Available online at www.sciencedirect.

com

ScienceDirect
International Journal of Project Management 34 (2016) 22 – 29
www.elsevier.com/locate/ijproman

Earned value project management: Improving the predictive

power of planned value
Hong Long Chen a,1 , Wei Tong Chen b,⁎, Ying Lien Lin c,2
a
Department of Business and Management, National University of Tainan, No. 33, Sec. 2, Shu-Lin St., Tainan 700, Taiwan
b
Department of Construction Engineering, National Yunlin University of Science & Technology, No. 123, Section 3, University Rd., Douliu, Yunlin 640 Taiwan
c
Department of Industrial and Information Management, National Cheng Kung University, Tainan City 701, Taiwan

Received 5 January 2015; received in revised form 12 September 2015; accepted 14 September 2015
Available online 9 October 2015

Abstract

Earned value project management (EVPM) is an effective tool for managing project performance. However, most studies on extensions and
applications of EVPM concentrate on improving final cost and duration estimates rather than improving upon the use of planned value (PV) to
predict earned value (EV) and actual cost value (AC). This study proposes a straightforward modeling method for improving the predictive power
of PV before executing a project. By using this modeling method, this study develops EV and AC forecasting models for four case projects. Out-
of-sample forecasting validation using mean absolute percentage error (MAPE) demonstrates that the proposed method improves forecasting
accuracy by an average of 23.66% and 17.39%, respectively, for EV and AC. This improvement on PV's predictive power prior to project
execution provides management with more reliable predictive information about EV and AC performance, allowing for effective proactive action
to ensure favorable performance outcomes.
© 2015 Elsevier Ltd. APM and IPMA. All rights reserved.

Keywords: Earned value management; Planned value; Earned value; Actual cost; Prediction

1. Introduction purposes and assumptions about information availability.

Earned value project management (EVPM), a method that
Project management studies widely assume that predictions employs scope, cost, and schedule to measure and communi-
of project performance guide management to engineer critical cate the real physical process of a project (Vanhoucke and
issues scientifically, allowing for proactive project performance Vandevoorde, 2007), is the most commonly used of the project
management. Several researchers have proposed predictive models performance forecasting approaches reviewed. EVPM produces
and approaches to better estimate project performance (Brandon, variance and performance indices for project costs and schedules,
1998; Brown, 1985; Chen, 2013, 2014; Cioffi, 2005, 2006a, and thus predicts project costs and schedules at completion,
2006b; Farris et al., 2006; Keil et al., 2003). The level of detail in providing early indications of expected project performance
these models and approaches varies considerably with their various results. A broad consensus exists among researchers and
practitioners (Anbari, 2003, 2004; Christensen, 1998, 1999;
Kim, 2014; Lipke, 2003; Lipke et al., 2009; Narbaev and
Marco, 2014; Project Management Institute, 2013) about the
⁎ Corresponding author. usefulness of EVPM for monitoring and predicting project
E-mail addresses: [email protected] (H.L. Chen), performance, and EVPM has become an important component of
[email protected] (W.T. Chen), [email protected]
(Y.L. Lin).
successful project management (Marshall, 2006, 2007; Marshall
1
Tel.: +886 6 260 6123x7741; fax: + 886 6 260 6173. et al., 2008), with considerable research on the extensions and
2
Tel.: +886 6 331 3236. applications of EVPM being published.

http://dx.doi.org/10.1016/j.ijproman.2015.09.008
0263-7863/00/© 2015 Elsevier Ltd. APM and IPMA. All rights reserved.
 H.L. Chen et al. / International Journal of Project Management 34 (2016) 22–29 23

