International Journal of Project Management
International Journal of Project Management
International Journal of Project Management
net/publication/260292210
CITATIONS READS
6 17,727
2 authors:
Some of the authors of this publication are also working on these related projects:
Earned Value Management: Integration of Cost and Time Contingencies into Estimates at Completion. View project
All content following this page was uploaded by Alberto De Marco on 21 January 2019.
estimate at completion
Elsevier - PrePrint
Abstract
Traditional Earned Value Management (EVM) index-based methods for Cost Estimate at
Completion (CEAC) of an ongoing project have been known for their limitations inherent with
both the assumption that past EVM data is the best available information and early-stage
unreliability.
a modified index-based formula predicting expected cost for the remaining work with the
Gompertz growth model via nonlinear regression curve fitting. Moreover, the proposed equation
accounts for the schedule progress as a factor of cost performance. To this end, it interpolates
into its equation an Earned Schedule-based factor indicating expected duration at completion.
The proposed model shows itself to be more accurate and precise in all early, middle, and late
The developed methodology is a practical tool for Project Managers to better incorporate the
progress status into the task of computing CEAC and is a contribution to extending EVM
research to better capture the inherent relation between cost and schedule factors.
Regression analysis.
1. Introduction
It is a forward looking tool to assist Project Managers with the task of making timely and
appropriate decisions about cost outcome of their in-progress projects (Fleming and
Koppelman, 2006).
For over four decades, Earned Value Management (EVM) has been used to forecast cost at
completion. This objective methodology integrates project cost, schedule and scope metrics into
a single measurement system. It is widely applied for measuring and analyzing project actual
status against its baseline, and for providing estimates of project cost and duration at completion
(De Marco and Narbaev, 2013). In particular, EVM is used to compute Cost Estimate at
Completion (CEAC), a top-down estimate of the project total cost based on the project’s status.
Within the EVM framework, several methods exist to compute CEAC, classified as either
In general, IB methods have an inherent limitation due to their only reliance on past
information: they assume that remaining budget is adjusted by a performance index (Fleming
and Koppelman, 2006; Kim and Reinschimdt, 2011). The second concern associated with the
traditional approach is that it provides unreliable cost forecasts early into a project life because
of few available EVM data (Fleming and Koppelman, 2006; Zwikael et al., 2000). In this
regard, some studies (Anbari, 2003; Cioffi, 2006; Kim et al., 2003; Lipke, 2004) simplified
researches (e.g., Kim and Reinschmidt, 2011; Lipke et al., 2009; Marshall et al., 2008)
employed statistics into EVM forecasting system to benefit from deeper analysis to support
decision making. Caron et al. (2013), Naeniet al. (2011), and Pajares and Lopez-Paredes (2011)
structural change in cost and schedule performance in the project dynamic environment.
With the purpose of overcoming the two mentioned weaknesses of IB approach and produce
more reliable CEAC, regression techniques have been regarded as an alternative to traditional
IB methods. Through their curve fitting process, regression techniques improve accuracy of the
CEAC, especially as they may use a combination of EVM data with Earned Schedule (ES) data
and provide more reliable forecasts early into the project life.
However, reported literature reveals that little advancement has been made in the area of
improving reliability of the IB approach via its refinement by regression techniques (Lipke et
al., 2009; Marshall et al., 2008; Tracy, 2005). Most studies integrating regression concepts into
IB approaches concern U.S. defense projects, which are complex in nature with large budgets
and long durations (Christensen et al., 1995; Lipke et al., 2009). In addition, within the EVM
To fill these gaps, a new regression methodology is proposed to provide more reliable
CEAC. The developed model overcomes the limitations inherent to traditional IB approaches.
In addition, the model regards project schedule as a factor of cost performance and, hence, takes
into account for the schedule progress, measured via the ES concept, to calculate CEAC. The
model equation is a classical IB formula modified with a Gompertz growth model function and
it integrates an ES-based factor to indicate the expected completion time used into the model.
The paper is structured as follows. Section 2 frames commonly used IB formulae for CEAC,
introduces a regression approach for cost S-curve fitting, and formulates a Gompertz growth
model to implement it to the proposed methodology. Section 3 designs the methodology and
establishes a framework for evaluating the model and comparing its estimated results with those
of IB formulae. In Section 4, we apply EVM data from nine projects to show application of the
proposed model, derive the study results, and present the role of ES in the developed
methodology. Section 5 explores the findings of the research and associated implications.
Section 6 presents the work contributions in advancing the body of knowledge and draws
In the EVM theory and practice, the calculation of the CEAC entails summing up two factors
(Eq. (1)), namely: the Actual Cost (AC) of performed work to the Actual Time (AT), and the
estimated cost of the remaining work. The second factor is a difference between the Budget at
Completion (BAC) and the Earned Value (EV) adjusted by a Performance Index (PI – a
The choice of a desirable PI depends on the project status and associated risks. Zwikael et al.
