Docstoc

Factory Physics

Document Sample
Factory Physics Powered By Docstoc
					                                    TM 792 Special 
                                        Topics 
                                    Decision Theory
                                      June 9, 2008 
                                         Spring 2008
                                Dr. Frank Joseph Matejcik
                                 Ch 9 cont. problems
                                 Ch 10 Methods for 
                                 Eliciting Probabilities
                                 Ch 11 Risk & 
                                 Uncertainty Management 
                                 Ch12 Decisions Involving 
                                 Groups of Individuals
Frank Matejcik SD School of Mines & Technology   1
          Agenda & New Assignment
 • Tentative Schedule
 • Assignment Ch 9: none, Ch 10: 3, Ch 11: 1
 • Chapters 9 problems, 10, 11, and start 12 GW
 • We will do the survey online
 • Decision Analysis for Management Judgment 
   3rd Edition Paul Goodwin & George Wright, 
   John Wiley  EU
 • Many slides & solutions provided by John Wiley.



Frank Matejcik SD School of Mines & Technology   2
                    Access & Overview

• Instructor: Dr. Frank J. Matejcik   CM 319
   – Work:  (605) 394-6066  roughly 10-3 M-F in May
   – Cell:    (605) 431-5731 I’ll try to keep it nearby
   – Home: (605) 342-6871 Call in June?
   – Frank.Matejcik@.sdsmt.edu

• Do the book, mostly

Frank Matejcik SD School of Mines & Technology   3
                  Web Resources
       • Class Web site on the HPCnet system
       • http://sdmines.sdsmt.edu/sdsmt/directo
         ry/courses/2008su/tm792M081
       • www.wileyeurope.com/go/goodwin&wright/ 
       • Streaming video ?  
         http://its.sdsmt.edu/Distance/ 
       • The same class session that is on the DVD 
         is on the stream in lower quality. 
         http://www.flashget.com/ will allow you to 
         capture the stream more readily and review 
         the lecture, anywhere you can get your 
         computer to run.

Frank Matejcik SD School of Mines & Technology   4
                    Tentative Schedule
        Chapters     Assigned                 
  12-May  1,2,3      e-mail, contact Discussion Q. 3 page 25 
  19-May  3,4,5      Ch 3 1, 3bc, 6 & Ch 4  5, 10
  26-May  Holiday
  29-May  5,6,7      Ch 5: 3, 8 & Ch 6: 4, 8 Holiday makeup
  02-June 7,8,9      Ch 7: 1, 3 & Ch 8: 3, 8
  09-June 9,10,11,12 Ch 9: none, Ch 10: 3, Ch 11: 1
  16-June 13,14
  23-June 15,16
  30-June Overview, Final

Attendance Policy: Help me work with you.
Frank Matejcik SD School of Mines & Technology   5
              Grading for this course
 • I haven’t been able to make up a midterm 
   in summer courses.
 • It is a small class so grading hw will be OK
 • Max of (70% Final, 30% hw), 
             (30% Final, 70% hw)
 • Final will have a study guide
 • I’ll be handling mailing, too.

