Earned Schedule …

an emerging enhancement to EVM

Walt Lipke
brought to you by the PMI MetSIG and Cheetah Learning

International Metrics Congress

2007 1
Brought To You By

 Cheetah Learning is Global Leader in PMP Exam

prep and PM Training with classes covering the
globe from the U.S. and Canada to South
America, Europe, New Zealand and Australia,
India and Asia, the Middle East and Africa.

 Our accelerated learning techniques speed

learning and retention and drive our 97% pass
rate on the PMP exam.

 Michelle LaBrosse, Founder & Chief Cheetah,

recently named one of 25 Influential Women in
PM in the world.

 Cheetah Learning strives to raise the profile of

PM with our webinars and our Know How
Network column which is published in over 200
publications around the world.

International Metrics Congress

2007 2
 Objective

• Introduce the Earned Schedule Concept

• Develop the Schedule Indicators
• Apply to Project Duration Prediction
• Apply to Schedule Analysis

International Metrics Congress

2007 3
 Earned Value Basics
BAC
PV
CPI  EV
AC

SPI  EV
PV
CV
$$ SV
AC

EV

Time
International Metrics Congress
2007 4
 EVM Schedule Indicators

• SV & SPI behave erratically for projects behind schedule

– SPI improves and equals 1.00 at end of project
– SV improves and concludes at $0 variance
• Schedule indicators lose predictive ability over the last third
of the project

International Metrics Congress

2007 5
 EVM Schedule Indicators

• Why does this happen?

– SV = EV – PV
– SPI = EV / PV
• At planned completion PV = BAC
• At actual completion EV = BAC
• When actual > planned completion
– SV = BAC – BAC = $000
– SPI = BAC / BAC = 1.00
Regardless of lateness !!

International Metrics Congress

2007 6
 Earned Schedule Concept

The cumulative value of ES is Time Now

$
found by using EV to identify
in which increment of PV the
cost value occurs.
A SVc
Σ PV

Σ EV B

ES SVt

1 2 3 4 5 6 7 8 9 10
7 months gone by, but the project only has “Earned Schedule” to Month 5
Which SV “Answers the mail?” $ behind or 2 months behind schedule?
International Metrics Congress
2007 7
 Earned Schedule Metric

• Required measures
– Performance Measurement Baseline (PMB) – the
time phased planned values (PV) from project start to
completion
– Earned Value (EV) – the planned value which has
been “earned”
– Actual Time (AT) - the actual time duration from the
project beginning to the time at which project status is
assessed
• All measures available from EVM

International Metrics Congress

2007 8
 Earned Schedule Calculation

• ES (cumulative) is the:
Number of completed PV time increments EV exceeds +
the fraction of the incomplete PV increment

• ES = C + I where:
C = number of time increments for EV  PV
I = (EV – PVC) / (PVC+1 – PVC)

International Metrics Congress

2007 9
 Interpolation Calculation

PVC+1 
I /1 mo = p / q
ES(calc)
I = (p / q)  1 mo
EV   q

$$
p p = EV – PVC
PVC  q = PVC+1 – PVC
I
1 mo
EV – PVC
ES I=  1mo
PVC+1 – PVC
May June July
Time

International Metrics Congress

2007 10
 Earned Schedule Indicators

• Schedule Variance:
SV(t) = ES – AT

• Schedule Performance Index:

SPI(t) = ES / AT
where AT is “Actual Time” – the duration from start to time now

• SV(t) and SPI(t) are time-based (months, weeks …)

International Metrics Congress

2007 11
 Earned Schedule Indicators

• What happens to the ES indicators, SV(t) & SPI(t), when the

planned project duration (PD) is exceeded (PV = BAC)?
They Still Work …Correctly!!
• ES will be  PD, while AT > PD
– SV(t) will be negative (time behind schedule)
– SPI(t) will be < 1.00
Reliable Values from Start to Finish !!