Despite the panoply of EVPM studies, prior studies (Anbari, Cioffi (2006b) provided EVPM with a novel formalism
2003, 2004; Cioffi, 2006a; Henderson, 2003; Lipke et al., 2009; derived from a scientific notation, allowing EVPM to extend
Vanhoucke and Vandevoorde, 2007) focus on enhancing the easily to predicting project cost (or earnings) and duration.
accuracy and reliability of predicting the final cost and duration Vandevoorde and Vanhoucke (2006) presented a generic forecast-
of a project. Relatively few address improving the predictive ing model based on the earned schedule (ES) method to predict
power of planned values, considered initial estimates of earned project duration, which is applicable in different situations,
values, and actual costs prior to project execution. As a result, including those involving a learning curve. Vanhoucke and
there appears to be a lack of models capable of enhancing the Vandevoorde (2007) further validated their generic forecasting
prediction accuracy of earned values and actual costs before model by performing extensive simulations on a large set of 3100
execution of projects. generated networks created with 30 project activities based on
The rest of the paper is organized as follows. Section 2 reviews nine simulation scenarios.
related EVPM studies, Section 3 presents the research question Subsequent work by Lipke et al. (2009) refined the EVPM and
and methodology, and Section 4 depicts the data, model building, ES methods to include the confidence limits of prediction, thereby
and validation. Section 5 presents the research summary and placing high and low bounds on forecasted costs (or earnings) and
conclusions. Section 6 describes research limitations and direction duration and thus improving project managers' abilities to make
for future research. informed decisions. Chou et al. (2010) developed EVPM into a
web-based visualized-implementation system, enabling managers
2. Background to monitor, evaluate, and estimate project financial and scheduling
performance by converting project data into manageable informa-
We first offer some definitions: planned value (PV) is the tion clusters. Hanna (2012) presented a case study to illustrate the
cumulative planned value as a continuous or periodic function use and applicability of EVPM in the electrical construction
summing the approved budget for accomplishing the scheduled industry. He concluded that early determination of probable project
work (activity, work package, or project) as a function of time. outcomes is possible with reasonable forecasting accuracy by
Earned value (EV) is the cumulative earned value either as a EVPM.
continuous or periodic sum of the approved budget for work Recently, Elshaer (2013) refined the EVPM and ES methods by
performed to date. Actual cost (AC) is the cumulative actual cost integrating activity-based sensitivity information into the earned
value either as a continuous or periodic sum of the actual cost of value calculations to remove and/or decrease false warning effects
work performed to date. PV, EV, and AC are respectively the caused by non-critical activities, and thus improved the forecasting
budgeted cost of work scheduled (BCWS), the budgeted cost of accuracy of project durations during project execution. Naeni et al.
work performed (BCWP), and the actual cost of work performed (2014) further developed EVPM into a fuzzy-based model using
(ACWP). linguistic terms and fuzzy theory, enabling management to
PV, EV, and AC are the fundamental metrics of the earned analyze, evaluate, and estimate project costs and scheduling
value project management (EVPM) that generates variance and performance under uncertainty during project execution.
performance indices for financial performance and schedules. More recently, Chen (2014) proposed a linear data-
EVPM forecasts project costs and schedules at completion, transformation formula and used data from 131 sample projects
providing early indications of expected project financial and to demonstrate that the formula significantly improves the
scheduling results based on current information. Aside from that, correlations between PV and EV and between PV and AC. He
the time-phased representation of PV is the performance- developed a mathematical model which he claimed can improve
management baseline of a project, PV is actually the predictor the prediction accuracy of EV and AC through further modeling
for both EV and AC. PV prior to the project's execution stage.
Specifically, PV predicts EV and AC prior to work performed Acebes et al. (2015) proposed a framework based on
at a given time in project. EV and AC are then measured EVPM, Monte Carlo simulation, and statistical learning
subsequent to work performed at that certain point of time. The techniques for project control under uncertainty. They used a
project manager then uses these measured values of EV and AC case project to demonstrate how to estimate the probability of
to evaluate project performance status. cost overruns and the expected project duration during project
Considerable research has been devoted to the extensions execution. They concluded that their framework can detect
and applications of EVPM for project cost performance and anomalies with regard to possible correlations between project
schedule control (Batselier and Vanhoucke, 2015; Brandon, time and cost.
1998; Brown, 1985; Christensen and Daniel, 1995; Colin and Vanhoucke and Colin (in press) assessed four multivariate
Vanhoucke, 2014; Kim, 2000; Kim et al., 2003). For example, regression methods for monitoring the activity level perfor-
Anbari (2003, 2004) shows the major aspects of EVPM and mance of an ongoing project from EVPM/ES observations.
provides logical extensions for forecasting project cost (or Based on a database of project networks, they concluded that
earnings) and completion time using several scenarios, such as using kernel principal component regression method with a
that the original cost (or time) estimate is flawed, that past cost radial base function kernel outperforms the other presented
(or time) performance is (or is not) a good predictor, and that regression methods.
the project will meet its original cost (or time) target upon In particular, Chen's (2014) EVPM forecasting approach
completion, regardless of prior performance. attempts to improve the forecasting accuracy of PV prior to
24 H.L. Chen et al. / International Journal of Project Management 34 (2016) 22–29