(2000) relate this choice to premises set by Project Managers in selecting the PI, from the belief
that all past cost deviations cancel into the future so that their projects can be accomplished
within the BAC to a pessimistic argument that the deviations will continue at the rate observed
so far. PMI (2011) provides four PIs to correct the remaining BAC (Table 1) with different
assumptions associated with actual project performance. Among these indexes the most
commonly used is the Cost Performance Index (CPI), which assumes that past cost performance
is the best available indicator of future cost outcome as a reasonable floor estimate. Anbari
(2003) states that an estimate obtained using a product of CPI and Schedule Performance Index
(SPI) is an indicator of the overall project health and is a ceiling CEAC to reflect both cost
deviation and schedule progress. Fig. 1 presents the EVM metrics addressed above and used in
this research.
Insert Table 1
Insert Fig. 1
Since this IB approach only relies on past information, it requires stability of the PI to
provide for reliable CEAC. In this regard, previous research carried out on defense projects
found that a cumulative value of CPI stabilizes by the time the project is 20 percent complete
and the forecast value does not vary by more than 10 percent from that point in time to
completion. The EVM community received this finding as a rule of thumb and generalized it as
being applicable for all types of projects. However, recent studies challenged this finding
attributing it to large-scaled and long duration defense and energy projects only (Henderson and
Zwikael, 2008; Lipke et al., 2009). They questioned whether the PI stability existed and found
that most projects from other industries (e.g., construction, software) with relatively small
budgets and short durations, achieved the PI stability by the second half portion of the project
life.
2.2. The regression-based approach and S-curve fitting
have been gaining acceptance by practitioners. The main feature of these methods is that they
describe a linear or nonlinear statistical relationship between a predictor (input) and response
(output) variables through their parameters (Bates and Watts, 1988). Parameters of a regression
model represent the behavior of project cost over the whole lifecycle.
Efforts put to apply regression models are greater than those needed for relatively simple IB
cost forecasting methods. However, claims have been made that they yield better estimates early
in the project life, while the IB approach is likely to be unreliable (Tracy, 2005).
work, expressed in units of costs, labor hours, progress percentage, etc., plotted against time
(PMI, 2008). The S-like shape of this curve represents work progress which has lower rate at
the beginning and end (steady patter) and higher rate in the middle (steeper pattern). In EVM,
such curves are used to display AC, EV, and Planned Value (PV) of a project over the time axis.
Cioffi (2005) proposed a parameterized S-curve tool for managing cost of an ongoing project: it
is the derivation of a modified logistics equation with minor mathematical assumptions Project
Managers can easily set. The model was validated using two projects and showed flexibility in
generating a desired smooth cost profile by selecting the strength of the rise of the curve and the
Depending on type, complexity, and nature of a project, the time-cost relationship can be
modelled using different mathematical equations (Warburton, 2011). Defining an equation for
the S-curve model requires considering some issues relevant to nonlinear regression analysis.
First, such models require defining initial values for their parameters and setting an algorithm
for the least squares (LS) approximation (Bates and Watts, 1988). In nonlinear regression, there
is no standard approach to specify initial values and one needs to know initial information (e.g.,
prior historical data, EVM data, variables relationship) in dealing with this task. This is because
points around the S-shaped curve follow a Gaussian distribution. With these concerns, nonlinear
curve fitting approximates values of the model parameters with LS method minimizing the sum
of squared errors of estimated and actual values. The proposed methodology applies the Gauss-
Newton algorithm for this iterative approximation, which converges not heavily depending on
initial values (Bates and Watts, 1988) of a nonlinear model within specified tolerance
thresholds.
A Gompertz growth model (GGM) has found wide application in many fields associated
with population growth studies, such as biology, economics, marketing, etc. The model
describes phenomena inherent to data with a growth pattern. It is extensively used in curve
The GGM generic function is given in Eq. (2). The α is a future value asymptote of the
model that represents the final cost (which is never attained) as time (x) tends to infinity (Seber
and Wild, 1989). The β parameter is the y-intercept indicating an initial budget size and the γ is
( x )
GGM( x ) αe [ e ]
(2)
This model features with the position of the inflection point at approximately 1/3 of the total
growth (GGM(x)=α/e) at time (x=β/γ) when its growth rate (GR=αγ/e) is the greatest. The
growth rate monotonically increases to a maximum before steadily declining to zero (Seber and
Insert Fig. 2
With regard to the project cost growth, the GGM shapes such growth considering the cost
behavior as follows. During the project initial stage, work progress is typically slow, it speeds
up close to the middle stage increasing the cumulated cost accrual associated with the progress
and accelerating the GR. Finally, as the project reaches its completion, there is less work
Recently, Trahan (2009) proposed to use the GGM as an industry proxy for future projects.
S-curves were developed using EVM data from a number of U.S. defense contracts. The
parameters of the S-curve model for complete projects were found by regressing normalized AC
values of the entire project life against respective time points. The majority of the sample
projects experienced cost overrun. Therefore, the normalization of AC and actual time values to
BAC and planned duration, respectively, produced the values for the S-curve model parameters
best suitable for overrun projects. Hence, the major finding was that the model is accurate to
Trahan’s work is one of the inspiring references of this proposed GGM. However, in contrast
with Trahan’s work, the method proposed in this paper presents a comprehensive
methodological approach, greater validity, increased practicability for ongoing projects, and
3. Methodology
The CEAC methodology proposed in this paper integrates ES concepts into its equation to
take into account the project work progress. The ES technique overcomes limitations inherent to
the EVM method when it comes to computing Expected Duration at Completion (EDAC) of a
project (Lipke, 2003). It measures the schedule progress in time units and eliminates a
deficiency of EVM-based SPI, which tends to unity as the project approaches its completion,
regardless of any early or late finish. As far as the accuracy of the ES method in computing
EDAC is concerned, comparative studies with EVM methods show that the ES technique
provides more accurate estimates than SPI-based calculations (e.g., Vandevoorde and
Vanhoucke, 2006).
