Available online at www.sciencedirect.

com

ScienceDirect
International Journal of Project Management 34 (2016) 1493 – 1504
www.elsevier.com/locate/ijproman

Estimating a project's earned and final duration

Roger D.H. Warburton a,⁎, Denis F. Cioffi b
a
Department of Administrative Sciences, Metropolitan College, Boston University, Boston, MA 02215, USA
b
Department of Decision Sciences, School of Business, The George Washington University, 2201 G Street, NW, Funger Hall, Suite 415, Washington, D.C. 20052, USA

Received 3 June 2015; received in revised form 9 August 2016; accepted 15 August 2016
Available online 1 September 2016

Abstract

Using standard Earned Value Management (EVM) protocols, the current approach to Earned Schedule (ES) is extended and formalized to
establish its rigorous, theoretical foundation. A precise definition is provided for what we term the project's earned duration, whose creation
completes the triad of planned, actual, and earned durations. The published ES formula emerges as a linear approximation, but is found to work
with some nonlinear cost profiles, and the conditions under which it gives both correct and incorrect duration estimates are noted. In the several
planned and earned value functional profiles examined, no approximations are required to derive an exact analytical expression for the final
duration; most duration formulas are straightforward and useful. The reliability and accuracy of the duration formulas are demonstrated with
several examples of real, nonlinear project data that represent large classes of projects. We conclude with practical guidance for project managers.
© 2016 Elsevier Ltd. APM and IPMA. All rights reserved.

Keywords: Earned value management; Earned schedule; Duration estimation; Schedule variance

1. Introduction 2010). However, Evensmo & Karlsen (2006) pointed out that the
resulting duration formula is based on linear cumulative planned
When a project goes into execution, many stakeholders wish and earned value cost curves. Why should such a linear theory
to know when the project will finish. Currently, while there work when real-world project cost curves are usually presented as
exists a formula to estimate a project's final duration, there is no S-shaped? Further, Batseliera & Vanhoucke (2015) suggest that
theory to support its use. Therefore, a project manager cannot duration forecasting methods have not been empirically proven.
know, for example, the accuracy of an estimate or, even, if the This is an example of the criticism by Koskela & Howell
formula is relevant to that particular project. This represents a (2002, p.293) that project management is a “narrow and implicit
major shortcoming in project management and our goal is to theory.” The theory is narrow because ES is a linear theory and
address this deficiency by providing a sound, formally justifiable it is implicit because the assumptions, such as linearity, are
approach to duration estimation. rarely acknowledged.
The current approach to duration estimation is based on Koskela & Howell (2002) also point out that project
Earned Schedule (ES) and several research studies claim that management is often dominated the dispatching, or thermostat,
ES works well in some instances (Vanhoucke & Vandevoorde, model and such criticisms are well known (Johnston & Brennan,
2006, 2007; Lipke et al., 2009). However, the concept of ES is 1996). In the primitive, thermostat model, project managers
not without its problems (Book, 2006; Kim, 2000). attempt to correct deviations from set parameters. Hofstede (1978)
Currently, ES begins with a geometrical construction that explains that this requires several assumptions: that there is a
defines the delay as the horizontal projection from the earned performance standard that can be defined and measured; that there
value curve to the planned value curve, see Fig. 1 (Lipke, 2003, is a causal relation between management actions and project
outcomes; and that management actions can return the project to
⁎ Corresponding author. the desired state.
E-mail addresses: [email protected] (R.D.H. Warburton), However, without a formal underlying theory that defines
dcioffi@email.gwu.edu (D.F. Cioffi). the relevant project observables, and the relations between

http://dx.doi.org/10.1016/j.ijproman.2016.08.007
0263-7863/00/© 2016 Elsevier Ltd. APM and IPMA. All rights reserved.
1494 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504

on real-world projects. We also wish to establish a theory that can

be applied to all projects, including those with nonlinear, S-shaped
cost profiles. Finally, our goal is to define the validity, range of
applicability, and accuracy to be expected from the use of a
specific duration formula. With such a theory, we should be able
to advise project managers on how to improve the accuracy of
their duration predictions.
Our research contribution is that we do, in fact, establish such
a theory and it is based on the standard underlying assumptions
for Earned Value Management (EVM), which means it should be
familiar to project managers and require little additional effort
in its use. Our research is also validated because the current
approach to ES is entirely contained in our model as the special
case of linear cumulative planned and earned value cost curves.
However, our new definition allows us to extend EVM into
new areas and to derive duration estimates for any type of realistic,
nonlinear cost curve. No approximations are required, which
Fig. 1. For the linear case, the definitions of the delay, δi, at the time, ti, and the
earned schedule, ES.
allows us to formally establish the validity and accuracy of all
duration estimation formulas. We explain that duration estimation
depends on structural assumptions about cost profiles and, so,
them, there is no guarantee that any specific action by a project understanding their shapes and parameters will lead project
manager will fix or, even, affect a particular observable. Thus, a managers to a better appreciation of the issues in duration
major shortcoming of the thermostat model is that, without an estimation.
underlying theory, the relevant project observables remain Finally, we estimate the duration for several well-known,
unknown. In the context of duration estimation, therefore, one nonlinear cost curves, which demonstrates the method's successful
does not know which parameters to measure or which to application to more realistic projects than the current linear version.
attempt to manage to effect change. By proposing a formal Even for nonlinear profiles, in most cases, a simple duration
theoretical foundation for duration estimation, we demonstrate formula emerges, validating the practicality of the theory. The
to project managers which quantities are important and should agreement between theory and practice shows that it is possible to
be measured, and which to prioritize to help them manage create a theory of duration estimation that has immediate practical
effectively. benefits. The theory is, for the most part, no more complex than the
As an example of common practice, we note that in many current version of ES, which is used for estimation and tracking.
types of large projects the cumulative cost profile follows an The paper proceeds with a review of the applicable literature,
S-shaped curve (Christian & Kallouris, 1991) and using such after which we undertake a critical evaluation of the existing ES
curves for cash flow projections can achieve accuracies of over concept (Section 2). In Section 3 we propose a new, formal
90% (Singh & Lakanathan, 1992). It is implicitly assumed, definition for the project delay, which allows us to derive exact,
therefore, that such cost curves have universal applicability and analytical expressions for the project duration. In Section 4, we
the success of such estimation tools suggests that there might be analyze several well-established, practical, nonlinear representa-
some fundamental, structural underpinning that is common to tions of the cost profile and derive exact expressions for the final
projects. If duration estimation is similarly to apply universally duration. In Section 5, we present practical project data for each of
to projects, that also requires the existence of some underlying the cost profiles and show that the relevant formula accurately
theoretical structure to give it legitimacy. predicts the final duration. We also explain for which profiles the
Another basic tenet of project management is that, during linear formula gives incorrect duration estimates, as well as the
execution, one can predict the project's future duration from degradation in duration accuracy from its use. Section 6 concludes
some initial data. There are a number of hidden assumptions with a review of the project management implications.
implied in this approach: that there exist project parameters that
can be measured; that these parameters remain constant (or at
least predictable) over the life of the project; and that the 1.1. Previous efforts examining duration prediction
estimation process is robust against uncertainties in the data.
Interestingly, industrial data strongly suggest that some project Standard EVM techniques yield reliable predictions only
parameters do indeed remain constant over time, e.g., error when applied to cost estimation (Vanhoucke & Vandevoorde,
rates, see McGarry et al. (1994). 2006; Christensen, 1993) and the problems of applying EVM to
The goal of this work is to eliminate this critical deficiency in schedule prediction are well known (Marshall, 2006; Book,
project management by establishing a formal theoretical founda- 2003, 2006). Lipke (2003, 2010) was one of the first to address
tion for duration estimation. A further goal is to verify that the this issue by defining a geometrical construction procedure for
theory is practically useful by demonstrating that it provides ES and Stratton (2007) later provided one of its first formal
estimates of the final duration early enough to be practically useful definitions.
 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504 1495

