Proceedings of the 2004 Winter Simulation Conference

R .G. Ingalls, M. D. Rossetti, J. S. Smith, and B. A. Peters, eds.

ABSTRACT
This paper introduces the Stochastic Project Scheduling
Simulation (SPSS) system and two additional useful statisti-
cal analysis tools. SPSS integrates CPM, PERT, and Dis-
crete Event Simulation (DES) scheduling methods into one
system. It can generate a CPM-based deterministic schedule,
a PERT-based probabilistic schedule, and a simulation-
based stochastic schedule for the project under study. Two
additional statistical analysis tools are added to assist users
in determining the number of simulation runs needed for a
given schedule network and testing the significance of dif-
ference between two independent simulation experiments.
1 INTRODUCTION
Due to the limitations of CPM and PERT scheduling meth-
ods, simulation has been applied to schedule networks by
considering random activity durations for estimating project
durations (Douglas 1978, Ahuja and Nandakumar 1985).
Moreover, simulation has been widely employed as a useful
tool for planning construction processes or operations (Mar-
tinez and Ioannou 1997, 1999, Ahuja 1984, Ioannou 1989,
Shi 1999, Hajjar and AbouRizk 1999).
A construction project may be consisted of hundreds
or thousands of activities. When simulation is applied to
schedule a project, the first question must be answered is
how many simulation runs must be conducted. Number of
simulation runs needed is project-specific. It is jointly af-
fected by many factors such as number of activities in the
project, relationships between activities, formulation of the
critical path(s), and distributions of activity durations.
A simulation cannot be executed indefinitely. Generally
speaking, a large number of simulation runs leads to a statis-
tically valid simulation results but it will consume a long
computing time. Although new computing technology has
made computing time not a major concern for small to me-
dium projects, computing time may be still crucial for large
projects, e.g., with thousands of activities. Reducing un-
needed simulation runs will improve the efficiency of a simu-
lation process. The validity of simulation results depends
upon the number of simulation runs. If it is terminated pre-
maturely, the obtained results may not objectively represent
the process under study and therefore be invalid. Properly de-
termining the number of runs needed for a given project
helps determine the minimal computing time and ensure
valid results. Moreover, when multiple simulation experi-
ments are conducted for the same schedule network, different
results may be obtained due to different streams of random
numbers. It is essential to examine the difference between
the results of the different experiments. They should be statis-
tically identical in order to valid the experiments.
This article discusses two important statistical factors
for schedule simulation: (1) determining number of simula-
tion runs for a given schedule network, and (2) testing the
significance of the difference between any two simulation
experiments. JAVA components are developed in the SPSS
system for carrying out the two functions.
2 THE STOCHASTIC PROJECT SCHEDULE
SIMULATION (SPSS) SYSTEM
The SPSS system provides a computer tool for analyzing a
schedule network by using CPM, PERT, and DES methods
(Lee 2004). CPM uses a deterministic approach, PERT
uses a probabilistic approach, and DES uses a discrete
event simulation approach for scheduling construction pro-
jects. Although PERT shows an advantage over CPM by
providing a probabilistic approach for representing activity
durations, a PERT schedule has found to be conservative
because the method systematically underestimates the
probability to complete a project in a given time (Halpin
and Riggs 1992, Shi 2001). The DES approach would im-
prove the accuracy of a PERT schedule. SPSS allows a
user to select any preferred methods freely and to compare
the schedules from the three methods for a given project.
SPSS can assist a user in estimating the probability of
completing a project within a given time. SPSS collects all


STATISTICAL ANALYSES FOR SIMULATING SCHEDULE NETWORKS


Dong-Eun Lee
Jonathan Jingsheng Shi

Construction Engineering and Management Program
Department of Civil and Architectural Engineering
Illinois Institute of Technology
3201S. Dearborn St.,
Chicago, IL 60616-3793, U.S.A.



