Available online at www.sciencedirect.

com

International Journal of Project Management 31 (2013) 579 – 588

www.elsevier.com/locate/ijproman

Impact of sensitivity information on the prediction of project's duration using

earned schedule method
Raafat Elshaer ⁎
Industrial Engineering Department, Faculty of Engineering, Zagazig University, Sharkia, Egypt

Received 13 February 2012; received in revised form 15 October 2012; accepted 16 October 2012

Abstract

Recently, the prediction of project duration has been investigated in earned value management (EVM) using three earned value methods, planed
value (PVM), earned duration (EDM) and earned schedule (ESM). The investigation has shown that ESM method outperforms on average the
other two methods and fails in case of wrong warning coming from non-critical activities which suffer from delays and/or ahead of schedule. The
objective of this paper is twofold: first, we study the impact of the activities' sensitivity information on the forecasting accuracy of the ESM
method. Second, we test the claim that in normal conditions the project performance indicator provided by ESM at higher work breakdown
structure is reliable. More precisely, activity based sensitivity measures are used as weighing parameters of the activities to improve the schedule
performance of a project by removing or decreasing the negative effect of wrong warning of the non-critical activities. The computational results of
a simulation study on a big benchmark projects reveal that the sensitivity information are capable of improving the forecasting accuracy of the
ESM method.
© 2012 Elsevier Ltd. APM and IPMA. All rights reserved.

Keywords: Earned value management; Project duration forecasting; Earned schedule; Critical activities; Time sensitivity measures