Vanhoucke & Vandevoorde (2007) reviewed the accuracy accuracy by including schedule progress as a factor in the cost
of forecasting methods, concluding that the cost performance performance. Chen et al. (2016) improved forecasting accuracy,
index (CPI) and the schedule performance index (SPI) provide but their model required a logarithm linear transformation of the
valuable information about trends in project performance. They planned value data and linear regression modeling. Warburton
then invoked the assumption that when corrective managerial (2014) used a trapezoidal labor profile, which is often used to
actions are implemented, the changes in the behavior of the describe construction projects, to derive accurate duration estimates
indexes are assumed to reflect the impact of management actions. early in the project. However, these studies are not supported by
Thus, they invoked Hofstede's (1978) general assumptions about formal theory.
management actions.
Vanhoucke (2012) used Monte-Carlo simulation on fictitious 2. A critical evaluation of current earned schedule
and empirical project data and concluded that networks with
greater parallelism have more variability than those with a more The usual equation for ES, for example found in Lipke
serial structure. Thus, more parallelism, which produces S-shaped (2003, 2010), is:
curves, degrades forecasting accuracy. This suggests that a more
EV i −PV u
powerful, nonlinear theory is required, which provides a strong ES i ¼ u þ Δ; where Δ ¼ : ð1Þ
motive for the current work. PV uþ1 −PV u
The predictions of the Average Planned Value method (Anbari, Fig. 1 explains the quantities: The subscript i represents the
2003) are accurate if the averages accurately represent the time- current time, ti, and the subscript u is an integer representing the
dependent quantities. Jacob & Kane (2004) observed that most number of time units where, if the project is behind schedule,
schedule prediction methods have similar accuracies because they the earned value is less than the planned value. The delay, δi,
apply the same basic parameters. Khamooshi & Golafshani (2014) represents a specific time interval: the horizontal projection back
criticized ES, along with other EVM-based measures, for using from the point ti on the earned value curve to the intersection with
monetary measures as a proxy for the true duration and argued the planned value curve. The mathematical issues associated with
that such measures may not accurately represent the duration's Eq. (1) are discussed in Appendix A.1.
progress. They, instead, proposed an Earned Duration Method The conventional argument for Δ is that since the intersection
(EDM) that decoupled the cost and schedule dimensions by using occurs between the times u and u + 1, it represents the appropriate
actual durations rather than their monetary proxies. However, all fraction of that time-interval. But the dimensionless ratio, Δ,
these methods continue the linear approach, again providing a presents an algebraic difficulty (Book, 2003, 2006). To clarify the
motive for a nonlinear theory. issue, we write the denominator in Eq. (1) as the slope of the
In a Monte-Carlo analysis, Vanhoucke & Vandevoorde (2007) planned value curve between u and u + 1:
showed that ES almost always outperformed other schedule
estimation techniques. Lipke et al. (2009) showed that ES worked PV uþ1 −PV u
mu ≡ : ð2Þ
well for 12 projects under study and suggested it was sufficiently ðu þ 1Þ−u
reliable for general application. While validating the usefulness
of the current ES method, these studies used the linear formula The standard schedule variance (SV) at time, u, is SVu ≡
without a theoretical justification. EVu − PVu, giving:
Elshaer (2013) suggested that while ES sometimes out- SV u
performed other methods, it failed when incorrect warnings ES u ¼ u þ : ð3Þ
mu
emerged from non-critical activities, i.e., when there was
parallelism. Vanhoucke (2012) confirmed that the network Thus, ESu depends on the point values of both the SV and PV
topology is a significant driver of variability, which again points cost rates (SVu and mu). But why should the estimated performance
to the importance of the relation between forecast accuracy and of the system at the current time depend on a single, point value of
the degree of nonlinearity, or parallelism, in the cost profile. mu only at the intersection time, u? Arguments could presumably
While the practical application of ES has been demonstrated be made for a dependence on some thoughtfully chosen average
in some cases, its theoretical foundation has been under- over previous points rather than the arbitrarily chosen points
developed. Evensmo & Karlsen (2006) addressed the linearity immediately around u.
assumption behind the current ES approach by proposing a Many projects follow nonlinear and S-shaped cumulative
cubic polynomial cost curve. Cioffi (2005) proposed an S-curve cash-flow curves for which the slope varies with time (Singh &
cost profile and demonstrated its use in predicting project costs and Lakanathan, 1992). Therefore, it is hard to understand how the
schedules. Warburton (2011) developed a time-dependent EVM instantaneous slope at one point can legitimately characterize
model for projects that follow the nonlinear Putnam-Norden- the entire project. At a minimum, one can argue that a longer
Rayleigh (PNR) profile (Putnam, 1978). While these studies time scale would at least average the noise in the data and yield
attempted to extend EVM, they all used empirical formulas a better estimate of the slope of the PV curve.
without an underlying theoretical justification. Thus, it is worth pointing out that Eq. (1) is actually based
Other authors modified the standard EVM approach. Narbaev on a hidden assumption: a linear slope in the local region,
& De Marco (2013) proposed a Gompertz-based growth model, which will generally not apply globally (Evensmo & Karlsen,
using nonlinear regression curve fitting, that improved forecast 2006). Since most realistic project cost curves are nonlinear,
1496 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504