International Metrics Congress

2007 12
 Schedule Variance Comparison
100 Early Finish Project 1

80 0.8

$ 60 0.6 Mo

40 0.4

20 SV($) SV(t)0.2

0 0
J F M A M J J A S O N D J F M
0 0
Late Finish Project

-100
-1

$ -200 Mo
-2
-300

-400 -3
J F M A M J J A S O N D J F M

International Metrics Congress

2007 13
 Schedule Index Comparison
1.12
Early Finish Project
1.10
1.08
1.06
1.04
1.02
1.00
SPI($) SPI(t)
0.98
J F M A M J J A S O N D J F M

1.30
Late Finish Project
1.20
1.10
1.00
0.90
0.80
0.70
J F M A M J J A S O N D J F M

International Metrics Congress

2007 14
 Late Finish Project
Commercial IT Infrastructure Expansion Project Phase 1
Cost and Schedule Variances
at Project Projection: Week Starting 15th July xx

CV cum SV cum Target SV & CV SV (t) cum

20 2

0 0

-20 -2
Stop wk 19
D ollars (,0 00 )

-40 -4
Sched wk 20 Re-start wk 26

W e eks
-60 -6

-80 -8

-100 -10

-120 -12

-140 -14

-160 -16
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34

Elapsed Weeks

International Metrics Congress

2007 15
 Earned Schedule – Key Points

• ES Indicators constructed to behave in an analogous

manner to the EVM Cost Indicators, CV and CPI
• SV(t) and SPI(t)
– Not constrained by PV calculation reference
– Provide duration based measures of schedule
performance
– Valid for entire project, including early and late finish
• Facilitates integrated Cost/Schedule project management
(using EVM with ES)

International Metrics Congress

2007 16
 Schedule Prediction

• Can the project be completed as planned?

– TSPI = Plan Remaining / Time Remaining
= (PD – ES) / (PD – AT)
where PD is the planned duration (time at BAC)
(PD – ES) = PDWR
PDWR = Planned Duration for Work Remaining
• …completed as estimated?
– TSPI = (PD – ES) / (ED – AT)
where ED = Estimated Duration
TSPI Value Predicted Outcome
 1.00 Achievable
 1.10 Not Achievable
International Metrics Congress
2007 17
 Schedule Forecasting

• Long time goal of EVM …Prediction of total project duration

from present schedule status
• Independent Estimate at Completion (time)
– IEAC(t) = PD / SPI(t)
– IEAC(t) = AT + (PD – ES) / PF(t)
where PF(t) is the Performance Factor (time)
– Analogous to IEAC used to predict final cost
• Independent Estimated Completion Date (IECD)
– IECD = Start Date + IEAC(t)

International Metrics Congress

2007 18
 Earned Schedule Terminology

ES = C + I number of complete
Earned Schedule EScum periods (C) plus an incomplete
Metrics portion (I)

Actual Time ATcum AT = number of periods executed

Schedule Variance SV(t) SV(t) = ES - AT

Schedule Performance
Indicators Index
SPI(t) SPI(t) = ES / AT

TSPI(t) = (PD – ES) / (PD – AT)

To Complete Schedule TSPI(t)
TSPI
Performance Index
TSPI(t) = (PD – ES) / (ED – AT)

Predictors Independent Estimate IEAC(t) IEAC(t) = PD / SPI(t)

at Completion (time)
IEAC(t) = AT + (PD – ES) / PF(t)
PF

International Metrics Congress

2007 19
 Independent Confirmation

• SPI(t) & SV(t) do portray the real schedule performance

• At early & middle project stages pre-ES & ES forecasts of
project duration produce similar results
• At late project stage ES forecasts outperform all pre-ES
forecasts
• The use of the SPI(t) in conjunction with the TSPI(t) has
been demonstrated to be useful for managing the schedule

Stephan Vandevoorde – Fabricom Airport Systems, Belgium

International Metrics Congress

2007 20
 Schedule Analysis with EVM?

• Most practitioners analyze schedule from the bottom up

using the network schedule, independent from EVM
….“It is the only way possible.”
– Analysis of the Schedule is overwhelming
– Critical Path is used to shorten analysis
(CP is longest path of the schedule)
• Duration prediction using Earned Schedule provides a
macro-method similar to the method for estimating Cost
– A significant advance in practice

• But, there’s more that ES facilitates ….