project execution. However, Chen's (2014) model implicitly where

assumes the duration estimate of a project prior to its execution to
Sðt ln ðPCbc ðtÞÞÞðt ln ðPCws ðtÞÞÞ
be or nearly be 100% accurate. Such an assumption not only is
β^PV ¼ PV PV
and ð4Þ
difficult to meet in practice but also makes the prediction Sðt ln ðPCbc ðtÞÞðt ln ðPCbc ðtÞÞÞ
PV PV
accuracy of PV subject to duration estimation. Thus, we suggest    bc 
^PV ¼ t ln PC ws
α ^
modeling extensions to improve the predictive power of PV prior PV ðt Þ −β PV t ln PC PV ðt Þ
to project executing. Enhancing PV's predictive power prior to
where
project execution is important, since it provides management
with more reliable predictive indications of project performance X  2
Sðt ln ðPCbc ðtÞÞðt ln ðPCbc ðtÞÞÞ ¼ PV ðt Þ
t ln PC bc
(EV and AC) before work begins, thus allowing an effective
X  2
PV PV

proactive action to advance favorable performance outcomes
PV ðt Þ
t ln PC bc
before it is too late. −
n
3. Research methodology
X    ws 
Sðt ln ðPCbc ðtÞÞðt ln ðPCws ðtÞÞÞ ¼ PV ðt Þ
t ln PC bc t ln PC PV ðt Þ
The previous section suggests that existing extensions of PV PV
X  X  ws 
EVPM are inadequate to improve the accuracy of EV and AC t ln PC PV ðt Þ
bc
t ln PC PV ðt Þ
−
predictions prior to project execution. In response to this limitation, n
this study poses the following research question: How can we
The equation for the predicted AC can be expressed as
enhance the predictive power of PV prior to project execution?
follows:
A two-fold methodology is adopted to answer this research
 

wp 
question. Based on Chen's (2014) logarithm linear transformation α^EV
^ þ EXP β^EV ln PC
^ ðt Þ
ac
(LLT), we first convert the PV data. The LLT equation proposed PC AC ðt Þ ¼ EXP EV ð5Þ
t
by Chen (2014) is one of the simplest transformations capable of
converting the relationship of PV, EV, and AC into a nearly linear where
relationship. This equation is found by the natural logarithm of Sðt ln ðPC^ wp ðtÞÞÞðt ln ðPCbc ðtÞÞÞ
project percentage completion multiplied by project time, which β^EV ¼ ð6Þ
EV EV
and
applies to both the dependent and independent variables: Sðt ln ðPC^ wp ðtÞÞÞðt ln ðPC^ wp ðtÞÞÞ
EV EV

 
wp 

 α^EV ¼ t ln PC bc ðt Þ −β^EV t ln PC
EV
^ ðt Þ
EV
Logarithm Linear Transformation ¼ t ln PC ij ðt Þ t ¼ 1; 2; …; n
ð1Þ where
X

wp 2
Sðt ln ðPC^ wp ðtÞÞÞðt ln ðPC^ wp ðtÞÞÞ ¼ t ln PC ^ ðt Þ
where i denotes the budgeted cost (bc), the work scheduled (ws), EV

X
wp 2
EV EV

the work performed (wp), or the actual cost (ac) of a project; j ^ ðt Þ

t ln PC EV
denotes the project's PV, EV, or AC; t denotes the time between −
project start and completion; ln is the natural logarithm; and n is n
ac
the project's duration in days. For example, PCAC (t) denotes the
percentage completion of the actual cost for a project's AC for X

wp   
Sðt ln ðPC^ wp ðtÞÞÞðt ln ðPCbc ðtÞÞÞ ¼ ^ ðt Þ
EV ðt Þ
t ln PC bc
wp t ln PC
time t; PCAC (t) denotes the percentage completion of the work EV
X

