Professional Documents
Culture Documents
Management
Journal
The Professional Journal of the Project Management Institute
Volume 33, Number 3 | September 2002
PAPERS
4 Research Report
Harry Stefanou, PhD
70 Calendar
72 Advertisers Index
Project Management Journal
PROJECT MANAGEMENT Book Development Editor PRODUCTION SERVICES PROVIDED
JOURNAL EDITOR Richard Schwartz; richard.schwartz@pmi.org BY IMAGINATION PUBLISHING,
Permissions Coordinator CHICAGO, IL, USA
Parviz F. Rad, PhD, PMP
Stevens Institute of Technology Insuk Choe; insuk.choe@pmi.org Senior Editor
Bookstore Administrator Ross Foti; rfoti@imaginepub.com
Book Review Editor
Kenneth H. Rose, PMP Regina Madonna; regina.madonna@pmi.org Assistant Managing Editor
Book Publishing Planner Lauren Strandquist; lstrandquist@imaginepub.com
PMI PUBLISHING STAFF Danielle Moore; danielle.moore@pmi.org Art Director
Publisher Doug Kelly; dkelly@imaginepub.com
Administrative Assistant
Linda Cherry; linda.cherry@pmi.org
Dotti Bobst; dotti.bobst@pmi.org Associate Art Director
Editor in Chief Tonya Weiland; tweiland@imaginepub.com
General e-mail:
Dan Goldfischer; dan.goldfischer@pmi.org
pmipub@pmi.org Associate Art Director
Publishing Support Specialist Theresa Rogulic; trogulic@imaginepub.com
Natasha Pollard; natasha.pollard@pmi.org
Publications Services Manager
Angela Kramer; akramer@imaginepub.com
EDITORIAL REVIEW BOARD PUBLICATION & MEMBERSHIP Project Management Institute, Publishing
Frank T. Anbari, PMP, The George Washington Department, Four Campus Boulevard, Newtown Square,
The Project Management Journal (USPS 8756-9728/
University; Bud Baker, Wright State University; Rick Pennsylvania 19073-3299 USA. Phone +610-356-4600,
02) is published quarterly (March, June, September,
Bilbro, The Innova Group, Inc.; David Christensen, fax +610-356-4647; e-mail: pmipub@pmi.org
December) by the Project Management Institute. PMJ
Cedar City, UT; David Cleland, University of Pittsburgh; Unless otherwise specified, all letters and articles
is printed in the USA by Cadmus Magazine, Richmond,
Helen S. Cooke, Cooke & Cooke; Dan H. Cooprider, sent to the PMI Publishing Department are assumed
VA. Periodical postage paid at Newtown Square, PA
Creative People Management; Jeffrey Covin, Indiana for publication and become the copyright property of
19073 USA and at additional mailing offices.
University; Steven V. DelGrosso, IBM; Deborah Fisher, PMI if published. PMI is not responsible for loss, dam-
Canadian agreement #40030957. Postmaster: Send
University of New Mexico; Vipul Gupta, Saint Joseph’s age, or injury to unsolicited manuscripts or other mate-
address changes to Project Management Institute,
University; Soren Hansen, Pennoni Associates of Ohio; rial.
Headquarters, Four Campus Boulevard, Newtown
Kenneth O. Hartley, PMP, Parsons Brinckerhoff; Francis Square, Pennsylvania 19073-3299 USA. Phone +610-
T. Hartman, The University of Calgary; Gary C. 356-4600, fax +610-356-4647. READER SERVICES
Humphreys, Humphreys & Associates; Lewis Ireland, The mission of the PMJ is to provide information Photocopies. PMJ has been registered with the
PMP, Project Technologies Corp.; Peter Kapsales, advancing the state of the art of the knowledge of pro- Copyright Clearance Center, Inc. Consent is given for
Bellcore; Lee R. Lambert, PMP, Lambert Consulting ject management. PMJ is devoted to both theory and copying of articles for personal or internal use, or for
Group, Inc.; Alexander Laufer, Technion-Israel practice in the field of project management. Authors the personal use of specific clients. This consent is
Inst. of Technology; Bill Leban, Keller Graduate are encouraged to submit original manuscripts that given on the condition that the copier pays through the
School of Management; Robert Loo, The University of are derived from research-oriented studies as well as Center the per copy fee stated in the code on the first
Lethbridge; Kenneth G. Merkel, PMP, University practitioner case studies. (See Notes for Authors in the page of each article for copying beyond that permitted
of Nebraska; James J. O’Brien, PMP, O’Brien-Kreitzberg back of this journal.) All articles in PMJ are the views by Sections 107 or 108 of the U.S. Copyright Law. The
& Assoc.; Michael D. Okrent, Agilent Technologies of the authors and are not necessarily those of PMI. appropriate fee should be forwarded with a copy of the
Inc.; John B. Phillips, Engineering Development Subscription rate for members is $14 per year and is first page of the article to the Copyright Clearance
Institute; Peggy C. Poindexter, Great Falls, VA; Tzvi Raz, included in the annual dues. Center, Inc., 222 Rosewood Drive, Danvers, MA 01923
Tel Aviv Unversity; Paul B. Robinson, The Eastern Claims for undelivered copies must be made no USA (tel: +508-750-8400; fax: +508-750-4744). The
Management Group; Arnold M. Ruskin, PMP, Claremont later than two months following month of publication. fee indicated on the first page of an article in this issue
Consulting Group; Avraham Shtub, Tel Aviv University; The publisher will supply missing copies when losses will apply retroactively to all articles published in the
Richard L. Sinatra, PMP, Potomac, MD; Larry A. Smith, have been sustained in transit and when the reserve journal, regardless of the year of publication. This con-
Applied Management Associates; Dwight Smith- stock will permit. sent does not extend to other kinds of copying, such as
Daniels, Arizona State University; James Snyder, PMI is a nonprofit professional organization whose for general distribution, resale, advertising and promo-
Springfield, PA; Paul Solomon, PMP, B-2 Earned Value mission is to serve the professional interests of its col- tion purposes, or for creating new collective works.
Management Systems; Robert G. Staples, Monroe, VA; lective membership by: advancing the state of the art Special written permission must be obtained from the
Walter Taylor, Delta Airlines; Charles J. Teplitz, University in the leadership and practice of managing projects publisher for such copying.
of San Diego; Veljko M. Torbica, PMP, Florida and programs; fostering professionalism in the man- Permissions. Requests to reprint articles pub-
International University; Walter A. Wawruck, agement of projects; and advocating acceptance of lished in PMJ must be made in writing to the publisher.
Vancouver, BC, Canada; Itzhak Wirth, Saint John’s project management as a profession and discipline. Reprints. Individual articles may be purchased
University; Janet K. Yates, San Jose State University Membership in PMI is open to all at an annual dues of through the Knowledge & Wisdom Department
$119 U.S. For information on PMI programs and mem- (www.pmi.org/k&wc) at a cost of $10.00 per article
PUBLICATIONS ADVISORY BOARD bership, or to report change of address or problems for members and $15.00 per article for nonmembers.
Linda C. Cherry, PMI Publisher; Greg Hutchins, Quality with your subscription, contact: Glossy Reprints. Requests for glossy reprints of
Plus Engineering; Sandy Jenkins, Sanjenco; William V. Project Management Institute, Headquarters, individual articles in quantities of 100 or more can be
Leban Jr., PMP, Keller Graduate School of Management; Four Campus Boulevard, Newtown Square, pennsyl- sent to pmipub@pmi.org.
Charles J. Teplitz, PMP, University of San Diego School vania 19073-3299 USA; tel: +610-356-4600; fax: Bulk Copies of Current Issues. Copies of the cur-
of Business Administration +610-356-4647; e-mail: pmihq@pmi.org rent PMJ can be obtained in quantities of 25 or more.
Orders must be placed 40 days prior to date of issue.
EDITORIAL & ADVERTISING SERVICES The cost is $5.50 per copy plus shipping.
Back Issues. Back issues are available on request
Address manuscripts and other editorial submissions,
at $20 each. Call +610-356-4600 for availability.
advertising and mailing list rental inquiries, and
requests for reprints/bulk copies/reprint permission to:
Rafi A. Ashrafi, University of Calgary, 2500 University Dr. NW, Calgary, Alberta T2N 1N4
Canada
I nformation systems (IS) and information technologies (IT) are the fastest growing
industries in developed countries. Huge amounts of money continue to be invest-
ed in these industries (Abdel-Hamid & Madnick, 1990). Due to pressure of time-to-
▼ market, there is a corresponding pressure to increase productivity. To maintain
Abstract a competitive edge in today’s fast-changing world, an organization’s success
For many enterprises, sustainable success is depends on effectively developing and adopting IS. Literature has discussed concern
closely linked to information systems (IS) and for problems related to IT/IS development and implementation.
information technologies (IT). Despite signifi-
According to Zells (1994) and other studies, approximately 85% of software pro-
cant efforts to improve software project suc-
cess, many still fail. Current literature indicates jects undertaken in Europe and North America are at level one of the Software
that most of the software project problems Engineering Institute’s capability maturity model (CMM). Level one is the lowest
are related to management, organizational, level of CMM. The challenges at level one are to have project planning, project man-
human, and cultural issues—not technical
problems. This paper presents results of a
agement, configuration management, and software quality assurance in place—and
survey of 36 software owners/sponsors, con- have them working effectively. To improve project delivery performance, a number of
tractors/suppliers, and consultants on 12 pro- organizations are adopting project management approaches and setting up project
jects. The empirical results address answers management offices (Barnes, 1991; Butterfield & Edwards, 1994; King, 1995; Munns
to questions related to success, performance
metrics, and project business drivers. A lack
& Bjeirmi, 1996; Raz, 1993; Redmond, 1991).