1283
Lee and Shi

of the results from the conducted simulation runs and con-
duct statistical analyses accordingly.
3 THE METHODOLOGY
Calculating the probability of completing a project in a
given time using simulation technique involves: (1) deter-
mining the activity durations, (2) conducting a simulation
experiment for the network and determining the critical
path in the network, (3) computing project duration with
the longest path, and (4) conducting multiple simulation
runs by repeating steps 1 to 3, and (5) computing mean
project duration, standard deviation, and probability to
complete the project in a desired time.
SPSS was implemented with the three scheduling
methods (CPM, PERT, and DES) as shown in Figures 1, 2,
3, respectively (Lee 2004). The same activity diagram is
shared by the three methods in SPSS as shown in Figure 4.
If the DES scheduling method is selected, the user
may enter a desired number of simulation runs as shown in
Figure 3. However, the program will calculate the mini-
mum number of simulation runs needed based on the given
confidence level. If the number of simulation runs is given
by the user, the system will return a confidence interval for
the mean project duration under a given confidence.
Three point estimates (optimistic, most-likely, and
pessimistic times) are used in PERT. For comparison pur-
pose, the same three point estimates are used for the ac-
tivities of the schedule network under simulation study.
Triangular distributions are used modeling for activity du-
rations ( in Figure 4).
A simulation experiment will be conducted for 120 runs
with simulation results to be collected from each run. Then,
the mean project duration and its confidence interval are cal-
culated for the samples in the experiment. Estimating the
confidence interval requires the mean () and standard de-
viation () of the population. Since the population standard
deviation is unknown, an unbiased estimate of the standard
deviation (S
x
) is used as the standard deviation of the popu-
lation (Mohammadi 2002). If a desired project duration is
given, SPSS can also calculate the probability for the project
to be completed in the given time as shown in Figure 3.
3.1 Determining the Number of Simulation Runs
The methodology for computing the minimal number of
simulation runs is based on the error term resulted from the
sample. SPSS first calculates the unbiased estimates of
mean and the standard deviation values. Then the confi-
dence interval of the mean is computed in the range be-
tween the lower bound exact mean (LBEM) and upper
bound exact mean (UBEM) (Ang and Tang 1975, Khisty
and Mohammadi 2001, Mohammadi 2002).


Figure 1: Deterministic CPM in SPSS
1284
Lee and Shi



Figure 2: Probabilistic PERT in SPSS

Figure 3: Stochastic Discrete Event Simulation in SPSS


1285
Lee and Shi

Triangular(o1,e1,p1) PERT(o1,e1,p1)
PERT(o2,e2,p2)
PERT(o3,e3,p3)
PERT(o4,e4,p4)
PERT(o5,e5,p5)
PERT(o6,e6,p6)
PERT(o7,e7,p7)
CPM(e1)
CPM(e2)
CPM(e3)
CPM(e4)
CPM(e5)
CPM(e6)
CPM(e7)
Activity CPM Mode Simulation Mode PERT Mode
a
b
c
d
e
f
g
* PERT(o,e,p)
Double PERT(o,e,p){
return (o+4*e+p)/6;
}
*Triangular(o,e,p): (refer to Lee 2004)
Double Beta (o,e,p){
if (rand()<(e-o)/(p-o)){
return o + sqrt((e-o)*(p-o)*rand());
}
else
return p-sqrt((p-e)*(p-o)*(1-rand())))
}
0.0 0.0
0.0 0.0
A
a: CPM(e1)/PERT(o1, e1, p1)
/Triangular (o1, e1, p1)
b: CPM(e2)/PERT(o2, e2, p2)
/Triangular(o2, e2, p2)
f: CPM(e6)/PERT(o6, e6, p6)
/Triangular(o6, e6, p6)
g: CPM(e7)/PERT(o7, e7, p7)
/Triangular (o7, e7, p7)
d: CPM(e4)/PERT(o4, e4, p4)
/Triangular(o4, e4, p4)
0.0 0.0
0.0 0.0
B
0.0 0.0
0.0 0.0
C
0.0 0.0
0.0 0.0
D
0.0 0.0
0.0 0.0
E
0.0 0.0
0.0 0.0
F
c: CPM(e3)/PERT(o3,e3,p3)
/Triangular(o3,e3,p3)
e: CPM(e5)/PERT(o5,e5,p5)
/Triangular(o5,e5,p5)
Triangular(o2,e2,p2)
Triangular(o3,e3,p3)
Triangular(o4,e4,p4)
Triangular(o5,e5,p5)
Triangular(o6,e6,p6)
Triangular(o7,e7,p7)
1 2