Frank Matejcik SD School of Mines & Technology   6
                             Ch 9: 3 
A chemical plant is due for a major overhaul and the 
manager has to make an assessment of a number of 
uncertainties associated with the project. These 
include the time the overhaul will to to complete (after 
35 days the losses of production caused by overhaul 
could have serious consequences for the company) 
and risks that there will be leakage of dangerous 
chemicals into local watercourses during the cleaning 
process. Comment on the extracts below from the 
manager's draft report for the overhaul in the light of 
Tversky and Kahneman's work on heuristics and 
biases. 
Frank Matejcik SD School of Mines & Technology   7
                             Ch 9: 3 
 (i) 'I assessed the most likely duration of the 
    overhaul to be 30 days. I then tried to take 
    an optimistic view and assumed that, if all 
    goes well, we could finish the work 5 days 
    earlier than this (i.e.. in 25 days). I then 
    took a pessimistic perspective and 
    estimated that the longest the project will 
    take is 34 days. I am therefore certain that 
    we should complete the overhaul within 35 
    days.'
Frank Matejcik SD School of Mines & Technology   8
                    Ch 9: 3 comment
 It is likely that the manager will have 
    anchored on the most likely duration of 30 
    days and that the use of the anchoring and 
    adjustment heuristic will have led to too 
    tight a range of possible project durations 
    when the optimistic and pessimistic 
    durations were estimated. He is therefore 
    likely to be overconfident that his 
    estimated range will include the actual 
    duration of the overhaul. 
Frank Matejcik SD School of Mines & Technology   9
                             Ch 9: 3 
 (ii) 'Essentially the overhaul will be split into 
    eight independent phases. I think the 
    chances of us completing each phase 
    without a pollution problem are high, say 
    90%. Overall, I therefore estimate that we 
    have almost a 90% chance of avoiding a 
    pollution problem during the project.'



Frank Matejcik SD School of Mines & Technology 10
                   Ch 9: 3 comment
 • The manager is anchoring on the 
   probability of the elementary event (i.e. the 
   90% probability that a given phase will be 
   completed without a pollution problem). 
   This is a common problem when 
   probabilities for conjunctive events need to 
   be estimated. The correct probability is 
   (0.9)8 = 0.43.

Frank Matejcik SD School of Mines & Technology 11
                             Ch 9: 3 
 (iii) 'There must be a high chance that there 
    will be no serious corrosion in the main 
    pump. The last five pumps we've 
    inspected at other plants were all corroded 
    and the chances of getting six corroded 
    pumps in a row must be very low indeed.'




Frank Matejcik SD School of Mines & Technology 12
                   Ch 9: 3 comment
 • The manager is manifesting the ‘gambler’s 
   fallacy’, i.e. he expects chance to be ‘self- 
   correcting’. Tversky and Kahneman argue 
   that this results from the use of the 
   representativeness heuristic. 




Frank Matejcik SD School of Mines & Technology 13
                             Ch 9: 3 
 (iv) 'I'm mindful of the theft of equipment we 
    had last week at our Briston plant. If we 
    don't take appropriate security precautions 
    I am virtually certain that we will lose 
    important equipment in this way during the 
    overhaul, with possible disastrous effects 
    on our ability to complete the project within 
    35 days.'

Frank Matejcik SD School of Mines & Technology 14
                   Ch 9: 3 comment
 • The use of the availability heuristic is 
   probably causing the manager to 
   overestimate the dangers of theft.




Frank Matejcik SD School of Mines & Technology 15
                             Ch 9: 3 
 (v) 'I estimated the probability of the West 
   boiler requiring repair to be about 10%.' 
   (On a later page:) 'Given the likelihood of 
   seepage into the pipe feeding the West 
   boiler, there must be a high chance of this 
   boiler being corroded. I therefore reckon 
   that there is a 50:50 chance that we will 
   have to repair this boiler as a result of the 
   seepage and corrosion.'
Frank Matejcik SD School of Mines & Technology 16
                    Ch 9: 3 comment
• The manager’s estimate is an example of the 
  conjunction fallacy where a specific event is 
  considered to be more probable than a general event 
  (of which the specific event is a part). In this case the 
  event in the second statement is a subset of the 
  event in the first statement and hence cannot be 
  more probable. Tversky and Kahneman argue that 
  this results from the use of the representativeness 
  heuristic. It can also result from the availability 
  heuristic where the specific event is more easily 
  imagined or recalled than the general event.
 Frank Matejcik SD School of Mines & Technology 17
                             Ch 9: 4 
           To what extent is it 
         reasonable to conclude 
         that human judgment in
          relation to probability 
                estimation is 
          fundamentally flawed.
Frank Matejcik SD School of Mines & Technology 18
              Chapter 10: 
             Methods for 
        Eliciting Probabilities
Frank Matejcik SD School of Mines & Technology 19
                 Preparing for probability 
                      assessment
 • Motivating – include discussion of 
   dileberate biases sells/production estimates

 • Structuring – clearly defined – units 
   commonly used – decision tree may help

 • Conditioning – discuss conditioning and 
   biases from chapter 9
Frank Matejcik SD School of Mines & Technology 20
              Direct assessment methods

 • Posing a direct question

   e.g. ‘What is the probability that the product
           will achieve break-even sales next
           month?’