this discrepancy introduces a question about the theoretical Thus, τ is the ratio of the planned to actual duration and its
applicability of Eq. (1). definition is consistent with traditional EVM index conventions,
Eq. (1) also presents a dimensional difficulty: while ES i.e., τ b 1 represents a delayed project.
and u have the dimensions of time, Δ is dimensionless. The At the current time, t, the delay, δ(t), is defined as the time
problem arises because the definition of Δ specifies the integers difference represented by the horizontal projection back from
u + 1 and u instead of the correct dimensional times, tu + 1 and tu. the point on the earned value curve at t to its intersection with
Unless the planned value curve has a constant slope, different the planned value curve:
time units will give different answers for ES and the resulting
duration estimate. Changing the time interval for the measure- C e ðt Þ ¼ C p ½t−δðt Þ: ð5Þ
ment of ES changes the answer for the duration, and studies using
Positive values for δ(t) represent accelerations while negative
ES under such conditions are suspect. The corrected formula is
values represent delays. An example of an acceleration, + δ(t1),
derived in Appendix A.1.
and a delay, − δ(t2), are shown in Fig. 2. One can see that for
In addition, towards the end of a project with a nonlinear
δ(t) b 0, the project has been accelerated because the current
cost curve, the cumulative planned value approaches a constant
earned value corresponds to a quantity that was to be obtained at
(the total budgeted cost). There, the expression for Δ in Eq. (1)
some later time.
becomes undefined because the slope of the curve goes to zero,
We now define the Earned Duration, Te(t), as the time from
e.g., see Fig. 2. With increasing time, the ratio SVu/mu approaches
the start of the project to that of the above intersection:
infinity, and ES becomes mathematically ill-defined.
T e ðt Þ ¼ t−δðt Þ: ð6Þ
3. Formally defining earned schedule
Thus, we have created a duration triad: planned total duration,
To eliminate the above difficulties, we propose new, formal T1; actual total duration, T1′; and, at any given time, t, the earned
definitions. Following Cioffi (2006a), we will streamline notation duration, Te(t). These durations are directly analogous to the
by using C (a mnemonic for cost) to denote cumulative cost data, quantities that compose the EVM cost triad: planned value, actual
thereby replacing EV with Ce, PV with Cp, and AC with Ca, except cost, and earned value.
in historical formulas. We distinguish between actual project data Because these new definitions are completely general, they are
and the associated, theoretical cost profiles. For example, the independent of the specific shape of the cost curves and can be
cumulative planned cost profile is denoted as Cp(t), whereas, at applied to any profile. They also eliminate the issues associated
time, ti, the project's planned cost data is denoted Cp(ti). with Δ in Eq. (1) because the quantity, δ(t), is well defined
We define the planned end point of the project as T1 and everywhere. It was simply the expression for Δ in Eq. (1) that
assume that, during execution, it ends at T1′. If the project is was undefined.
delayed, T1′ N T1, and if the project is accelerated, T1′ b T1. The We follow standard EVM, which says that as each activity is
total planned cost is Cp(T1). completed, it earns its planned value, even if there is a cost
We define the useful quantity, τ, as, increase or a delay in completing the activity, e.g., if additional
resources have to be applied (Project Management Institute,
T1 2013, 2011). If new, unplanned work is proposed (i.e., a scope
τ¼ : ð4Þ
T 01 increase), the project manager must re-plan the project and
develop a new planned cost profile, Cp′(t). All of the formulas in
this paper then apply with Cp′(t) replacing Cp(t).
At the end of the project, when all activities are completed,
the total earned value equals the total planned value (Project
Management Institute, 2011). Therefore, one can always draw a
line from the earned value curve to the planned value curve,
which ensures that both the delay, δ(t), and the earned duration,
Te(t), exist everywhere.

3.1. The linear cost profile

Although the difficulties with Eq. (1) manifest themselves

most strongly with nonlinear profiles, to illustrate our method in a
simple system, we begin with a linear profile. After gaining this
understanding, we progress to the more complicated nonlinear
cost curves.
Consider a project with a constant planned spending rate. For
example, if labor dominates the spending, a constant planned
Fig. 2. Two examples of the earned duration: Te(t2) for a delay (at t2) and Te(t1) spending rate would occur if the number of people projected to
for an acceleration (at t1). work on the project is constant and if they work a constant
 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504 1497

number of hours in each time period. The cumulative planned and an expected duration” (Project Management Institute, 2013).
earned values are then linearly proportional to time: Triangular cost rate profiles result in non-linear cumulative cost
curves and, interestingly, give support both for and against the
t t
C p ðt Þ ¼ C p ðT 1 Þ and C e ðt Þ ¼ C p ðT 1 Þ 0 ; ð7Þ use of the linear duration estimation formula.
T1 T1 Consider triangular profiles for planned and earned value
where Cp(T1) is the total planned cost, and we have invoked the rates; neither need be isosceles—see Figs. 3 and 4. The planned
standard EVM assumption that the total earned value equals the profile peaks at Πp and the total planned duration is, as above, T1.
total planned value at the end of the project (Project Management The earned value cost rate peaks at Πe and the actual final
Institute, 2011). The definition for δ(t), via Eq. (5), then gives duration is T1′. The constants, f (planned) and g (earned),
δ(t) = t(1 − τ). The earned duration at time, t, is defined as Te(t) ≡ represent the fraction of the total duration at which the peak in
t − δ(t), which gives Te(t) = t τ. Using Eq. (4), the expression for the cost rate occurs (0 b f , g b 1) and f and g can be used to create a
the project's new final duration is: variety of project cost profiles.
We again follow the standard EVM approach and begin with
T1 t T1 the case where f = g, i.e., the peaks in the planned and earned cost
T 01 ¼ ¼ : ð8Þ rates occur at the same fraction of their total durations, i.e., Πp
T e ðt Þ τ
occurs at time, f T1, and Πe at f T1′. In Appendix A.2, we show
Knowing the planned duration, T1, one measures the time that, even for these nonlinear profiles, the delay grows linearly
delay, δ(t), between the earned and planned curves from the project with time and the prediction of the final project duration, T1′, is
data, and the estimate of the project's new final duration follows. identical to the estimate in the linear case, Eq. (8). Further, in this
Using inverse function notation, the delay is δ(t) ≡ t − f = g case, the linear formula is valid for the entire project duration.
Cp− 1[Ce(t)]. In the linear case, from Eq. (7), we find Cp− 1(x) = The quantity, f, is an example of an internal parameter that
xT1/Cp(T1) and Ce(t) = Cp(T1)t/T1′, from which Eq. (8) immedi- characterizes the cost profile. However, it is important to note that
ately follows. f does not appear in the final duration formula, so that a project
We also note that if there were scope creep, the earned value manager does not actually need to know the value of f to find the
would satisfy Ce(T1′) N Cp(T1). Unless the planned value is also duration for the triangular profile. One simply calculates ES, from
adjusted, Cp(T1) is not canceled from Eq. (8), which shows that, the delay between the curves, and uses Eq. (8).
in the presence of scope creep, the linear duration formula may We next consider the case where f ≠ g, which means that the
not hold. planned and earned value rates peak at different fractions
through the project; see Fig. 3. In this case, Eq. (8) is no longer
3.2. The SPI for the linear profile valid and this can most easily be seen in the region 0 b t ≤ f T1,
where the duration formula is (see Appendix A.2):
The classic Schedule Performance Index (SPI) can also be used sffiffiffi
to estimate the project duration. Using the linear profiles, Eq. (7) 0 T1 t f
T 1 ð f ; gÞ ¼ : ð9Þ
gives SPI = Ce(t)/Cp(t) = T1/T1′ = τ, which is a constant. While a T e ðt Þ g
constant SPI is only a feature of linear cost profiles, we have
established that it is possible to derive a duration formula in terms We emphasize that for the triangular profile, selecting
of the SPI, as well as its conditions of applicability (here, linear different sets of parameters for the planned and earned curves
profiles). In fact, this will turn out to be possible for all cost profiles.
Lipke (2010) revised the definition of the SPI to derive an
Independent Estimate at Completion (IEAC), which gives the
same estimate for the duration as our new definition. However,
the above derivation of T1′ = T1/τ (Eq. (8)) makes it clear that
IEAC is theoretically valid only for the linear case.
Summarizing, then, the final duration can be measured from
the delay, δ(t), and from the SPI. Once T1′ (or τ) is known at any
time, t, the evolution of the system is determined. The accuracy
of the duration prediction will depend on the match of the
project data to the linear cost profile, an issue that is discussed
in the conclusions.