wp X 
EV EV

performed for a project's AC for time t. ^ ðt Þ
t ln PC EV t ln PC bc EV ðt Þ
In the second stage, we propose a mathematical modeling −
n
procedure that combines time-series and linear regression analysis,
starting with the following simple time-series regression model: Model restructuring for remedial measures of seasonal cycles
ensures that our parameter estimates are efficient. Because
Y t ¼ α þ βX t þ εt ð2Þ the effects of seasonal cycles require tests of a higher-order
autoregressive process, the study incorporates the generalized
where Yt is the matrix whose tth element is y(t) = t ln(PCji(t)), Durbin–Watson statistic (Vinod, 1973) to detect the effects of
denoting a response variable at time t; Xt is the matrix whose tth seasonal cycles (or autocorrelated errors that occur when a
element is x(t) = t ln(PCji(t)), denoting explanatory variables at correlation between two or more consecutive error terms is
time t; α is the y-intercept; β is the slope; and εt is the vector of present), specified as:
random errors. The equation for the predicted EV is expressed as X
n  2
follows: ^vt −^vt−p
t¼pþ1
  dp ¼ X ð7Þ
^PV
 
n
^ wp ðt Þ ¼ EXP α
PC þ EXP ^PV ln PC bc ðt Þ
β ð3Þ ^v2t
EV PV
t t¼1
 H.L. Chen et al. / International Journal of Project Management 34 (2016) 22–29 25

Table 1
Sample projects used to demonstrate the development of EV and AC models.
Project Project description Type of contract Category Budget at Cost at Planned Actual
name completion completion duration duration
ST Steel tank project for chemical waste storage Fixed-price contract Early finish over budget NTD a 3,172,140 NTD 3,212,850 262 days 212 days
GDP Gas duct and pipe project for a manufacturing Fixed-price contract Early finish under budget NTD 11,000,000 NTD 10,200,000 345 days 231 days
company
DP Distance pipe project for an engineering Fixed-price contract Late finish over budget NTD 1,803,500 NTD 1,998,500 107 days 231 days
company
RPS Raw and product silo project for a Fixed-price contract Early finish over budget NTD 29,103,200 NTD 31,988,200 180 days 177 days
manufacturing company
a
New Taiwan dollars, where 1 New Taiwan dollar = 0.032819 U.S. dollars.

where dp is the pth order Durbin–Watson statistic and ^vt , t = 1 to project, was completed early and within budget. GDP's planned
n are the residuals from ^v ¼ Y−X β^ . In situations where costs, actual costs, planned duration, and actual duration are,
autocorrelation is present, Yule–Walker estimates (used for respectively, NTD 11,000,000, NTD 10,200,000, 345 days,
autocorrelation correction in regression analysis) are then and 231 days.
performed to estimate the set of autoregressive parameters The DP project, a distance pipe construction project, was
(Shumway and Stoffer, 2000) for model restructuring, specified finished behind schedule and over budget. DP's respective
as: planned costs, actual costs, planned duration, and actual duration
are NTD 1,803,500, NTD 1,998,500, 107 days, and 231 days.
^ −1
^¼R ^ 0^r p
^ 2w ¼ ^r ð0Þ−ϕ
^p ; σ The RPS is a raw and product silo project that was finished ahead
ϕ p ρ ð8Þ
of schedule but was over budget. The respective planned costs,
actual costs, planned duration, and actual duration of RPS are NTD
where ϕ ^ 0 ¼ ðϕ ^ 2 …ϕ
^ 1ϕ ^ p Þ is the vector of autoregressive 29,103,200, NTD 31,988,200, 180 days, and 177 days. These
^
parameters, Rp ¼ f^ ρðk−1Þgpj; k¼1 is the matrix of the autocorre- sample projects are summarized in Table 1. Due to space
lation function (ACF), ρ ^0p ¼ ð^ρð1Þ^
ρð2Þ⋯^ ρðpÞÞ is the ACF vector, limitations, this study only presents a project's complete
σ
^ w is the autoregressive variance, and^r p ¼ ð^r ð1Þ^r ð2Þ⋯^rðpÞÞ is the
2
modeling process, but the rest are available from the authors
autocovariance vector. upon request.

