Available online at www.sciencedirect.

com

International Journal of Project Management 29 (2011) 1082 – 1090

www.elsevier.com/locate/ijproman

A time-dependent earned value model for software projects

Roger D.H. Warburton ⁎
Metropolitan College, Boston University Boston, MA 02215, USA

Received 23 August 2010; received in revised form 16 February 2011; accepted 17 February 2011

Abstract

This paper proposes a formal method for including time dependence into Earned Value (EV) management. The model requires three
parameters, which map directly to the fundamental “triple constraint” of scope, cost, and schedule: the reject rate of activities, the cost overrun
parameter, and the time to repair the rejected activities. Time dependent expressions for the planned value, earned value, and actual cost are
derived, along with the cost performance index (CPI) and schedule performance index (SPI). The model is built on the well-established Putnam–
Norden–Rayleigh (PNR) labor rate profile, which is a useful representation for large software projects. We apply the model to a well-known
software dataset, demonstrating how to estimate the project's final cost, which converges faster to the correct answer with less variability than
standard Estimate-at-Completion (EAC) calculations. The model also accurately predicts the required revised labor profile and the new schedule.
© 2011 Elsevier Ltd. and IPMA. All rights reserved.

Keywords: Earned Value Management; Cost estimation; CPI; SPI

1. Introduction (e.g., dollars or labor hours). The general method was

established by Cioffi (2006a), who showed how to convert,
The Practice Standard for Earned Value Management for example, attorney minutes or bulldozer days into the
(EVM) (Project Management Institute, 2005a) provides insight common unit of dollars.
into and explanation of the basic elements of EVM. Marshall EVM provides project managers with triggers or early
(2006) suggests that EVM is effective in providing significant warning signals of project trouble, which were found to be
predictors of project status. EVM metrics were also shown to be reliable as early as 15% into a project (Christensen and Heise,
important contributors to the successful administration of 1992). Data from completed Air Force contracts established that
contracts (Marshall et al., 2008). the cost performance index (CPI) did not change by more than
On the other hand, Kim (2000) found that few projects were 10% once the contract had reached the 20% completion point,
using EVM, and the main reasons were that it was “not needed regardless of the type or phase of the defense contract, weapon
on small projects” and it was “hard to apply.” Kim also pointed system, or the military service involved (Christensen and Payne,
out that computer tools and training significantly improved the 1991).
acceptance and performance of EVM, and that the literature Therefore, a significant overrun which continues more than
suffers from an “over reliance on anecdotal data.” 20% into the project indicates that the project is unlikely to meet
The key concept is the Earned Value (EV), which converts the budgetary goals, and customers can reliably conclude the
project accomplishments from physical units of measure (e.g., project is in trouble. Christensen et al. (1995) clearly established
miles of roadway or deliverables completed) to financial units that performance measurement data (such as the CPI) have
predictive value. Christensen and Payne (1991) added that if
anything, the CPI tended to decline, and so using the CPI
actually provides a lower bound for the final cost estimate.
⁎ Tel.: +1 617 358 3583; fax: +1 617 353 6840. Somewhat overlooked is what we will refer to as
E-mail address: [email protected]. “instantaneous” values for the CPI. For example, less than
0263-7863/$ - see front matter © 2011 Elsevier Ltd. and IPMA. All rights reserved.
doi:10.1016/j.ijproman.2011.02.008
 R.D.H. Warburton / International Journal of Project Management 29 (2011) 1082–1090 1083

