topic 7 - uncertainty mgmt.pptx
TRANSCRIPT
Uncertainty Management
SCO 6041 – Class 7
Uncertainty Management
• Agenda
– PERT Analysis
– Monte-Carlo Analysis
The Role of Chance …
Source: Dilbert Comic Strips Archivehttp://www.dilbert.com/strips/
A Roadmap
Uncertainty Risks Ambiguity
An understanding that important aspects of the project are not determined, but rather random variables with probability distributions. Examples: payoffs, task durations, etc.
A knowledge that the project depends on the outcomes of certain events in the future that can be fully described by their probability distributions. Examples: Market entry of competitors, outcome of drug trials, etc.
An awareness that the project is influenced by unforeseeable events, which cannot be predicted, and are not even recognized at the time of project planning.
Focus of this chapter
Key Ideas of This Class
• Instead of taking task durations as determined, we treat them as random variables
• We can then calculate the likelihood of finishing the project by a certain time
• This can serve as a basis for determining the ‘right’ project buffer
Topic Objective: Establish better estimates of task duration, buffers, and project completion
Central Assumption
0
0.01
0.02
0.03
0.04
0.05
0.06
5 6 7 8 9 10
Prob
abili
ty
Duration
Task Durations follow a Beta Distribution (right skewed)
Maximum Time (Pessimistic Estimate)
Minimum Time (Optimistic Estimate)
Mode(Most likely time)
Mean(Average Time)
Bounded by Min & Max (in contrast to Normal distribution)
0
0.01
0.02
0.03
0.04
0.05
0.06
5 6 7 8 9 10
Prob
abili
ty
Duration
Central AssumptionTask Durations follow a Beta Distribution (centered)
Maximum Time (Pessimistic Estimate)
Minimum Time (Optimistic Estimate)
Mean(Average Time)
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
5 6 7 8 9 10
Prob
abili
ty
Duration
Central AssumptionTask Durations follow a Beta Distribution (left skewed)
Maximum Time (Pessimistic Estimate)
Minimum Time (Optimistic Estimate)
Mean(Average Time)
Mode(Most likely time)
9
PERT
• Program Evaluation and Review Technique (PERT)
• Developed by U.S. Navy in 1950’s– Navy Special Projects Office– Polaris submarine weapon system
“…. With units of time as a common denominator, PERT quantifies knowledge about the uncertainties involved in developmental programs requiring effort at the edge of, or beyond, current knowledge of the subject – effort for which little or no previous experience exists. ….”
— Willard Fazar (Head, Program Evaluation Branch, Special Projects Office, U. S. Navy), The American Statistician, April 1959.
PERT Analysis
Step 1: Instead of asking for task durations, ask for optimistic, pessimistic and most likely
Task Optimistic Most Likely PessimisticA 1 2 3B 7 8 9C 11 13 15D 8 10 12E 10 15 23F 4 7 14G 2 4 11H 3 4 7I 16 21 35J 4 5 6K 2 2 2L 2 2 2M 8 10 14N 1 1 1
Optimistic = Minimum Time
Most Likely = Mode
Pessimistic = Maximum Time
PERT Analysis
Step 2: Calculate the mean and variance of each task duration using the following formulas:
4*6
Optimistic Pessimistic MostLikely
22
36Pessimistic Optimistic
PERT Analysis
Step 2: Calculate the mean task duration, and the variance of each task duration
Task Optimistic Most Likely Pessimistic Mean VarianceA 1 2 3 2 .11B 7 8 9 8 .11C 11 13 15 13 .44D 8 10 12 10 .44E 10 15 23 15.5 4.69F 4 7 14 7.67 2.78G 2 4 11 4.83 2.25H 3 4 7 4.33 .44I 16 21 35 22.5 10.03J 4 5 6 5 .11K 2 2 2 2 0L 2 2 2 2 0M 8 10 14 10.33 1N 1 1 1 1 0
PERT AnalysisStep 3: Schedule the Project Using Mean Durations