The value of ES is obtained by projecting to actual date the EV curve onto PV curve
assuming that the current EV should actually have been earned at that projected time
ES(x)=C(x)+I(x) (3)
Where C and the subscript c denote the number of total time units for which EV exceeds PV
and the incremental portion I(x) = (EV(x) - PVc)/(PVc+1 - PVc) which is more or equal to 0 and
SPIt(x)=ES(x)/AT (4)
Thus, the resulting EDAC when the project is at time (x) is the ratio of Planned Duration
(PD) to SPIt(x). As the proposed approach utilizes the ES concept to consider schedule impact
in CEAC, the model uses the inverse of SPIt(x), which is the ratio of EDAC to PD.
For the purpose of better understanding the proposed equation, this inverse ratio is referred
to as Completion Factor (CF). The CF indicates EDAC yielded to unity and it can also be
If the value of the CF, based on work progress to date, is greater than 1.00 it indicates that a
project is likely to be delivered late, whilst less than 1.00 shows an early finish.
This section develops the new methodology following three steps. First, the values of the
three parameters of the GGM (Eq. (2)) are found through nonlinear regression analysis.
Then, the new CEAC formula is introduced with integrating parameters of GGM to calculate
CEAC.
Finally, we further modify the CEAC formula with the purpose of reflecting schedule
progress on cost performance. To this end, the ES-based CF is integrated into the formula. Here,
the CEAC equation has two variants: a base one without integrating the CF, and an ES-based
Recently, Narbaev and De Marco (2013) provided comparative study on this CEAC
methodology integrating four growth models (Bass, Gompertz, Logistic, and Weibull) into its
equation. They found that GGM is the best statistically valid model converging to approximate
values of its parameters in nonlinear regression curve fitting. In addition, the GGM generates
more accurate CEAC for early and middle stages of the project life. This work provides further
extended applicability and reliability of the previous model by providing accurate late estimates,
analysis of forecast precision, model timeliness, and integration of the influence of schedule
In particular, the proposed GGM is compared with four different index-based performance
indexes, is applied when the project is 80 percent complete, is tested the narrowness of the
forecast error, and is proven its reliability over time, which refers to as generating more accurate
and precise cost estimates in both the early and middle and late completion stages of a project.
The first step in developing the methodology is to find the three GGM parameters through
nonlinear regression curve fitting. For this, both time (a predictor variable) and cost (a response
variable) units are normalized to input into the GGM equation. The normalization of all the
values of time points to unity (1.00) assumes a project is 100 percent time complete (i.e.,
PD=1.00). Each next time point is a cumulated portion of this unity with the final time point
representing PD (1.00) of a project. These values represent a predictor variable (x) of the GGM.
Each time point (x), a value of the predictor variable, has a corresponding cost point, a value of
the response variable. These corresponding cost points are formed as follows. The values of AC
from time zero (x=0) to AT are normalized to unity (i.e., BAC=1.00) while the values of PV
from AT onto project completion with the final value of the normalization representing BAC
(1.00, i.e., 100 percent complete). Then, the normalized values of to date AC and PV are
combined to form the values of the response variable (y) in the GGM.
Finally, each time point (x) of the GGM equation (Eq. (2)) has its corresponding cost value
(y) to run the nonlinear regression with the GGM. This allows finding the values for the three
fitting parameters. Both time and cost units have final values equaling 1.00 (PD=1.00 for time
The following requirements are taken into account for the GGM equation in the nonlinear
regression curve fitting: the normalization of the predictor and response variables and what the
three parameters represent an initial value for these parameters is 1.00 with the confidence level
95% and the approximation algorithm the Gauss-Newton (which converges the parameter
values not heavily depending on their initial values). Then, via running this regression
procedure, the values of the three parameters are obtained: the α asymptote, the y-intercept β,
and γ-scale. The Minitab® software tool is used for this task.