 • Marking a point on a 0 to 1 scale
 • Also, urn with balls

Frank Matejcik SD School of Mines & Technology 21
                   The probability wheel




Frank Matejcik SD School of Mines & Technology 22
         An Assessment method for probability 
                   distributions
 Step 1: Establish the range of values within 
   which the decision maker thinks that the 
   uncertain quantity will lie.
 Step 2: Ask the decision maker to imagine 
   scenarios that could lead to the true value 
   lying outside the range.
 Step 3: Revise the range in the light of the 
   responses in Step 2.

Frank Matejcik SD School of Mines & Technology 23
         An Assessment method for probability 
                   distributions

 Step 4: Divide the range into six or seven roughly equal 
   intervals.
 Step 5: Ask the decision maker for the cumulative 
   probability at each interval. This can either be a 
   cumulative 'less than' distribution (e.g.. what is the 
   probability that the uncertain quantity will fall below each 
   of these values?) or a cumulative 'greater than' (e.g.. 
   what is the probability that the uncertain quantity will 
   exceed each of these values?), depending on which 
   approach is easiest for the decision maker.
 Step 6: Fit a curve, by hand, through the assessed points. 

Frank Matejcik SD School of Mines & Technology 24
         An Assessment method for probability 
                   distributions
 Step 7: Carry out checks as follows.
 (i) Split the possible range into three equally likely intervals 
     and find out if the decision maker would be equally 
     happy to place a bet on the uncertain quantity falling in 
     each interval. If he is not, then make appropriate 
     revisions to the distribution.
 (ii) Check the modality of the elicited distribution (a mode is 
     a value where the probability distribution has a peak). 
     For example, if the elicited probability distribution has a 
     single mode (this can usually be recognized by 
     examining the cumulative curve and seeing if it has a 
     single inflection), ask the decision maker if he does have 
     a single best guess as to the value the uncertain quantity 
     will assume. Again revise the distribution, if necessary.
Frank Matejcik SD School of Mines & Technology 25
           Assessment methods for probability 
                    distributions


 • Graph drawing…
 •  pdf or cdf




Frank Matejcik SD School of Mines & Technology 26
               The method of relative heights




Frank Matejcik SD School of Mines & Technology 27
            The method of relative heights 
                    (continued)




Frank Matejcik SD School of Mines & Technology 28
           Assessing the validity of subjective 
           probabilities def page 288, earlier




Frank Matejcik SD School of Mines & Technology 29
             An event tree sometimes needed




Frank Matejcik SD School of Mines & Technology 30
                           A fault tree




Frank Matejcik SD School of Mines & Technology 31
                  Using a log-odds scale




Frank Matejcik SD School of Mines & Technology 32
                   Ch 10: problem 2
 Use the probability method to elicit a 
  colleague’s distribution for his or her 
  marks in their next decision analysis 
  assignment.




Frank Matejcik SD School of Mines & Technology 33
                  Chapter 11: 
                   Risk and 
                  Uncertainty 
                  Management
Frank Matejcik SD School of Mines & Technology 34
   • Rather than passively accepting risks, 
      many managers see it as their duty to 
      take actions to reduce them 
   • Indeed, many would go further than this 
      and say that their role is also to identify 
      opportunities


Frank Matejcik SD School of Mines & Technology 35
         Sources of uncertainty at Two Valleys 
                company (value tree)




Frank Matejcik SD School of Mines & Technology 36
         Estimated values for uncertain factors




Frank Matejcik SD School of Mines & Technology 37
           Cumulative probability distributions for 
                annual profits at two sites




Frank Matejcik SD School of Mines & Technology 38
          Results of risk management actions




Frank Matejcik SD School of Mines & Technology 39
              Tornado diagram for Littleton




Frank Matejcik SD School of Mines & Technology 40
            Tornado diagrams for Callum Falls
                        scales? 