1. Introduction and Vanhoucke (2006) give a summary of time measurement

Earned value management (EVM) is a management tech-
nique for project performance monitoring. A detailed explana-
tion of EVM basis can be found in Anbari (2003), Fleming and
Koppelman (2010) and PMI (2008). Based on the earned value
parameters, to the best of our knowledge, only three project
duration forecasting methods are in the literature: the planned
value method (PVM) developed by Anbari (2003), the earned
duration method (EDM) developed by Jacob (2003) and
extended by Jacob and Kane (2004), and the earned schedule
method (ESM) developed by Lipke (2003). Recently, the main
concepts of EVM and the three methods have been reviewed
with an illustrated example in Vanhoucke (2010a). Vandevoorde
⁎ Tel.: +20 012 2023 3347; fax: +20 055 230 4987.
E-mail address: r_h_elshaer@zu.edu.eg.
0263-7863/$36.00 © 2012 Elsevier Ltd. APM and IPMA. All rights reserved.
http://dx.doi.org/10.1016/j.ijproman.2012.10.006
and the aforementioned forecasting methods. They also compare
the three methods and test them to a simple one activity project
and a real-life data set. Based on these data, they conclude that
the three methods give similar results in the early and middle
stages. However, the ESM provides better results for monitoring
project progress at the final stage of the project. Vanhoucke and
Vandevoorde (2007a) also investigate the reliability of the three
EVM methods to forecast a project's final duration on a wide
and diverse set of fictitious project networks. They conclude that
the ESM indeed performs on average better than the PVM and
EDM methods in case of no false warning effects coming from
non-critical activities. Therefore, in the sake of helping the
professional community in getting a reliable prediction and
avoiding managerial effort spending in false warning, studying
the impact of the activities' sensitivity information on the
forecasting accuracy of the ESM method is the first objective of
this paper.
580 R. Elshaer / International Journal of Project Management 31 (2013) 579–588
Jacob and Kane (2004) argue that the well-known perfor-
mance measures of EVM are true indicators for project
performance as long as they are used on the activity level,
and not on the control account level or higher work breakdown
structure (WBS) levels. Whereas, in Vandevoorde and Vanhoucke
(2006) and Vanhoucke and Vandevoorde (2007a), the EVM
performance measures are calculated on the project level, and not
on the level of each activity. They claim that effects (delays) of
non-performing activities can be neutralized by well-performing
activities (ahead of schedule) at higher WBS levels. Testing the
validation of this claim is the second objective of this paper.
Because of doing the project performance measurement at the
cost account level which might result in covering potential
problems, project performance indicators provided by ESM act as
early warning signals to detect problems and/or opportunities.
Based on these indicators, Vanhoucke (2010b) investigates
whether the activity sensitivity measures are able to distinguish
between highly sensitive and insensitive project activities in order
to steer the focus of the project tracking and control phase to those
activities that are likely to have the most beneficial effect on the
project outcome. More precisely, he uses the following four
activity sensitivity measures from Martin (1965), Williams (1992)
and PMI (2008) to guide the managerial corrective actions when
anticipating delays in the project: 1 — criticality index (CI), 2 —
significance index (SI), 3 — schedule sensitivity Index (SSI), and
4 — cruciality index (CRI) based on: 4.a — Pearson's
product-moment (CRIr), 4.b — Spearman's rank correlation
(CRIρ), and 4.c — Kendall's tau rank correlation (CRIτ). For
more information the formulas of the four measures and how to
compute them are described with an illustrated example in
Vanhoucke (2010b). He extends his work in Vanhoucke (2011a)
in which he presents two alternative project tracking methods.
First, the top-down project tracking method relies on earned value
management project performance data (based on ESM) that
triggers to the need for corrective actions that require a drill down
to the lowest WBS levels. Second, the bottom-up tracking method
is based on schedule risk analysis that reveals sensitivity
information of each activity and assumes that the focus should
lie on only the highly sensitive parts of the project. His
computational experiment shows that a top-down project tracking
approach is highly efficient for project networks with a serial
activity structure while a bottom-up approach performs better in a
parallel structured project network. Moreover, based on a
Monte-Carlo simulation study on fictitious and empirical project
data (Vanhoucke, 2011b), the topological structure of a project
network considers a main driver for the variability and efficiency
of the project control phase. Also, the efficiency of project control
can be significantly improved by combining a top-down and a
bottom-up project control approach which consider a strong
reason of this study. In the same direction Lipke et al. (2009) use
the statistical prediction and testing methods to improve the
reliability of ESM in forecasting the final cost and duration.
Motivated by the above studies and the common knowledge
that the critical path analysis gives a good project duration estimate
(see, Schonberger, 1981; Gutierrez and Kouvelis, 1991), in this
paper, we investigate the impact of emerging activities' sensitivity
information into the earned value calculations on the project time
performance indicator of ESM method. In doing so, the influence
of activity based sensitivity measures on the forecasting accuracy
of ESM is examined in three levels: 1 — overall forecasting
accuracy, 2 — during project progress, and 3 — at a certain project
network structure.
The paper is organized as follows: in Section 2 a proposed
methodology to combine activity sensitivity information in the
earned schedule calculations is explained. A simulation study,
the used project dataset and the simulation scenarios, is
described in Section 3. Section 4 shows the forecasting
accuracy formulas and the computational results. The compu-
tational results are as follows: the impact of the sensitivity