4. Research results 4.2. Model building for EV forecasting

4.1. The data Table 2 summarizes the parameter estimates and Durbin–
Watson (DW) statistics of ST's EV prediction using Eqs. (3)
Data to support analysis were collected on four sample projects ^PV and β^PV
and (4). As shown in the left side of Table 2, both α
using case study research protocols (Yin, 2003), including the use are highly significant with p-values of b 0.001, capable of
ws
of multiple data sources (where possible) to ensure the quality of explaining 99.89% of the variation in the t ln(PCPV (t)) data. As
the data collected. The names of these projects have been changed reported in the right side of Table 2, the Durbin–Watson
at the request of the companies involved. The collected data statistics (with a value of 0.0033) suggest a significantly
include detailed scheduling, budget, cost, contract, and charac- positive autocorrelation in the model on the basis of the 0.05
teristic data for the sample projects. threshold value.
The ST project, a chemical waste storage tank, was finished Yule–Walker estimates were thus performed to estimate the
within schedule but was over budget. ST's respective planned set of autoregressive parameters for model restructuring, shown
costs, actual costs, planned duration, and actual duration are, in the left portion of Table 3. The center portion of Table 3
respectively, NTD 3,172,140, NTD 3,212,850, 262 days, and shows the test results of Durbin–Watson statistics after Yule–
212 days. The GDP project, a gas duct and pipe construction Walker estimates, indicating that autocorrelation is largely

Table 2
Parameter estimation and Durbin–Watson (DW) statistics of ST's predicted EV, PĈwp
EV(t).

Dependent variable: t ln(PCws

PV(t)) Durbin–Watson statistics
Variables Coefficient Order DW Probability b DW Probability N DW
Intercept α^PV = −0.00397⁎⁎⁎ 1 0.0033 b0.01 1.00
t ln(PCbc
PV(t)) β^PV = 0.98450⁎⁎⁎
Total R2 0.9989
⁎P b 0.05.
⁎⁎P b 0.01.
⁎⁎⁎P b 0.001.
26 H.L. Chen et al. / International Journal of Project Management 34 (2016) 22–29

Table 3
Parameter estimation and Durbin–Watson (DW) statistics of ST's predicted EV, PĈwp
EV(t), subsequent to Yule–Walker estimators.

Yule–Walker estimates Durbin–Watson statistics after Yule–Walker estimates Parameter estimation after
Yule–Walker estimates
Source A Value of source A Source B Value of source B Order DW Probability b DW Probability N DW Variables Coefficient
SSE 0.00004 DFE 255.00000 1 0.5230 b 0.01 1.00 Intercept α^PV = −0.00395⁎⁎⁎
MSE 0.00001 Root MSE 0.00044 t ln(PCbc
PV(t)) β^PV = 0.99100⁎⁎⁎
SBC − 3272.63090 AIC −3297.60930
Regress R2 0.99540 Total R2 0.99999 Total R2 0.99999
Note: SSE = sum of squares for error; SBC = Schwarz's Bayesian criterion; DFE = degree of freedom for error; AIC = Akaike's information criterion.
⁎P b 0.05.
⁎⁎P b 0.01.
⁎⁎⁎P b 0.001.

alleviated from a value of 0.0003 to a value of 0.5230 for the While pattern-matching logic supports analysis of empirical
Durbin–Watson statistics, although it is still significant. observations, a weakness of the approach is that it does not
The left portion of Table 3 reports the parameter estimates of provide a quantitative measure of the improvement achieved by
ST's EV prediction after Yule–Walker estimates. The model our models. In contrast, MAPE provides a measure for evaluating
ws
explains 99.99% of the variation in the t ln(PCPV (t)) data, out-of-sample forecasting accuracy, which is calculated as:
which is 0.10% more before model restructuring.
1X
MAPE ¼ jðyt −^yt Þ=yt j ð9Þ
4.3. Model building for AC forecasting n
where yt is the actual value of data at time t in the out-of-sample
Table 4 reports the parameter estimates and Durbin–Watson forecasts, ŷt is the predicted value for yt, and n is the number of
(DW) statistics of ST's AC prediction using Eqs. (5) and (6). The days of out-of-sample forecast data. When autocorrelated errors
left portion of the table reveals that α ^PV and β^PV are highly are not present, the respective EV and AC response models
significant with p-values of b 0.001; it explains 99.99% of the are denoted as PĈEV wp ac
(t) (Eq. (3)) and PĈAC (t) (Eq. (5)). The EV
bc wp
variation in the t ln(PCEV (t)) data. As the right side of Table 4 and AC response models are respectively denoted as PĈEV (t)MR
ac
shows, the Durbin–Watson statistics (with a value of 1.9420) and PĈAC(t)MR after model restructuring for the effect of
wp ac
suggest an insignificant positive or negative autocorrelation in the autocorrelated errors on PĈEV (t) and PĈAC (t).
model on the basis of the 0.05 threshold value. Therefore, this Fig. 1 plots the PV, EV, and AC curves of the ST project
study concludes that there is no need for model restructuring. prior to data transformation. Fig. 2 plots the transformed PV,
EV, and AC curves of the ST project using Chen's (2014)
4.4. Evaluation of forecast accuracy LLT = t × ln(PCji(t)). Fig. 3 depicts pattern matching among
ST's actual EV (red curve), predicted EV (black curve) created
wp
To assess the accuracy of our EV and AC forecasting with PĈEV (t)MR, and predicted EV (blue curve) created with
models, data were analyzed using two independent approaches: PV. Fig. 4 depicts pattern matching among ST's actual AC (red
ac
pattern-matching logic (Trochim, 1989) and the mean absolute curve), predicted AC (black curve) by PĈAC (t), and predicted
percentage error (MAPE) (Chen, 2010). Pattern-matching logic AC (blue curve) using PV.
compares an empirically based pattern with a predicted one. For As seen in Figs. 3 and 4, ST's respective predicted EV (black
wp ac
this study, the historical EV and AC for each project were used curve) and AC (black curve) with PĈEV (t)MR and PĈAC (t)
as empirical bases for comparison with EV and AC predictions appear to have smaller deviations from its actual EV and AC
generated from the original PVs and our EV and AC response than those (blue curves) by PV.
models. If certain patterns of forecasts and those of historical The study develops similar EV and AC forecasting models
EV and AC are closely matched, a strong causal inference can for the other three projects. Table 5 summarizes the analysis of
be made. the out-of-sample forecast data sets from the EV and AC