of alignment on these critical issues emerge Current literature on software projects shows that most of the software problems
consistently by phase as well as across are of a management, organizational or behavioral nature, not technical (Johnston,
the entire project. The results of this study 1995; Martin, 1994; Whitten, 1995).
also are compared with others that span
seven additional industry sectors. As a result, A survey of high-tech firms showed that if project management improved, time and
the authors have developed an approach that cost could be reduced by more than 25% and profits would increase by more than 5%
links project critical success factors (CSFs) to (Fisher, 1991). This has since been validated by use of Stratigically Managed Aligned
corporate strategy, and project metrics to the
Regenerative Transitional (SMART) project management, based on internal bench-
CSFs. An important finding of this study is the
critical need to identify and manage realistic marking by the companies involved in the field trials.
expectations of the stakeholders to achieve
perceived project success. Objectives of the Study
In this paper, the authors report findings on current project management practices in
the IT/IS industries. The purpose of the study was to find out what practices are
Keywords: information systems; information
technology; managing stakeholder expecta-
important to IT/IS industries in successfully accomplishing their projects. Do they
tions; critical success factors; software use proven project management practices? Whatever the IT/IS industries regard as
project management important for the success of their projects, do they measure it? What are the project
drivers? Are these three important elements aligned with each other? The authors
©2002 by the Project Management Institute investigated these questions not only for various phases of a project, but also from
2002, Vol. 33, No. 3, 5–15
8756–9728/02/$10.00 per article + $0.50 per page
the perspective of three major stakeholders. These stakeholders include an owner or
sponsor, a major contractor or supplier, and a consultant for the same project.
Minimum
Scope Changes
Change Management
Technology and
Expertise Termination
Project Plan
Success Factors
Execution
Business Purpose
Top Management
Support Planning
Project Mission
Definition
Communication
Owner’s Consultation
Owner’s Approval
0 1 2 3 4 5
Importance
Figure 1. Project Success Factors by Phases
■ How consistent are the three major stakeholders (owner, that things did not get out of hand by the time the execution
contractor, and consultant) in their perceptions of CSFs, met- stage rolled around because, at that point, the project has enough
rics, and project priorities? momentum that it becomes quite difficult to get it back on track.
■ Are the perceived CSFs consistent with the metrics used and Table 1 shows the most important overall CSFs and project
the project priorities identified by these stakeholders? metrics as identified by all the three stakeholder groups over
An average of all scores of the responses in the appropriate four project phases. The respondents showed inconsistencies
survey groups was calculated. The most important characteristics between what they identified as the project success factors and
then were defined as those that had the highest average score: what were used as project metrics. It was observed that in some
■ CSFs by Phases. Figure 1 shows the 10 most important cases that respondents in the same project agreed on the
CSFs over the four phases of projects; importance of certain CSFs, but they did not agree on how the
■ CSFs by Stakeholders. The 10 most important CSFs by CSFs were measured. In other cases, respondents agreed on the
stakeholders group are shown in Figure 2; importance of these factors, but they indicated that they did
■ Project Metrics by Phases. Figure 3 shows the 10 most not have a metric established for measuring them.
important project metrics for four phases of the projects inves- It also was observed that project owners, contractors, and
tigated; consultants do not have a clear understanding of the methods
■ Project Metrics by Stakeholders. The most important pro- that are used on their projects to measure how well project
ject metrics according to each of the stakeholders are shown in goals are met. Although there is some agreement as to which
Figure 4; factors are important to the success of the project, there also
■ Project Priority Ranking by Phase. Figure 5 shows project should be agreement on how to measure success. If project
priority ranking during four phases of the projects studied; metrics are not clearly understood, it is difficult to determine
■ Project Priority Ranking by Stakeholders. Figure 6 shows the level of success of the project. Each individual involved in
the most important project priority rankings by stakeholders. the project may have a different opinion as to how successful
From the results, it was concluded that the value of metrics as the project is, depending on his or her own measurement.
a predictive tool was not fully exploited by the project teams. It Another important point is that success factors considered to
may have been possible to place more importance on key met- be important to a project should be measured in some
rics earlier on in the project. This could have been done to ensure way during execution to determine ahead of time whether
Minimum
Scope Changes
Change Management
Technology and
Expertise
Business Purpose
Consultant
Top Management
Support
Owner
Project Mission
Communication
Owner’s Consultation
Owner’s Approval
0 1 2 3 4 5
Importance
Figure 2. Success Factors by Stakeholders
New Technology
Accepted
Responsibilities
Assigned
Resources Supplied
Activities Schedule
Planning
Scope Defined
Definition
Deliverables Identified
Milestones Met
On Time
0 1 2 3 4 5
Importance
Figure 3. Project Metrics by Phases
Resources Supplied
Within Budget
Contract
Completion Defined
Metrics
Consultant
Activities Schedule
Deliverables Identified
Milestones Met
On Time
0 1 2 3 4 5
Importance
Figure 4. Project Metrics by Stakeholders
performance objectives will be met. If the project stakeholders ■ Responsibility breakdown structures, work breakdown
and the team do not formally measure the factors that they structures, and CSFs were not well utilized;
deem to be most important, they cannot hope to predict ■ The owners did not have control, monitoring or feedback
its success and take corrective action as required. The project systems independent of those used by the contractors and/or
stakeholders and the team may be spending their time on mea- consultants;
suring less important factors that will lead it to an incorrect ■ Time taken to align stakeholders on what is important to
ongoing measurement of whether or not the project is a success. the project probably would help improve communication,
reduce rework, and enhance the possibility of success;
Summary of Findings ■ The alignment of project metrics with project success factors
Based on the results, the authors found: and priorities appears to be an opportunity for improvement
■ The ratings for a particular project success factor did not in the software industry.
change very significantly between different phases;
■ Throughout all project phases, there was general agreement Recommendations
among survey participants that a project mission, consultation It is widely accepted that there is room for improvement in the
with the project owner, good communication, and the avail- delivery of software projects including new software develop-
ability of resources are important factors for project success; ment, upgrades, or implementation. Many of the specific stud-
■ Participants on each project agreed on certain project suc- ies in this area suggest either what the problems may be or what
cess factors, but they tended to either disagree on how the suc- needs to be in place for success. While this is useful informa-
cess factor should be measured, or they did not attempt to tion, it does not help the practitioner with the question: “How
measure the success factor at all; do I achieve greater success?” This study set out to link the
■ Project metrics were not fully utilized as a predictive tool symptoms for success or failure with what constructive action
but rather as a measure of how well the project performed to may be needed to achieve such success. These recommenda-
that point in time. This often is too late to allow effective cor- tions, which have been tested on live projects to validate them,
rective action; make that critical link. Based on internal benchmarks in the test
■ The owners of the projects agreed unanimously that it is very companies, savings in time and cost of between 10% and 30%
important for the project to meet the needs of the end user; were matched by improved quality and end-user acceptance.
Career Development
and Training
Termination
End-User Satisfaction
Priorities
Execution
Cost Planning
Definition
Performance
Time
0 1 2 3 4 5
Importance
Figure 5. Project Priorities by Phase
Team Development
Career Development
and Training
Consultant
Cost
Owner
Performance
Time
0 1 2 3 4 5
Importance
Figure 6. Project Priorities by Stakeholders
1 Owner is informed of the project status and Project completed on time or ahead
his/her approval is obtained at each stage of schedule
2 Owner is consulted at all stages of develop- Milestones are identified and met
ment and implementation
4 The project has a clearly defined mission The scope of the project is clearly
defined and quantified
6 The project achieves its stated business Project completion is precisely defined
purpose
10 The project is completed with minimal A specific new technology is adopted and
and mutually agreed scope changes accepted by end users
The recommendations that follow represent the four most software industry, it may just be a matter of learning what tools
significant elements identified and tested in this study: are available and how to use them properly to raise the number
■ Link your project to corporate business strategy; of successful software projects to an acceptable level.
■ Align major stakeholders on key issues; The authors hope that this study will help in:
■ Simplify project controls and metrics; ■ Considering a holistic approach for the project;
■ Make sure effective communication and expectation ■ Understanding what is important for success;
management is maintained throughout the project life. ■ Understanding the dynamics of project drivers and priori-
Greater detail on how these aspects are implemented can be ties and that these may shift over time;
found in Hartman (2000). ■ Getting and maintaining alignment of major stakeholders
including the immediate project team on all important strate-
Conclusions gic and tactical issues;
Although the projects surveyed were rated as successes, some ■ Realizing better planning and more effective control;
projects lacked defined goals or defined metrics to measure this ■ Accomplishing a successful project with satisfied stake-
success. If the owner, contractor, and consultant on a project all holders, project teams, and customers.
have different ideas of what success is and how success will be Some general guidelines for how this may be achieved
measured, it is unlikely that everyone (or possibly anyone) will have been offered. The suggested approaches to achieving
be satisfied when the project is completed. There are many tools project success have been tested on live projects with consis-
that can be utilized to ensure a successful project. For the tently successful outcomes.
This article is copyrighted material and has been reproduced with the permission of PMI. Unauthorized reproduction of this material is strictly prohibited.