measures on the accuracy of the project duration is investigated
in Section 4.1. Section 4.2 and Section 4.3 show the impact of
the sensitivity measures during the project progress and at
different project network structures on the accuracy of the
prediction respectively. Section 5 is left to conclusions.
2. Proposed methodology
The forecasting of a project duration using the ESM can be
estimated by the following equation: EDAC ¼ AT þ PD−ES SPI ðt Þ ,
where the future performance of the remaining time, PD − ES,
can be corrected towards the current SPI (t) trend. where:
EDAC estimated duration at completion.
AT actual time, is the duration at which the EV accrued is
recorded.
PD project planned duration.
ES earned schedule, ¼ t þ PVEV−PV t
tþ1 −PV t
, where t satisfies that
EV ≥ PVt and EV b PVt + 1. ES identifies the time at
which the amount of earned value (EV) accrued
should have been earned (Lipke, 2003).
PVt planned value at time instance t, ¼ ∑i PV i;t , where PVi,t
is the planned value of activity i at time instance t.
EV earned value at actual time, ¼ ∑i EV i;AT where EVt,AT
is the earned value of activity i at actual time AT.
SPI (t) schedule performance index,=ES/AT.
In this study, we try to improve the forecasting accuracy of
ESM by removing/decreasing the false warning effects caused
by the non-critical activities. The proposed methodology to
fulfill this aim is to merge the sensitivity information into the
earned value calculations as follow. Let αi be the weighting
factor of activity i, EV α ¼ ∑i αi :EV i;AT be the weighted earned
value at actual time, and PV α;t ¼ ∑i αi :PV i;t be the weighted
planned value at time instance t. The earned schedule, ESα, and
schedule performance index, SPI (t), based on the weighting
factor α can be easily calculated. Also, the project duration can
be estimated using ESMα at weighting factor α, (i.e. EDACα).
In this study, the four sensitivity measures mentioned in the
above section will be used as weighting factors. Formally, α ∈
{1, CI, SI, SSI, CRIr, CRIρ, CRIτ}, where α = 1 means that the
project duration estimation is based on no distinction among
activities and all the activities have an equal weight in the total
earned value calculations. That is, ESM1 ≡ ESM.
 R. Elshaer / International Journal of Project Management 31 (2013) 579–588
3. A simulation study
This section presents the test design for the simulation study to
measure the ability of the four sensitivity measures in improving
the forecast accuracy of the ESM method. Section 3.1 presents
the project networks used in this study and Section 3.2 describes
three proposed simulation scenarios.
3.1. Project data set
The project test data set consists of 4100 generated
activity-on-node networks with 30 activities and random activity
durations and costs, and with a well-considered topological
network structure. These projects have been generated by the
RanGen project network generator (Demeulemeester et al.,
2003; Vanhoucke et al., 2008) and have been used in many
project scheduling and control studies (e.g. Vandevoorde and
Vanhoucke, 2006; Vanhoucke and Vandevoorde, 2007a,b;
Vanhoucke, 2010a,b, 2011a,b) mentioned in the introduction
of this paper. A crucial parameter used during the project
network generation is the serial/parallel (SP ∈ [0,1]) indicator,
measuring the closeness of a project network to a complete serial
(SP = 1) or parallel (SP = 0) network. AD, LA and TF are other
indicators used during the project network generation. Where
AD ∈ [0,1], activity distribution indicator measures the
distribution of project activities along the levels of the project,
LA ∈ [0,1], length of arcs indicator measures the length of each
precedence relation (i, j) in the network as the difference between
the level of the end activity j and the level of the start activity i
and TF ∈ [0,1], topological float indicator measures the
topological float of a precedence relation as the number of
levels each activity can shift without violating the maximal level
of the network as defined by SP.
3.2. Simulation scenarios
The realized duration di of activity i is drawn from a triangular
distribution with the following characteristics, di ~ Tri (a, b, c),
where a, b and c represent the minimum, the mode, and the
maximum value of the distribution respectively. To investigate
the impact of sensitivity information in removing/decreasing the
false warning effect of non-critical activities and improving the
accuracy of forecasting a project's duration, three scenarios are
proposed for the random variation of activity durations. In these
scenarios we distinct between critical and non-critical activities
aiming in generating false warning signals due to increasing or
decreasing on the realized duration of the baseline non-critical
581
non-critical (critical) activities may change to critical
(non-critical) activities. Consequently, the percentage of the
baseline critical activities existence decreases when compared
with Scenario 2.
2 Scenario 2: random activity time deviations: both critical
and non-critical activity durations follow: di ~ Tri (0.8 di, di,
1.2di), which might result in activity durations which are
ahead, on or behind schedule.
3 Scenario 3: di ~ Tri (0.5 di, di, di). This left-tailed distribution
gives false warning signals that the project will be finished
ahead of schedule. This scenario emphasizes the high
existence of the baseline critical activities when compared
with Scenario 2.
In Scenarios 1 and 3 we measure the capability of the new
methodology to improve the forecast accuracy of the ESM
method by removing the false effects of non-critical activities.
In scenario 2 we try to verify the claim that in the cost account
level the ESM method could get a good forecasting for the
project duration because the well-performing activities neutral-
ized the non-performing activities.
4. Computational tests and results
In this section, we display extensive results for the three above
mentioned simulations. This section is divided into three
subsections for evaluating the impact of sensitivity measures on
the performance of ESM method. The performance, as proposed
by Covach et al. (1981) and used in Vanhoucke and
Vandevoorde (2007a) and Vanhoucke (2010a), includes forecast
accuracy, timeliness and stability respectively as follow: in
