150K

    L2 $300K L3 $100K

    w, 0.2

    L4 -$100K

    CV

    O
    =$195K

    1.0

    L1 $150K

    DA 2. The Value of Perfect Information

    12

    General Tree
    If the DM faces uncertainty in a decision (uncertainty nodes after the decision node), the impact of perfect information will be evaluated by redrawing the tree and reordering the decision and chance nodes.

    The evaluation of perfect information is done a priori. The DM has not yet consulted the clairvoyant. The DM is considering whether to actually do it.
    DA 2. The Value of Perfect Information 13

    Summary and Observations


    We have developed single-attribute, multi-stage sequential Decision Trees. The model is useful to a single decision maker. Decision Criterion: Maximize the EMV. Maximizing the EMV is not the best decision criterion.
    DA 2. The Value of Perfect Information 14

    You might also like