Scheduling Programs With Repetitive
Projects Using Composite Learning
Curve Approximations
Jean-Pierre Amor, University of San Diego, School of Business Administration, 5998 Alcala
Park, San Diego, CA 92110–2492 USA
P rograms that deliver a relatively small number of similar products arise in a variety
of industries. In the defense industry, the budgetary reductions that followed the
end of the Cold War and the increasing complexity and cost of major weapon sys-
tems have led to the procurement of shrinking quantities of combat ships and air-
craft from the prime contractors. In the aerospace industry, the nature of space mis-
sions and their related activities imply fairly small orders for spacecraft, satellites, and
rocket boosters. And in the housing industry, it has long been a common practice to
build the more expensive homes in relatively small tracts. These types of pro-
grams also occur in the provision of certain services, such as management con-
▼ sulting, the upgrade of existing equipment, the change of software, or the intro-
duction of a process improvement or of a new monitoring system.
Abstract Frequently, these types of programs consist of a one-time order for a product that
Programs that require executing a modest
the contractor has never produced before, and due to the complexity of the product,
number of similar projects arise in several
industries, such as aerospace, construction, each “unit” requires the execution of a distinct project. Thus, the scheduling of such
and defense. The scheduling of such programs programs—that is programs with repetitive projects—is very complicated. It also is
often involves deciding how many projects will very important because the cost per unit produced can be quite high. The program
run simultaneously (in parallel) to “optimally”
trade-off resource utilization and penalty costs.
manager, who is responsible for providing an accurate delivery schedule to all parties,
One approach for scheduling such programs is must make the development of this schedule the most important planning activity.
to perform a quick approximation and follow-
ing it later with a “full-blown” procedure. By What the Problem Entails
assuming a common learning rate for all pro-
ject activities, the quick approximation,
The scheduling of a program is used to make an a priori estimate of its overall dura-
although computationally inexpensive, is not tion and cost. This enables the manager to set a completion date, to budget resources,
sufficiently accurate to eliminate the need for and to allocate resource usage. Scheduling also permits control over the timing of
the full procedure, which requires so much ongoing activities to ensure the timely completion of each project.
data tracking and so many calculations that it
is discouraging to practitioners. The approxi- The scheduling of programs with repetitive projects often involves balancing
mation proposed in this paper is significantly two opposite tendencies. On the one hand, when the contractor has little or no
more accurate than the quick approximation, previous experience with the product, it is imperative to take advantage of the
while requiring only slightly more computation,
learning phenomenon and execute as many projects as possible with the same
making it more valuable for program managers.
resources, e.g., individuals, crews, teams, materials, or equipment. This lessens the
cost of the resources used because the project performance time decreases as the
Keywords: schedule development; learning
curves; multiproject planning; estimating number of repetitions increases. In the extreme case, all the projects would be per-
formed in a single sequence—all in series. On the other hand, the need to deliver
©2002 by the Project Management Institute
2002, Vol. 33, No. 2, 16–29
each unit by its contracted due date to avoid a penalty cost often requires that sev-
8756–9728/02/$10.00 per article + $0.50 per page eral projects be executed simultaneously. In the extreme case, all the projects
would be performed simultaneously—all in parallel. Thus, a major challenge
R Paint L, M 3 95%
W Finish S, T, V 0 —
Note. From “Case and Readings in Production and Operations Management,” by J.C. Latona and J. Nathan (1994), Boston, MA: Allyn and Bacon; and
“Improving CPM’s Accuracy Using Learning Curves,” by C.J. Teplitz and J.P. Amor, 1993, Project Management Journal, 24, pp. 15–19.
Table 2. Tasks, Precedences, Times and Improvement Factors for Housing Example
Calendar Days
two types of errors arise, which eventually affect the schedule
100 Three
and cost estimations.
Parallel
First, the composite learning curve of the M activities [which
800
is not log-linear even if, as is usually assumed, the individual
Four
activity learning curves are log-linear (Amor & Teplitz, 1998)]
Parallel
is approximated by a log-linear curve, which coincides with the 600
180
Resource Tgnt (0.90)
$1,500,000
160
Hire IDEAL
Resource Days
140
Fire
Dollars
Scnt (1–30)
$1,000,000
Penalty 120
Tgnt (0.85)
100
Incentive
$500,000
Tgnt (0.80)
TOTAL 80
60
1 3 5 7 9 11 13 15 1 4 7 10 13 16 19 22 25 28
Figure 3. Cost Curves: IDEAL – All Types of Costs Figure 4. Learning Curves: Composite – All Activities
through these two points. Thus, the “secant approximation” gen- full-blown procedure. This tool combines the CPM, individual
erates, a postiori, an approximate learning rate, rp, for the project. task learning curves, and the “secant approximation” with
The computational requirements of this method (two CPMs, Shtub’s cost model and search heuristic. The resulting package
each with M activities, and two M learning curve calculations) are dramatically reduces the computational workload required by
not very different from those of Shtub’s “tangent approxima- Shtub’s approach (quick/tangent approximation followed by
tion.” For the composite learning curve of the M activities, the the streamlined full-blown procedure), while providing suffi-
secant approximation generates, a postiori, another approximate ciently accurate and robust results. This last point is illustrated
learning rate, rs, which can be viewed as that of an all-in-series in the next section, which compares the results from the two
network of the project’s M activities. This learning curve is used in approximation methods (for those practitioners who might
the resource utilization cost calculations (see Appendix 1). consider using only the quick approximation and ignore the
The direction of the divergence between a composite follow-up) to the “ideal” results of the full-blown procedure.
learning curve and its “secant approximation” is always the
same. Because the composite curve is convex (Amor & Results
Teplitz, 1998), the secant approximation always overesti- A spreadsheet is used to integrate the various elements of the
mates it. The amount of divergence cannot be bounded the- author’s approximation procedure for scheduling sequences of
oretically, but Amor and Teplitz demonstrate, via examples similar projects. An example from the house construction
from several industries, that the accuracy of the approxima- industry, developed by Amor and Teplitz (1998), is used to:
tion is well within 4%. In fact, the accuracy is within 2% in ■ Compare the accuracies of the tangent and secant approxi-
most cases (Amor & Teplitz, 1998). mation methods;
For comparison purposes, the numbers of required cal- ■ Examine the robustness of these methods over a range of
culations for the various procedures are summarized in plausible scenarios.
Table 1. Note that the number of calculations associated This example is used because it is conservative; it generated the
with the “secant approximation” is independent of N, the largest error (3.3%) among the various examples reported. The
number of projects in the program—an attractive feature author’s results indicate that the secant approximation method is
from a practitioner’s point of view. consistently more accurate than Shtub’s tangent approximation.
Therefore, an efficient program-scheduling tool is developed House Construction Example. The author uses the multi-
in this research effort by implementing the Amor and Teplitz unit construction program presented by Amor and Teplitz
(1998) approximation method in the context of Shtub’s (1991) (1998). This example had been drawn from Latona and
Number of projects N 30
Acitivity first performance tj 8.1; 2.8; 11.8; 17.7; 2.0; 21.4; 8.8;
time (days) 5.9; 8.1; 20.1; 8.8; 2.0; 5.9; 13.2;
5.9; 2.0; 2.0; 4.2; 21.4; 4.4; 4.0;
10.1 (based on data from Table 2)
IDEAL IDEAL
Resource Days
$1,400,000 80
Dollars
$1,200,000 60
Tgnt (0.85) Tgnt (0.85)
$1,000,000
Tgnt (0.80) 40
Tgnt (0.80)
$800,000 20
1 3 5 7 9 11 13 15 1 4 7 10 13 16 19 22 25 28
Nathan (1994), with the learning curve data extracted from program. Normally, of course, a given task is performed on
Teplitz and Amor (1993) and adapted to that situation. more than one house at a time (within any given sequence),
Table 2 lists the tasks, precedence, and times at standard resulting in a shorter program duration. While this assumption
reference quantity (SRQ) and rates of improvement associated avoids the complexity of dealing with “overlapping” projects, it
with the construction of a single home. The SRQ normally rep- does tend to overstate the beneficial impact of incorporating
resents a unit beyond which a fully experienced worker is con- learning curves (Amor & Teplitz, 1997).
sidered to demonstrate no perceptible future improvement. Because, in this paper, the house construction example is
In this example, the SRQ is assumed to be unit 100. Although used with a cost model, additional data are obtained from
there are 23 tasks listed in the table, only the first 22 are time- Shtub’s paper (1991) and from informal conversations with a
consuming activities; the twenty-third task simply represents building contractor. These data then were adapted to the cur-
the project completion. Based on their precedence relation- rent example. Table 3 lists the values of all the inputs for the cost
ships, the tasks are arranged in a moderately complex network model defined in Appendix 1. It includes the values that are
(see Figure 1). Coincidentally, there are 22 possible paths used later in the robustness excursions. For the base case, ana-
through the network, all which could be a critical path at any lyzed in the next section, Contract Schedule B (first 10 houses
time throughout this repetitive program. due on day 180, next 10 houses due on day 270, and last 10
The author examined the construction of a 30-house subdi- houses due on day 360) and a penalty cost of $50/day are used.
vision and, as done earlier (Amor & Teplitz, 1998), made two This table presents the information in the same order as it is
simplifying assumptions. The first assumption is that all work- introduced in Appendix 1 under “Assumptions and Notation.”
ers and crews are new hires with no previous experience with Comparative Accuracy Results. In this section, the author
their upcoming tasks. This assumption serves two purposes: compares the accuracy of the tangent and secant approxima-
■ If all workers are equally experienced/inexperienced, it is tions methods for the base case in the house construction
not necessary to maintain learning curves depicting the actual example. The streamlined, full-blown procedure is used to
repetitive position of each worker; obtain the ideal results, to which the approximated results were
■ The “early stage” of a learning curve is the most demanding compared. In the Figures 2, 3, 4, 5, 6, 7, 9 and 10, the curves cor-
on the accuracy potential of the approximations. responding to the full-blown procedure are labeled “IDEAL.”