Section 4.1, we investigate the impact of the activity sensitivity
measures on the overall forecasting accuracy of the ESM for the
three proposed scenarios. In Section 4.2, we investigate the
impact of the activity sensitivity measures on the behavior of the
ESM forecasting along the completion stages of the project
execution. Based on the four indicators that used to generate each
network in the dataset, in Section 4.3 we discuss the influence of
the activity sensitivity measures at a different topological project
network on the forecast accuracy of ESM.
Following Vanhoucke and Vandevoorde (2007a) to deter-
mine the forecast accuracy, we calculate:
100 T RD−EDAC rp
The mean percentage error ðMPEÞ as : ∑rp¼1
T RD
The mean absolute percentage error ðMAPEÞ as :
activities. To do so, the planned critical activity will follow the  
following symmetric triangular distribution di ~ Tri (0.8di, di,
1.2di), where di represents the planned duration of activity i in the
project. In the same time the planned non-critical activity will
follow one of the following three scenarios:
1 Scenario 1: di ~ Tri (di, di, 1.5di). This right-tailed distribution
leads to delays in the planned non-critical activities which give
false warning signals that the project may finish behind
schedule. In this scenario also on average some baseline
100 T RD−EDAC 
rp
∑rp¼1
T RD
where T is used to refer to the total number of reporting periods
over the complete project horizon, RD denotes the realized
project duration and EDAC rp is used to denote the estimated
duration at completion in reporting period rp (rp = 1,2…,T). The
MPE-values lower (larger) than zero give an indication of an
over-estimation (under-estimation) of the forecasting metrics
for the actual project duration, RD.
582 R. Elshaer / International Journal of Project Management 31 (2013) 579–588
The mean percentage presence of planned critical activities
∑ CIi
(MPPPCA) is calculated as: i∈C jC j , where C denotes the set of
critical activities in the baseline schedule, and CIi is the crit-
icality index of activity i. MPPPCA shows the effect of the
variation in the baseline non-critical activity durations in the
existence of the baseline critical activities through the proposed
scenarios. This metric will be used only when investigating the
impact of the sensitivity measures on the forecasting accuracy
of the ESM at a different network structure (see Section 4.3).
4.1. Impact of sensitivity measures on the forecast accuracy of
ESM
In the current section we discuss impact of activity
sensitivity information as a weighting factor on the forecast
accuracy of ESM. The left-hand side of Fig. 1 shows that on
average the CI, CRIr, CRIρ, or SI improves the forecasting
accuracy of the ESM in scenarios 1 and 3 and, gets almost the
same performance in scenario 2. Moreover, although ESMSSI is
the second best in scenarios 1 and 3, it gets the worst result in
scenario 2. Along the three scenarios, whereas ESMCRIτ gets the
worst results, ESMCI is considered to be the best. We can also
observe that the difference between the performance of ESMSSI
and ESMCI is almost fixed which reveals a systematic error in
the performance of ESMSSI. The right-hand side of Fig. 1 shows
that the CI and SSI are more reliable than the other sensitivity
measures in removing the wrong effect of the non-critical
activities. In scenario 2 also, ESM 1 performs as well as ESMCI
by which the claim that at cost account level the non-performing
activities are neutralized by well-performing activities is proved.
4.2. Impact of sensitivity measures on the ESM during project
progress
In order to investigate the effect of the sensitivity measures
on the accuracy of time forecasts of ESM during the project
progress, the project horizon is divided into 10 reviewing
periods. Each reviewing period represents 10% completion of
3
MAPE/MPE (%)
0
Scenario 1 Scenario 2
-1
MAPE
-2
-3
ESM1 ESMSSI
Fig. 1. Impact of the sensitivity measures on the forecast accuracy (MAPE and MPE) of ESM for the three scenarios.
the realized duration. The simulation results of the three
scenarios are shown in Fig. 2.
From the results, we can obviously notice the following:
1 — all graphs show that the forecast accuracy improves as
approaching to the end of the project. This observation is in line
with Vanhoucke and Vandevoorde (2007a) and Vanhoucke
(2010a). 2 — through all review periods, in scenarios 1 and 3,
CI, SSI, CRIr, SI and CRIρ improve the performance of the
ESM, where ESMCI is considered the best and ESMCRIτ is the
worst. Whereas, in scenario 2, we find that ESM using all the
sensitivity measures, except SSI in the first reviewing period, is
almost doing as well as ESM1. This emphasized the claim of
Vanhoucke and Vandevoorde. 3 — For the three scenarios, at
the beginning of the project ESMSSIa shows bizarre results
which show the source of error in the performance of ESMSSI
we mentioned in the above section. When drilled down in the
results we found at the first reviewing period of some projects
the prediction of the final duration is zero. As a sequence,
MAPE and MPE rise to 100% and weaken the average forecast
accuracy of ESMSSI as shown in Section 4.1 for the ten
reviewing periods and this subsection for first reviewing period.
4.3. Influence of the sensitivity measures on the ESM at
different network structure
Vanhoucke and Vandevoorde (2007a) and Vanhoucke
(2010a) found from the results of the investigation of the
three predictive project duration methods (PVM, EDM, ESM)
that the network structure influences the forecasting accuracy.
Also, in Vanhoucke (2010b, 2011b), the author observed that
the reliability of the schedule risk analysis (SRA) increases
when the presence of parallel activities in the project increases.
In this section, we investigate the influence of the sensitivity
measures on the forecast accuracy of ESM at a different
topological structure of a project network. Due to the bizarre
results of SSI in the above section at the first review period, the
calculations in the current section will depend only on the last
nine review periods of the project. The degree of improvement
Scenario 3 Scenario 1 Scenario 2 Scenario 3
MPE
ESMCI ESMSI ESMCRIr ESMCRIρ ESMCRIτ
 R. Elshaer / International Journal of Project Management 31 (2013) 579–588 583