10% of contracts have 3-month stable CPI values. That means schedule. In the literature, this issue has been addressed in two
almost all measured values of the CPI, continually recomputed ways:
over short, 3-month intervals, were found to change (were
unstable). Less than one third of projects have stable 6-month 1. Method #1: Converting SV into time units (Anbari, 2003).
CPIs. Christensen (1992) established that despite the variability, One first calculates an average of the actual costs spent per
the continually updated 3-month averages provided the most time-period, called the spend rate (AC Rate), and the average
reliable estimate of the final cost. planned value per time-period, called the planned value rate,
These ideas suggest that the instantaneous CPI changes (PV Rate). PV Rate is defined as the baseline budget at
overtime, and as Christensen and Payne (1991) observed, only completion (BAC) divided by the baseline schedule at
stabilizes because of its cumulative nature. Here we should completion (SAC), which converts SV into time units, and is
carefully distinguish two types of variation in the CPI: 1) referred to as the time variance (TV).
Statistical fluctuations in the data, i.e., the inherent uncertainty There are problems with this method: e.g., PV Rate is
and variation in project data. 2) The existence of a functional assumed constant over the life of the project, and yet the
time dependence. It is this latter variation, which is rarely labor profile typically follows an ‘S’ curve.
discussed, that we will analyze in this paper. 2. Method #2: Measuring the time delay between the earned
A functional time dependence is important because the CPI value and the planned value cost curves (Fleming and
is used to compute the estimate at completion, EAC, and a Koppelman, 2005). This is referred to as the Schedule
changing CPI implies a changing EAC. We suggest, however, Variance Method, SVM. A quantity called the “Earned
that if the EAC is to be a useful concept, it should not change Schedule” is determined by drawing a horizontal line from
overtime. Customers want to know the final budget, and might the EV curve backward (or forward) to the PV curve, and
be understandably upset at a continually changing estimate of interpreting the distance on the horizontal time axis as a
the final project cost. We will demonstrate that there are several measure of the schedule delay (or acceleration). This method
reasonable time-dependent shapes for the CPI, but that the has the advantage of defining the schedule delay in time units.
resulting EAC is in fact a constant. However, there are problems with this method also. The
In general, methods for calculating the EAC depend on the dimensionless ratio in the formula for ES presents an algebraic
assumptions made about the future performance of the project difficulty as pointed out by Book (2006).
versus the historical, established performance to date. The
Project Management Institute (2004) provides three Vanhoucke and Vandevoorde (2007) extensively reviewed
approaches, based on three different sets of assumptions: (1) the accuracy with which the above methods forecast the total
when the original estimates are flawed; (2) when past project duration, and concluded that SVM generally outperforms
performance is not a good predictor of future performance; other forecasting methods. This is not altogether surprising
and, (3) when past performance is a good predictor of future because SVM is an instantaneous metric, and one continually re-
performance. estimates the change in schedule based on the project data to date.
There are a number of issues in the theory of EVM. For This is in agreement with Christensen (1992), who established
example, as pointed out by Cioffi (2006a), the schedule that the continually updated 3-month averages provided the most
variance, SV, is not really a variance, which has statistical reliable estimate of the final cost. This is in contrast to Method #1,
implications, but is really a schedule “difference.” Further, SV which is defined in terms of global quantities, which are assumed
is in dollars (not weeks or months), strange units for a schedule constant (BAC and SAC). However, neither approach defines how
quantity. Cioffi (2006b) also proposed a revised formalism for the parameters should evolve over time, e.g., how does the Earned
EVM in an attempt to address the problem that “the historically Schedule evolve over time?
arcane terminology and calculational notation have stood as Vanhoucke and Vandevoorde (2007) concluded that graphs
road blocks to its embrace by the management community.” of CPI and SPI over time provide valuable information about
Another criticism of EVM is that schedule performance trends in project performance. When corrective managerial
index, SPI, is inherently a function of time, but the form of the actions are implemented, the changes in the behavior of the
time dependence is unknown. This can be most easily be seen indexes are assumed to reflect the impact of management
by examining the behavior of SPI towards the end of the project. actions. However, since SPI(t) always → 1 at the end of the
As the last few activities are completed, the earned value project, management cannot claim any credit for the improve-
approaches the planned value, i.e., EV → PV, and therefore, ment in SPI. Therefore, how does one determine if management
SPI → EV/PV → 1 (Kerzner, 2006; Lipke, 2003). This is true actions actually changed the SPI?
even if the project is late, in which case the SPI → 1 after the We will also make use of labor rate profiles, an approach that
planned completion date. A reasonable question, therefore, is: was pioneered by Putnam (1978). The Putnam–Norden–
How and when does SPI → 1. Rayleigh curve, now known as PNR, describes the labor rate
A goal of this paper is to improve the theory of EVM by over time for software projects. The PNR curve is well-
including time dependence into the definitions of all quantities. established and has significant realism. Also, the mathematics is
By knowing the time dependence of CPI and SPI, and tractable, without too much complexity. The PNR curve does
measuring the mat defined points in time, we show that it is not work so well for projects that are incrementally developed
possible to estimate precisely the project's final cost and (Conte et al., 1986).
1084 R.D.H. Warburton / International Journal of Project Management 29 (2011) 1082–1090