4. Nonlinear cost rate profiles

4.1. The triangular cost rate

A triangular cost rate profile is often used when the peak

spending rate can be estimated reliably, e.g., “duration estimates Fig. 3. Triangular cost rate profiles for planned value (black, thin, f = 0.3) and
based on three points with an assumed distribution provide earned value (black, thick, g = 0.3; gray, g ≠ 0.3).
1498 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504

Instead of finding the SPI proportional to τ (as it was in the

linear case), we find that it is proportional to its square and also
depends on f and g. Nevertheless, in this region, we can again
determine an SPI-based duration estimate:
rffiffiffiffiffiffiffiffi
0 f =g
T1 ¼ T1 0 b t ≤ f T 1: ð11Þ
SPI
Eq. (11) illustrates several general characteristics of SPI-based
duration estimates. First, it depends on the parameters of the
cost profile (here, f and g). Also, the estimate is only valid up to
the peak in the planned value (t ≤ f T1). However, this is the
practically valuable time range, as project managers wish to
determine a revised duration as early as possible.

4.2. The Putnam-Norden-Rayleigh profile

We now analyze a theoretically and practically important

profile, the Putnam-Norden-Rayleigh (PNR) curve. Multiple
studies show that cumulative earned value project data closely
follow the PNR curve (Warburton, 1983; Basili & Beane, 1981;
Fig. 4. Cumulative planned and earned values for triangular profiles with the
same notation as Fig. 3. The arrow shows where the f = g = 0.3 curves transition
Lee, 2002; Gallagher & Lee, 1996). Davis et al. (2009) rigorously
from pre-peak to post-peak; see Appendix 5.1. demonstrated that PNR-based methods efficiently estimated costs
and schedules for projects in the DoD Contract Analysis System,
which includes 107 Research and Development contracts from
( f and g) results in different duration formulas. If a naive 1970 through 2002. While business practices have evolved, the
user did not realize that the project's triangular earned value PNR model explains later cost variations as well as it did in 1970.
rate was following a different triangular curve from the planned The PNR profile is theoretically powerful because it provides
rate ( f ≠ g), using the linear duration formula would be in error. an analytical representation of a nonlinear cost profile over
That the duration formula depends on the parameters of the the project's entire life. The cumulative planned value curve is
planned and earned value cost profiles is an important assumption (Putnam, 1978; Warburton, 2011):
that it is already built into traditional EVM, although it is rarely " !#
t2
stated explicitly. For example, in the traditional, ES-based linear C p ðt Þ ¼ C p ðT 1 Þ 1− exp − 2 ; ð12Þ
case, one must specify that the cumulative planned and earned 2T p
value curves are both linear, but with different slope parameters.
where Cp(T1) is the total cost; and Tp is the time when the planned
However, the linear duration formula does not contain the slope
value rate curve, C_ p ðtÞ , reaches its maximum. (By way of
parameters. Somewhat surprisingly, the duration formula in the
comparison, in the previous triangular case, the peak was at
triangular case also does not contain the parameters that charac-
Tp = fT1.) The same PNR expression is used for the earned curve,
terize the cost triangle (Πp, Πe, and f = g), so the project manager
but Tn represents the time of the peak and Tn N Tp, so the earned
does not need to know anything about them to use the duration
curve is delayed. Cp(T1) is the total planned cost, and, invoking
formula.
the standard EVM assumption that, at the end of the project,
Thus, triangular profiles illustrate both positive and negative
the total planned value equals the total earned value (Project
aspects of the linear duration formula. The f = g case shows the
Management Institute, 2011), the earned value curve is:
linear formula applies beyond its linear origins and into more
  
realistic, nonlinear regimes. Also, the linear duration formula t2
applies over the entire project. On the other hand, when f ≠ g, C e ðt Þ ¼ C p ðT 1 Þ 1− exp − 2 : ð13Þ
2T n
the linear duration formula no longer applies and significant
errors may occur in its use; see Section 5.1. In Appendix A.3 we use the definition of δ(t) from Eq. (5) to
derive the duration estimation formula for the nonlinear PNR
4.1.1. The SPI in the triangular case profile. The delay and the earned duration are both identical to
A duration estimate can be also obtained in the triangular case those of the linear and triangle ( f = g) cases. Again, we note that
from the SPI. Early in the project, when 0 b t b f T1, the planned the parameters of the cost curve (Tp and Tn) do not appear in
and earned rates are given by Eqs. (19) and (20), so that: the duration formula, so the project manager does not need to
estimate them to determine a duration.
Fig. 5 illustrates that, even though the planned and earned
Πe t 2 2 f T 1 f
SPI ¼ 0 ¼ τ2 : ð10Þ curves are nonlinear, the time difference between the two curves,
2gT 1 Πp t 2 g the delay, δ(t), increases linearly with time. Since the PNR
 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504 1499

where y∞ is the final cost, higher values of α result in steeper

S-curves, γ = exp (αTp), and Tp is the peak in the cost rate curve,
y_ ðtÞ. The mathematical analysis is presented in Appendix A.5.
For practical applications, see Cioffi (2005, 2006b). Using the
definition of the delay, ye(t) = yp[t − δ(t)], gives:
2   3
−αt −αðt−δðt Þ−T p Þ
1 ð 1−e Þ 1 þ e
T n ¼ t þ ln 4 −15; ð16Þ
α 1−e−αðt−δðtÞÞ