11

Scenario 1 (MAPE (%))

9

1
0%-10% 10%-20% 20%-30% 30%-40% 40%-50% 50%-60% 60%-70% 70%-80% 80%-90% 90%-100%
-1
10
Scenario 2 (MAPE (%))

9
8
7
6
5
4
3
2
1
0
0%-10% 10%-20% 20%-30% 30%-40% 40%-50% 50%-60% 60%-70% 70%-80% 80%-90% 90%-100%
11
Scenario 3 (MAPE (%))

1
0%-10% 10%-20% 20%-30% 30%-40% 40%-50% 50%-60% 60%-70% 70%-80% 80%-90% 90%-100%
-1
% of Project progress
ESM1 ESMSSI ESMCI ESMSI ESMCRIr ESMCRIρ ESMCRIτ

Fig. 2. Influence of the sensitivity measures on the forecasting accuracy (MAPE) of ESM for the three scenarios during project progress.

in the accuracy of prediction is measured by both MAPE and
MPE. Also, for the three scenarios, the distribution of the
baseline critical activities at every network structure will be
measured by MPPPCA as maximum, mean and minimum of
percentage presence of planned critical activities.
4.3.1. Sensitivity measures and SP indicator
In this section we investigate the influence of the sensitivity
measures on the forecasting accuracy of ESM at different
values of SP. Fig. 3 displays the MAPE, MPE and MPPPCA of
the three scenarios for the networks with varying values for the
SP indicator, ranging from 0.1 to 0.9, in steps of 0.1.
The results of Fig. 3 can be summarized as follows: 1—
ESM method over-estimates (under-estimates) the actual
project duration in scenario 1 (scenario 3). The forecasting
accuracy is improved by increasing SP (increasing serial
activities) which is in line with Vanhoucke and Vandevoorde
(2007a) and Vanhoucke (2010a), 2 — for scenarios 1 and 3, All
sensitivity measures, except CRIτ, improve the forecasting
accuracy of the ESM up to SP b 0.5. When SP ≥ 0.5, the
improvement in the performance of ESM can be ignored. Also,
ESMCI and ESMSSI are considered the best in removing/
minimizing the wrong effect of non-critical activities. In
scenario 2, ESM using the sensitivity measures, except CRIτ,
almost performs as well as ESM1, and 3 — the impact of the SP
indicator on the percentage presence of the baseline critical
activities for the three scenarios is shown in Fig. 3. The
MPPPCA curve reveals that the possibility of losing an activity
its baseline criticality decreases when the presence of serial
activities in the project increases. Also, even under high
stability of critical activities of a baseline schedule the early
finishing of non-critical activities lead to wrong project
prediction as shown in scenario 3.
Due to disappearing the false warning effects of non-critical
activities (as discussed above and shown in Fig. 3) on the time
forecasts of ESM when SP ≥ 0.5, we will investigate the impact
of the activity sensitivity information for the other topological
network factors only at SP = 0.2 in the following sections.
4.3.2. Sensitivity measures and AD indicator
When AD = 0, the number of activities is uniformly
distributed over all levels. When AD = 1, there is one level
584 R. Elshaer / International Journal of Project Management 31 (2013) 579–588