Table 4
Parameter estimation and Durbin–Watson (DW) statistics of ST's predicted AC, PĈac
AC(t).

Dependent variable: t ln(PCbc

EV(t)) Durbin–Watson statistics
Variables Coefficient Order DW Probability b DW Probability N DW
Intercept α^EV = 0.00251⁎⁎⁎ 1 1.9420 0.26 0.68
t ln(PĈwp
EV(t)) β^EV = 1.0080⁎⁎⁎
Total R2 0.9999
⁎P b 0.05.
⁎⁎P b 0.01.
⁎⁎⁎P b 0.001.
 H.L. Chen et al. / International Journal of Project Management 34 (2016) 22–29 27

Fig. 1. Scatterplot of the project ST's PV, EV, and AC data. Fig. 3. Pattern matching: comparison of project ST's actual EV, predicted EV by
PĈwp
EV(t), and predicted EV by PV.

models for projects ST, GDP, DP, and RPS. Due to space linear transformation, this study combines time-series and
limitations, the complete modeling details and related figures regression analysis to develop PV into an EV response model
wp
for those projects are available from the authors upon request. for predicting PĈEV (t). Through modeling the relationship
wp bc
To evaluate the improved predictive performance, the between PĈEV(t) and PCEV (t) depicted in the EV response
MAPE values are then compared to the forecasting results of model, this study further develops EV into an AC response model
ac
the projects' PV curves, where a project's PV curve is the for predicting PĈAC (t). The results show that the proposed
predictor of EV and AC prior to project execution. models improve the predictive performance of EV and AC by
As Table 5 shows, the respective MAPE values of the EV and 23.66% and 17.39%, respectively.
AC response models for projects ST, GDP, DP, and RPS are Adding to the benefits of current EVPM prediction models
(10.80%, 17.99%), (19.61%, 19.71%), (9.69%, 57.60%), and that provide management with predictions of project schedules
(10.94%, 4.17%), respectively. In contrast, those directly using and costs at completion (Anbari, 2003, 2004; Cioffi, 2006b;
PV curves are (12.33%, 30.89%), (81.62%, 73.00%), (20.16%, Henderson, 2003; Lipke et al., 2009; Vanhoucke and
59.38%), and (31.55%, 5.77%). The proposed models improved Vandevoorde, 2007), this study contributes a methodology
EV and AC forecasts by 23.66% and 17.39%, respectively, capable of improving predictive performance of PV. Specifically,
compared to the initial estimates by PV. This comparison the present research extends the state of knowledge about the
suggests that the proposed models could be viable for enhancing predictive power of PV by further modeling PV prior to project
the predictive performance of PV, thus providing more accurate execution. As such, this paper provides a fundamental improve-
EV and AC predictions before executing a project. ment in EVPM.