Frank Matejcik SD School of Mines & Technology 41
         Tornado diagram for Callum Falls 
                   and Littleton 




Frank Matejcik SD School of Mines & Technology 42
                         Brainstorming

 •    Do not criticize ideas 
 •    Encourage participants to put forward 
      any idea that they can think of 
 •    Aim to generate large quantities of ideas
 •    Encourage people to combine or modify 
      ideas that have already been put forward
  


Frank Matejcik SD School of Mines & Technology 43
        • Table 11.3 Ideas

        • How can we reduce fixed costs? 
        • How can we increase open market 
          demand?
        • How can we reduce variable costs 
          per unit?
        • How can we increase the chances of 
          the contract being signed? 

Frank Matejcik SD School of Mines & Technology 44
            The effectiveness of risk management 
                          measures




Frank Matejcik SD School of Mines & Technology 45
           Alternative to brain storming
               Ishikawa Diagrams ?
 • http://webpages.sdsmt.edu/~fmatejci/EIPI/EIPIh
   ome.html




Frank Matejcik SD School of Mines & Technology 46
           Chapter 12: 
        Decisions Involving 
       Groups of Individuals


Frank Matejcik SD School of Mines & Technology 47
               Mathematical aggregation:
             The production manager’s model




Frank Matejcik SD School of Mines & Technology 48
                 Mathematical aggregation:
                  The accountant’s model




Frank Matejcik SD School of Mines & Technology 49
                 Mathematical aggregation:
                   The ‘average’ model




Frank Matejcik SD School of Mines & Technology 50
            Aggregating judgments in general


 • Most studies suggest that equal weighting 
   of individual judgments works as well as 
   more complex methods

 • Often, little is to be gained by averaging 
   the estimates of more than 5 or 6 people



Frank Matejcik SD School of Mines & Technology 51
            Aggregating probability judgments




Frank Matejcik SD School of Mines & Technology 52
         Aggregating preference orderings: 
               Condorcet’s paradox


 Member                       Preference ordering
 Edwards                        A  >  B  >  C
 Fletcher                       B  >  C  >  A
 Green                          C  >  A  >  B




Frank Matejcik SD School of Mines & Technology 53
               Arrow’s impossibility theorem: 
                      Four conditions
 • Method  must produce a transitive group 
   preference order
 • If every member prefers an option then so 
   must the group
 • Group choice between 2 options does not 
   depend on preferences for any other 
   option
 • There is no dictator

Frank Matejcik SD School of Mines & Technology 54
              Aggregating values and utilities

 Destination           Person A Person B              Average
 Rio de Janeiro        100           50             75
 San Francisco           40         100             70
 Toronto                   0           0              0




Frank Matejcik SD School of Mines & Technology 55
    Measuring individuals’ strengths of preference 
                on a common scale




Frank Matejcik SD School of Mines & Technology 56
              Aggregating values and utilities

 Destination           Person A Person B              Average
 Rio de Janeiro        100           50             75
 San Francisco           40         100             70
 Toronto                   0           0              0




Frank Matejcik SD School of Mines & Technology 57
              Unstructured group processes: 
                       Groupthink


     • the tendency of groups who have
       been working together for some time to
       make poor decisions
       - because social pressures to conform
         and avoid conflict lead to suppression
         of contradicting opinions


Frank Matejcik SD School of Mines & Technology 58
          Conditions  leading to groupthink