Cioffi (2005) proposed that the solution to a differential 2. A formal model for EVM
equation frequently used in ecology can successfully
reproduce project labor ‘S’ curves. Unfortunately, the In this paper, we will explicitly indicate that the CPI and SPI
parameters characterizing the curve are not obviously related are functions of time, so that their definitions become:
to projects: a shape parameter, and the mid-point slope of the
EV ðt Þ EV ðt Þ
curve. CPI ðt Þ = SPI ðt Þ = : ð1Þ
Parr (1980) had already analyzed a similar family of curves, AC ðt Þ PV ðt Þ
which bear his name. Parr described a software project as
Parr (1980) first derived the analytical formula for the PNR
starting out with a fixed number of problems to be solved, and
curve, as follows: A project involves solving some fixed set of
ended all problems were solved. Parr's problems (nodes) have
problems, and W(t) denotes the proportion of the problems
tree-diagram dependencies: Some nodes are leaves of the tree,
already solved at time, t. W(t) is normalized to 1:0 at the end of
and some have dependent nodes, which cannot be started until
the project, and the completion of the project is simply the
the parent nodes are completed.
exhaustion of the problem space. The “skill” available for
Parr was unaware of the ideas of modern project manage-
solving problems is denoted as p(t), and Parr proposed that the
ment, he was attempting to describe a 1980s software
rate at which project progress is made should be proportional
development. Parr's work could have wider relevance to the
jointly to the skill and the number of tasks left to be completed:
underlying theory of projects if we re-interpret his nodes as
activities to be completed with finish-to-start constraints. This dW ðt Þ
may throw light on the question of why, if projects are unique, = pðt Þd ð1−W ðt ÞÞ: ð2Þ
dt
do they have a common underlying mathematical structure?
More commonly, why do supposedly unique software projects It is then argued that the choice that best fits the observations
have a common labor profile? from actual projects is a linear learning, p(t) = at. Parr's
While we are taught that projects are unique, what project assumption of a linear learning rate was somewhat arbitrary,
staff do in terms of solving problems (in Parr's terminology) since learning is conventionally described by a power law
might lead to the characterization of projects in more (Meredith and Mantell, 2005). Nevertheless, the PNR curve was
fundamental terms: Project staff complete activities, which practically useful, and organizations adopted it to predict
open up other activities to be worked on. This is an exciting software costs and schedules. With the assumption of linear
idea, suggesting that some fundamental mathematics might learning, 2 can be rearranged to give:
apply more widely to projects, even possibly beyond software. 

dW ðt Þ at 2
However, to establish how these ideas apply, and to which = atdt W ðt Þ = 1−exp − : ð3Þ
1−W ðt Þ 2
classes of projects, requires extensive research and a much
longer discussion. One then assumes management applies personnel resources
McGarry et al. (1994) carefully analyzed the PNR and Parr (labor) proportional to the number of problems available to be
labor rate curves, and concluded that while the Parr curve worked on. The application of too little labor will not complete
often fits the raw data better, noise in the data makes it the available work, and too much results in excess staff. Thus,
difficult to choose between the models. They also noted that management applies labor proportional to the available work
the Parr curve requires four parameters to PNR's two, so it is rate, dW(t) = dt, which results in the well established PNR labor
not surprising that it is a better fit. Unfortunately, the rate curve:
parameters in the Parr curve are not easily determined, and
so the curve is not very effective for resource estimation and dW ðt Þ  
PNRðt Þ = = at exp −at 2 = 2 : ð4Þ
tracking. This may help to explain why the Parr curve fell out dt
of favor. When plotted cumulatively over time, all of the
Without loss of generality, we can define the total labor as
above-mentioned labor rate profiles result in the typical ‘S’
the total number of activities, N. Also, the time of the labor
curve. However, it has been pointed out that using
peak, T, is easily seen to be, a = 1 = T2. Both model parameters
instantaneous labor rate curves is frequently more useful
then describe useful project parameters, and are easily
than using the cumulative ‘S’ curves (Cioffi, 2005).
determined in practice. This results in the more familiar form
Management actions are often reflected in the labor rate
of the PNR curve:
curve as clear departures from smoothness, while the
 
cumulative ‘S’ curve washes them out (Warburton, 1983). Nt t2
The rest of the paper is as follows. We next present the PNRðt Þ = 2 exp − 2 : ð5Þ
T 2T
Time-Dependent EV model based on the PNR labor curve.
The time dependent cost and schedule performance indexes,
CPI(t) and SPI(t), are calculated. We examine the model's 2.1. Planned value
practical applicability by estimating the final cost and
schedule for a software project with real-world data. Finally, When a project is planned, the time-phased budget is
we provide some conclusions and project management developed by summing the time-phased labor contributions of
implications. the scheduled activities, which is the labor rate curve. If the project
 R.D.H. Warburton / International Journal of Project Management 29 (2011) 1082–1090 1085

is in the planning phase, then the rate of completion of activities is

simply the planned labor over time, pv(t). If necessary, we can use
the approach detailed in Cioffi (2006b) to convert back and forth
between completed activities measured in labor hours and the
number of completed activities. Thus, the instantaneous planned
value is the work rate, i.e., the PNR curve:
 