5. Summary and conclusions 6. Limitations and future research

This paper presents a step-by-step methodology for enhancing While the results of this study provide the potential for
the predictive power of PV. Based on Chen's (2014) logarithm improving the predictive power of PV prior to project execution,

Fig. 2. Scatterplot of the PV, EV, and AC data of project ST using Chen's Fig. 4. Pattern matching: comparison of project ST's actual AC, predicted AC
(2014) logarithm linear transformation = t × ln(PCij(t). by PĈacAC(t), and predicted AC by PV.
28 H.L. Chen et al. / International Journal of Project Management 34 (2016) 22–29

Table 5 Chou, J.S., Chen, H.M., Hou, C.C., Lin, C.W., 2010. Visualized EVM system
Comparison of the MAPE values of the prediction models to those of the for assessing project performance. Autom. Constr. 19 (5), 596–607.
projects' PV curves. Christensen, D., 1998. The costs and benefits of the earned value management
process. Acquis. Rev. Q. 5, 373–386.
Project Prediction MAPE MAPE of theprojects' Improved
Christensen, D., 1999. Using the earned value cost management report to
model PV curves MAPE
evaluate the contractor's estimate at completion. Acquis. Rev. Q. 19,
ST PĈwp
EV(t)MR 10.80% 12.33% 1.52% 283–296.
PĈac
AC(t) 17.99% 30.89% 12.90% Christensen, D., Daniel, V., 1995. Using earned value for performance
GDP PĈwp
EV(t)MR 19.61% 81.62% 62.01% measurement on software development projects. Acquis. Rev. Q. 2,
PĈac
AC(t) 19.71% 73.00% 53.28% 155–171.
DP PĈwp
EV(t)MR 9.69% 20.16% 10.47% Cioffi, D.G., 2005. A tool for managing projects an analytic parameterization of
PĈac
AC(t) 57.60% 59.38% 1.78% the S curve. Int. J. Proj. Manag. 23 (3), 215–222.
RPS PĈwp
EV(t)MR 10.94% 31.55% 20.62% Cioffi, D.G., 2006a. Completing projects according to plans an earned-value
PĈac
AC(t) 4.17% 5.77% 1.59% improvement index. J. Oper. Res. Soc. 57 (3), 290–295.
Note: Computation of MAPEs is based on 5%, 10%, 15%, 20%, 25%, 30%, Cioffi, D.G., 2006b. Designing project management: a scientific notation and an
35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, and 95% improved formalism for earned value calculations. Int. J. Proj. Manag. 24
completion of project time. The average improvements of the prediction models (2), 215–222.
over the PV curves in EV and AC forecasting performance are 23.66% and Colin, J., Vanhoucke, M., 2014. Setting tolerance limits for statistical project
17.39%, respectively. control using earned value management. OMEGA Int. J. Manag. Sci. 49,
107–122.
Elshaer, R., 2013. Impact of sensitivity information on the prediction of
project's duration using earned schedule method. Int. J. Proj. Manag. 31 (4),
some limitations of this study point to opportunities for future 579–588.
research. First, while this study uses four sample projects with Farris, J.A., Groesbeck, R.L., Van Aken, E.M., Letens, G., 2006. Evaluating the
different conditions to demonstrate the proposed methodology relative performance of engineering design projects: a case study using data
and the research results are statistically sound, making general- envelopment analysis. IEEE Trans. Eng. Manag. 53 (3), 471–482.
Hanna, A.S., 2012. Using the earned value management system to improve
ized claims based on the results would require further empirical electrical project control. J. Constr. Eng. Manag. 138 (3), 449–457.
testing. Specifically, more research is necessary to generalize the Henderson, K., 2003. Earned schedule: a breakthrough extension to earned
proposed methodology and broaden the area of its applicability. value theory? A retrospective analysis of real project data. The Measurable
Second, although this study presents the proposed modeling News, Summer: 13–23.
Keil, M., Rai, A., Cheney Mann, J.E., Zhang, G.P., 2003. Why software
method systematically and in detail, project managers and/or
projects escalate: the importance of project management constructs. IEEE
engineers may still have difficulty applying the method in Trans. Eng. Manag. 50 (3), 251–261.
practice. Future research could further develop the modeling Kim, E.H., 2000. A study on the effective implementation of earned value
method into a software system to increase ease of use for management methodology Doctoral dissertation The George Washington