100%
MPPPCA
80%

60%

40%

20%
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

9.00
8.00
7.00
MAPE (%)

6.00
5.00
4.00
3.00
2.00
1.00
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

10.00
8.00
6.00
4.00
MPE (%)

2.00
0.00
-2.00
-4.00
-6.00
-8.00
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Scenario #1 Scenario #2 Scenario #3
SP
ESM1 ESMSSI ESMCI ESMSI ESMCRIr ESMCRIρ ESMCRIτ

Fig. 3. Influence of the sensitivity measures on the forecasting accuracy (MAPE and MPE) of the ESM for the three scenarios at different SP values.

with a maximal number of activities, and all other levels
contain a single activity. Fig. 4 displays the MAPE, MPE and
MPPPCA of the three simulation scenarios for the networks
with varying values for the AD indicator, ranging from 0.2 to
0.8, in steps of 0.2 at SP = 0.2.
The results of Fig. 4 can be summarized as follows: 1 —
ESM shows that over-estimation and under-estimation are in
scenarios 1 and 3 respectively. Also, the mean percentage
presence of the planned critical activities increases when the
AD increases. Moreover, the distribution of the MPPPCA
becomes narrow with increasing AD, for the three scenarios.
2 — In scenarios 1 and 3 all sensitivity measures except CRIτ
significantly improve the forecast accuracy of ESM over all AD
values. In addition, while ESMCI is considered the best in
decreasing the wrong effect of non-critical activities for
scenarios 1 and 3, ESM with all sensitivity measures except
SSI and CRIτ performs as well as ESM1 in scenario 2.
4.3.3. Sensitivity measures and LA indicator
When LA equals 0, the network has many precedence
relations between two activities on a level far from each other,
and hence the activity can be shifted further in the network.
When LA equals 1, many precedence relations have a length of
one, resulting in activities with immediate successors on the
next level of the network, and hence little freedom to shift.
Fig. 5 displays the MAPE, MPE and MPPPCA of the three
simulation scenarios for the networks with varying values for
the LA indicator, ranging from 0.2 to 0.8, in steps of 0.2 at SP =
0.2. From Fig. 5 we observe the following: 1 — the wrong
effect of the non-critical activities on the forecasting accuracy
of ESM1 decreases with increasing LA. Whereas, in scenario 1,
MPPPCA almost remains constant and its distribution narrows
in increasing LA, in scenarios 2 and 3 MPPPCA deceases with
increasing LA. These results are in line with the definition of
LA. 2 — For all scenarios, also we obtain the same results
 R. Elshaer / International Journal of Project Management 31 (2013) 579–588 585

100%

M PPPCA (%)
80%

60%

40%
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8

7.00
6.00
MAPE (%)

5.00
4.00
3.00
2.00
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
5.00

3.00

1.00
MPE (%)

-1.00

-3.00

-5.00

-7.00
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
Scenario #1 Scenario #2 Scenario #3
AD
ESM1 ESMSSI ESMCI ESMSI ESMCRIr ESMCRIρ ESMCRIτ

Fig. 4. Influence of the sensitivity measures on the forecasting accuracy (MAPE and MPE) of ESM at different AD values and SP = 0.2 for the three scenarios.