Figure 4: The Schedule Diagram and Activity Durations

The width of the confidence interval depends on the
critical value (t
|2
) and S
x
. A rough estimate of S
x
could be
obtained from the conducted simulation runs. This estimate
is used to determine the minimal number of simulation
runs (N) (Haas 2004) based on a 95% confidence level. As
shown in Figure 3, the 95% confidence interval of the
mean is in the range of [26.41, 27.19] for the given sched-
ule problem with 120 runs. Figure 3 also shows that the
minimal number of runs, N should be in the range
91 . 9 37 . 9 N
. As the computation algorithms sug-
gested, the conducted simulation runs (i.e., 120 in this
case) are sufficient when Ns high range is less than the
conducted runs. In other words, it is appropriate to use 120
runs for this project. If the obtained range of N is greater
than the number of conducted simulation runs, extra simu-
lation runs will be conducted.
3.2 Testing the Significance of the Difference
between Two Simulation Experiments
To test the difference between the mean project duration
values of two independent simulation experiments, t-Test
is implemented in SPSS. Testing the difference between
the variances of two independent simulation runs can also
be conducted to check if they are from the same population
as shown in Figure 5 by using the F-test. The testing re-
sults for the given example are shown in Figure 5.


Figure 5: t-Test Results on the Mean Project Duration
4 A CASE STUDY
For the given example as shown in Figure 1, the SPSS sys-
tem calculates (1) a deterministic project duration of 27
days by using the CPM method; (2) a mean project dura-
tion of 27 with a standard deviation of 2.0 by using PERT;
and (3) and a mean project duration of 26.79 with a stan-
dard deviation of 2.12, and a confidence interval
(
85 . 26 61 . 26
) for the project duration at the 95%
confidence level.
After eleven (11) independent simulation experiments
(each with 120 runs) were conducted, the results are sum-
marized in Figure 6. The mean project durations (
total
) fall
in the range
38 . 26 08 . 26
total

with an overall
mean value of 26.22. If one particular experiment is exam-
ined, the project durations () of the 120 runs vary in the
range,
00 . 28 50 . 23
as shown in Figure 7.
The t-Test shows no significant difference in the
mean project durations and F-test on the sample variances
indicates that they could be from the same population
(

level of 0.005) for the two experiments as shown in

Table 1. In conclusion, the two independent simulation ex-
periments are not significantly different.





1286
Lee and Shi


Figure 6: Trend Analysis of Mean Project Durations


Figure 7: Comparison of Two Different Simulation
1287
Lee and Shi

Table 1: t-Test Results on Mean Project Duration
1st experiment 2nd experiment
Mean 26.18 25.96
Standard deviation 2.37 2.26
No. of Observations 121 121
Degrees of freedom
(v)
240.0
Critical Value (t: one-
tailed)
0.748
P(x>t) 0.228
Alpha level (