International Metrics Congress

2007 21
 Facilitates Drill-Down Analysis

• ES can be applied to any level of the WBS, to include task

groupings, such as the Critical Path (CP)
– Requires creating PMB for the area of interest
– EV for the area of interest is used to determine its ES
• Enables comparison of forecasts, total project (TP) to CP
– Desired result: forecasts are equal
– When TP forecast > CP forecast, CP has changed
– When CP > TP, possibility of future problems

International Metrics Congress

2007 22
 ES Bridges EVM to the Schedule

BAC
$$
PV

EV

SV(t)

Time ES AT PD
International Metrics Congress
2007 23
 How Can This Be Used?

• Tasks behind – possibility of impediments or constraints can

be identified
• Tasks ahead – a likelihood of future rework can be identified
• The identification is independent from schedule efficiency
• The identification can be automated

PMs can now have a schedule analysis tool

connected to the EVM Data!!

International Metrics Congress

2007 24
 Current Usage & Recognition

• EVM Instructors
– Performance Management Associates, Management Technologies,
George Washington University, University of Florida …
• Boeing Dreamliner, Lockheed Martin, US State
Department, Secretary of the Air Force
• Several Countries - Australia, Belgium, United Kingdom,
USA ….(Japan, Switzerland, Sweden, Spain, Brazil, India, …)
• Applications across weapons programs, construction,
software development, …
• Range of project size from very small and short to extremely
large and long duration
• Inclusion of Emerging Practice Insert into PMI - EVM
Practice Standard (2004)

International Metrics Congress

2007 25
 Summary

• Derived from EVM data … only

• Provides time-based schedule indicators
• Indicators do not fail for late finish projects
• Application is scalable up/down, just as is EVM
• Schedule prediction is better than any other EVM method
presently used
• Application is growing in both small and large projects
• Practice recognized as “Emerging Practice”
• Facilitates bridging EVM analysis to include the Schedule

International Metrics Congress

2007 26
Thank You!

Walt Lipke

Email: [email protected]
Thank You from the MetSIG…The
Information Highway For the Metrics
of the World

International Metrics Congress

2007 27
PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

Earned Schedule Forecasting Method Selection1

Walt Lipke
PMI Oklahoma City Chapter

Abstract
Recent research indicates Earned Schedule (ES) forecasting from Earned Value
Management data is generally improved when the performance factor, PF=1, is used.
However, the use of the ES schedule performance index, SPI(t), remains as accepted
practice and, at times, provides the better deterministic forecast. It is postulated that
there may be recognizable conditions for which one method is preferable. This paper is
an investigation to discern those conditions.

Introduction

The authors, Batselier and Vanhoucke (B&V) examined several methods of forecasting
in their paper, “Empirical Evaluation of Earned Value Management Forecasting
Accuracy for Time and Cost” [Batselier et al, 2015]. Their research comprehensively
evaluated forecasting using Earned Value Management (EVM) data from 51 projects,
predominantly construction. The results of the research demonstrated the use of
performance factor, PF=1, with the Earned Schedule (ES) method often provides the
more accurate deterministic forecast.

Independently, the B&V finding was corroborated, albeit with a smaller amount of data
[Lipke, 2017]. Of the 16 projects examined, ES forecasting using PF=1 was shown to
provide better results for 12. Beyond affirming the B&V research, the application of
PF=1 to statistical forecasting was examined. Additionally, an observation was made
that when performance variation is small, forecasting with the ES schedule performance
index, SPI(t), is usually better.

The good showing by PF=1, in both cited papers, was surprising and unexpected.
Although PF=1 more often provides the better forecast, curiosity was created as to why
it is not always so; there are several instances when SPI(t) is preferred. Because the
examinations involved real data, there was some concern as to whether the data sets

1
How to cite this paper: Lipke, W. (2019). Earned Schedule Forecasting Method Selection; PM World Journal, Vol.
VIII, Issue I (January).

© 2019 Walt Lipke www.pmworldlibrary.net Page 1 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

represented a very localized set of conditions, or if there were anomalies. That is, do
peculiarities within the data examined cause the forecasting performance of PF=1 to be
enhanced? For instance, management actions, such as re-plans, can obscure
performance and cause the SPI(t) value to be close to 1.0. Of course, when the index
value is close to 1.0, it is very likely that PF=1 provides the better forecast.