Nt t2
pvðt Þ = PNRðt Þ = exp − : ð6Þ
T2 2T 2

The conventional definition of planned value, PV(t), is the

cumulative sum of the instantaneous planned value, pv(t):
Z t
 
t2
PV ðt Þ = pvðsÞds = N 1−exp − 2 : ð7Þ
0 2T
Fig. 2. Cumulative planned value, earned value, and actual cost for the PNR
Throughout this paper, we will use lower case for instantaneous labor profile.
quantities and upper case for cumulative quantities. The
instantaneous planned value is plotted in Fig. 1 as the solid rejection rate, we have been unable to find reliable data on the
curve. The cumulative planned value is plotted in Fig. 2, and shows average delay experienced by individual activities. Therefore,
the typical S shape. the reasonableness of this assumption needs to be evaluated by
comparing real project data with the predictions of the model.
2.2. Earned Value An example of the process is as follows: Suppose that at the
end of week 2 (t = 2), 5 activities were due to be completed, but
We now assume that a constant fraction, r, of the activities that one was rejected for some reason (r = 20%). This activity will be
were supposed to be completed at time, t, were rejected for some delayed for rework, and take an extra 3 weeks to be completed
reason and require extra work. Project managers should be able to (τ = 3). The rejected activity will be completed in week 5, when
estimate the reject rate early in the project as the initial activities are it finally earns its planned value.
completed, and some are rejected. We assume that the reject rate is In each interval, a fraction, r, of the activities were rejected, but
constant over the life of the project, which appears reasonable. For the other activities, (1 − r), were completed and earned value. At
example, software project data suggests that error rates remain the beginning of the project, in the interval t b τ, the earned value is
constant over the life of a project (McGarry et al., 1994). In a simply the fraction of activities that were successfully completed.
fascinating example of project longevity, Cioffi (2006c) analyzed For t N τ, the earned value has two contributions: the fraction of
the development of the Oxford English Dictionary, which activities successfully completed at t, and those from t − τ that
appeared to follow the same ‘S’ curve for seven decades! were delayed and are now complete:
The completion of rejected activities will be delayed by a 
time, τ, which is assumed to be constant. That is, on average the ev1 ðt Þ = ð1−rÞ:pvðt Þ t≤τ
evðt Þ = : ð8Þ
rejected activities all take the same time to repair. Unlike the ev2 ðt Þ = ð1−rÞ:pvðt Þ + r:pvðt−τÞ t N τ

We can now calculate the cumulative earned value by

integrating the instantaneous values.
8
>
> EV1 ðtÞ = ∫t0 ev1 ðsÞds
 2 
>
>
>
> t
>
> = N ð1−rÞ 1−exp t≤τ
>
> 2T 2
>
>
>
> EV ð t Þ = EV ð τ Þ + ∫ t
ev ð s Þ
< 2 1 τ 2 
EV ðtÞ = t2 ð9Þ
>
> = N −N ð1−rÞexp
>
> 2T 2
>
>
>
> !
>
>
>
> ðt−τÞ2
>
> −rN exp − tNτ
: 2T 2
We note that for large t,

EV ðt→∞Þ→N ; ð10Þ
which merely says that at the end of the project, all of the
Fig. 1. Instantaneous planned value, earned value, and actual cost for the PNR activities were completed. However, the final earned value only
labor profile. becomes equal to the final planned value after the planned end
1086 R.D.H. Warburton / International Journal of Project Management 29 (2011) 1082–1090

of the project. This can be seen in Fig. 1, where the earned value
curve is delayed relative to the planned value curve. The
cumulative versions are shown in Fig. 2, where both the planned
and earned value curves approach the same value because the
total number of activities in the project remains the same.
If extra activities had been added (i.e., scope creep occurred),
then the earned value would approach a higher value than the
planned value. This assumption is not terribly restrictive,
because the project manager should know quite early on if the
project is increasing in size. If so, the project manager simply re-
estimates the total labor, N, and possibly adjusts the schedule.
The important point is that the project manager can analyze
potential options from a well-defined, validated model.