The second assumption is that no house in a parallel sequence To use the tangent approximation method, an a priori value
is started until the previous house has been completed, i.e., for the common learning rate, r, must be selected. This value is
only x houses are under construction at any given time in the used for approximating:
■ The composite learning curve of all the activities, which is Ideal Completion Times and Costs. Figure 2 displays the
required for calculating the resource utilization cost; ideal house completion times for up to six parallel sequences of
■ The project composite learning curve, which is needed for projects as obtained with the full-blown procedure. Note that this
calculating the delivery dates and the penalty costs. figure illustrates the rapid increase in completion times and thus
Shtub furnishes no guidance for selecting a good value of r. in the likelihood of penalty costs as the number of sequences
In his example, he uses 0.85 and, because the activity learning decreases. Figure 3 displays the various program cost curves as
rates are not provided, one can only assume that 0.85 is a rel- functions of the number of parallel sequences used. Because
atively central value for the individual rates he has in mind. In there are not daily incentive “costs” in this example, the incentive
the author’s housing example, the largest and smallest learning “cost” curve lies completely on the horizontal axis. Because the
rates are 0.95 and 0.70 respectively; their mean is 0.861, their initial number of crews is one for all the activities, the hire and
median is 0.875, their mode is 0.90, and their Delionback fire cost curves simply increase linearly. Note how the resource
weighted average (NASA, 1975)—at the first unit—is 0.831. utilization cost and penalty cost curves “pull” in opposite direc-
From among the activities that are on the critical path of the tions and, essentially, determine the U shape of the total cost
first project, the largest and smallest learning rates still are 0.95 curve. For this base case, the optimal number of sequences to
and 0.70, respectively; their mean is 0.844, their median is operate is four, which yields a minimum total cost of $1,421,277.
0.90, their mode is 0.90, and their Delionback weighted aver- Comparative Resource Utilization Costs. Figure 4 dis-
age (NASA, 1975)—at the first unit—is 0.804. Which one of plays the composite learning curves for all M activities. Note
these numbers would make a good, a priori estimate for r? that the secant method approximates the ideal curve quite
The initial estimate for r can affect greatly the computations well. However, the tangent method can yield a broad spec-
in the tangent approximation method and, hence, the com- trum of approximations between Tgnt (0.80) and Tgnt
parative accuracy results. Consequently, the author was reluc- (0.90). As expected, the Tgnt (0.85) curve diverges from the
tant to select only one value for r. Based on the potential choic- ideal curve less than the other two tangents, but it does not
es listed, the author used three values: 0.80, 0.85, and 0.90. fit the ideal curve as well as Scnt (1–30). Figure 5 displays
Because they are extreme among the calculated measures of the resource utilization cost curves as functions of the num-
centrality and to develop empirical bounds on the error of the ber of parallel sequences used. Again, the ideal curve is well
tangent approximation method, the values of 0.80 and 0.90 approximated by the secant method, and the Tgnt (0.80)
were selected. The value of 0.85 also was used because it is and Tgnt (0.90) curves bound a broad spectrum of approx-
Shtub’s choice and because it is halfway between 0.80 and imations. Note, however, the convergence of the tangent
0.90. In the Figures 4, 5, 6, 7, 9, 11, 12, 13, and 14, the curves curves as the number of sequences increases. This is due to:
corresponding to these three estimates are labeled “Tgnt ■ The inverse relationship between the number of projects
(0.80),” “Tgnt (0.90),” and “Tgnt (0.85).” in a sequence and the number of sequences used;
As mentioned earlier, the secant approximation method pro- ■ The fact that increasingly shorter portions of the early
duces a postiori approximate learning rates, rs and rp, for the all-in- part of the composite learning curve (of all the activities) are
series and project networks. These rates are derived from the used as the number of sequences increases.
slopes of the log-linear secants (to the composite learning curves) Comparative Penalty Costs. Figure 6 displays the project
“passing through” projects/houses 1 and 30. In this example, composite learning curves. Although the vertical scale is dif-
these rates turn out to be 0.844 and 0.831, respectively. In the ferent from that of Figure 4, the general observations are
Figures 4, 5, 6, 7, 9, 11, 12, 13, and 14, the curves corresponding similar, as should be expected, for both sets of curves. Figure
to a secant approximation are labeled “Scnt (1–30).” 7 displays the penalty cost curves as functions of the
number of parallel sequences used. Again, the ideal curve is estimated the duration of several programs (in various indus-
well approximated by the secant method, and the Tgnt tries), each one using a single sequence of projects.
(0.80) and Tgnt (0.90) curves bound a broad spectrum of Consequently, it does not seem beneficial to work with their
approximations, especially when the number of sequences other examples. Instead, to investigate the robustness of these
is small. As with Figure 5, there is a convergence of the tan- approximation methods, the author conducted a series of
gent curves. This is due to similar, though slightly more excursions from the base case by varying two key parameters
complex, reasons as for the resource utilization cost curves. over relevant ranges of values. These two parameters are the
Comparative Hire and Fire Costs. Figure 8 displays the contract schedule and the daily penalties. The various other
hire and fire cost curves. These curves do not depend on parameters are far less significant.
either of the composite learning curves (all M activities or Four contract schedules are examined, including the one
projects). Consequently, they are the same for all the meth- that is used in the base case. From the tightest to the loosest
ods being compared. schedule, they are labeled A, B, C, and D; their details are
Comparative Total Costs. Figure 9 displays the total cost shown in Table 3. For each contract schedule, five daily penal-
curves as functions of the number of parallel sequences used. ties are examined; they range from $0 to $500. (As mentioned
As expected from the results, the ideal curve is well approxi- earlier, the base case uses Contract Schedule B and a $50/day
mated by the secant method. However, the tangent method penalty.) Because, when there are no penalty and incentive
yields a broad range of possible approximations between Tgnt costs the total program cost is independent of the contract
(0.80) and Tgnt (0.90). Of course, Tgnt (0.85) provides a bet- schedule, there are 17 data points/excursions (including the
ter fit, but it is not a good as that of Scnt (1–30). base case) for use in the robustness analysis.
Comparative Optimization Results. For each method and Ideal Minimum Total Cost. Figure 10 displays the ideal mini-
parameter value discussed, the optimal number of parallel mum total cost, TC*, for each of the 17 excursions mentioned, as
sequences (x*), its associated minimum total cost (TC*), and obtained with the streamlined full-blown procedure. It also indi-
the error and percentage error from the full-blown procedure cates in parentheses under each data point the optimal number of
are presented in Table 4. parallel sequences, x*, which yields the plotted value of TC*.
Clearly, the secant approximation method results in much As expected, for any given penalty rate, the tighter the contract
smaller cost estimation errors than the tangent approximation schedule, the more sequences are needed to minimize the total
method. Although Scnt (1–30) would recommend using five cost—because there is a greater likelihood of incurring penalties.
sequences rather than four, the error associated with following Naturally, the tighter the schedule, the larger the minimum total
that recommendation would result in an error of only $173 cost. Also as expected, for any given contract schedule, as the
($1,421,277 with four sequences in IDEAL vs. $1,421,450 with penalty rate increases, the number of sequences needed to mini-
five sequences in IDEAL) after the fact, i.e., after program mize the total cost tends to increase. The larger the penalty rate,
implementation and assuming that the ideal curve turns out to the larger the minimum total cost—until x* becomes greater
be a perfect forecast. Thus, for the base case example, the secant than the number of sequences beyond which there would be no
method provides an approximation to Shtub’s full-blown pro- more late projects. This turns out to be the case for Contract
cedure, which is superior to almost any reasonable application Schedule D, when x* becomes 3, because for that relatively loose
of the tangent approximation method. However, what about schedule there are no late projects when x is greater than 2.
the robustness of these results?