Certainly management actions can perturb the forecast and assuredly there are
disturbances imbedded in the data. Nevertheless, examination of the data from the
second paper did not reveal significant issues causing PF=1 to be preferred. It is now
believed that the results from the two studies are reasonably broad-based and
representative.

Nevertheless, there is reason to explore; there may be performance characteristics

useful for identifying which method is more likely to provide the better forecast. We
know by inspection that when performance is close to planned, PF=1 should provide the
better forecast. As well, it is easily shown that when performance has little variation, the
forecast from SPI(t) is better.

From these observations, the research proposition is made:

Two characteristics determine the selection of the more accurate forecasting method,
either PF=1 or SPI(t):

1) Performance deviation from SPI(t) = 1.0

2) Periodic variation in ES

This paper examines simulated and real data to seek and establish the selection criteria
for choosing between PF=1 and SPI(t) forecasting methods. Presuming selection
criteria can be established, it is reasoned ES deterministic forecasting will yield
increased accuracy and thereby improve decision-making by project managers.

Earned Schedule Review

Earned Schedule is dependent upon EVM; the ES measure is derived from the accrued
earned value (EV) and the performance measurement baseline (PMB) [Lipke, 2003].2
As shown in figure 1, “…the idea is to determine the time at which the EV accrued
should have occurred.” The time duration from project start to the point on the PMB

2
It is presumed the reader has knowledge of Earned Value Management.

© 2019 Walt Lipke www.pmworldlibrary.net Page 2 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

where the planned value (PV) equals the EV accrued is the earned portion of the
planned duration (PD), i.e. ES.

As diagrammed in the figure, determining the value of ES is a simple matter of

interpreting the graph. In practice, however, a calculation method is applied. Its
description is explained, in detail, in the Earned Schedule book [Lipke, 2009].

Having the capability to determine ES and the actual time (AT) at which the EV is
reported, the time-based indicators for schedule performance index, SPI(t), and
schedule variance, SV(t), are created:

SPI(t) = ES / AT
SV(t) = ES – AT

The index, SPI(t), describes the efficiency of achieving the PD for the time invested.
When the value of SPI(t) is 1.0 or greater, schedule performance is good. And, when
the indicator value is less than 1.0, performance problems need investigating and
correcting. Similarly, when the value of SV(t) is positive, performance is ahead of
schedule and when negative, performance is lagging.

Figure 1. Earned Schedule Concept

© 2019 Walt Lipke www.pmworldlibrary.net Page 3 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

Beyond assessing current status, the forecasting of project duration and completion
date is made possible using the formulas below:

1) IEAC(t) = PD / SPI(t)

2) IEAC(t) = AT + (PD – ES) / PF

IEAC(t) is the independent estimate at completion (time), i.e. the forecast duration;
adding the computed duration to the project start date can be used to forecast the
completion date. The PF term, appearing in formula 2, is a performance factor; for
example, PF=1, could be used when believed it better represents the expected future
execution efficiency.

Methodology

The approach to determine criteria for forecasting selection was a two-step process.
The first step simulated project performance, applying variation over a range of values
and biasing the outcome from early to late finish. Thirty nine scenarios, representing
combinations of variation and bias, were induced in the execution of twenty five
simulated projects. The simulated projects were limited to a final duration no greater
than 50 periods.

For the simulation, fixed values and variable multipliers were used to create periodic ES
values (ESp). The fixed values ranged from 0.15 to 0.80, while the variable multipliers
were 0.1, 0.5, and 0.9. Each was applied through a random number, ranging from zero
to one. To further explain, let’s use the fixed value 0.8; ESp is either 1.8 or 0.2,
dependent upon whether the bias is plus or minus. That is, the fixed value is either
added to, or subtracted from 1.0. When a variable multiplier is used, it is coupled to a
random number to produce the value added or subtracted from 1.0 in the calculation of
ESp.