where Tn is the peak in the earned value rate curve. The values for
Tp and α can be determined from a fit to the planned value data,
using either the cumulative or rate curve. The delay, δ(t), is
determined in the usual way, as the difference between the
cumulative planned and earned value curves, and the new Tn is
then determined from Eq. (16).
For the Cioffi curve, we used of the same value for α, but
different values for Tp and Tn. T1 and T1′ are determined in terms
of Tp and Tn in the same way as for the end of the PNR project,
see Appendix A.4. We can also determine T1′ from the SPI, but
Fig. 5. For the nonlinear PNR profile, the delay, δ(t), grows linearly with time the expression is not useful as it is complex and valid only very
(as it did in the linear case). early in the project.
The key point is that the linear duration formula does not apply
to the Cioffi profile and we will explore the practical impact in
expressions are valid for all time, the duration estimate is valid for Section 5.3.
the life of the project.
A complication associated with the PNR rate profile is that, 5. Practical applications
mathematically, it never reaches zero and, so, we need to define
the project's end time. In Appendix A.4, we prove that the 5.1. The triangular profile
duration estimate is independent of an arbitrary choice of the
project end time. The cost rates were shown in Fig. 3 and the cumulative costs
That two, quite different, nonlinear profiles give the same in Fig. 4. We showed that it is incorrect to use the linear duration
duration expression as the linear case reinforces the idea that formula for triangular profiles for which f ≠ g and Fig. 6 shows the
duration estimation has wide practical value and that the linear size of the duration errors that occur in that case. The duration
formula actually applies beyond its linear origins. This is further
explored in the conclusions.

4.2.1. Using the SPI to estimate the project delay

Early in the project, we have (see Appendix A.3):
 2
T1 T1
SPI→ 0 ⟹ T 01 ¼ pffiffiffiffiffiffiffiffi : ð14Þ
T1 SPI

While the expression for the revised duration, T1′, based on

the delay, δ(t), is the same for the linear, triangular ( f = g) and
PNR profiles, the SPI-based duration expressions are different.
Therefore, we again find that determining the duration from the
SPI requires knowledge about the cost profile.

4.3. The Cioffi S-curve

We now explore the Cioffi (2005) profile (presented in a

slightly modified notation), which produces a markedly different
expression for the duration. The cumulative Cioffi cost profile is:

1−e−αt
yð t Þ ¼ y∞ ; ð15Þ Fig. 6. Percentage errors in the duration estimate when using the linear formula
1 þ γe−αt for triangular profiles; f = 0.3 and various values of g.
1500 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504

error is zero when f = g because the linear formula applies

throughout (Appendix A.2). The duration is overestimated when
g N f, and can be significant. When f ≠ g, only towards the very
end of the project does the linear duration formula eventually
approach the correct answer.

5.2. PNR

Fig. 7 shows a project for which the planned, actual, and

earned cost rates follow the PNR curve (Warburton, 2011). Both
cost rate data (gray) and the cumulative cost data (black) are
shown. The planned duration was 118 weeks and the actual final
duration was 142 weeks. Duration predictions are shown in Fig. 8,
where the curve labeled “PNR” uses Eq. (8) and the curve labeled
Δ uses Eq. (1). Fig. 8. Duration estimates over time when using the linear duration formula for
In the early stages of the project, funds are spent but little the nonlinear PNR profile, which is theoretically correct. Δ from Eq. (1).
credit is earned. Therefore, early values of Ce(ti) were small and
resulted in large, erratic duration estimates. After about 40 weeks
(30% of the total planned duration), the duration error is less than
(labeled “Cioffi”) uses the exact formula, Eq. (16), which is
~ 10%. We emphasize that the linear duration formula is being
accurate to within about 10% after 20% of the total planned
applied to a highly nonlinear data set, which, for the PNR profile,
duration and within a few percent after 25%.
we proved to be theoretically legitimate.
The smooth gray curve (labeled “Linear”) represents the
theoretical duration predictions that result from using the
5.3. Cioffi (incorrect) linear formula. That the linear formula gives incorrect
duration estimates is further confirmed by its application to the
To illustrate the impact of the complex duration expression for OED data (labeled “Δ”), which appear to be following the linear
the Cioffi profile, Fig. 9 shows duration predictions computed for linear curve.
the Oxford English Dictionary (OED) project. Cioffi (2006b) For data that follow the Cioffi profile, only the appropriate
proved that the OED data were an excellent fit to the cumulative nonlinear formula gives the correct duration. Thus, the linear
cost curve, Eq. (15), with an R value of 0.995. The black curve duration formula must be used with considerable caution.

Fig. 7. The cumulative data for planned value, Cp, earned value, Ce, and actual cost, Ca for the PNR profile showing a delay and cost overrun. The rate data are C_ p and
C_ e . The PNR profile (dashed line, Ce(t)) is a fit to the Ce data.
 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504 1501