Comparative Errors Under
Comparative Robustness Results: the Various Contract Schedules
Excursions from the Base Case Figures 11, 12, 13, and 14 display the percentage error of each
The base case is derived from the example in Amor and Teplitz approximation method for each daily penalty under Contract
(1998) that turned in the largest error (3.3%) when they Schedules A, B, C, and D. That is, for each excursion, the
$1,400,000
$120,000
Tgnt (0.90) Hire
$1,200,000
$100,000
IDEAL Fire
$1,000,000
$80,000
Dollars
Dollars
Scnt (1–30)
$800,000
$60,000
$600,000 Tgnt (0.85)
$40,000
$400,000
Tgnt (0.80)
$20,000
$200,000
$0 $0
1 3 5 7 9 11 13 15 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
$2,600,000
Tgnt (0.90) $1,800,000 A: 120/180/240
$2,400,000
Scnt (1–30)
Dollars
Dollars
$2,000,000 $1,400,000
C: 120/405/540
Tgnt (0.85)
$1,800,000 D: 360/540/720
$1,200,000
$1,000,000
$1,400,000
$1,200,000 $800,000
1 3 5 7 9 11 13 15 1 100 200 300 400 500
method, that procedure will make the scheduling of sequences Appendix 1. The Cost Model and its Characteristics
of similar projects even more valuable for program managers. Assumptions and Notation
For the program:
References ■ The program consists of N identical projects, each result-
Amor, J.P., & Teplitz, C.J. (1993). Improving CPM’s accuracy ing in the completion of one unit of the product;
using learning curves. Project Management Journal, 24 (4), 15–19. ■ Each project, j, refers to one product unit; j = 1 through j =
Amor, J.P., & Teplitz, C.J. (1997). On the relative impacts N. And “J” represents the set of all projects within the program;
of learning and overlap on the scheduling of programs with ■ A due date, Dj, is specified for each project according to
repetitive projects. Proceedings of the Twenty-Sixth Annual the contract schedule;
Meeting of the Western Decision Sciences Institute, Kamuela, HI. ■ A delivery date, Tj, eventually occurs for each project as a
Madison, WI: Omnipress, 729–732. result of executing the program;
Amor, J.P., & Teplitz, C.J. (1998). An efficient approxima- ■ A penalty rate, pj, applies to each project completed
tion procedure for project composite learning curves. Project beyond schedule, e.g., in $/day;
Management Journal, 29 (3), 28–42. ■ An incentive rate, gj, applies to each project completed
Badiru, A.B. (1995). Incorporating learning curve effects ahead of schedule, e.g., in $/day.
into critical resource diagramming. Project Management
Journal, 2 (2), 38–45. For the projects:
NASA. (1975). Guidelines for application of learning/cost ■ Each project consists of a network of M activities;
improvement curves (Report TM X-64968). Washington, DC: ■ Each activity, i, is performed by one unit of the specific
Delionback, L.M. resource type required for that activity, e.g., one worker, one
Latona, J.C., & Nathan, J. (1994). Cases and readings in produc- crew; i = 1 through i = M. And “I” represents the set of all
tion and operations management. Boston, MA: Allyn and Bacon. activities within a project;
Shtub, A. (1991). Scheduling of programs with repetitive ■ A duration, ti, applies to each activity when performed for
projects. Project Management Journal, 22 (4), 49–53. the first time (by one unit of its associated resource type);
Shtub, A., LeBlanc, L.J., & Cai, Z. (1996). Scheduling pro- ■ A learning curve exponent, bi, corresponds to the learning
grams with repetitive projects: A comparison of a simulated rate, ri, of resource type i. (bi = log ri / log 2);
annealing, a genetic and a pair-wise swap algorithm. ■ A resource cost rate, ki, applies to each resource unit when
European Journal of Operational Research, 88, 124–138. it is in use, e.g., in $/day;
–10 –10
–20 –20
1 100 200 300 400 500 1 100 200 300 400 500
Figure 11. Percent Error in Min TC – Contract Schedule A Figure 12. Percent Error in Min TC – Contract Schedule B
(120/180/240) (180/270/360)
■ Each resource type has an associated hiring cost, hi, e.g., The Cost Model
in $/crew;
■ Each resource type has an associated firing cost, fi, e.g., TC(x) = RC(x) + HC(x) + FC(x) + PC(x) – IC(x)
in $/crew;
■ Each resource type is available in a certain quantity, e.g., N/x
number of crews, ci, at the beginning of the program, e.g., ci where: RC(x) = x∑ ∑ ki ti (j)bi if N/x is an integer or
∋
j=1 i I
= 0 through ci = N, and must returned to that level at the end
of the program. [N/x]+
= (x – [N/x]+ x + N) ∑ ∑ ki ti (j)b ∋
i
j=1 i I
For the cost functions and the optimization process:
■ x represents the number of (equal/near equal) parallel [N/x]+–1
j=1 i I
■ Tj(x) represents the delivery date of project j when using
x parallel sequences;
■ RC(x) represents the resource utilization cost when using HC(x) = ∑ hi Max{x – ci,0}
∋
i I
x parallel sequences;
■ HC(x) represents the hiring cost when using x
parallel sequences;
■ FC(x) represents the firing cost when using x FC(x) = ∑ fi Max{x – ci,0}
∋
parallel sequences; i I
Scnt (1–30)
Scnt (1–30)
Error in Min TC (%)
Tgnt (0.80)
Tgnt (0.80)
0 0 Tgnt (0.90)
Tgnt (0.90)
–10 –10
–20 –20
1 100 200 300 400 500 1 100 200 300 400 500
Figure 13. Percent Error in Min TC – Contract Schedule C Figure 14. Percent Error in Min TC – Contract Schedule D
(270/405/540) (360/540/720)
This article is copyrighted material and has been reproduced with the permission of PMI. Unauthorized reproduction of this material is strictly prohibited.
A Measure of Software
Development Risk
Gary Klein, The University of Colorado at Colorado Springs, College of Business and
Administration, 1420 Austin Bluffs Parkway, Colorado Springs, CO 80933–7150 USA
T. Selwyn Ellis, Louisiana Tech University, College of Administration and Business, Department
of Computer Information Systems, Ruston, LA 71272–0001 USA
Position
IS executive 8 6 14
IS manager 12 10 22
IS project leader 44 39 83
Other IS professionals 18 5 23
Work experience
1–5 years 5 3 8
6–10 years 12 9 21
11–15 years 20 5 25
16–20 years 20 9 29
21 or above 25 33 58
Gender
Male 65 48 113
Female 18 12 30
2–3 members 1 1 2
4–5 members 13 3 16
6–10 members 22 21 43
11–15 members 12 9 21
16–20 members 15 14 29
21–25 members 18 13 31
represents a more parsimonious representation of observed co- anticipated due to the diametrically arranged adjectives between
variances, it should be accepted over the baseline as a “true” the software development risks and project success dimensions.
representation of model structure. Essentially, it confirms that software risk does dampen project
Assessment Question 2. Are software development risks success. In fact, the strength of the correlation shows that risk
significantly associated with project success? should be measured and controlled to improve success and that
A correlation between software development risk and project the measure proposed is a good indicator of total risk. The
success was found to be –0.22 using a confirmatory factor results confirm the authors’ anticipation that software develop-
analysis model (see Appendix 1). This negative relationship was ment risk adversely affects the outcomes of a project.
1. Need for new hardware 28. Great intensity of conflicts among team members
2. Need for new software 29. Great intensity of conflicts between users and
3. Large number of hardware suppliers team members
4. Large number of software suppliers
Extent of changes brought
Application size
30. The system requires a large number of users’
5. Large number of people on team tasks to be modified
6. Large number of different stakeholders 31. The system will lead to major changes in the
on team, e.g., information systems staff, users, organization
consultants, suppliers, customers
7. Large project size Resources insufficient
8. Large number of users will be using this system
9. Large number of hierarchical levels occupied by 32. In order to develop and implement the system,
users who will be using the system, e.g., office the scheduled number of people per day is
clerks, supervisors insufficient
33. In order to develop and implement the system,
Lack of team’s general expertise the dollar budget provided is insufficient
10. Ability to work with uncertain objective Lack of clarity of role definitions
11. Ability to work with top management
12. Ability to work effectively as a team 34. Role of each member of the team is not
13. Ability to understand human implications clearly defined
of a new system 35. Role of each person involved in the project is not
14. Ability to carry out tasks effectively clearly defined
36. Communications between those involved in the
Lack of team’s expertise with the task project are unpleasant
Lack of clarity of 34. Role of each member of the team is not clearly defined 0.78
role definition 35. Role of each person involved in the project is not clearly defined 0.79
36. Communications between those involved in the project
are unpleasant 0.77
Lack of user 41. Users are not very familiar with system development 0.67
experience on 42. Users have little experience with the activities to be supported
systems development by the future system 0.81
43. Users are not very familiar with this type of application 0.77
45. Users are not familiar with data processing as a work tool 0.71
Lack of 20. Users have a negative opinion about the system meeting their needs 0.77
user support 21. Users are not enthusiastic about the project 0.79
23. Users are not available to answer the question 0.72
24. Users are not ready to accept the changes the system will entail 0.79
25. Users slowly respond to development team requests 0.69
Lack of team 13. Ability to understand human implications of a new system 0.69
expertise 15. In-depth knowledge of the functioning of user department 0.70
16. Overall knowledge of organizational operation 0.75
17. Overall administrative experience and skill 0.75
18. Expertise in the specific application area of the system 0.71
Quantitative Application in the Social Science. Thousand Oaks, development. Journal of Management Information Systems, 13
CA: Sage Publications. (1), 145–166.
