Professional Documents
Culture Documents
Scenario Analysis
Scenario Analysis
Background
ABSTRACT isk management is one of the nine Knowledge Areas within the Project
durations, especially if only a limited occurring), but only for the scenarios iterations of activity j
number of activities have significant in which activity j occurs. This total is (Again, these could be made condi-
uncertainty in their durations. then divided by the probability that tional if necessary.) The calculations
activity j occurs to get the condition- are straightforward, consistent with the
Let us define: al expected value or mean. earlier discussion.
i network scenario i
N(i) = network scenario i We will use similar notation and Examples
calculations to define:
p(i) = probability of network 1. Random Event
scenario i — Consider the project network given as
TS j|occur = expected total slack of
oj = probability that activity j Table 1. Suppose that this problem
activity j, given that aj actually
will actually occur includes a random event, such as the
occurs
— result of pending legislation that could
d j = expected duration of activity j
(considering its duration to be 0 require an additional testing activity,
If activity j occurs in all scenarios, we a4. There are three possible scenarios:
if it does not occur)
can also similarly define and calculate: (Denoted as N(1))—no additional
LCP(i) = critical path length for testing (legislation does not pass),
network scenario i — with a probability of 0.5
— ESj = expected early start time for
LCP = expected critical path length (Denoted as N(2))—minimal addi-
activity j
over all network scenarios tional testing, duration of 4, with a
—
EFj = expected early finish time probability of 0.3
The probability that each activity for activity j (Denoted as N(3))—significant
will occur is just the sum of the proba- — additional testing, duration of 11, with
bilities of all scenarios in which that LSj = expected late start time for a probability of 0.2
activity occurs. The calculations of activity j In scenarios 2 and 3, a4 has imme-
expected quantities are standard — diate predecessors of a2 and a3, and is
LF j = expected late finish time for
expected value calculations: the value an immediate predecessor of a6.
activity j
in each scenario is multiplied by the
probability of that scenario, and then (If aj did not occur in some sce- We can represent this situation
all of the results are added together. narios, we could also make these con- schematically with a modified version
a2 a5
a0 a1 a4 a7 a8
a3 a6
Figure 1: Schematic micro-approach project network graph for the random event example
a0 a1 a7 a8
a3 a6
p(1)=.5
a2 a5
d4(2)=4
p(2)=.3
a0 a1 a4 a7 a8
a3 a6
p(3)=.2
a2 a5
d4(3)=11
a0 a1 a4 a7 a8
a3 a6
a0 a1 a4 a6 a8 a9
a5 a7
a5 a7
(1)
N
(2) a2 a4 a5
N
d0 =0 d1=4 d2 =1 d8 = 6 d9 = 0
N (3)
a0 a1 a3 a8 a9
(4)
N d6 =9 d7 = 5
N
(5) a6 a7
(6)
N
Figure 5: Macro-approach diagram for looping Figure 6: Basic network diagram for uncertain activity duration example
and uncertain durations examples
fashion, a4 involves a casting process connections to each other. As men- not fail the first time, fails only the first
that may not work (the cast may break tioned at the end of the previous sec- time, or fails twice) are assumed to be
if the materials are not thick enough). tion, we will use double subscripts, independent of each other2. Then
The original activity is a “quick and where the first subscript represents there are six different scenarios to con-
dirty” attempt to do the casting, with the original initial activity from the sider (three cases in the second loop
an estimated 60% chance of success. If base case, and the second subscript for each of the two cases of the first
the process fails, the activity will be represents which iteration of that loop). The probability for each sce-
repeated, but more time will be taken activity we are discussing. This nota- nario is just the product of the corre-
to do it right (duration of 7 vs. 5). Even tion will apply to all definitions we sponding two probabilities for the two
then, we assume a 10% chance of fail- have made that involved a subscript loops in that scenario. Thus, the proj-
ure the second time around, forcing a of j, where applicable. (For activities ect could also be conceived as shown
third repetition of the casting activity, that are not involved in looping, we in Figure 5, where N(1), N(2),…, N(6) are
this time with a duration of 9. We could make all of them have a second the standard network diagrams for the
assume the chances of a third failure subscript of 1, but that seems unnec- six scenarios.