project management practitioners. University.
Kim, B., 2014. Dynamic control thresholds for consistent earned value analysis
and reliable early warning. J. Manag. Eng. 31 (5), 1–9.
Acknowledgments Kim, E.H., Wells, W.G., Duffey, M.R., 2003. A model for effective
implementation of earned value management methodology. Int. J. Proj.
The authors gratefully acknowledge financial support from Manag. 21 (5), 375–382.
the Ministry of Science and Technology of Taiwan ROC. Lipke, W. (2003). “Schedule is different.” The Measurable News, March:
10–15.
Lipke, W., Zwikael, O., Henderson, K., Anbari, F., 2009. Prediction of project
References outcome: the application of statistical methods to earned value management
and earned schedule performance indexes. Int. J. Proj. Manag. 27 (2),
Acebes, F., Pereda, M., Poza, D., Pajares, j., 2015. Stochastic earned value 400–407.
analysis using Monte Carlo simulation and statistical learning techniques. Marshall, R.A., 2006. The contribution of earned value management to project
Int. J. Proj. Manag. 33 (7), 1597–1609. success on contracted efforts: a quantitative statistics approach within the
Anbari, F.T., 2003. Earned value project management method and extensions. population of experienced practitioners. Proceedings of the International
Proj. Manag. J. 34 (4), 12–23. Research Conference on Organizing by Projects, Xi'an, China, October 11-
Anbari, F.T., 2004. Earned value project management method and extensions. 13, pp. 739–751.
IEEE Eng. Manag. Rev. 32 (3), 97–110. Marshall, R.A., 2007. A quantitative study of the contribution of earned value
Batselier, J., Vanhoucke, J., 2015. Evaluation of deterministic state-of-the-art management to project success on external projects under contract Doctoral
forecasting approaches for project duration based on earned value dissertation, Lille Graduate School of Management, Lille, France.
management. Int. J. Proj. Manag. 33 (7), 1588–1596. Marshall, R.A., Ruiz, P., Bredillet, C.N., 2008. Earned value management
Brandon, D.M., 1998. Implementing earned value easily and effectively. Proj. insights using inferential statistics. Int. J. Manag. Proj.Bus. 1 (2),
Manag. J. 29 (2), 11–18. 288–294.
Brown, J.W., 1985. Evaluation of projects using critical path analysis and Naeni, L.M., Shadrokh, S., Salehipour, A., 2014. A fuzzy approach for the
earned value in combination. Proj. Manag. J. 16 (3), 59–63. earned value management. Int. J. Proj. Manag. 32 (4), 709–716.
Chen, H.L., 2010. Using financial and macroeconomic indicators to forecast Narbaev, T., Marco, A.D., 2014. An earned schedule-based regression
sales of large development and construction corporations. J. Real Estate model to improve cost estimate at completion. Int. J. Proj. Manag. 32 (6),
Financ. Econ. 40 (3), 310–331. 1007–1018.
Chen, H.L., 2013. Logit models for early warning of distressed capital projects. Project Management Institute, 2013. A guide to the project management body
J. Bus. Econ. Manag. 14 (Suppl. 1), 145–167. of knowledge. Project Management Institute, Newtown Square, PA.
Chen, H.L., 2014. Improving forecasting accuracy of project earned value Shumway, R.H., Stoffer, D.S., 2000. Time series analysis and its application.
metrics: linear modeling approach. J. Manag. Eng. 30 (2), 135–145. Springer-Verlag, New York.
 H.L. Chen et al. / International Journal of Project Management 34 (2016) 22–29 29

Trochim, W., 1989. Concept mapping for evaluation and planning. Eval. Vanhoucke, M., Vandevoorde, S., 2007. A simulation and evaluation of earned value
Program Plann. 12 (1), 1–11. metrics to forecast the project duration. J. Oper. Res. Soc. 58 (10), 1361–1374.
Vandevoorde, S., Vanhoucke, M., 2006. A comparison of different project duration Vinod, H.D., 1973. Generalization of the Durbin–Watson statistic for higher
forecasting methods using earned value metrics. Int. J. Proj. Manag. 24 (4), 289–302. order autoregressive process. Comput. Stat. 2 (2), 115–144.
Vanhoucke, M., Colin, J., 2015. On the use of multivariate regression methods Yin, R.K., 2003. Case Study Research: Design and Methods. Sage Publications,
for longest path calculations from earned value management observations. California.
OMEGA Int. J. Manag. Sci. (in press).