2.3. Actual cost

We assume that initially, the labor was expended according

to the plan, but that some activities were not completed when Fig. 3. The actual cost, AC(t), for increasing values of the cost over-run
parameter, c.
the planned labor was used up. Therefore, for t b τ, the actual
cost is assumed equal to the planned value. The costs to repair
the rejected activities will be incurred in the future. For the The cumulative actual cost as a function of time is shown in
interval t N τ, the instantaneous actual cost, ac(t), includes the Fig. 2. The actual and planned costs are identical up to time t = τ.
work performed on both the activities that were successfully After that, the actual cost increases as the cost of the rejected
completed at t, and those that were rejected at t − τ and are now activities accumulates. The planned and earned values do not
complete. The fractional extra work required is defined as c, so depend on the cost overrun parameter, only the actual costs
the instantaneous actual cost is: depend on c (Fig. 3).
8 The Total Actual Cost of the project is predicted to be:
>
> ac1 ðt Þ = pvðt Þ t≤τ
>
> ð Þ ð Þ ð Þ
>
>
ac 2 t = pv t
+ r c
dd pv t−τ
 Total Actual Cost = AC ðt→∞Þ = N ð1 + rcÞ: ð13Þ
>
< N t2
acðt Þ = = 2 texp − 2 ð11Þ
> T 2T !
>
> Once we determine the parameters, r and c, we can estimate
>
> ðt−τÞ 2
>
> + rcðt−τÞexp − t N τ: the final total cost of the project. We note that this estimate is a
: 2T 2 constant because it only depends on the fundamental project
parameters, which are constants. Therefore, it is possible to
The term r·pv(t − τ) represents the fraction of activities estimate the final cost from an early determination of the
rejected previously, and c accounts for the additional repair parameters, r and c.
work (c = 1·0 represents a 100% overrun). The instantaneous
actual costs are shown in Fig. 1. 3. Cost and schedule performance indices
An example of the process is as follows: At week 2 (t = 2),
five activities were due to be completed. One of these activities We now have the functions necessary to calculate CPI(t) and
was rejected for some reason (r = 20%). This activity is delayed SPI(t). Fig. 4 shows the shape of the CPI(t) and SPI(t) curves
for rework and takes an extra 3 weeks to complete; it is when r and τ are constant and the cost overrun parameter, c,
successfully completed in week 5 (τ = 3). The actual cost for varies. It is clear that slightly different shapes for CPI(t) are
week 5 includes two contributions: The planned labor amount possible.
for week 5; and the activity rejected in week 2 incurs extra CPI(t) immediately falls to (1 − r) because in each interval
repair costs in week 5, consuming an extra 25% (c = 0·25). some activities are rejected. The rejected activities incur repair
costs, and are completed after a time, τ. The changing costs
8 change the CPI(t). For small values of c, the CPI rises, while for
>
> AC1 ðtÞ = ∫t0 ac

1 ðsÞds 
larger values, it declines.
>
>
>
> t2 t≤τ SPI(t) also falls initially to (1 − r), but then slowly climbs
>
> = N 1−exp −
>
> 2T 2 back to 1.0 at the end of the project, as expected. In fact, it is
>
>
< AC2 ðt Þ = AC1 ðτÞ + ∫t ac2 ðsÞds easy to show that SPI(t → ∞) → 1 is an explicit property of this
AC ðtÞ =
τ 2  ð12Þ
> t model. SPI(t) depends only on the repair delay, τ, and the reject
>
> = N 1−exp − 2
>
> 2T tNτ
rate, r.
>
> " !#
>
> ð Þ 2 Fig. 5 shows the shape of the SPI(t) curve when r changes,
>
> t−τ
>
: + Nrc 1−exp − : holding c and τ constant. When the reject rate, r, changes, the
2T 2
general characteristics of the SPI(t) curves remain the same.
 R.D.H. Warburton / International Journal of Project Management 29 (2011) 1082–1090 1087

Table 1
Initial and final values for the CPI and SPI, and the estimate of the final cost.
Parameter t→0 t→∞
1
CPI 1−r
1 + rc
SPI 1−r 1
Final cost N N (1 + rc)

SPI to its projected performance. Positive managerial actions

should result in improvements to the fundamental parameters,
i.e., a lowered reject rate, a reduced cost overrun, and a
decreased time to repair. Genuine parameter improvements will
be reflected in detectable changes in the shapes of the CPI and
SPI curves. Smooth CPI and SPI curves are a signal that the
underlying project parameters have changed very little.
The initial and final values of CPI, SPI, and the final
estimated cost are summarized in Table 1.
Fig. 4. The time dependence of SPI and CPI for varying values of the parameter,
c, for the PNR labor model. r = 0.25, delay parameter, τ = 10.
4. Analysis of a software project