The second step requires computing CEAC by using Eq. (6). This equation is the refined
version of a classical IB formula as previously given in Eq. (1). The difference is that Eq. (6)
calculates the remaining expected cost by regression analysis, while the IB formula adjusts it
with a PI. The second summand is an estimate to complete a project. It is equal to the product of
BAC times the difference of the two values of GGM (Eq. (2)): when a project is 100 percent
time complete (the result of the GGM function when time (x) is 1.00) and at AT (the result of
Finally, the GGM is modified to consider possible influence of work progress on CEAC. The
main assumption of this refinement is that a favorable schedule efficiency tends to improve the
final cost, while a poor schedule progress may increase the final cost. To this end, in Eq. (6), the
value of x=1.00 (which implies that a project finishes on time) is replaced by the CF (the ratio of
EDAC to PD). This is less than1.00 if a project is ahead of schedule and greater than 1.00 if a
This section provides the framework for assessing the quality of the proposed methodology
and analyzing the influence of schedule progress through ES-based CF on CEAC. The
Among the two criteria to assess the quality of a cost forecasting method, accuracy is
regarded as the most often used and important one (Yokum and Armstrong, 1995). This study
measures CEAC accuracy by a percentage error (PE) and the mean absolute percentage error
(MAPE) for early, middle, and late stages. PE is the difference between CEAC and Cost at
underestimation and a positive value the overestimation. MAPE is referred to as the average of
the absolute values of differences between CEAC and CAC over the number of projects tested
(Bates and Watts, 1988). Eq. (8) and Eq. (9) are used to compute these measures:
CEAC - CAC
PE% 100% (8)
CAC
The second criterion of the model is precision, defined as the narrowness of a forecast error.
the values of prediction errors from the average forecast within the population (Seber, 1989).
SD is computed by Eq. (10), which takes the square root of the variance (the average of the
squared differences between the PE of an individual project and mean of the PEs). A smaller
value of SD indicates that cost estimates calculated by a particular model are closer to its MPE
(PE i MPE) 2
SD% i 1
% (10)
n
The accuracy of EVM cost forecasting methods should also be reliable over a certain period
or the entire project life. This property of cost forecast is defined as timeliness and shows
reliability in accuracy of cost forecasting (Kim, 2007). From a practical perspective, Project
Managers may be more concerned about timeliness in cost forecasting as it implies reliability in
cost forecasting and provides a project team with warning signals about the final cost outcome
(Kim, 2007; Kim and Reinschmidt, 2011). Teicholz (1993) defines it as accuracy of estimates
during the first half of project duration. Vandevoorde and Vanhoucke (2006) evaluate it
correlating to changes in EDAC accuracy over the project’s final stage. These works report
timeliness analysis with regard to accuracy of estimates. This paper adds also analysis of the
precision timeliness. In this regard, this paper defines the timeliness of the proposed CEAC
methodology as a property describing more accurate and precise CEAC over the three forecast
stages. From a practical perspective, this may be of great importance to Project Managers as it
Finally, based on these two criteria, estimates using the proposed model are compared with
As discussed earlier in the paper, another important advantage of the proposed methodology
is the ability to appropriately capture the influence of the schedule progress into CEAC. The
EVM approach is known as an objective method that assists project managers in the task of
system (PMI, 2008). This implies that changes in one element of this triangle may cause
changes in the other/others. One of basic prerequisites of EVM approach is that work scope
remains as it is throughout the project life. On the contrary, the scope of work is revised when
complimentary activities are added into a project upon approval of change orders from a project
owner. Such scope change is subject to project’s rescheduling leading to potential changes in all
components of the measurement system including a work breakdown structure, the performance
measurement baseline and so forth. In such a case, the EVM system is revised according to
these changes.
In line with these considerations, the proposed CEAC model considers the possible influence
of work progress on CEAC. This relation is reflected through the integration of ES-based CF
which is related to SPIt and, hence, a measure of time-based schedule efficiency (PMI, 2011).
For this, at some time(x) if the project schedule efficiency is favorable (CF<1.00) this shows
that the final cost tends to improve, while a poor efficiency (CF>1.00) would influence increase
of the final cost. However, it is noted that this cannot be generalized for those ongoing projects
that are subject to adjustments and corrective actions as measures to speed up the work
progress, such as in the case of activity crashing or fast tracking. This usually results in cost
increase or significant changes to the original scope and schedule network (e.g., re-baselining),
The model considers this relationship by replacing the 100 percent time completion value
with the value of the CF in its equation (Eq. (7)). In other words, the model generates more
accurate and precise estimates when it takes into account for the schedule progress.
This study uses EVM data of nine construction projects selected from qualified reported
literature. Five out of nine projects are delivered with cost overruns and six report schedule
delays. They all are small to medium-scale projects with average BAC close to 8 million US
project, a percent value for the budget completion at early, middle, and late stages cannot be a
predetermined percentage. As a consequence, we define the range for these stages as 10-25%,
45-65%, 70-95%, respectively (Narbaev and De Marco, 2013). Below we demonstrate the
stepped procedure using EVM data of Project 1 to forecast CEAC when the project is in its
early stage.
The first step is to determine values of the three GGM’s parameters through the nonlinear
regression curve fitting. Table 2 provides initial absolute values for time (column Time point)
and cost data (column PV and AC). Then these time points are normalized to unity (assuming
PD=15 is 1.00) and PV and AC values to unity (assuming BAC=3,725,000 euro is 1.00). These
normalized time (variable x) and cost (variable y) points are reported down on the column
Predictor and Response in Table 2 and are input data for the GGM equation to run the fitting
process. The AC-PV values are combined values of AC from time zero (x=0) to AT and of PV
from AT to BAC=15. To compute the early stage CEAC for Project 1, month 4 is chosen as
time for the early stage estimation time when 20.90% of the BAC is earned. The requirements
one should take into account when running the nonlinear regression are considered in Section
3.2 above.