where all sensitivity measures except CRIτ improve the
forecasting accuracy of ESM from the wrong effect of
non-critical activities. Whereas ESMCI is considered the best
through all scenarios and LA values, ESMCRIτ is considered the
worst one.
4.3.4. Sensitivity measures and TF indicator
When TF equals 0, the network structure is 100% dense and
no activities can be shifted within a given SP value. When TF
equals 1, the network consists of one chain of activities without
topological float while the remaining activities have a maximal
float value.
Fig. 6 displays the MAPE, MPE and MPPPCA of the three
simulation scenarios for the networks with varying values for
the TF indicator, ranging from 0.2 to 0.8, in steps of 0.2 at SP =
0.2. The results of Fig. 6 can be summarized as follows: 1 —
ESM shows that over-estimation and under-estimation are in
scenarios 1 and 3 respectively. MPPPCA, the mean percentage
presence of the planned critical activities, increases and its
distribution becomes narrow with increasing TF, for the three
scenarios. 2 — In scenarios 1 and 3 all sensitivity measures
except CRIτ significantly improve the forecast accuracy of
ESM over all TF values. In addition, while ESMCI is
considered the best in decreasing the false warning effects of
non-critical activities for scenarios 1 and 3, ESM with all
sensitivity measures except SSI and CRIτ performs as well as
ESM1 in scenario 2. 4 — From Figs. 4 and 6 we can obviously
observe that the behavior of sensitivity measures on the
forecasting accuracy of ESM is similar for AD and TF.
4.4. Summary
The above discussions in Sections 4.1–4.3 state that the
following four sensitivity measures CI, CRIr, SI and CRIρ are
capable of improving the performance of earned schedule
method in predicting the final project's duration whatever the
topological structure of the project network is. Also, as shown
in Figs. (1–6) the ESMCI has the best performance and is
capable of removing/decreasing the error caused by the false
warning signals of the non-critical activities. Also, in scenario
586 R. Elshaer / International Journal of Project Management 31 (2013) 579–588

100%
90%
M PPPCA (%) 80%
70%
60%
50%
40%
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8

7.00
6.50
6.00
5.50
MAPE (%)

5.00
4.50
4.00
3.50
3.00
2.50
2.00
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
5.00
4.00
3.00
2.00
MPE (%)

1.00
0.00
-1.00
-2.00
-3.00
-4.00
-5.00
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
Scenario #1 Scenario #2 Scenario #3
LA
ESM1 ESMSSI ESMCI ESMSI ESMCRIr ESMCRIρ ESMCRIτ

Fig. 5. Influence of the sensitivity metrics on the forecasting accuracy (MAPE and MPE) of ESM at different LA values and SP = 0.2 for the three scenarios.

2, ESMCI performs as well as ESM1 and this indicates the
validity of the claim that, in normal conditions, the project
performance indictor provided by ESM1 at higher work
breakdown structure is reliable. ESMSSI suffers from troubles
in scenario 2 and its performance is lower than ESMCI returning
to the following: SSIi = SDi ⋅ CIi/PSD where SSIi, SDi, and CIi
are schedule sensitivity index, standard deviation and criticality
index of activity i respectively, and PSD is the project standard
deviation. The main problem of SSI as a weighting factor is that
SSI drops to zero when the activity duration is stable (i.e. SDi =
0) even if it has high criticality index (i.e. CIi = 1).
5. Conclusions
In this paper we investigate the impact of four activity
sensitivity measures in the forecasting accuracy of ESM
method. Based on extensive simulations on a large set of
generated networks, our findings can be concluded in the
following lines: through all topological project network
structures ESMCI, the earned schedule method based on the
criticality index (CI), is considered the most reliable metric in
forecasting a project's final duration, where ESMCI out-
performs ESM in case of false warning coming from
non-critical activities and performs as well as ESM in case of
normal conditions. Therefore, we recommend the practitioners
to use the ESMCI and the traditional ESM in the same time in
predicting a project's final duration. By doing so, any false
warning due to delays and/or ahead in the non-critical activities
will be eliminated and detected at the cost account level
without drilling down to lower work breakdown structure
levels. Consequently, the professional community can get a
reliable project time prediction and the project manager effort
to find a corrective action will be minimized. The above
conclusion also emphasizes the validity of the claim that, in
normal conditions, the project performance indictor provided
by ESM at higher work breakdown structure is reliable.
 R. Elshaer / International Journal of Project Management 31 (2013) 579–588 587