are negligible. essarily cumbersome.) This way of thinking of the prob-
This problem could be dia- One way to represent this problem lem again makes it clear that we can do
grammed most directly as shown in schematically, similar to what we did the overall analysis by simply analyz-
Figure 3. Because of these types of for our first example, is shown in ing each scenario, then combining the
looping, certain activities may end up Figure 4. results. We will number the scenarios
being repeated once or twice in this Let us assume that the two cases so that they have the following inter-
example. We could always give these for the first loop (whether the test fails pretations:
repetitions different activity number or not) and the three cases for the sec- • Scenario 1: First loop test is suc-
names, but this would not reflect their ond loop (whether the casting does cessful the first time; second loop
3 Activity Prob. Occur Exp. Crit. Exp. Crit. Occur Exp. TS Occur we will instead simply summarize the
results in Table 7.
4 A0 1 1 1 0
Notice, for example, that activities
5 A1 1 1 1 0
a2,1 and a3,1 (the first occurrence of the
6 A2 1 1 1 0 prototype production and testing
7 A3 1 0 0 3.5 activities) are critical in scenarios 1, 4,
8 A5 1 0.5 0.5 1.9 and 5, but not in scenarios 2, 3, and 6
9 A6 1 0.5 0.5 2.5 (in which both activities have total
slack values of 2, 11, and 8, respective-
10 A7 1 1 1 0
ly). On the other hand, a4,1 (first occur-
11 A8 1 1 1 0
rence of the casting activity) is exactly
12 A4 0.5 0.5 1 0 the opposite: it is critical in scenarios
Table 5: Random event network: probability of occurrence, expected criticalities, and expected 2, 3, and 6, but not in scenarios 1, 4,
total slack for each activity and 5 (where total slack values of that
activity are 5, 8, and 1, respectively).
casting is successful the first time. 2002 to analyze each of these scenar- Using the probabilities given in
• Scenario 2: First loop test is suc- ios, and passed the results to Excel. Table 7, we can now do the calculation
cessful the first time; second loop Rather than providing the complete of the expected critical path length:
casting fails the first time, but is results for each individual scenario as
successful the second time. in the Random Event example above, —
LCP =0.480(37)+0.288(39)+
• Scenario 3: First loop test is suc- (0.120+0.072)(40)+(0.032+0.008)
cessful the first time; second loop
casting fails the first and second aj Pj dj (48)=38.592
times, but is successful the third We can also calculate the expected
time. a0 --- 0 criticalities and total slack for each activ-
• Scenario 4: First loop test fails a1 {a0} 6 ity over all of the scenarios, as defined
the first time, but is successful the earlier. These are given in Table 8.
second time; second loop casting a2 {a1} 3 In the base case, the first loop is
is successful the first time. a3 {a2} 4 critical. If the first loop is successful,
• Scenario 5: First loop test fails whether or not the second loop fails
the first time, but is successful the a4 {a1} 5 once or twice, the second loop is criti-
second time; second loop casting a5 {a1} 10 cal. If the first loop fails, it is critical
fails the first time, but is successful unless the second loop fails twice (a
the second time. a6 {a3} 11 very low-probability event).