Task A2 days0 slack
Task B8 days0 slack
Task C13 days0 slack
Task D10 days0 slack
Task E15.5 days
3 slack
Task F7.67 days
3 slack
Task G4.83 days
3 slack
Task J5 days
42.5 slack
Task H4.33 days
0 slack
Task I22.5 days
0 slack
Task K2 days
20.5 slack
Task L2 days
45.8 slack
Task M42.5 days44 slack
Task N1 day
0 slack
0 2
0 2
2 17.5
5 20.5
2 10
2 10
10 23
10 23
17.5 25.2
20.5 28.2
23 33
23 33
25.2 30
28.2 33
2 7
44.5 49.5
33 37.3
33 37.3
37.3 59.8
37.3 59.8
37.3 39.3
57.8 59.8
7 9
55.8 57.8 7 17.3
49.5 59.8
59.8 60.8
59.8 60.8
ES EF
LS LF
Ear
liest
S
tart
Ear
liest
Fi
nish
Note: I have assumed that a second Product Architect was hired (no resource conflict) and that K has to come after L (no resource conflict for product manager). See case.
PERT AnalysisStep 4: Calculate Mean & Standard Deviation of Critical Path
Task A2 days0 slack
Task B8 days0 slack
Task C13 days0 slack
Task D10 days0 slack
Task H4.33 days
0 slack
Task I22.5 days
0 slack
Task N1 day
0 slack
0 2
0 2
2 10
2 10
10 23
10 23
23 33
23 33
33 37.3
33 37.3
37.3 59.8
37.3 59.8
59.8 60.8
59.8 60.8
Mean(μ) 2 8 13+ + 10+ 4.33+ 22.5+ 1+ 60.8=
Variance(σ2) .11 .11 .44+ + .44+ .44+ 10.03+ 0+ 11.6=
Standard Deviation
(σ)3.4√𝟏𝟏 .𝟔 =Assume independence between tasks
(allows variances to be additive)
PERT AnalysisStep 5: Plot cumulative normal distribution
50 54.000000000000158.000000000000162.000000000000266.000000000000270.00000000000030
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Project Duration
Cum
ulati
ve P
roba
bilit
y
PERT Analysis
Hint: To plot the cumulative normal distribution in Excel, create (and then plot) the following Table:
X-Axis Y-Axis
50 =NORM.DIST(50,μ,σ,1)
51 =NORM.DIST(51,μ,σ,1)
… …
70 =NORM.DIST(70,μ,σ,1)
PERT Analysis
Step 5: Why Normal Distribution?
Central Limit Theorem: The sum of several random variables with any (though identical) distribution is approximately normally distributed
PERT AnalysisStep 5: Plot cumulative normal distribution
50 54.200000000000158.400000000000162.600000000000266.80000000000020
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Project Duration
Cum
ulat
ive
Prob
abili
ty
There is a 90% chance of the project finishing in less than 65 days.
PERT AnalysisStep 5: Plot cumulative normal distribution
50 54.200000000000158.400000000000162.600000000000266.80000000000020
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Project Duration
Cum
ulat
ive
Prob
abili
ty There is a 70% chance of the project finishing in less than 63 days.
In Excel, use =NORMINV(0.70,60.8,3.4)
PERT AnalysisStep 6: Understand Uncertainty
61
Likelihood of Finishing the Project
on Time
Total Time until Project is Due
50%
62 60%
63 70%
64 80%
65 90%
69 99%
PERT AnalysisStep 7: Planning Project Buffers
61
Likelihood of Finishing the Project
on Time
Total Time until Project is Due
50%
6260%
6370%
6480%
6590%
6999%
0
Project Buffer (Days)
1
2
3
4
8
PERT - Criticisms
• PERT assumes that the critical path remains constant – but uncertainty in task durations implies that critical paths may change
• PERT assumes that task durations are independent of each other – but since the same people were involved in creating different estimates, and since the same resources may be involved in different tasks, this assumption may not hold
Monte Carlo Simulations
• A lot of times we have to make assumptions simply to know what a distribution looks like. If uncertain events interact in complex ways, we often have no clue what the real underlying distribution looks like – but we can simulate it!