Insert Table 2
The GGM equation generated by Minitab® for EVM data of Project 1 (Table 2) is given in
Eq. (11). To calculate CEAC for the early stage, four months into the project execution when x
is 0.267 (in Table 2), this GGM equation result is 0.241 (or 24.10% of the project BAC). The
interpretation of the values of the three parameters (addressed in Section 2.3) is as follows. The
ratio of the β parameter to the γ parameter gives time percent complete point (x=β/γ) when the
cost growth rate is maximum which is 43.70% for Project 1with resulting cumulative cost of
44.20% (GGM(x)=α/e) of the BAC. Finally, the α asymptote value of 1.202 implies that, as
(1 .212 2 .773 x )
GMM( x ) 1 .202 e [ e ]
(11)
Step 2 computes the project CEAC for its base case using Eq. (6). For this purpose, we
additionally compute the value of GGM equation when x=1.00 (100% time complete). As
previously explained, the remaining of the project BAC must be adjusted by the difference of
the two values of GGM: when a project is 100 percent time complete and at AT. From this,
GGM(1.00)=0.974 and we use the refined version (Eq. (6)) to calculate CEAC of Project 1. The
methodology finds the project CEAC with PE=-6.96, which means that it is underestimating its
CAC.
Step 3 is about taking into account schedule progress based on the assumption that the
schedule is a factor of cost performance and, hence, it has its impact on the estimate. Therefore,
we include the projects CF into Eq. (6). It is noted that the project has PD=15 months,
EDAC=16 months at AT=4, and actual duration (AD) of 16.25 months. At month four, CF
equals 1.083 and it replaces x=1.00, as given in Eq. (7). It produces a new CEAC that considers
work progress: it is closer to CAC value with PE=-2.91 (4.05% of improvement over PE=-6.96
This section presents the CEAC computations for the nine sample projects and provides an
analysis of accuracy and precision of the estimates together with an assessment of the role of the
Table 3 allows an evaluation of the projects’ CEAC accuracy in PE for the early stage
computed applying the proposed method (Eq. 6 and Eq.7 for base and ES-based cases,
respectively) in comparison with the IB method (Eq. 1). It adjusts the remaining portion of BAC
by the four PIs (CPI, CR, CI, and MA) introduced in Table 1. Accordingly, the IB method
produces four different CEACs, which vary in assumptions associated with future cost
performance (Anbari, 2003; PMI, 2011). The detailed information on how these PEs are
CEAC results calculated by GGM (either base or ES-based cases) are more accurate than those
computed by four IB formulae. When comparing the cost estimates of two GGM cases, it
appears that the integration of CF into the model improves the model’s forecasting capability
(Project 1, 2, 3, 5, 7, 8, and 9). Overall, GGM allows overcoming the critical limitation of
traditional IB formulae to accurately determine early CEAC. The IB methods generate mixed
results difficult to interpret. However, the following can be concluded: all of the traditional
formulae provide PE above 10.00 for Project 2 and 7 and above 5.00 for most of the cases.
Insert Table 3
Table 4 provides results of the estimates’ accuracy (computed by Eq. 9) and precision
(computed by Eq. 10). Overall, the results show that the proposed model’s estimates are more
accurate (in MAPE) and precise (in SD) than those of the index-based formulae. Also, the
integration of the schedule progress into the model equation leads to improving the CEAC
accuracy. Therefore, the ES-based GGM appears to be the most accurate and precise model in
all the execution stages. This allows concluding that schedule is a factor of cost behavior and
delay/advance in the work progress has its respective influence on the project cost outcome.
Insert Table 4
With regard to the model timeliness, as addressed in Section 3.3, it implies providing for
more accurate and precise CEAC in early, middle, and late completion stages. The results of
both accuracy and precision suggest that the ES-based GGM meets this criterion producing
more accurate (Fig. 3a) and precise (Fig. 3b) estimates in all of the stages. Finally, another
pattern noticed is that all the compared models (except the GGM base case) improve both
Insert Figure 3
This section aims to test the effect of the schedule impact on the CEAC accuracy and to find
out if there is a relation between the work progress and the estimate accuracy. To accomplish
this analysis, we calculate CEAC for both cases (both without and with the CF). Then, the study
analyzes possible existence of a relationship between CEAC and project cost outcome
(underrun, on budget, overrun). Table 5 reports the results of the analysis: values of CF
computed at the forecast time, PE of CEAC for the two cases, and the projects’ cost outcome. A
CF greater than 1.00 indicates the project is experiencing a schedule delay at the time of
estimation and warns that this poor schedule efficiency may be influencing increase in the final
cost.
Insert Table 5
Overall, the ES-based case appears to be more accurate in all three stages than the case when
the work progress is not considered. From this table it can be drawn that the integration of the
CF into the forecasting model results in more accurate cost estimates. In particular, this
improves the estimates accuracy of seven projects in the early, five in the middle, and six in the
late stages.
proposed CEAC methodology for the early, middle, and late stages is provided in Fig. 4a, b, and
c, respectively. Closer to zero line values of PE in the chart implies more accurate estimates. On
average, integration of CF into the GGM (Eq. (7) against Eq. (6)) improves the cost estimates
from MAPE=7.28 to 4.20, 5.11 to 3.47, and 5.48 to 3.22 in the early, middle, and late execution
phases, respectively.