    Ø High group cohesiveness
    Ø Insulation of group
    Ø Lack of methodological procedures for
       searching for and appraising options
    Ø Directive leadership
    Ø High stress with a low degree of hope of
       finding a solution better than the one
       favoured by the leader or other influential
       person

Frank Matejcik SD School of Mines & Technology 59
                 Symptoms of groupthink

   Ø Illusion of invulnerability: excessive optimism; 
     taking of extreme risks
   Ø Collective rationalization
   Ø Belief in group’s inherent morality
   Ø Stereotypes of rivals and enemies as evil, 
     weak and stupid
   Ø Direct pressure on dissenters
   Ø Self-censorship – minimizing importance of 
     one’s doubts
   Ø Shared illusion of unanimity 
   Ø Self-appointed ‘mindguards’ – who protect 
     group from adverse information

Frank Matejcik SD School of Mines & Technology 60
           Some consequences of groupthink

   • Incomplete survey of alternative courses of 
     action and objectives

   • Failure to examine risks of preferred choice

   • Poor information search

   • Selective bias in processing available 
     information

   • Failure to work out contingency plans in case 
     things go wrong
Frank Matejcik SD School of Mines & Technology 61
                    The Delphi method

    • Designed to obtain estimates from groups of 
      people without the biasing effect of face-to-
      face discussion and to ensure the airing of 
      diverse views

    • Does this by restricting inter-personal 
      interaction between the group members and 
      controlling information flow



Frank Matejcik SD School of Mines & Technology 62
                     The phases of Delphi

     1. Panelists provide estimates anonymously
     2. Results of this polling are tallied and
        statistics of group’s opinions are fed back to
        panelists
     3. A re-polling takes place
     4. Process is repeated until a consensus
        emerges or no further changes of opinion are
        evident
     5. Median of the group’s estimate in the final
        round is then used as their estimate

Frank Matejcik SD School of Mines & Technology 63
                  Advantages of Delphi

    • Allows larger no. of participants than group or
      committee meeting
    • Panelists can be geographically dispersed
    • Panelists can make estimates free from undue
      pressures from group or dominant individuals
    • Avoids influence of potentially irrelevant factors
      like status of person proposing an estimate
    • Anonymity means panelists can change estimates
      between rounds without fear of losing face

Frank Matejcik SD School of Mines & Technology 64
                  Rationale for Delphi

    After feedback, improved accuracy is thought to
    result from:

    • opinion changes in the ‘swingers’ who change less
      firmly grounded opinions
      and
    • opinion stability of the ‘holdouts’ who are
      assumed to be more accurate than the swingers



Frank Matejcik SD School of Mines & Technology 65
                    Limitation of Delphi

 • There is little information sharing between 
   panelists -this does not help individuals to 
   construct an alternative theory or scenario 
   in order to produce a revised estimate
 • The moderator has a great deal of 
   influence. Interests must be reviewed



Frank Matejcik SD School of Mines & Technology 66
                Decision conferencing
     • Brings together group processes, decision
       analysis and IT in intensive 2-3 day session

     • Involves small group of decision makers with
       a decision analyst (facilitator)

     • Simple models often used so model is clear to
       all participants and shared understanding of
        problem generated

     • Participants gain a sense of common purpose
       and commitment to action
Frank Matejcik SD School of Mines & Technology 67
               Why conferencing should avoid 
                        groupthink

    • Participants are not on home ground

    • The group is carefully composed of people
      from all perspectives

    • The facilitator is a neutral outsider who is
      aware of the unhelpful affects of groupthink

    • A decision analysis model is used to structure
      the discussion and make assumptions explicit
Frank Matejcik SD School of Mines & Technology 68
            Can Web 2.0 innovations
           create alternative methods?
 • Social Networking style sites
 • Modified Blogs
 • Model creating software?

 • The Decision Conference can have heavy 
   transit usage costs



Frank Matejcik SD School of Mines & Technology 69
 • Done for this week.




Frank Matejcik SD School of Mines & Technology 70

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:2
posted:7/23/2013
language:English
pages:70