MacCallum, R.C. (1986). Specification searches in covariance Schunk, D.H. (1989). Self-efficiency and cognitive skill
structural modeling. Psychological Bulletin, 100 (1), 107–120. learning. In C. Ames & R. Ames (Eds.). Research on Motivation
McFarlan, F.W. (1981). Portfolio approach to informa- in Education (Vol. 3). San Diego: Academic Press.
tion systems. Harvard Business Review, 59 (5), 142–150. Schwalbe, K. (2000). Information technology project
Meyer, R.L. (1998). Avoiding the risks in large software management. Cambridge, MA: Course Technology.
system acquisitions. Information Strategy, 14 (4), 18–33. Zmud, R.W. (1980). Management of large software
Nidumolu, S. (1995). The effect of coordination and development efforts. MIS Quarterly, 4 (1), 45–55.
uncertainty on software project performance: Residual per-
formance risk as an intervening variable. Information Appendix 1
Systems Research, 6 (3), 191–219. Determination of Factor Structure
Pressman, R.S. (1992). Software engineering: A practitioner’s First, in an exploratory factor analysis using principal compo-
approach (3rd ed.). New York: McGraw-Hill. nent analysis (PCA), if a given item has a meaningful loading,
Saleem, N. (1996). An empirical test of the contingency i.e., loading greater than 0.45 on more than one, that item is
approach to user participation in information system dropped. Likewise, items with less than a 0.45 loading were
2 0.77
Technology
Acquisition
3
7 Project
0.79 Size
8
0.83
15 0.78
0.97
Lack of Team
16 Experience
0.56
17
0.43
0.84
18 20
0.83
21
Lack of User
23 Support
0.84
0.22 0.46
24
0.72 0.43 0.42
25
34 0.97
Lack of Role
0.94
35 Definition
0.54
36
0.30
41
42 0.85
Lack of User
0.92 Experience
43
0.62
45
Table A4. Properties of the Revised First-Order Software Development Risk Model
dropped. The variables remaining after the PCA are shown in complexity may best be measured by direct observation rather
Table A3. All the remaining items loaded as expected, except than psychometric scales (Jiang, Klein, & Pick, 1995).
that the two a priori dimensions “the lack of team’s general Subsequent confirmatory factor analysis (CFA) determined
expertise” and “lack of team’s expertise with the task” indicat- construct validity (MacCallum, 1986). The expectation was
ed being combined. Four dimensions of the software develop- that each of the developed scales in Table A3 uniquely measures
ment risk did not have a single item with a significant load- its associated factor and that the system of factors measures
ing—including intensity of conflicts, extent of changes different aspects of software development risks. The
brought, resource insufficiency, and application complexity— Covariance Analysis of Linear Structural Equations (CALIS)
and were not considered in further analysis. Application procedure of Statistical Analysis Systems (SAS), version 6.12,
was utilized as the analytical tool for testing the measurement A second-order factor model captures correlations among
and structural equation models. the first-order constructs. The efficacy of such a structure can
One important assumption of confirmatory factor modeling be tested using a comparative methodology for higher-order
is multivariate normality. Because multivariate normality is diffi- factor models (Bollen, 1989; Joreskog, 1993). A second
cult to test, it is recommended that univariate normality among order analysis also was conducted using a CFA over the six
variables be initially tested (Anderson, Hair, & Tatham, 1992). identified factors. The fit indices showed a good fit with
Such testing was accomplished through examination of the AGFI = 0.81, CFI = 0.94, NNFI = 0.93, and the Chi-
moments around the mean of each variate’s distribution (Bollen, square/degrees of freedom = 1.5. Further empirical support
1989). Among the variables of this study, analysis of these statis- for acceptance of the higher-order factor structure is found in
tics suggests no serious departures in univariate normality. the significance of estimated parameters as well as the
When a CFA model provides a reasonably good approxi- amount of variance explained by the structural equations. All
mation to reality, it accounts for the observed relationships in structural equation parameters exhibit significantly high t-
the data set. Four fit indices are typically used to identify over- values (all > 3.18, p = 0.01). Specifically, the CFA path
all goodness of fit in a CFA: between overall software risks and its underlying first-order
■ Comparative fit index (CFI); dimensions are 0.61 for technical acquisition, 0.58 for pro-
■ The Bentler-Bonett non-normed fit index (NNFI); ject size, 0.69 for team expertise, 0.53 for user support, 0.44
■ Adjusted goodness-of-fit index (AGFI); for role definition, and 0.40 for user support.
■ The Chi-square divided by the degrees of freedom In the project success measurement model initial
(< 3 preferred). estimation, item 5, due to a lack of significant loading, was
Values greater than 0.90 are desirable for CFI and NNFI and deleted from the model. The final model described indicat-
greater than AGFI should be equal or greater than 0.80. ed a good fit AGFI = 0.87, CFI = 0.97, NNFI = 0.95, and the
The CFA measurement model is shown in Figure 1. The Chi-square/degrees of freedom = 2.6. Internal consistency
items correspond to those of Table A3. Table A4 provides a (reliability) was assessed using a measurement model and
summary of the analysis. In the initial phase of model estima- reported in Table A5 (Joreskog & Sorbam, 1989). The com-
tion, several items (7, 13, 17, 18, and 23) were deleted due to posite reliability (0.92), Cronbach (1951) alpha (0.90), and
a significant cross-loading with other constructs. In the subse- variance extracted estimate (0.65 > 0.50) indicate strong
quent tests for discriminant validity (discussed in the following internal consistency of this construct. In summary, the
section), items 3 and 4 were deleted due to a lack of significant project success measure has strong validity and reliability.
reliability. The final model, Chi-square/degrees of freedom = It has been suggested that the efficacy of second-order
1.41 < 3, AGFI = 0.81, CFI = 0.95, and NNFI = 0.94, indicates models be assessed through examination of the target (T)
a good fit between the model and data. In addition, the coefficient [T = Chi-square (baseline model)/Chi-square
significance of all parameter estimates, e.g., t-value > 3, shown (alternative model)] (Hocevar & Marsh, 1985). This coeffi-
in Table A4 support the revised model. cient has an upper bound of 1.0 with higher values implying
This article is copyrighted material and has been reproduced with the permission of PMI. Unauthorized reproduction of this material is strictly prohibited.
A Hybrid Intelligent System
to Facilitate Information System
Project Management Activities
Hamid R. Nemati, Department of Information Systems and Operations Management,
The University of North Carolina at Greensboro, 440 Bryan Building, P.O. Box 26165,
Greensboro, NC 27014–6165 USA
Dewey W. Todd, Department of Decision Sciences, Georgia State University, University Plaza,
Atlanta, GA 30303–3083 USA
Paul D. Brown, Department of Decision Sciences, Georgia State University, University Plaza,
Atlanta, GA 30303–3083 USA
Any Tasks
> 80 Hours?
Enter Percentage
of Total Estimate
Yes
for Which Tasks are
No > 80 Hours
What is the
Expected Duration
< 2 Weeks of the Project?
2–4 1–3
> 3 Months
Weeks Months
Acceptable Estimate
Acceptable Estimate Highly Questionable
Add 15% Management Acceptable Estimate
Adjust Management Estimate – Needs
Overhead Adjust Contingency
Overhead to be Reviewed
Add 10% Contingency
Note. The flow diagram begins with the box labeled “Similar project done in the past?” It then flows through the qualifiers and variables and finally to
the choices. The flow diagram does not attempt to display or determine any weights used in the decision process. All alternatives are treated equally
and displayed as such.
Figure 1. Flow Diagram Showing the Influences and Relationships Among Qualifiers and Variables with the Choices
Comfort level These weights only come into play if “Similar Project” is No. The numbers 1 through
4 correspond to the four selections.
Tasks > 80 This selection by itself does not impact the weights.
Percentage > 80 If there are tasks greater than 80 hours, the choice from the 0 to 100 range
is divided into four discrete weights and applied.
Review frequency These weights only come into play if “duration” is greater than two weeks.
The numbers 1 through 4 correspond to the four selections.
Experience of estimator The choice from the 0-to-100 range is divided into five discrete weights and applied.
Experience of programmer The choice from the 0-to-100 range is divided into five discrete weights and applied.
Table 1. Summary of Each Possible Choice and How It Would be Influenced by Each of the Qualifiers and Variables
increasing productivity and quality. The development of an ES Rule Set Definition. The elements that make up the ES for
involves the construction of a problem-specific knowledge project estimate validation and contingency are shown in
base by acquiring knowledge from experts or documented Figure 1. The qualifiers for this exercise are:
sources. The most widely used form of representing knowledge 1. Was the estimate developed from a similar past project
in the knowledge base is using rules. The system then uses the (similar implies that at least 40% to 60% of the tasks would
rules in the knowledge base to draw conclusions similar to be roughly similar in nature)? A yes to this question takes the
those drawn by the experts. user to Qualifier 3, otherwise to Qualifier 2;
The first component of this intelligence system is an ES. It is 2. How comfortable are you with the estimates considering
designed to help management validate estimates of the deliv- the lack of familiarity with the tasks? The user has four
ery time and quality of a given project. After extensive review “fuzzy” options here ranging from “very comfortable” to
of project management activities and conducting interviews “uncomfortable.” These responses have a weighted effect on
with a number of project leaders, a set of rules was compiled. the outcome;
This was accomplished according to steps: 3. Are any of the tasks in the project estimated to be more
1. The qualifiers used to decide the accuracy of the estimate than 80 hours? There is an “80 hour rule” that basically says
were written up; that any estimate more than 80 hours is wrong (imprecise).