100%
90%

M PPPCA (%)
80%
70%
60%
50%
40%
30%
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
8.00

7.00

6.00
MAPE (%)

5.00

4.00

3.00

2.00
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
6.00

4.00

2.00
MPE (%)

0.00

-2.00

-4.00

-6.00

-8.00
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
Scenario #1 Scenario #2 Scenario #3
TF
ESM1 ESMSSI ESMCI ESMSI ESMCRIr ESMCRIρ ESMCRIτ

Fig. 6. Influence of the sensitivity measures on the forecasting accuracy (MAPE and MPE) of ESM at different TF values and SP = 0.2 for the three scenarios.

Acknowledgment
The author would like to thank Prof. Mario Vanhoucke for
supporting and providing datasets.
References
Anbari, F.T., 2003. Earned value project management method and extensions.
Project Management Journal 34 (4), 12–23.
Covach, J., Haydon, J., Reither, R., 1981. A Study to Determine Indicators and
Methods to Compute Estimate at Completion (EAC). ManTech Interna-
tional Corporation, Virginia, p. 30.
Demeulemeester, E., Vanhoucke, M., Herroelen, W., 2003. A random network
generator for activity-on-the-node networks. Journal of Scheduling 6, 13–34.
Fleming, Q., Koppelman, J., 2010. Earned Value Project Management, fourth
ed. PMI, Newtown Square, Pennsylvania.
Gutierrez, G., Kouvelis, P., 1991. Parkinson's law and its implications for
project management. Management Science 37, 990–1001.
Jacob, D., 2003. Forecasting project schedule completion with earned value
metrics. The Measurable News, pp. 7–9 (March).
Jacob, D., Kane, M., 2004. Forecasting project schedule completion with
earned value metrics revisited. The Measurable News, pp. 11–17 (summer).
Lipke, W., 2003. Schedule is different. The Measurable News, pp. 31–34 (Summer).
Lipke, W., Zwikael, O., Henderson, K., Anbari, F.T., 2009. Prediction of
project outcome: the application of statistical methods to earned value
management and earned schedule performance indexes. International
Journal of Project Management 27, 400–407.
Martin, J., 1965. Distribution of the time through a directed acyclic network.
Operations Research 13, 46–66.
Project Management Institute (PMI), 2008. A Guide to the Project Management
Body of Knowledge (PMBOK Guide), third ed. Newtown Square, Pennsylvania.
Schonberger, R., 1981. Why projects are “always” late: a rationale based on
manual simulation of a PERT/CPM network. Interfaces 11, 65–70.
Vandevoorde, S., Vanhoucke, M., 2006. A comparison of different project
duration forecasting methods using earned value metrics. International
Journal of Project Management 24, 289–302.
Vanhoucke, M., 2010a. Using activity sensitivity and network topology
information to monitor project time performance. Omega 38, 359–370.
Vanhoucke, M., 2010b. Measuring time: improving project performance using
earned value management. International Series in Operations Research and
Management Science, vol. 136. Springer.
Vanhoucke, M., 2011a. On the dynamic use of project performance and
schedule risk information during project tracking. Omega 39, 416–426.
Vanhoucke, M., 2011b. Measuring the efficiency of project control using
fictitious and empirical project data. International Journal of Project
Management 30 (2), 252–263.
588 R. Elshaer / International Journal of Project Management 31 (2013) 579–588

Vanhoucke, M., Vandevoorde, S., 2007a. A simulation and evaluation of
earned value metrics to forecast the project duration. Journal of the
Operational Research Society 58, 1361–1374.
Vanhoucke, M., Vandevoorde, S., 2007b. Measuring the accuracy of earned
value/earned schedule forecasting predictors. The Measurable News 26–30
(Winter).
Vanhoucke, M., Coelho, J., Debels, D., Maenhout, B., Tavares, L., 2008. An
evaluation of the adequacy of project network generators with systemati-
cally sampled networks. European Journal of Operational Research 187,
511–524.
Williams, T., 1992. Criticality in stochastic networks. Journal of the Operational
Research Society 43, 353–357.