case of linear cost curves. Thus, we are extending, rather than

replacing the current approach.
We proved that the same duration formula applies to the
constant, triangular ( f = g), and PNR profiles. Thus, the ES-based
duration formula may be more general than its linear origins and
large classes of nonlinear projects may legitimately use it. Further,
for those nonlinear profiles, we showed that the linear formula
is valid over the project's entire life. These results may help
explain why the linear approach has proved effective for projects
with nonlinear cost curves, e.g., Vanhoucke & Vandevoorde
(2007).
Cost profiles are represented by structural parameters, e.g., the
time of the peak in the cost rate. However, the project manager
does not need to know these internal parameters, because they do
not occur in the duration formula; they are only required in its
derivation. Therefore, the project manager does not need to
Fig. 9. Duration estimates from the OED data: the exact Cioffi solution from Eq. (16) estimate them to determine a duration. For example, we rigorously
(black, labeled “Cioffi”); from Eq. (1) (gray, labeled “Δ”); and the theoretical linear
formula (smooth gray curve, labeled “Linear”).
proved that the parameters designating the peaks in the PNR and
triangular profiles do not appear in the duration formula.
These parameter assumptions are already built into traditional
EVM. For example, the traditional, ES-based duration formula
6. Conclusions requires the specification of the slopes of the linear cumulative
planned and earned value curves. However, these slope parameters
We established a sound theoretical basis for project duration do appear in the duration formula.
estimation using the standard definitions of EVM, which resulted While the derivation of some duration formulas was complex,
in a duration triad (planned, earned, and actual duration) most of the resulting formulas are quite straightforward and
analogous to the EVM cost triad. The entire theory is completely require little mathematical skill to use them. Also, since the entire
general and applies to any, even non-linear, cost curves. We theory was built on standard EVM, it should be familiar and
derived the duration formulas for several cost profiles; no accessible to practitioners. As a result, project managers should,
approximations were required and, so, the results are exact. The for example, be capable of deciding effectively between the
duration estimation formulas are summarized in Table 1. degraded accuracy of a simpler, linear cost profile and the
We also showed that it is possible to derive duration estimates improved accuracy of a more complex, nonlinear version. Also,
from early SPI data. However, the SPI-based duration formulas the theory shows the importance of the peak in the cost rate,
are not as useful, because, for nonlinear profiles, the duration which suggests that more attention be paid to it during execution.
estimates depend on the parameters of the cost profile and, These are examples of Koskela and Howell's (2002) idea that
typically, apply only in the early stages of projects. However, the theory is condensed knowledge, because it empowers novices to
theory provides the justification for the role of the SPI, when it do what experts formerly did.
can be used, and the expected accuracy. This shows how useful Changes in scope could be easily handled. The addition of
results emerge from the interaction of theory and practice. unplanned activities results in a revised plan with a new planned
We established that the current theory of duration estimation, duration. Delays in implementation would result in a new earned
via ES, is linear, which is a restrictive assumption. However, all value curve with a new final duration. All of the formulas in the
previous research on ES is contained in our model as the special paper would simply apply to the revised planned and earned
curves.
We compared our proposed theory to well-known, real-
Table 1 world projects and the theory provided accurate estimates for
Cost rate profiles with their corresponding duration formulas. the final duration. Therefore, the predictions of our theoretical
Cost rate profile Sample Exact duration, T1′ T1′ from early SPI model were validated by the agreement between theory and
Linear T1t/Te(t) T1/SPI practice.
While the Cioffi profile has a complex duration formula,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Fig. 9 confirms that it produces accurate duration estimates. On
Triangle gT1t/( fTe(t)) T1 g=ðf SPIÞ
the other hand, using the linear formula on such projects gives
pffiffiffiffiffiffiffiffi incorrect (not just inaccurate) answers. (This is also true for the
PNR T1t/Te(t) T 1 = SPI triangular, f ≠ g, case.) Therefore, project managers must use
caution when using the linear duration formula.
Cioffi T1′ from Eq. (16) N/A The presented duration estimation methods should not be
considered as alternatives to Critical Path and Tracking Gantt
techniques. Instead, they offer additional techniques that should
1502 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504

help check the estimates from other tools. For small projects, scale is in days: tu = 28 and tu + 1 = 35, so that tu + 1 − tu = 7. In which
schedule control using a Gantt chart is appropriate. However, for case, Eq. (1) will give an incorrect answer for Δ.
larger projects with parallelism, the potential number of critical The problem arises because the definition of Δ specifies the
paths can grow, making it difficult to determine the duration. In quantities as PV(u + 1) − PV(u) and, unless the planned value
such situations, the tools presented may be an effective, additional curve has a constant slope, different time units will give different
option. answers. If the planned value curve is not linear, then the two
Several issues might be explored in future research efforts, slope calculations will not necessarily agree.
e.g., how the effect of uncertainty, or noise, in the cost data
affects the accuracy of the model's predictions. A related issue A.2. The triangular profile
is to determine what types of cost profiles apply to which types
of projects and industries and to provide guidance on the The triangular planned value cost rate profile is expressed
selection of an appropriate cost profile. It would be useful to by:
include the dynamic nature of project management, e.g., the 8
impacts on duration estimates of changes to the cost profile >
> Πp t
< 0 ≤t ≤ f T 1
f T1
during execution occasioned by factors such as scope growth. C_ p ðt Þ ¼ Πp ð19Þ
One might also extend this work to other profiles such as >
>
: ðT 1 −t Þ f T 1 ≤ t ≤T 1 ;
the trapezoidal profile often used in construction (Warburton, T 1 ð1− f Þ
2014). where f is the fraction of total planned duration at which the
In conclusion, we demonstrated that, for all projects, there peak in the triangle occurs, i.e., Πp occurs at time, fT1. We next
is a fundamental relation between the project's cost profile and assume the project is delayed during execution, ending at
its duration estimate. We validated the theory by comparing its T1′ N T1. Now, however, we have a choice for the parameters in
predictions to several real-world projects. A project manager the earned value rate, which need not be the same as those for
need not understand the technical details, but should understand the planned value rate. If the earned value rate peaks with the
the high level implications, e.g., duration accuracy, relevant value Πe at time, g T1′, where g ≠ f, then,
cost profiles, etc. Most of the duration formulas are quite
8
simple and require little additional effort to be practically > Πe t
>
< 0 0 ≤t ≤ g T 01
useful. gT
_C e ðt Þ ¼ 1 ð20Þ
>
> Πe 0

Appendix A. Mathematical derivations : 0 T 1 −t g T 01 ≤t ≤T 01 :

T 1 ð1−gÞ

A.1. The Expression for Δ To obtain the cumulative functions, Cp(t) and Ce(t), we
integrate Eqs. (19) and (20). For example, the cumulative planned
Fig. 1 defines the terminology for the derivation of the value is:
expression for Δ. The earned schedule at time, ti, is defined as the 8
geometrical projection back from the earned value curve, Ce(ti) to >
> Πp t 2
< 0 ≤t ≤ f T 1
the planned curve, Cp(ti − δi), which occurs between the planned C p ðt Þ ¼
2 f T1  
value data points at times, tu and tu + 1. The data points used in the >
> Πp t 2
f Πp T 1
: T 1 t− − f T 1 ≤ t ≤ T 1:
calculation are designated by filled circles. T 1 ð1− f Þ 2 2ð1− f Þ
The tangent of the angle, α, between the planned curve and ð21Þ
the horizontal at tu, is,:
The curves in Figs. 3 and 4 are scaled so that the project plan
ends at T1 = 1 and the actual project is delayed, ending at T1′ = 1.5.
C p ðt uþ1 Þ−C p ðt u Þ C e ðt i Þ−C p ðt u Þ We follow the standard approach to EVM, which is that when
¼ ; ð17Þ
t uþ1 −t u Δ individual planned activities are completed, they earn their
planned value. While delays and cost increases may occur, there
which gives, is no scope creep, so that the total final planned value equals
the total final earned value (SPI → 1). Therefore, when T1 b T1′,
Πe b Πp.
C e ðt i Þ−C p ðt u Þ We begin with the case where f = g and use the definition for
Δ¼ ðt uþ1 −t u Þ: ð18Þ δ(t) from Eq. (5). We define two regions: region 1: Cp(t) in
C p ðt uþ1 Þ−C p ðt u Þ
0 b t b f T1 and Ce(t) in 0 b t b f T1′; and region 2: Cp(t) in
f T1 b t b T1 and Ce(t) in f T1′ b t b T1′.
The above eq. is only equal to the usual expression for Δ The standard EVM assumption is that all work is completed
(Eq. (1)) when tu + 1 − tu = 1. This is not true if the units of the and the total planned value equals the total earned value.
calculation are changed, when an incorrect answer can result. For Therefore, we have Πe/Πp = T1/T1′, which means that the two
example, suppose the data are in weeks so that tu = 4 (week 4) and curves transition from region 1 to region 2 at the same y-values,
tu + 1 = 5. Now suppose the same data points are used, but the time i.e., at Cp( fT1) = Ce( fT1′). This is indicated in Fig. 4, where the
 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504 1503