We now show how the above theory works in practice. We

When r is decreased, the curves move closer together; and as r use historical data from a well-known software project—see
increases, the curves move farther apart. When the reject rate, r, Fig. 6 (Warburton, 1983). The goal is to determine whether
changes, the general characteristics of the SPI(t) curves remain early in the project life cycle we can accurately determine the
the same. When r is decreased, the curves move closer together; model parameters and then estimate the cost overrun and
and as r increases, the curves move farther apart. We remind the schedule delay. We also wish to compare the answers to
reader that even though CPI(t) and SPI(t) are changing as the traditional methods.
project proceeds, the underlying project parameters are Figs. 7 and 8 show the early project data; weeks 1–20. Fig. 7
constant. presents the instantaneous data, while Fig. 8 presents the same
Many authors (e.g., Fleming and Koppelman (2005); data in cumulative form, which smoothes out the noise.
Vanhoucke and Vandevoorde (2007)) have observed that However, in both cases it is apparent that the project has fallen
since SPI → 1.0 at the end of the project, it is not useful as an behind the plan, which is shown by the dotted curve. The gray
indicator over the last third of the project. That conclusion boxes denote the actual labor hours recorded by project
assumes the time dependent behavior of SPI(t) is unknown. engineers on their weekly time-sheets.
With this model in hand, we understand how SPI(t) changes The black boxes denote the earned value–the completed
over time, and we can compare the actual performance of the deliverables–which was reconstructed after the project was
finished. However, we can use this earned value data as if it had
been accumulated in real-time as the project proceeded. Since

Fig. 5. The time dependence of SPI for varying values of the reject rate Fig. 6. Calibration project data: the planned labor (dotted line), actual cost (gray
parameter, r, for the PNR labor model. c = 0.33, delay parameter, τ = 10. line), and earned value (black line).
1088 R.D.H. Warburton / International Journal of Project Management 29 (2011) 1082–1090

Fig. 9. Estimating the final cost using EAC (gray) and the model (black).

Fig. 7. Early calibration project data (instantaneous): the planned labor (dotted
The first few modules were just 1–2 weeks late. However, after
line), actual cost (gray box), and earned value (black box). about 8–10 weeks, the delays became more significant and the
average delay settled in at around 10 weeks (τ = 10).

completed deliverables earn discrete quantities of labor hours, 5. Predicting the final cost
the earned value curve shows discrete vertical jumps.
Next we estimated the three model parameters. From the We now use the standard estimate at completion (EAC)
number of modules requiring rework, we estimated the reject formula (Project Management Institute, 2004) for the final
rate by simply recording the number of modules sent back for project cost:
rework. Somewhat surprisingly, almost all of the modules were
delivered late, and required extra work to complete satisfacto- EV ðTotal Þ−EV ðt Þ
EAC = AC ðt Þ + : ð14Þ
rily, resulting in a reject rate of 95% (r = 0.95). CPI ðt Þ
We also determined the amount of extra work required to
complete the rejected modules, which gave an estimate of the The EAC uses the actual cost to date, AC(t), plus the
cost overrun parameter, which was fairly consistent at 30% remaining earned value divided by the CPI. This assumes that
(c = 0:30). That is, on average, almost all modules required an progress in the future will be the same as the historical progress.
extra 30% to complete. Finally, we estimated the average time The EAC prediction over time is shown in Fig. 9, along with the
delay associated with the completion of the rejected activities. prediction of the model for the first 30 weeks of the project.
Both estimates converge quite quickly to the actual final cost
after about 20 weeks, or 15% of the way through the project.

Fig. 10. A comparison the predictions of the model (smooth curves) with the
Fig. 8. Early calibration project data (cumulative): the planned labor (dotted actual project data. The plan (dotted), actual cost (gray), and earned value
line), actual cost (gray box), and earned value (black box). (black).
 R.D.H. Warburton / International Journal of Project Management 29 (2011) 1082–1090 1089