Nine bias values were used, ranging from 0.1, at 0.1 increments, through 0.9. Bias is
produced by comparing a random number to the bias value; when the random number
is less than one minus the bias value, the periodic ES variation is subtracted from 1.0.
Conversely, the variation is added. Bias values above 0.5 cause SPI(t) to be greater
than 1.0, while values below 0.5 induce SPI(t) to be less than 1.0.

Utilizing the specified variation and bias values, four sets of examination scenarios were
created. The first set viewed large variation over the full range of bias values. The

© 2019 Walt Lipke www.pmworldlibrary.net Page 4 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

second set examined the range of variation multipliers in three subsets of bias values
(0.1, 0.5, and 0.9). Similarly, the third set applied fixed variation multipliers (0.8, 0.5, and
0.2) in three subsets of bias values. Lastly, the fourth set examined combinations of
variation and bias to clarify results observed in the three control sets.

For the second step, the tabulated results were inspected for patterns favoring one of
the forecasting methods, PF=1 or SPI(t). A process of filtering and graphical analysis
was applied for deducing possible forecast method selection rules.

The simulation results for the 39 scenarios were evaluated using the Mean Absolute
Percent Error (MAPE) for each forecasting formula. The MAPE for a specific scenario is
the computed average across the performance results of the 25 simulated projects.

The scenarios having MAPE favorable to PF=1 were grouped, while those favorable to
SPI(t) were likewise placed together. Scatter plot graphs were created for each
grouping. The scatter plots are the paired values of the two proposed selection
characteristics: natural logarithm (ln) of SPI(t) and the standard deviation () of the
natural logarithm of ESp. Just as for MAPE, the two characteristics are represented by
the average across the simulations for each scenario.

The patterns recognized in the graphs were used to propose forecasting selection rules.
The proposed forecasting method selection rules were then tested using both simulated
and real data.

Results/Analysis

An example of the results from the simulations of the 39 scenarios is represented in

table 1. The inputs creating distinctive performance scenarios include the following
variables:

1) SPI(t) – the initial Schedule Performance Index (time) value for the simulation
2) PD - Planned Duration
3) Fixed Var – Fixed Variable, the value of the fixed variation
4) F or V - choice between fixed (F) or variable (V) type variation
5) Var Mult – Variation Multiplier, the multiplier value when V is chosen
6) Bias “+” - bias value to induce late to early finish performance
7) Trigger – performance stability value

© 2019 Walt Lipke www.pmworldlibrary.net Page 5 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

The trigger value was used to determine whether performance stability had influence in
the analysis and selection of the forecasting method. There is no further discussion of
trigger value; applying performance stability of 0.10 did not significantly impact MAPE
results.

Table 1. Scenario Results

As well, the output values from the scenario simulations, Output Averages and Error
Results, are depicted in table 1. The definitions of the sub-headings are:

1) FD – Final Duration
2) lnESp  –the standard deviation of the natural logarithm of the periodic values of
ES
3) lnSPI(t) – natural logarithm of cumulative value of SPI(t)
4) MAPE(1) – mean absolute percent error for PF=1 forecasting
5) MAPE(S) – mean absolute percent error for SPI(t) forecasting

A few general observations can be made from analysis of table 1. As expected, the
Variation Multiplier induces lnESp  values; as the multiplier value increases, so does
the standard deviation. And, similarly, Bias is reflected in the lnSPI(t) values; bias of 0.1
induces negative numbers (late finish), while 0.9 creates positive numbers (early
completion).

Further analysis of table 1 indicates there may be correlation between lnESp  and
MAPE; error increases as variation becomes larger. Additionally, for scenario results
having lnSPI(t) values near 0.0, MAPE(1) is seen to be less than MAPE(S); i.e., when
SPI(t) is very near the value of 1.0, PF=1 provides the better forecast, although the error
difference is very small. This finding was an expectation.

© 2019 Walt Lipke www.pmworldlibrary.net Page 6 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

Scatter Plots

With the simulation results for the 39 scenarios, two outcome data sets were created.
One contained those having more accurate forecasts from using performance factor
PF=1. Of course, the second set contained results when SPI(t) forecasting was more
accurate. For each of these sets, scatter plots were made, graphing the paired values of
lnSPI(t) and lnESp . The scatter plots are shown in figures 2 and 3.