arrow shows where both curves transition from region 1 to region values (Tp and Tn) rather than the end times (T1 and T1′). In
2. Using Eq. (5) in region 1 gives, Appendix A.4, we prove the exact result that, for the PNR curve,
the end time is proportional to the peak, so that, Tp/Tn = T1/T1′ = τ
Πp t 2 Πe t 2 and the duration estimate is:
ðt−δðt ÞÞ2 ¼ : ð22Þ
2f T1 2 f T 01
t T1
Therefore, the earned duration, Te(t) = t − δ(t), follows again as T 01 ¼ : ð29Þ
Te(t) = t τ, which is the same as in the linear case (Eq. (8)), even T e ðt Þ
though the cost profile is nonlinear. In region 2, the expression for
δ(t) is: Even though the PNR profile is non-linear, this is the same
! formula as in the linear case. Also, it is exact and valid for the
 
Πp ðt−δðt ÞÞ2 Πe 0 t2 entire project's duration. We note that Cp(Tp) = 0.393 Cp(T1)
T 1 ðt−δðt ÞÞ− ¼ T t− : and Ce(Tn) = 0.393 Ce(T1′). Therefore, the peaks in the planned
ð1− f ÞT 1 2 ð1− f ÞT 01 1
2
and earned value rate curves occur roughly 40% through both
ð23Þ the planned and actual evolution.
We can again estimate the schedule delay from the SPI:
This results in a quadratic equation for δ(t),

T2
C e ðt Þ C p ðT 1 Þ 1− expð−t 2 = 2T 2n
−δðt Þ þ δðt Þð2t−2T 1 Þ−t þ 2T 1 t− 1
2 2T 01 t−t 2 ¼ 0; ð24Þ
2 2
SPI ðt Þ ¼ ¼    : ð30Þ
T 01 C p ðt Þ C e T 0
1− expð−t 2 = 2T 2
1 p

which has two solutions. Discarding the negative solution, we

find, δ(t) = t(1 − τ) and the duration formula is again identical to Early in the project, where t ≪ Tp and t ≪ Tn, the exponents
that in the linear case, Eq. (8). Therefore, for f = g, the linear in the exponentials are small, giving:
duration formula is valid for the project's entire life.
We now analyze the case where f ≠ g, which means that the T 2p T1
planned and earned value rates peak at different fractions through SPI→ ⇒ T 01 ¼ pffiffiffiffiffiffiffiffi ; ð31Þ
T 2n SPI
the project; see Fig. 3. In this case, Eq. (8) is no longer valid and
this can most easily be seen for the region 0 b t b f T1, where
where we have used the fact that Tp and Tn are proportional to
Ce(t) = Cp[t − δ(t)] gives:
T1 and T1′.
Πe t 2 Πp ðt−δðt ÞÞ2
¼ ; ð25Þ A.4. Defining the project end time
2gT 01 2f T 1
which gives, The PNR profile never reaches zero and, so, we need to
 pffiffiffiffiffiffiffiffi define the meaning of the project end time. We define the
δðt Þ ¼ t 1−τ f =g ; ð26Þ planned project end time, T1, as the time where the planned cost
has reached Cp(T1) = (1 − ϵ)Cp(∞), where Cp(∞) is the cost as
and the formula for the duration becomes: t → ∞, and ϵ is a fraction. Then, from the cumulative profile, we
sffiffiffi have at the project planned end time, T1,
T t f
T 01 ð f ; g Þ ¼
1
: ð27Þ  h  i
T e ðt Þ g
ð1−ϵÞC p ð∞Þ ¼ C p ð∞Þ 1− expð− T 21 = 2T 2p : ð32Þ
Therefore, when f ≠ g we have a different duration formula.
If one inappropriately used the linear duration formula, the Te(t) pffiffiffiffiffi to represent ϵ = exp (− k) and, simplifying, gives,
It is useful
term would now be a different function of time and generate a T 1 ¼ T p 2k. Therefore, the planned end point is proportional to
different value for the duration, T1′(est), with the resulting the peak time. Similarly, assuming the earned curve pffiffiffiffiffi ends at the
percent error (T1′(est) − T1′)/T1′ × 100. See Fig. 6. same fraction, ϵ = exp (− k), we have T 01 ¼ T n 2k . Therefore,
Tp/Tn = T1/T1′ and δ(t) = t(1 − τ).
A.3. The PNR profile Since ϵ does not appear in the expression for δ(t), the duration
formula does not depend on the selection of the arbitrary end
Using Eq. (5) for δ(t) for the PNR profiles, we have: point, ϵ. Further, the result is exact as long as the end points of the
   " !# planned and earned curves are set to the same value for ϵ, which
0
t2 ðt−δðt ÞÞ2 is reasonable.
C e T 1 1− exp − 2 ¼ C p ðT 1 Þ 1− exp − ;
2T n 2T 2p As a specific example, suppose the planned project end were
defined as 0.99Cp(∞), so that ϵ = 0.01, k = − ln ϵ = 4.605, and
pffiffiffiffiffi
ð28Þ
T 1 ¼ T p 2k ¼ 3:03T p . Therefore, a useful approximation for
which gives δ(t) = ± t(1 − Tp/Tn). This formula differs slightly the PNR project duration is ~ 3 times the peak time, and this
from the linear case in that δ(t) is now in terms of the peak applies to both the planned and earned curves.
1504 R.D.H. Warburton, D.F. Cioffi / International Journal of Project Management 34 (2016) 1493–1504

A.5. The Cioffi profile Elshaer, R., 2013. Impact of sensitivity information on the prediction of
project's duration using earned schedule method. Int. J. Proj. Manag. 31 (4),
579–588.
We set γp = exp (αTp) for the planned curve and γn = exp (αTn) Evensmo, J., Karlsen, J.T., 2006. Earned value based forecasts-some pitfalls.
for the delayed, earned value curve. The derivative of the planned AACE Int. Trans. EV6, 1–5.
curve, y_ p ðtÞ, peaks at Tp and the derivative of the earned value Gallagher, M.A., Lee, D.A., 1996. Final-cost estimates for research and development
curve, y_ e ðtÞ, peaks at Tn N Tp. programs conditioned on realized costs. Mil. Oper. Res. 2 (2), 51–65.
Hofstede, G., 1978. The poverty of management control philosophy. Acad.
The definition of the delay, δ(t), is, ye(t) = yp(t − δ), which
Manag. Rev. 3 (3), 450–461.
gives. Jacob, D.S., Kane, M., 2004. Forecasting Schedule Completion Using Earned
Value Metrics Revisited. The Measurable News Vol. 1(11) Summer.
1− expð−αt Þ 1− expð−αðt−δðt ÞÞÞ Johnston, R.B., Brennan, M., 1996. Planning or organizing: the implications of
¼