This is consistent with the findings of Christensen and Heise terms, e.g., pv(t − τ). More research is required to determine if
(1992), who found that EAC was a decent predictor of the final the model can be reformulated to include schedule accelera-
cost after about 20% of the project, but only for projects for tions. We have been unable to obtain project data on the average
which the CPI was constant. The predictions of the model delay experienced by individual activities, so another topic for
appear to have less scatter, and converge faster to the true final future research is to evaluate the reasonableness of the constant
cost. In fact, by week 10, the model's cost prediction is within τ assumption by calibrating the model, and determining its
about 5% of the correct answer. range of applicability.
For projects with a declining CPI, the EAC formula merely By analyzing early project data, we demonstrated that project
provides a lower bound for the cost (Christensen and Payne, managers could easily estimate the values of the model's three
1991). This model improves on this by giving a constant parameters. Predictions of the final cost and schedule were then
estimate for the EAC, even though the CPI and SPI are time- available, and allowing for noise in the data, these predictions
dependent. were constant over the life of the project, an inherently useful
The revised labor curve that results from the model is shown property of a prediction. We conclude that the three parameters
in Fig. 10. The dotted line shows the plan. The model predicts are in some sense fundamental, while quantities such as CPI and
quite well the shape of the revised labor curve (gray). The SPI are shown to be derived functions of time. The final cost
smooth curve is the revised labor prediction from the model, prediction from the model appears to converge to the true value
while the jagged curve (gray) is the actual labor. The smooth faster than standard EAC calculations. Also, the model predicts
black line is the model's predicted earned value, which is a the new schedule and a revised labor profile, which can be used
reasonable representation of the actual earned value (jagged for staff planning.
black line). The model predicts the time dependence of the SPI(t), and
Therefore, the model is useful in determining future labor further, the property that SPI(t) → 1 is an explicit property of the
needs. Furthermore, the revised staffing curve predicts the new model. Project managers can measure SPI(t) over time and
schedule for the project. While the EAC formula estimates the distinguish between inherent changes (those due to the time
final cost, it predicts neither the revised staff profile nor the new dependence of SPI(t)), and genuine changes in project
schedule. In summary, it appears that the predictions of the performance, which are a result of significant managerial
model are more reliable and more practically useful than the actions. Here, “significant actions” are those that actually
EAC. reduce the reject rate, the cost overrun rate, and the time to
repair. Without such significant actions, the project will simply
6. Conclusions follow its established path.
We used the PNR labor profile, but the model does not
A major challenge in project management is balancing the depend on that profile. In the earned value, any labor profile
“triple constraint” of scope, cost, and schedule. We presented a could be used for pv(t):
model with three parameters, which map directly to the 
fundamental triple constraint: the reject rate, r, is a measure of ð1−rÞd pvðt Þ t≤τ
evðt Þ = ð15Þ
the scope quality, as it characterizes the rate at which activities ð1−rÞd pvðt Þ + rd pvðt−τÞ t N τ:
do not meet their designed objectives, and must be reworked.
The cost overrun parameter, c, measures the cost to repair the The derivation of PNR assumes a linear learning rate. For
rejected activities, and so maps directly to the project cost software projects, this is not unreasonable as engineers become
constraint. The time delay parameter, τ describes the time more familiar with the project, the documentation, and the tools.
associated with repairing the rejected activities, and so maps Interesting future research questions include: Can the same form
directly to the project's schedule constraint. There are no be used for other types of projects? Which projects might also
extraneous or phenomenological parameters. follow such a form?
We assumed that the parameters are constant over the life of For many types of large projects the cumulative labor profile
the project. Industrial data strongly suggests that project error follows an ‘S’ curve (Christian and Kallouris, 1991). The
rates do indeed remain constant over time (McGarry et al., application of ‘S’ curves for cash flow projections can achieve
1994), which suggests that the assumption that r is constant is accuracies of over 90%, and the shape of the ‘S’ curve budget
reasonable. The Oxford English Dictionary appeared to follow versus time is a quick way to judge performance (Singh and
the same ‘S’ curve for decades (Cioffi, 2006c), suggesting that Lakanathan, 1992). This suggests another intriguing property of
its underlying project parameters were constant. Therefore, it the model: The behavior of the CPI and SPI curves over time for
seems a reasonable assumption that the model parameters are projects with an ‘S’ curve labor profile might be similar to those
indeed constant. shown in Fig. 4, and the model could therefore have wider
We assumed that the model parameters were all positive. The applicability. This claim should be evaluated by studying other
reject rate, r, is inherently positive. A positive value for c results labor profiles, such as construction, Cioffi curve, etc.
in a cost overrun, but without change the model will work just Another issue for future research is to compare the model
as well if c is negative, representing a cost under run. with the “timebased SPI(t)” calculation described in the
However, the model will not work as presented for schedule “Practice Standard for Earned Value Management” (Project
accelerations (negative values of τ) because of the time delay Management Institute, 2005b). Indeed, the current model
1090 R.D.H. Warburton / International Journal of Project Management 29 (2011) 1082–1090