Monte Carlo Simulations
• Example of a known distribution: Rolling a dice
1 2 3 4 5 6
-0.0499999999999997
2.91433543964104E-16
0.0500000000000003
0.1
0.15
0.2
0.25
0.3
0.35
Outcome
Prob
abili
ty
Expected Value: 3.5
Monte Carlo Simulations
• What does the distribution of the maximum of two dice look like?
D1 D2 Score D1 D2 Score D1 D2 Score D1 D2 Score D1 D2 Score D1 D2 Score
1 1 1 2 1 2 3 1 3 4 1 4 5 1 5 6 1 6
1 2 2 2 2 2 3 2 3 4 2 4 5 2 5 6 2 6
1 3 3 2 3 3 3 3 3 4 3 4 5 3 5 6 3 6
1 4 4 2 4 4 3 4 4 4 4 4 5 4 5 6 4 6
1 5 5 2 5 5 3 5 5 4 5 5 5 5 5 6 5 6
1 6 6 2 6 6 3 6 6 4 6 6 5 6 6 6 6 6
Monte Carlo Simulations
• What does the distribution of the maximum of two dice look like?
1 2 3 4 5 60.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Outcome
Prob
abili
ty
Expected Value: 4.47
Monte Carlo Simulations
• Suppose I offer you the following gamble: I get to roll a dice. You get to roll two dice, record the maximum, then re-roll the dice, record the minimum, and your score is the maximum-minimum. If I beat you, I get $1 from you, if you beat me, you get $1 from me.
Is this a good gamble?
• We can simulate this by throwing a lot of dice and recording the results. Fortunately, there is software that can do this for us …
Monte Carlo Simulations
• This is the Probability Distribution of your Score
Monte Carlo Simulations
• Your Probability of Winning: 23%• Your Probability of Drawing: 13%• Your Probability of Loosing: 64%
Simulating the Echelon Release Project
• Histogram
Simulating the Echelon Release Project
• Parameter Estimate Comparison
60.8
Monte Carlo SimulationPERT
61.3Mean
Standard Deviation 3.4 3.8
Simulating the Echelon Release Project
• Project Duration
61
PERT Likelihood of Finishing on Time
Total Time until Project is Due
50%
62 60%
63 70%
64 80%
65 90%
69 99%
Monte Carlo Likelihood of
Finishing on Time
50%
59%
68%
75%
82%
97%
PERT underestimates the project duration, since the method ignores that critical paths can change!
The Path-Changing Issue …
Task
Task
Task
Task
Task
Task
Task
Project A
Task
Task
Task
Task
Task
Task
Task
Project B
In which project would you be more distrustful of PERT estimates?