Insert Fig. 4
5. Discussion
Three main findings of the proposed CEAC model can be highlighted with regard to its
First, the comparative analysis (Section 4.2) of CEAC found by the proposed nonlinear
GGM and the four simple IB formulae proves that the proposed model generates more accurate
CEAC (Table 3). The cost estimates provided by the ES-based GGM are the most accurate:
4.20, 3.47, and 3.22 for the early, middle, and late stages, respectively. CEAC generated by
Base GGM and the traditional approach with CPI, CI, and MA produced the same results
between these four cases (Table 4). Lastly, the estimates computed by CR-based method are the
worst among the methods (MAPE=16.22, 10.19, and 5.72 for the three stages, respectively).
Overall, with regard to accuracy of CEAC the following concerns are worthwhile to note. The
test results show that a pure and simple traditional IB approach to forecast the expected project
final cost might be inadequate. CEAC comprises a sum of two components: to date actual cost
and the remaining portion of cost to complete. The first limitation is associated with the
assumption that a traditional IB technique is backward looking and relies on past EVM
information only. Therefore, the index-based approach adjusts the remaining BAC by PI
proposing a project will continue in its current progress until its completion. This forces Project
Managers to accept the belief that cost performance is stable for the rest of the project life.
However, the project implementation environment has uncertainty and changes in project cost
outcome that decrease as a project tends to completion. This change is reflected in the
cumulative value of CPI, which gets stabilized as a project progresses. This indicates stability in
project execution and, subsequently, ensures for more stable values of CEAC by the end of a
project. This leads to integrate the interpretation of the first finding with the second limitation of
traditional IB approaches. For a project in its early life, when few EVM data are at hand, this
technique is unreliable as it makes extrapolations from few time points for the rest of the
Second, the proposed model provides more precise estimates. Precision refers to the
narrowness of the forecasting error. On average, the cost estimates provided by the ES-based
GGM are the most precise: MAPE=5.27, 3.42, and 3.17 while the worst estimates are produced
by the CR-based method as MAPE=11.82, 5.41, and 4.77 for the early, middle, and late stages,
respectively (Table 4). Overall, unlike the IB approach, the developed CEAC model gives more
accurate and precise estimates as it adjusts the remaining portion of CEAC (second summand of
Eq. 6) by the GGM via nonlinear curve fitting, whereas the IB method achieves this adjustment
interpolates intrinsic properties inherent to growth models into its formula (Eq. 6) and takes into
account for a combination of AC (from a project start to AT) and PV (from AT onto
Another point to remark is about the GGM property to maintain more accurate and precise
estimates in all the three stages compared to the traditional approach, as it means reliability in
the forecasting. We defined this property of the model as timeliness, referred to as the ability to
give warning signals about the final cost outcome of a project. The timeliness of accuracy and
precision means reliability in CEAC forecasting. Moreover, in the ES-based GGM case,
accuracy (Fig.3a) and precision (Fig.3b) appeared to improve over time from early to late stage
estimation, with decreasing values of MAPE (4.20, 3.47, and 3.22) for accuracy and SD (5.27,
3.42, and 3.17) for precision (Table 4). The characteristic of this decrease in the estimates’
errors proves the model to be considered as viable. This feature of the model suggests that the
observations in MAPE and SD tend to converge to the actual result at completion. In addition,
the explanation of this tendency lies in the nature of the ES approach suggesting improvement
of duration estimates as SPIt stabilizes and a project approaches its completion. In addition to
this, it makes use of nonlinear regression. The growth model via the regression combines to date
AC data with future PV for which its three parameters show the relationship between past,
current and future project performance and progress. This second finding collaborates with the
previous research (Christensen et al., 1995; Marshall et al., 2008; Tracy 2005) which reported
the advantage of the nonlinear regression modelling over the conventional IB methods in EVM
system. Our comparative analysis shows that CEAC computed by the four IB formulae produce
abrupt and, hence, unstable estimates. The values of MAPE of these four formulae vary from as
small as 4.32 to as large as 16.22 and SD ranging from 5.92 to 20.2. Therefore, the proposed
model’s timeliness property is another practical advantage over the IB approach. We remark
that, in most projects regardless of their nature, budget, and duration, estimates by a traditional
approach stabilize by the second half of the project life or at late stage. The results of timeliness
of accuracy and precision are in accordance with the findings of previous research in the field.
For example Henderson and Zwikael (2008) showed that the PI values (CPI and SPI(t))
converged to their respective final values as the project gets closer to completion.