)
0.005
5 CONCLUSIONS
SPSS can support construction scheduling by providing a
more accurate prediction of the completion time. The activ-
ity data inputted for a project can be shared by the three
scheduling methods (CPM, PERT, and DES). The system
will provide three different schedules for the project so that
the user can compare the differences of these schedules and
estimate the potential schedule risks involved in the project.
In summary, the system can (1) forecast the probabil-
ity of completing a project at a given duration; (2) compare
the schedule results from CPM, PERT, and DES methods;
(3) determine an appropriate number of simulation runs for
a given project, (4) to calculate the probability of complet-
ing a project at a given confidence level, and (5) test the
significance of difference between two independent simu-
lation experiments.
If the system allows interactive changes of the critical
value (t
|2
) of the t-distribution and the acceptable error
term ( ), its analytical capability can be enhanced with
different confidence values and error terms.
Moreover, more research is required to examine how
project durations are effected by skewness and/or variabil-
ity of activity durations (e.g., high, medium, or low vari-
ability). It is also noted that a project duration may be-
come more complicated if there are multiple critical or
close-to critical paths in the schedule network. Corre-
sponding t-test methods should be considered.
REFERENCES
Ahuja, N. Hira. (1984). Project management: techniques
in planning and controlling construction projects.
Wiley, New York.
Ahuja, N. T. H., and Nandakumar, V. (1985). Simulation
model to forecast project completion time. Journal of
Construction Engineering and Management, ASCE,
111(4), 325-342.
Ang, Alfredo. H-S., and Tang, Wilson. H. (1975). Prob-
ability Concepts in Engineering Planning and Design:
Volume I Basic Principles. Analytical Models of
Random Phenomena, John Wiley & Sons, New York.
80-169.
Douglas, E. David. (1978) PERT and simulation Proceed-
ings of the 1978 Winter Simulation Conference. Harold
J. Highland, IEEE Press, Piscataway, N.J, 89 98.
Haas, J. Peter., (2004) Introduction to Simulation [online].
Available online via <http://www.stanford.
edu/class/msande223/handouts/lecturen
otes01.pdf> (accessed Apr 20, 2004).
Hajjar, Dany., and AbouRizk, Simaan. M. (1999). Sim-
phony: An Environment for Building Special Purpose
Construction Simulation Tools. Proceedings of the
1999 Winter Simulation Conference. P. A. Farring-
ton, H. B. Nembhard, D. T. Sturrock, and G. W. Ev-
ans, eds. IEEE, Piscataway, N.J, 998-1006.
Halpin, D. W. and Riggs L.S. (1992). Planning and
Analysis of Construction Operations. John Wiley &
Sons, New York, N.Y.
Ioannou, P.G. (1989). UM_CYCLONE Users Guide.
Dept. of Civil Engineering, The University of Michi-
gan, Ann Arbor. Mich.
Khisty, C. Jotin., and Mohammadi, Jamshid. (2001). Fun-
damental of System Engineering with Economics,
Probability, and Statistics. Principles of probability,
Prentice Hall. Upper Saddle River, N.J, 131-150, 294
313.
Lee, Dong-Eun. (Accepted 2004). Probability of Project
Completion Using Stochastic Project Scheduling
Simulation (SPSS). Journal of Construction Engi-
neering and Management, ASCE.
Martinez, C. Julio., and Ioannou, G. Photios. (1997)
State-Based Probabilistic Scheduling Using
STROBOSCOPEs CPM Add-On Proceedings, Con-
struction Congress V, ASCE, Stuart D. Anderson,ed,
Minneapolis,MN, 438-445.
Martinez, J.C., and Ioannou, P.G. (1999). General Pur-
pose Systems for Effective Construction Simulation.
Journal of Construction Engineering and Manage-
ment, ASCE, 125(4), 265-276.
Mohammadi, Jamshid. (2002). Statistical Analysis of En-
gineering Data lecture note in summer 2002. Dept. of
Civil Engineering, Illinois Institute of Technology,
Chicago, IL.
Shi, J. (1999). Activity-Based Construction (ABC) Mod-
eling and Simulation Method. Journal of Construc-
tion Engineering and Management, ASCE, 125(5),
354-360.
Shi, J. Jonathan. (2001). Activity-Based Construction
(ABC) Modeling and Simulation Dept. of Civil Engi-
neering, Illinois Institute of Technology, Chicago, IL.
AUTHOR BIOGRAPHIES
DONG-EUN LEE, P.E., ISO QMS ASSESOR. is a Ph.D.
Candidate and Research Assistant in the Construction Engi-
1288
Lee and Shi

neering and Management Program in the Department of
Civil and Architectural Engineering at Illinois Institute of
Technology. Before he joined IIT in 1999, he had worked
for Hyosung Group <http://www.hyosung.com> for
5 years as project engineer and had acted as an registered
ISO 9000 Quality Management System Assessor
<http://www.kab.or.kr>. He is a registered Pro-
fessional Engineer in Korea. He has a B.S. in Civil and
Architectural Engineering from Chung-Ang University in
Seoul (1993), and a M.S. in Civil Engineering from Illinois
Institute of Technology (2000). Lee defended his Ph.D. the-
sis in Summer 2004. Currently, he is working for the NK
Construction Consultant, Inc., New York as a project sched-
uler. His research interests include construction business
process modeling and automation and the development and
application of Workflow Management Systems using simu-
lation. His other research interests include construction
scheduling and planning method, simulation, and mathe-
matical programming. His e-mail address is <leedong1@
iit.edu>.

JONATHAN JINGSHENG SHI, Ph. D., P. E. is an As-
sociate Professor in the Department of Civil and Archi-
tectural Engineering at Illinois Institute of Technology in
Chicago. He received his Ph.D. in Civil Engineering from
the University of Alberta in 1995. His research interests
in simulation are focused on automated modeling, simula-
tion, and optimization in construction application. His
current researches has been extended to construction
automation and innovation as investigating fundamental
issues relating to construction enterprise resource plan-
ning systems (Construct-ERP) and development of enter-
prise wide system architecture. His other research inter-
ests include neural networks, construction scheduling and
planning method, and construction productivity and im-
provement. He has served as committee member of Con-
struction Research Council and Computing in Construc-
tion, ASCE and CMAA. His email and web addresses are
<shi@iit.edu> and his web address is <http://
www.iit.edu/~jshi>.
1289