: ð33Þ
1 þ expð−αðt−T n ÞÞ 1 þ exp −α t−δðt Þ−T p theories of activity for management of operations. OMEGA Int. J. Manag.
Sci. 24 (4), 367–384.
Khamooshi, H., Golafshani, H., 2014. EDM: earned duration management, a
After some straightforward algebra, we find: new approach to schedule performance management and measurement. Int.
J. Proj. Manag. 32 (6), 1019–1041.
2   3
−αt −αðt−δðt Þ−T p Þ Kim, E.H., 2000. A Study of the Effective Implementation of Earned Value
1 4ð1−e Þ 1 þ e Management Methodology PhD thesis The George Washington University,
T n ¼ t þ ln −15: ð34Þ
α 1−e−αðt−δðtÞÞ Washington, D.C.
Koskela, L., Howell, G., 2002. The Underlying Theory of Project Management
Is Obsolete. Proceedings of the PMI Research Conference. PMI, Seattle,
The values for Tp and α are determined from a fit to the planned Washington, USA, pp. 292–302 June.
Lee, D., 2002. Norden–Raleigh Analysis: A Useful Tool for EVM in Development
value curve—see Cioffi (2006b). As the project proceeds, the Projects. The Measurable News, pp. 21–24 March.
earned value curve evolves and, at any time, t, the delay, δ(t), can Lipke, W., 2003. Schedule is different. The Measurable News, pp. 31–34
be found. The peak in the earned duration curve, Tn, is then Summer.
determined from Eq. (34) in terms of the known parameters, δ(t), Lipke, W., 2010. Earned Schedule. Lulu (R) Publishing, Lexington, KY.
Tp, and α. The final duration, T1′, is then determined using the Lipke, W., Zwikael, O., K., H., Anbari, F., 2009. Prediction of project outcome.
The application of statistical methods to earned value management and
process described in Appendix A.4, but applied to the Cioffi earned schedule performance indexes. Int. J. Proj. Manag. 27, 400–407.
curve. Section 5.3 illustrates the duration estimation process for Marshall, R.A., 2006. The Contribution of Earned Value Management to
the OED project. Project Success on Contracted Efforts: A Quantitative Statistics Approach
within the Population of Experienced Pratitioners PhD thesis Lille Graduate
References School of Management.
McGarry, F., Pajerski, R., Page, G., Waligora, S., Basili, V., Zelkowitz, M.,
1994. Software Process Improvement in the NASA Software Engineering
Anbari, F.T., 2003. Earned value project management: method and extensions.
Laboratory. Technical Report, CMU/SEI-94-TR-22, ESC-TR-94-022.
Proj. Manag. J. 34 (4), 12.
Narbaev, T., De Marco, A., 2013. An earned schedule-based regression model
Basili, V.R., Beane, J., 1981. Can the Parr curve help with manpower
to improve cost estimate at completion. Int. J. Proj. Manag. 32, 1007–1018.
distribution and resource estimation problems? J. Syst. Softw. 2, 59–69.
Project Management Institute, 2011. Practice Standard for Earned Value
Batseliera, J., Vanhoucke, M., 2015. Evaluation of deterministic state-of-the-art
Management. 2nd edition. Project Management Institute, Newtown Square, PA.
forecasting approaches for project duration based on earned value management.
Project Management Institute, 2013. A Guide to the Project Management Body
Int. J. Proj. Manag. 33, 1588–1596.
of Knowledge (PMBOK (R) Guide). 5th edition. Project Management
Book, S., 2003. Issues Associated with Basing Decisions on Schedule Variance in
Institute, Newtown Square, PA.
an Earned Value Management System. National Estimator, pp. 11–15 Fall.
Putnam, L.H., 1978. A general empirical solution to the macro software sizing
Book, S., 2006. “Earned Schedule” and its Possible Unreliability as an Indicator.
and estimating problem. IEEE Trans. Softw. Eng. 4 (4), 345–361.
The Measurable News, pp. 24–30 Spring.
Singh, S., Lakanathan, G., 1992. Computer-Based Cash Flow Model. Proceedings
Chen, H.L., Chen, W.T., Lin, Y.L., 2016. Earned value project management:
of the 36th Annual Transactions of the American Association of Cost Engineers,
improving the predictive power of planned value. Int. J. Proj. Manag. 34,
pp. R.5.1–R.5.14.
22–29.
Stratton, R.W., 2007. Applying Earned Schedule Analysis to EVM Data for
Christensen, D.S., 1993. The estimate at completion problem: a review of three
Estimating Completion Date. The Association for Advancement of Cost
studies. Proj. Manag. J. 24, 37–42.
Engineering Proceedings, pp. 41–45.
Christian, J., Kallouris, G., 1991. An expert system for predicting the cost–time
Vanhoucke, M., 2012. Measuring the efficiency of project control using
profiles of building activities. Can. J. Civ. Eng. 18, 814.
fictitious and empirical project data. Int. J. Proj. Manag. 30 (2), 252–263.
Cioffi, D.F., 2005. A tool for managing projects: an analytic parameterization of
Vanhoucke, M., Vandevoorde, S., 2006. A comparison of different project
the S-curve. Int. J. Proj. Manag. 23, 215–222.
forecasting methods using earned value metrics. Int. J. Proj. Manag. 24,
Cioffi, D.F., 2006a. Designing project management: a scientific notation and an
289–302.
improved formalism for earned value calculations. Int. J. Proj. Manag. 24,
Vanhoucke, M., Vandevoorde, S., 2007. A simulation and evaluation of earned
136–144.
value metrics to forecast the project duration. J. Oper. Res. Soc. 58, 1361–1374.
Cioffi, D.F., 2006b. Subject Expertise, Management Effectiveness, and the
Warburton, R.D.H., 1983. Managing and predicting the costs of real-time
Newness of a Project: the Creation of the Oxford English Dictionary.
software. IEEE Trans. Softw. Eng. SE-9 (5), 562–569.
Project Management Institute Research Conference. Project Management
Warburton, R.D.H., 2011. A time-dependent earned value model for software
Institute, Montreal, Canada July.
projects. Int. J. Proj. Manag. 29, 1082–1090.
Davis, D., Christle, G., Abba, W., 2009. Using the Rayleigh Model to Assess Future
Warburton, R.D.H., 2014. A new schedule estimation technique for construc-
Acquisition Contract Performance and Overall Contract Risk (Volume I).
tion projects. Organ. Technol. Manag. Constr. Int. J. 6 (3), 1075–1082.
Technical Report. CNA.