2. The variables that would be used with the qualifiers were The authors’ domain expert recommended that tasks greater
determined; than 80 hours be broken down, if possible. If no, the user
3. The possible choices were determined; continues to Qualifier 4, otherwise to Variable 1 to determine
4. A flow diagram was developed to show how these elements a percentage;
would interact to move from qualifying to choosing (Figure 1); 4. What is the expected duration of the project? This question
5. A table was developed (using Lotus 1-2-3® ) as a matrix of is asked to determine the need for management review meet-
the possible effects of each qualifier’s choice and the range of ings. It has been shown conclusively that frequent reviews pos-
values in the variables and how they would affect the certainty itively influence the overall development time because delays
factor for each choice (Table 1 and Table 2); are caught and corrected early. If the project duration is less
6. The elements were all entered into EXSYS Professional® than two weeks, the user proceeds to Variable 2. There are three
according to the matrix; other choices ranging from “two to four weeks” to “greater than
7. The functional screens were developed using EXSYS Editor three months.” These values do not currently affect the weight-
and an editor called Brief®; ing but may in the future. The user proceeds to Qualifier 5;
8. Multiple iterations of sample estimates were tested to vali- 5. How often will the project be reviewed by management?
date the system. If once a week is selected, there is no weighting, but the other
Comfort level 1 = NC 1 = NC 1 = NC 1 = NC 1 = NC
2 = –2 2 = –2 2=3 2=3 2=3
3 = –6 3 = –6 3=8 3=8 3=8
Tasks > 80 NC NC NC NC NC
Percentage > 80 0 – 33 = –1 0 – 33 = –1 NC 0 – 33 = 2 0 – 33 = 2
34 – 66 = –3 34 – 66 = –3 34 – 66 = 4 34 – 66 = 4
> 66 = –6 > 66 = –6 > 66 = 8 > 66 = 8
Duration NC NC NC NC NC
Review frequency 1 = NC 1 = NC 1 = NC 1 = NC 1 = NC
2 = –2 2 = –2 2=3 2=3 2=3
3 = –6 3 = –7 3=7 3=8 3=8
Experience of estimator 0 – 33 = –7 0 – 33 = –7 0 – 33 = 7 0 – 33 = 7 0 – 33 = 8
34 – 66 = –3 34 – 66 = –3 34 – 66 = 3 34 – 66 = 3 34 – 66 = 4
> 66 = NC > 66 = NC > 66 = NC > 66 = NC > 66 = NC
Experience of programmer 0 – 33 = –7 0 – 33 = –7 0 – 33 = 7 0 – 33 = 7 0 – 33 = 8
34 – 66 = –3 34 – 66 = –3 34 – 66 = 3 34 – 66 = 3 34 – 66 = 4
> 66 = NC > 66 = NC > 66 = NC > 66 = NC > 66 = NC
set using the training algorithm. The adjustment to the interconnections between more than the two I/O layers of the
weight set can be done in either a single update, in which the network change in the learning process.
weights are changed every time the network is presented The network has additional layers, hidden from the envi-
with a new case data or an epoch update, in which the net- ronment. The most widely multilayer neural network used
work changes the weights only after it receives the whole network today is the Backpropagation neural network
training data set. For faster training, it is possible to specify (Chauvin & Rumelhart, 1995; Rumelhart, Hinton, &
an epoch size. For example, an epoch size of five means that Williams, 1986). The basic idea in a Backpropagation net-
the network adjusts the weights only after it has been pre- work is that the neurons of a lower layer send their outputs
sented the entire training set data five times. The training up to the next layer. It is a neural network that is very good
continues until a prespecified condition is reached. Training in generating input-to-output mappings based on computa-
a neural network is an iterative process. The developer goes tions of interconnection of its nodes. It is a fully connected,
through the designing system, training the systems and eval- feed forward, hierarchical multilayer network, with hidden
uating the system repeatedly. layers and no intraconnections.
In a neural network, neurons are grouped into layers, or Figure 2 shows a typical Backpropagation neural network.
slabs, that include neurons of the same type. There also are dif- The network shown has four input neurons, four output neu-
ferent types of layers. The input layer consists of neurons that rons, and a hidden layer with three neurons. The weights
receive input from the external environment. The output layer between output layer and the hidden layer are updated based
consists of neurons that communicate to the user or external on the amount of error at the output layer, using generalized
environment. The hidden layer consists of neurons that only delta rule (an iterative steepest descent procedure, minimizing
communicate with other layers of the network. The authors the least-mean-square error measure between the desired and
define a multilayer neural network as a network in which the actual outputs). The error of the output layer propagates back
Hidden
Layer
Stinchcombe, & White, 1989). To build the prediction model Because this produced a satisfactory network for predicting
for “Score,” building two prediction models actually was actual hours, the authors then developed the second neural
necessary. The first, which in theory was based on the vari- network, using the same database. This one was called “Score”
ables “manager” through “programmer experience,” was and contained a neural network for predicting the score of a
developed to predict actual hours, because this is, of course, given project based on all of the previous variables plus the
important in predicting how close the actual hours will actual hours. Here, quality is an input from the user, because
come to the estimate. For example, a project that completed no ES has yet been developed for this function. The network
exactly on time and achieved a perfect quality rating would using the training set and testing it using the testing data set,
have a score of 100. A score of less than 100 would indicate the authors obtained a high correlation between the model
a project that suffered from quality of less than 100%, actu- and the predicted actual hours (correlation coefficient was
al hours exceeding its estimate, or a combination of both. A 0.738 and root MSE was 0.0898).
score higher than 100 indicates a project with high quality or
with actual hours that were less than estimated, or some Future Development of Hybrid Systems
combination of both. In this section, the authors provide a vision for future devel-
The authors experimented with using ANN models to opment of intelligence hybrid systems for project manage-
predict the actual hours required for a project. This neural ment and QA of information systems projects. The authors
network model consisted of the 15 input variables and one provide a breakdown project management/QA processes
output variable. In training the network, the authors and analyze the details of development of individual com-
employed the BackPack Neural Network software package ponents of an AI-based hybrid system that can be used to
using its default settings. The network had one hidden layer support these processes.
with 10 neurons, an epoch size of five, and 20,000 as the The steps that are typically taken in an IT development
maximum number of iterations. Training the network project and the associated recommended AI systems for each
using the training set and testing it using the testing data piece are:
set, the authors obtained a high correlation between the ■ Team Member Profiles. Recommended AI Approach:
model and the predicted actual hours [correlation coeffi- Expert System. This is a crucial step in developing a tactical
cient was 0.9895 and root mean squared errors (MSE) was estimating system. In studying project management, one
0.03350]. The results indicates that the neural network does not have to delve too deeply to realize how much influ-
model is a powerful predictive tool. ence the make-up of a team has on the completion of the
This article is copyrighted material and has been reproduced with the permission of PMI. Unauthorized reproduction of this material is strictly prohibited.
Bud Baker, Wright State University, Department of Management, 270 Rike Hall, Dayton, OH
45435–0001 USA
This article is copyrighted material and has been reproduced with the permission of PMI. Unauthorized reproduction of this material is strictly prohibited.
Table 1. Separating the 39 Processes Belonging to the Nine Knowledge Areas by Planning Processes and Other Processes
be identified for each planning process. For example, the major based on the learning curve theory, which has proved ongoing
product of the scope definition process is the work breakdown improvement as a function of the number of repetitions
structure (WBS). Table 2 lists the major products for all plan- (Griffith, 1996; Snead & Harrell, 1994; Yiming & Hao, 2000;
ning processes. Watson & Behnke, 1991).
Participants in the study were project managers and others
The Study who are involved in project management activities. The 282
A field study was conducted to evaluate the extent of the pro- participants came from different project management work-
ject manager’s involvement in the planning processes and to shops administered as part of internal or external training sem-
evaluate their quality. A major problem in designing this study inars. The portion of the questionnaire that they were asked to
was to establish a way to evaluate the extent to which planning complete, which served as the database for this study, appears
processes were used in projects and their quality level. For this in Appendix 1.
purpose, the following assumption was made: The quality of a The following scale was used for evaluating the intensity of
process is a function of the frequency with which it is used to use of the different products:
obtain the major product of the process. This assumption is ■ 5 = The product is always obtained;
■ 4 = The product is obtained quite frequently; ■ Medium quality areas, to which cost, quality, and procure-
■ 3 = The product is obtained frequently; ment belong. The score for this group is around 3;
■ 2 = The product is seldom obtained; ■ Poor quality areas, to which risk and communications
■ 1 = The product is hardly ever obtained; belong. The score of both is around 2.3.
■ 9 = The product is irrelevant to the projects I am involved in;
■ 0 = I do not know whether the product is being obtained. Analysis and Discussion
Table 3 summarizes the results. As may be seen from Table Assuming that for successful completion of a project, all
3, the quality of the processes ranges from a low of 2.0 for processes in the nine knowledge areas should be high quali-
qualitative risk analysis up to 4.2 for activity duration estimat- ty, we should discuss ways to improve the poor performance
ing. In other words, qualitative risk analysis is hardly practiced. areas. Performance in a specific area is a function of the pro-
There is a high correlation among the quality of products ject manager’s know-how of that area, the know-how of
belonging to the same knowledge area; that is, they are either other professionals, such as functional managers, who are
low or high. For example, the quality of all risk processes is involved in the specific process, and with the project manag-
below 2.9, whereas the quality of all the scope processes is er’s ability to affect the area and its attendant processes.
above 3.5. The quality of each knowledge area is calculated by In general, functional managers are accountable for the
the average quality of the processes belonging to it, as present- proper execution of the specific work packages assigned to
ed in Table 3, and charted in descending order in Figure 1. them, whereas the project manager is responsible for integra-
The three groups of knowledge areas identified in Figure 1 are: tion and infrastructure-related work packages and activities. For
■ High quality areas, to which integration, scope, time, and example, a project manager is heavily involved with the process
human resources belong. The score for this group is around 4; of establishing the WBS for the whole project and is directly
Table 3. Quality of Major Planning Processes Within the Different PMBOK® Guide Knowledge Areas
accountable for it. However, a process such as quantitative risk within the same organization, much care is taken to include all
analysis, which must be done separately for each individual relevant expectations in the contract, to ensure that the external
work package, requires heavy involvement of the functional suppliers follows all requirements. The combination of employ-
manager and those of his employees who are responsible for ees working on a work package without the relevant know-how
executing the work package (assuming a matrix organization). in relevant risk management processes, and a project manager
Therefore, to perform the process properly, all the involved par- without the ability to endow them with this knowledge makes
ties should possess know-how of risk management methods. it unrealistic to expect effective implementation of project man-
Process know-how means being familiar with the required agement processes. In other words, if a certain process is to be
inputs, the tools and techniques used, and the desired outputs carried out in an appropriate manner, the people involved in the
of the process. If this is not the case, then risk management or process need the relevant know-how.