Simulating the Echelon Release Project
• Likelihood of Being Critical (Criticality)
A
Likelihood of Being CriticalTask
100%B 80%C 80%D 80%E 20%F 20%G 20%H 100%I 100%J 0%K 0%L 0%
M 0%N 100%
These estimates give you a good idea which tasks are critical and require more intense monitoring and control
Simulating the Echelon Release Project• Contribution to Variance in Project Duration (Cruciality)
A
ContributionTask
1.1%B 0.5%C 2.9%D 3.3%E 2.1%F 0.8%G 0.5%H 3.7%I 84.9%J 0%K 0%L 0%
M 0%N 0%
These estimates give you a good idea which tasks drive uncertainty in project duration, and therefore where to target risk reduction efforts
0
1
2
3
4
5
6
0 1 2 3 4 5 6 7 8
Dura
tion
Task
B
Duration Task A
0123456789
10
0 1 2 3 4 5 6 7
Dura
tion
Task
B
Duration Task A
Task Correlations
0
2
4
6
8
10
12
14
0 1 2 3 4 5 6 7 8
Dura
tion
Task
B
Duration Task A
High, positive correlation
No correlation
High, negative correlation
- Estimates were made by same person- Problems from one task transfer to the
next- Same resources work on multiple tasks
- Tasks are independent and their forecasts were not prepared by the same person
- Task boundaries are not well defined- Team members will try to pre-empt
other people’s work
Variance and CorrelationsTwo Independent Random Variables
7 7.4 7.8 8.2 8.6 9 9.4 9.810.2
10.6 1111.4
11.812.2
12.6 130
0.050.1
0.150.2
0.250.3
0.350.4
0.45
7 7.4 7.8 8.2 8.6 9 9.4 9.810.2
10.6 1111.4
11.812.2
12.6 130
0.050.1
0.150.2
0.250.3
0.350.4
0.45
1717.4
17.818.2
18.6 1919.4
19.820.2
20.6000000000001
21.0000000000001
21.4000000000001
21.8000000000001
22.2000000000001
22.6000000000001
23.00000000000010
0.050.1
0.150.2
0.250.3
0.350.4
0.45
μ1 = 10, σ1 = 1 μ2 = 10, σ2 = 1
μ = 20, σ = 1.41
1 2
2 21 2
10 10 20
2 1.41
Variance and CorrelationsTwo Correlated Random Variables
7 7.4 7.8 8.2 8.6 9 9.4 9.8 10.2
10.6 11 11
.411
.812
.212
.6 130
0.050.1
0.150.2
0.250.3
0.350.4
0.45
7 7.4 7.8 8.2 8.6 9 9.4 9.8 10.2
10.6 11 11
.411
.812
.212
.6 130
0.050.1
0.150.2
0.250.3
0.350.4
0.45μ1 = 10, σ1 = 1 μ2 = 10, σ2 = 1
μ = 20, σ = 1.73
1 2
2 2 21 2 12
21
1
10 10 20
2
2 1
2 1
3 1.73
ρ = 0.5
1717.4
17.818.2
18.6 1919.4
19.820.2
20.6000000000001
21.0000000000001
21.4000000000001
21.8000000000001
22.2000000000001
22.6000000000001
23.00000000000010
0.050.1
0.150.2
0.250.3
0.350.4
0.45
Variance and CorrelationsTwo Correlated Random Variables
• Summing independent random variables reduces the standard deviation, since summing always means that high and low values cancel out
• As the correlation between random variables increases, so does the standard deviation of their sum, since a higher correlation implies less ‘canceling out’ of high and low values
40
A Note on Correlation Matrices
• Suppose that the correlation between A and B is 0.9, and the correlation between B and C is 0.9. Could A and C be uncorrelated?
• You can easily end up with ‘inconsistent’ correlation matrices. Crystal Ball will ‘fix’ them for you by altering your correlations to be as close as possible to a valid correlation matrix.
Simulating the Echelon Release Project
• Project Duration
61
PERT LikelihoodTotal Time until Project is Due
50%
62 60%
63 70%
64 80%
65 90%
69 99%
Monte Carlo Likelihood
50%
59%
68%
75%
82%
97%
Ignoring correlations between task durations also underestimates the project duration
Monte Carlo Likelihood with
Correlations
52%
58%
64%
69%
74%
88%
PERT or Monte Carlo?
• PERT analysis is easy to carry out and understand– Available in MS Project– Can be easily done in Excel
• Monte Carlo simulations can be more accurate and allow more modeling flexibility– Requires specialized software– Requires some spreadsheet development
Key Take-Aways
• Uncertainty– Simple variations in key project parameters can be
easily incorporated into planning
• PERT Analysis– Simple and reasonably effective
• Monte Carlo Simulations– Sophisticated and flexible