• Scenario 6: First loop test fails a7 {a4,a5} 8 Notice that the loop nature of this
the first time, but is successful the example means that activities 2 and 3
second time; second loop casting a8 {a6 ,a7} 13 sometimes occur twice, and activity 4
fails the first and second times, a9 {a8} 0 can occur as many as three times.
but is successful the third time. Based on the original data, we can
Again, we used Microsoft Project Table 6: Loop network: base case compute the average total time spent
analysis and contingency planning. and conditional activity criticality and vides exact results, as well as simplicity,
During the project planning process, a slack, as well as aggregated expected and offers advantages in analyzing and
series of project network scenarios can durations, and early and late start and interpreting the impact of uncertainty,
be identified, each with an assessed finish times for repeated activities especially when the number of critical
probability of occurrence. These sce- resulting from looping or stochastic uncertainties is relatively small. The
narios might differ according to the branching. A second benefit is greater proposed approach can handle uncer-
results of uncertain events that could accessibility and likelihood of the use tain activity durations and finite loops
with varying durations for each repeti-
3 Activity Exp. Crit. Exp. Crit. Occur Exp. TS Occur tion, allows the durations of activities
4 A0 1 1 0
to be dependent upon others, and pro-
vides the ability to summarize critical
5 A1 1 1 0
path analysis information about
6 A2 0.52 0.52 0.96
repeated and uncertain activities. For
7 A3 0.4 0.4 1.2 these reasons, we believe that our
8 A4 0.12 0.12 2.3 approach is more manageable and
9 A5 0.12 0.12 2.3 comprehensive than existing methods
10 A6 0.6 0.6 1.2
for project managers analyzing the
effects of uncertainty on their project
11 A7 1 1 0
schedule.
12 A8 1 1 0
13 A9 1 1 0 References and Notes
Malcolm, D. G., Rosenboom, J.
Table 13: Uncertain activity durations network: overall expected criticalities and total slack for
each activity H., Clark, C. E., & Fazar, W. (1959).
Application of a technique for research
occur during the course of the project, of uncertainty analysis in project plan- and development project evaluation.
uncertain (possibly dependent) activi- ning, since the modest data needs and Operations Research, 7(5), 646-669.
ty durations, finite loops, or any com- straightforward analysis are focused on Neumann, K. (1990). Stochastic
bination of these three. key scenarios driving schedule uncer- project networks: Temporal analysis,
We have presented a scenario tainty. Examples were presented to scheduling, and cost minimization.
approach for modeling and analyzing illustrate the approach, including ran- Berlin: Springer-Verlag.
these uncertain project situations. An dom events, loops, and dependent ran- Pollack-Johnson, B., & Liberatore,
advantage of our approach is that it dom activity times. M. J. (2003). Analytical techniques in
generalizes standard project network The scenario method proposed project planning and control: Current
analysis methods, such as critical path here is a direct extension of existing practice and future research directions.
analysis, to the network scenario level. project scheduling procedures, and Unpublished manuscript, Villanova,
Our approach also leads to the devel- uses software and techniques familiar PA: Villanova University, Department
opment of new project network uncer- to (and easily accessible by) almost all of Decision and Information
tainty measures, including expected project managers. This approach pro- Technologies.
BRUCE POLLACK-JOHNSON earned a BA in sociology with a minor in education from Brandeis University,
an MA in Applied Mathematics from Temple University, and an MS and PhD in Operations Research from
the University of Pennsylvania. He has taught at Oberlin College, and is currently an Associate Professor
of Mathematical Sciences at Villanova University. He has published dozens of papers on project man-
agement, forecasting, educational modeling, and on teaching applied mathematics, as well as a two-vol-
ume text on business calculus and finite mathematics (partially funded by grants from FIPSE, NSF, and
Prentice Hall). His current research is on modeling uncertainty in project scheduling. He is a member of
INFORMS and MAA.
MATTHEW J. LIBERATORE, PhD, is the John F. Connelly Chair in Management and Professor of Decision and
Information Technologies at Villanova University. He previously served as Chair of the Department of
Management and as Associate Dean. Dr. Liberatore has published over sixty journal articles in the fields
of project management, management science, information systems, and research and engineering man-
agement. He recently co-authored a text, Decision Technology: Modeling, Software, and Applications,
published by Wiley. Dr. Liberatore serves on the editorial boards of the American Journal of Mathematical
and Management Sciences and IEEE Transactions on Engineering Management. He is a member of PMI,
INFORMS, and the Decision Sciences Institute.