any other relevant processes can’t be effectively handled on a Table 4 divides the planning processes according to the
work package level or on the integrated level. A project man- individual who expends most effort in properly executing a
ager functioning in a matrix environment gets work packages process (the project manager or the functional manager), keep-
done by “contracting out” to internal suppliers, who are the ing in mind that the project manager is held responsible for all
functional managers in the organization, and purchasing other processes. In general, in a matrix organization, a functional
work packages from external sources via procurement process- manager is responsible for carrying single work package relat-
es. Because procurement requires the signing of formal docu- ed activities, whereas a project manager deals with work pack-
ments, as compared to working with functional managers ages that are integrated in nature. For example, a functional
4.0
3.9
4 3.8
3.7
3 2.9
Planning Quality
2.3 2.3
0
s
ns
n
st
ity
sk
e
t
en
ce
ag
io
op
Co
tio
al
Ri
at
em
ur
Ti
er
Sc
Qu
ica
gr
so
Av
ur
te
un
Re
oc
In
m
Pr
an
m
Co
m
Hu
Knowledge Areas
manager carries the major responsibility for activity definition, The situation is not the same for the communications
whereas the project manager has to integrate all project activi- area, because its only planning process—communications
ties into activity sequencing, and so on. planning—is an integrated work package with which the pro-
Keep in mind that a project manager is regarded as a func- ject manager is the most heavily involved. Although other
tional manager when he or she works on the professional work professionals, such as functional mangers, must be involved
packages that fall within his or her direct area of accountabili- in the communications planning process, the project manag-
ty. For example, the project manager is directly accountable for er should lead the overall effort and have a strong command
the scope-definition process, the major output of which is the of the relevant tools and techniques. Therefore, lack of know-
WBS. However, the project manager will not be able to execute how on the part of the functional manager is not the main
this process out without the cooperation and the know-how of reason for poor performance in this area. The explanation has
functional managers. The same holds true for all processes, to be sought elsewhere.
regardless of the manager who is directly accountable. According to the PMBOK® Guide, “Communications
Further, the two areas that belong to the “poor quality” group, planning determines the information and communications
that is, risk management and communications, displayed poor needs for the stakeholders: Who needs what information,
results. Because risk management is part of the PMBOK® Guide when they will need it, and how it will be given to them”
and, therefore, also part of the Project Management Professional (p.119). This area is probably the most difficult for the
(PMP®) examination, one may assume that project managers are project manager to plan, because it requires getting present
familiar with it. However, functional managers do not have any and future information needs from all the stakeholders. Very
formal risk management training, and one may assume that they few formal tools and techniques are available to the project
suffer from lack of the requisite know-how with regard to risk manager for supporting the communications area. PMBOK®
management processes. It is therefore not surprising that func- Guide offers only one unstructured tool, namely, stakeholder
tional managers are of little help in performing risk management analysis. The lack of proper and easily accessible tools cou-
processes. The project manager is thus left alone to struggle with pled with the difficult task of identifying future information
it, with limited success according to the results of this study and needs of stakeholders makes communication a very difficult
others (Ibbs & Kwak, 2000; Mullaly, 1997). process to plan.
Table 4. Mapping of the 21 PMBOK® Guide Planning Processes According to the Individual (Project or Functional
Manager) Who Performs Most of the Activities Required to Execute the Planning Process
The areas that belong to the middle quality group are cost, tional manager also should get intensive but adapted project
quality, and procurement. From Table 4, we can identify management training. The agent for such a fundamental
the manager who needs most of the know-how required change in the organizational culture can’t be the project man-
to properly execute each of the processes. While the project ager alone. It is essential that it be sponsored at a high level
manager leads the process in the quality area and the func- of the organization and even treated as a project by itself.
tional manager in the procurement area, both are involved in
the cost area. Therefore, the key to success in these areas is References
different tailor-made training modules for the project and the Cleland, D.I. (1994). Project management: Strategic design
functional manager. and implementation. New York: McGraw-Hill.
Fleming, Q.W., & Koppelman, J.M. (1994). The earned
Conclusion value concept: Back to the basics. PM Network, 8 (11), 27–29.
As the person who is fully accountable for the success of the Griffith, T.L. (1996). Negotiating successful technology
project as a whole, the project manager is responsible for implementation: A motivation perspective. Journal of
overcoming the difficulties encountered in guaranteeing Engineering & Technology Management, 13 (1), 29–53.
that all planning processes are properly executed. To resolve Ibbs, C.W., & Kwak, Y.H. (2000). Assessing project man-
the problems, the project manager should identify the agement maturity. Project Management Journal, 31, 32–43.
events that have a negative impact on the successful com- Kimmons, R.L. (1990). Project management basics.
pletion of the project and develop explicit mitigating plans Houston, TX: Kimmons-Asaro Group Ltd. Inc.
to accommodate them. Mullaly, M. (1998). 1997 Canadian project management
In some areas, such as communications and quality, the baseline study. Proceedings of the Project Management
project management community should develop better tools Institute’s 29th Annual Symposium, Long Beach, CA. Newtown
and techniques to support the project manager’s efforts. In Square, PA: PMI, 375–384.
other areas, such as risk and cost, more emphasis should be PMI Standards Committee. (2000). A guide to the project
placed on the training of functional managers in the use of management body of knowledge. Newtown Square, PA: Project
the relevant tools and techniques. In other words, the func- Management Institute.
This article is copyrighted material and has been reproduced with the permission of PMI. Unauthorized reproduction of this material is strictly prohibited.
Cover to Cover
Book Review Editor, Kenneth H. Rose, PMP
Successful Information System Implementation: Reviewing earlier partial definitions, Pinto and Millet add
The Human Side, Second Edition what’s needed for a full picture of implementation success.
by Jeffrey K. Pinto and Ido Millet Their three-part definition starts with system traits, under
which come two measures: system quality (“The system
Reviews should be prepared using MS Word and may be submitted by e-mail (preferred)
or on disk. Submissions should include the name, title, company, address, phone/fax/
e-mail, and brief (one or two sentence) biosketch of the reviewer. Reviews should be
submitted to:
PMI reserves the right to edit all material submitted for publication.
September 16–17 Successful Project Management for IT Professionals. Atlanta, GA, USA. For more information,
visit www.TLTraining.com or call +908-789-2800.
September 18–19 Successful Project Management for IT Professionals. Ft. Lauderdale, FL, USA. For more informa-
tion, visit www.TLTraining.com or call +908-789-2800.
September 24–27 Third International NAISO Symposium on Engineering of Intelligent Systems and 2002
Workshop on Information Systems for Mass Customization. Sponsored by the Natural and
Artificial Intelligence Systems Organization. University of Malaga, Malaga, Spain. For more infor-
mation, visit www.icsc-naiso.org/conferences/eis2002/index.html.
October 2–4 First UK International Performance Management Symposium. Bristol, U.K. For more informa-
tion, visit www.mtc.aust.com/symposium/uk2002/main.html.
October 3–10 PMI® 2002 Annual Seminars & Symposium. Sponsored by the Project Management Institute.
Henry B. Gonzales Convention Center, San Antonio, TX, USA. For more information,
visit www.pmi.org/pmi2002.
October 10–11 Managing People in Projects. Princeton, NJ, USA. For more information, visit www.kepner-
tregoe.com or call +609-921-2806.
October 10–11 Managing People in Projects. San Francisco, CA, USA. For more information, visit www.kepner-
tregoe.com or call +609-921-2806.
October 14–16 Technical Project Management. New York, NY, USA. Sponsored by the American Management
Association. For more information, visit www.amanet.org.
October 16–18 Technical Project Management. San Francisco, CA, USA. Sponsored by the American
Management Association. For more information, visit www.amanet.org.
October 18 Project Management: The Journey, Not the Destination. PMI South Florida Chapter
Professional Development Seminar 2002. Fort Lauderdale, FL, USA. For more information,
visit www.southfloridapmi.org.
October 18 Puerto Rico Chapter Project Management Annual Symposium. Tropimar Convention Center,
Isla Verde, Carolina, Puerto Rico. For more information, e-mail j_rodriguez@actpr.com.
November 11–15 Project World. Navy Pier, Chicago, IL, USA. For more information, visit www.projectworld.com.
November 13–16 Professional Development Days. Minneapolis, MN, USA. Sponsored by the PMI Minnesota
Chapter. For more information, visit www.pmi-mn.org.
November 14 PMI Information Technology & Telecommunications SIG member teleconference. For more
information, visit www.pmi-ittelecom.org.
November 16 Project Management in the Government. Ritz Carlton, San Juan, Puerto Rico.
Index of Advertisers
Mindjet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C2
Primavera . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . C4
Organizational behavior
Organizational development
Software engineering
Construction engineering
Communications
General management
Papers from researchers and practitioners in allied fields should have a project management slant.
For more information on publishing in PMJ, please see the Notes for Authors on the PMI Web site at www.pmi.org.
Questions about submissions may be addressed to the PMJ Editor at natasha.pollard@pmi.org or via mail to:
PMJ Editor
PMI Publishing Dept.
Four Campus Boulevard
Newtown Square, PA 19073 USA