A third finding of this study is that asserting schedule progress as a factor of future cost
improves both accuracy and precision of the developed model. EVM is a system that integrates
project cost, schedule, and scope. In this regard, schedule is known as a factor of project cost
performance. Advance/delay in work progress has its relative influence on cost behavior. The
majority of projects experience impact of schedule progress on their final cost. Therefore, our
In addition, this research provides a novel and extended contribution toprevious research on
the development of the GGM. In particular, the research conducted by Narbaev and De Marco
(2013) provided for information about adapting the GGM to EVM, comparing it with other
growth models in the field, constructing the best fit S-curve for CEAC with the integration of
ES concepts for early and late stage estimates, and, finally, applying the model to compute
CEAC of a set of construction projects. This paper continues that work as part of its future
research and provides a comparison with the four IB methods, the calculation of CEAC for a
late stage, the analysis of precision of CEAC, and finds the model timeliness property as being
reliable in providing most accurate and precise estimates throughout project life.
Finally, this study aims at the diffusion of the GGM for a variety of projects and various
industries.
6. Conclusion
Reliability in forecasting CEAC has long been a focus of many comprehensive research and
comparative studies. This is because accurate and precise CEAC forecasts throughout the
project life equip the project team with essential information in taking prompt preventive
actions and assisting on timely completion within available budget. The traditional approach
using PI in EVM has been in use to assist in this task for over four decades with little change.
However, this technique relies on EVM data only and merely calculating CEAC by adjusting
the remaining BAC by a PI (Kim and Reinschimdt, 2011; Lipke et al., 2009). In addition, IB
method may produce inaccurate and unreliable CEAC in the early stages of projects due to little
This paper proposes a new CEAC methodology to forecast the final cost of ongoing projects.
modified by integration of the GGM via a nonlinear regression analysis. This combined
approach produces more accurate and precise CEAC effective for all stages of the project life
because it meets the property of timeliness. The proposed methodology overcomes the reported
Practical implications arise from the proposed CEAC model as a tool for Project Managers
to better incorporate the progress status into the task of forecasting the final cost of an ongoing
project. In particular, the inclusion of the ES-based CF, which indicates the expected
completion time, takes into account for any schedule advance or delay into the estimate. This
implies that schedule is a factor of cost performance and has large influence on CEAC. The
results of comparative analysis without (the base case) and with (the ES-based case) CF in the
model equation show that the second case generates more accurate and precise forecasts in all
three completion stages of a project. This cost-schedule relationship represented in the model
equation is a contribution to extending the EVM research and practice to better capture the
In addition, the field literature reports that application of traditional IB methods (originally
developed as a tool to manage complex and large projects) is questionable when it comes to
CEAC of small-sized and short-duration projects. On the contrary, the proposed forecasting tool
demonstrates applicability to small and medium-sized projects, such as those used for the
sample test.
Moreover, it is an effective and practicable method for early CEAC when as few as three
time points are available. Thus, it is effective for short lifespan projects with small-sized
budgets.
To pursuit the advantages of this methodology, future research is directed toward evolving
the theoretical framework and providing for extended applicability. On the one hand, with the
opportune to integrate the given CEAC calculation method with uncertainty and risk analysis
able to capture major exceptional risk events into CEAC formulation, as well as include expert
managerial belief as a potential source of corrective actions affecting future project performance
and CEACs. With this regard, it should also be comprised within the framework of the system
behavior theory.
Another prospective research direction drives towards understanding the impact of reducing
the number of the parameters for the GGM, as it will ease understanding the time-cost/schedule
relationship being described by the model, as well as it will make the model easier-to-adopt by
field practitioners.
Finally, the method is proposed to practitioners for application to a larger variety of projects
Glossary
Actual Cost (AC). The realized cost incurred for a project during a specific time period.
Actual Time (AT). The number of time periods from the start of a project to a project status
date.
Budget at Completion (BAC). The sum of all the budgets established for the work to be
Completion Factor (CF). The Earned Schedule (ES)-based factor which is the ratio of
Expected Duration at Completion to Planned Duration (PD). It is inversely related to the time-
based Schedule Performance Index (SPIt). The value of CF less than 1.00 indicates a project is
ahead of schedule (favorable condition), more than 1.00 – behind schedule (unfavorable
Composite Index (CI) – a Performance Index (PI) expressed as the sum of the portions of
Cost at Completion (CAC). The realized cost incurred for a whole project. Actual Cost
Cost Estimate at Completion (CEAC). The expected total cost of completing a project. It
is expressed as the sum of the Actual Cost (AC) to date and the remaining portion of Budget at
Completion (BAC minus AC) adjusted by a Performance Index (PI) or Gompertz Growth
Model (GGM).
Cost Performance Index (CPI) – a Performance Index (PI) expressed as the ratio of Earned
Critical Ratio (CR) – a Performance Index (PI) expressed as the product of Cost
Earned Value (EV). The cumulative to date measure of the work performed, expressed in
Expected Duration at Completion (EDAC). The expected total time for completing a
project. It is expressed as the ratio of Planned Duration (PD) to the time-based Schedule
which accounts for the slow initial growth of cost accrual, accelerating the growth rate on the
Index-Based (IB) method. A method to compute Cost Estimate at Completion (CEAC) that
Least Squares (LS) method. A method that determines the best fit of the observed values in
the data fitting. The method minimizes the sum of the squared forecast errors; the error which is
the difference between an observed value and the fitted value generated by the Gompertz
Mean Absolute Percentage Error (MAPE). The average of the absolute values of
Moving Average (MA) – a Performance Index (PI) expressed as the averaged CPI which is
the ratio of sums through three periods beginning with the most recent period and going
backwards.