provides theoretical support for the idea that “SPI(t) stabilizes Conte, S.D., Dunsmore, H., Shen, V.Y., 1986. Software Engineering Metrics and
15% into the project” and that one can estimate the end-date Models. Benjamin/Cummings Publishing Company, Inc., Menlo Park, CA.
Fleming, Q.W., Koppelman, J.M., 2005. Earned Value Project Management,
quite well. It would be interesting to compare the schedule slip 3rd Edition. Project Management Institute, Newtown Square, PA.
calculated using the time-based SPI(t) method with the model Kerzner, H., 2006. Project Management: A Systems Approach to Planning,
presented here. The ease of use of the two approaches could be Scheduling, and Controlling, 9th Edition. John Wiley & Sons, New York,
compared along with their relative accuracy. NY.
Kim, E. H., 2000. A study of the Effective Implementation of Earned Value
This model could provide a useful contribution to a project
Management Methodology. Ph.D. thesis, The George Washington Univer-
manager's toolbox by providing more reliable estimates of a sity, Washington, DC.
project's final cost overrun and schedule delay. Despite the Lipke, W., 2003. Schedule is different. The Measurable News, pp. 31–34.
preliminary nature of the model, a number of interesting Marshall, R. A., 2006. The contribution of Earned Value Management to project
features emerged. While much research remains to be done, it success on contracted efforts: A quantitative statistics approach within the
appears that this model is a useful starting point and shows population of experienced practitioners. Ph.D. thesis, Lille Graduate School
of Management.
promise. Marshall, R.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.
References 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.
Project Management Journal 34 (4), 12. Meredith, J.R., Mantell Jr., S.J., 2005. Project Management: A Managerial
Book, S., 2006. The Measurable News Spring. Approach, 6th Edition. John Wiley & Sons, Hoboken, NJ.
Christensen, D.S., 1992. Determining an accurate Estimate at Completion. Parr, F.N., 1980. An alternative to the Rayleigh curve model for software
National Contract Management Journal 25 (1), 17–25. development effort. IEEE Transactions on Software Engineering SE-6 (3),
Christensen, D.S., Heise, S.R., 1992. Cost performance index stability. National 291–296.
Contract Management Journal 25 (1), 7–15. Project Management Institute, 2004. A Guide to the Project Management Body
Christensen, D.S., Payne, K., 1991. Cost performance index stability: fact or of Knowledge (PMBOK (R) Guide), 3rd Edition. Project Management
fiction? Vol. Proceedings of the 1991 Acquisition Research Symposium. Institute, Newtown Square, PA.
Defense Systems Management College, Fort Belvoir, VA, pp. 257–266. June. Project Management Institute, 2005a. Practice Standard for Earned Value
Christensen, D.S., Antolini, R.C., McKinney, J., 1995. A review of EAC Management.
research. Journal of Cost Analysis and Management Spring 41–62. Project Management Institute, 2005b. Practice Standard for Earned Value
Christian, J., Kallouris, G., 1991. An expert system for predicting the costtime Management.
profiles of building activities. Canadian Journal of Civil Engineering 18, 814. Putnam, L.H., 1978. A general empirical solution to the macro software sizing
Cioffi, D.F., 2005. A tool for managing projects: an analytic parameterization of and estimating problem. IEEE Transactions on Software Engineering 4 (4),
the S-curve. International Journal of Project Management 23, 215–222. 345.
Cioffi, D.F., 2006a. Completing projects according to plans: an earned-value Singh, S., Lakanathan, G., 1992. Computer-based cash flow model. Proceedings
improvement index. The Journal of the Operational Research Society 57, of the 36th Annual Transactions of the American Association of Cost
290–295. Engineers R.5.1–R.5.14.
Cioffi, D.F., 2006b. Designing project management: a scientific notation and an Vanhoucke, M., Vandevoorde, S., 2007. A simulation and evaluation of Earned
improved formalism for earned value calculations. International Journal of Value Metrics to forecast the project duration. The Journal of the
Project Management 24, 136–144. Operational Research Society 58, 1361–1374.
Cioffi, D.F., 2006c. 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. Project software. IEEE Transactions on Software Engineering SE-9 (5), 562–569.
Management Institute Research Conference, Montreal. Project Management
Institute. July.