Percentage Error (PE). A measure of forecast accuracy which is the difference between
Cost Estimate at Completion (CEAC) and Cost at Completion (CAC) expressed as a percentage
of CAC with a negative value suggesting underestimation and a positive value – overestimation.
Performance Index (PI). A measure of cost efficiency of budgeted resources. Four types of
Errors (PE) from the Mean Absolute Percentage Error (MAPE) which is computed by taking the
square root of the variance (the average of the squared differences between PE of an individual
project and mean of PE). A smaller value of SD indicates that cost estimates calculated by a
particular model are closer to its mean forecast error and suggest more precise estimates.
The α parameter. A parameter of the Gompertz Growth Model (GGM) that represents the
The β parameter. A parameter of the Gompertz Growth Model (GGM) that is the y-
The γ parameter. A parameter of the Gompertz Growth Model (GGM) that governs the cost
growth rate.
References
Anbari, F.T., 2003. Earned value project management method and extensions. Project
Bates, D.M., Watts, D.G., 1988.Nonlinear regression analysis and its applications. John
Caron, F., Ruggeri, F., Merli, A., 2013.A Bayesian approach to improve estimate at completion
Christensen, D.S., Antolini, R.D., McKinney, J.W., 1995. A review of estimate at completion
Cioffi, D.F., 2005. A tool for managing projects: an analytical parameterization of the S-curve.
Cioffi, D.F., 2006. Designing project management: A scientific notation and an improved
formalism for earned value calculations. International Journal of Project Management, 24,
136-144.
De Marco, A., Narbaev, T., 2013. Earned value-based performance monitoring of facility
Fleming, Q.W., Koppelman, J.M., 2006. Earned value project management, third ed. Project
Henderson, K., Zwikael, O., 2008. Does project performance stability exist? A reexamination of
CPI and evaluation of SPI(t) stability. Cross Talk – Journal of Defense Software
Kim B.C., Reinschmidt, K.F., 2011. Combination of project cost forecasts in earned value
Kim, E., Wells, W.G., Duffey, M.R., 2003. A model for effective implementation of earned
382.
Lipke, W., 2003. Schedule is different. The Measurable News, Summer, 31-34.
Lipke, W., 2004. Independent estimates at completion: Another method. Crosstalk – The
Lipke, W., Zwikael, O., Henderson, K., Anbari, F., 2009. Prediction of project outcome: the
Marshall, A., Ruiz, P., Bredillet, C.N., 2008. Earned value management insights using
inferential statistics. International Journal of Managing Projects in Business, 1(2), 288 - 294.
Naeni, L.M., Shadrokh, S., Salehipour, A. 2011. A fuzzy approach for the earned value
Narbaev, T., De Marco, A. 2012. Combination of a nonlinear regression model and earned
schedule to forecast a project final cost. In: Proceedings of the 4th EVM Europe Conference
Pajares, J., Lopez-Paredes, A., 2011. An extension of the EVM analysis for project monitoring:
The cost control index and the schedule control index. International Journal of Project
PMI, 2011. Practice Standard for Earned Value Management, second ed. Project Management
PMI, 2008. A Guide to the Project Management Body of Knowledge, fourth ed. Project
Seber, G.A., Wild, C.J., 1989. Nonlinear regression. John Wiley&Sons, New York, NY.
Teicholz, P., 1993. Forecasting final cost and budget of construction projects. Journal of
Tracy, S.P., 2005. Estimate at completion: A regression approach to earned value. M.S. thesis,
Trahan, E.N., 2009. An evaluation of growth models as predictive tools for estimates at
completion (EAC). M.S. thesis, Air Force Institute of Technology, Wright-Patterson AFB,
OH.
Vandevoorde, S., Vanhoucke, M., 2006.A comparison of different project duration forecasting
methods using earned value metrics. International Journal of Project Management, 24(4),
289-302.
Warburton, R.D.H. 2011. A time dependent earned value model for software projects.
Yokum, J.T., Armstrong, J.S., 1995. Beyond accuracy: Comparison of criteria used to select
Zwikael, O., Globerson, S., Raz, T., 2000. Evaluation of models for forecasting the final cost of
Schedule
AC EV d l
delay
EV onto PV
ES AT PD EDAC Time
Growth, $
α
GGM(x)
=α/e
0 x=β/γ Time
fig 2
18
16
14
12
10
8
6
4
2
0
Early Middle Late
Base ES CPI CR CI MA
SD, %
25
20
15
10
0
Early Middle Late Stage
Base ES CPI CR CI MA
3b
PE, %
15
10
5
0
1 2 3 4 5 6 7 8 9
-5
-10
-15
-20
Project
Base ES
number
4a
PE, %
10
5
0
1 2 3 4 5 6 7 8 9
-5
-10
-15
-20
Project
Base ES
number
4b
PE, %
8
6
4
2
0
-2 1 2 3 4 5 6 7 8 9
-4
-6
-8
-10
-12
-14
Project
Base ES
number
4c