Figure 2. Scatter Plot of PF=1 Set

The scatter plot of figure 2 contains results from 16 scenarios. Clearly, the majority of
data points lie between lnSPI(t) values of 0.1 and -0.1. Two of three points outside of
this range are near the -0.1 value, while the third appears to be an outlier. All of the
results in figure 2, favor PF=1 forecasting by a small percentage, usually less than 2
percent.

Figure 3 is the plot of 23 results favoring SPI(t) forecasting. With one exception, the
points reside in the lnESp  range of 0.0 to 0.8. The one point outside of the range has
extremely high variation with an extremely low value for SPI(t). The error results for this
point are excessively high; 29.4 percent for PF=1 and 16.7 percent for SPI(t). Similar to

© 2019 Walt Lipke www.pmworldlibrary.net Page 7 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

analysis of figure 2, this point is considered an outlier. As well, it is observed that all of
the points, excluding the outlier, lie in the region of SPI(t) values between, 0.6 and -0.6.

Figure 3. Scatter Plot of SPI(t) Set

Examining the scenario results graphed in figure 3, the following findings were
compiled:

1) Three points have nearly the same value of error with PF=1 forecasting
2) Seven points indicate superior SPI(t) forecasting by 10 percent or greater
3) Six points indicate better SPI(t) forecasting in the range of 5 to 10 percent
4) Remaining seven points have a small difference from PF=1 forecasting, generally
between 1 to 3 percent difference

Thus, from analysis of the data for the two graphs, 26 of 39 scenarios produce very
similar forecasts. Approximately 18 percent of the SPI(t) forecasts are significantly
better, while another 15 percent show moderate improvement.

The implication from the above analysis is PF=1 deterministic forecasting is generally
good. This statement further confirms the results from the two cited studies. However,

© 2019 Walt Lipke www.pmworldlibrary.net Page 8 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

there are instances when PF=1 results are significantly less accurate. Obviously then,
there is value in being able to discern when SPI(t) forecasting should be used.

Using the observations from figures 2 and 3 scatter plots, selection rules are proposed
in table 2. Fundamentally, project managers should use SPI(t) when the indicator’s
value has some amount of separation from 1.0 and when performance variation is not
large. When this is not the situation, PF=1 is preferred. It is to be noted that when out of
control performance is encountered, PF=1 is the recommended choice. As identified in
table 2, the out of control condition is characterized by extreme performance values for
lnSPI(t) and lnESp . For this condition, it is usually the case that neither method
provides accurate forecasts; generally, the SPI(t) method is worse.

Table 2. Selection Rules

Selection Rules Testing

The proposed rules for selecting between the SPI(t) and PF=1 forecasting methods,
specified in table 2, were tested using both simulated and real data, 10 projects each.
Analysis information from the testing is compiled in tables 3 and 4. The attributes of
interest are identified below:

1) Planned and Final Duration

2) The maximum and minimum values for lnSPI(t) and lnESp 
3) The number of correct and incorrect selections made, at both zero and five
percent tolerance. When the SPI(t) and PF=1 forecasts are within tolerance (i.e.,
difference between MAPE values), the selection from application of the rules is
counted as correct.
4) The number of out of control conditions encountered
5) Four forecasting error modes are tabulated for comparison: Err-SPI(t), Err-PF1,
Err-Selection, and Err-Best.

For clarity, each error mode is computed as MAPE. The terminology is defined below:

© 2019 Walt Lipke www.pmworldlibrary.net Page 9 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

1) Err-SPI(t) is the error when forecasts are made using only SPI(t)
2) Err-PF1 is the forecasting error when only PF=1 is used
3) Err-Selection is the error when the forecasting selection rules are applied
4) Err-Best represents the least forecasting error possible; i.e., the method (PF=1 or
SPI(t)) producing the better forecast is selected at every period of performance

From review of table 3, Selection Rules Test – Simulated Data, the use of the selection
rules appears to provide good results in comparison to the other forecasting modes. For
eight of the 10 projects, the error shown for Err-Selection is within 1.6 percent of the
best forecast, Err-Best. The largest difference is 3.6 percent shown for project 6.

Four projects (1, 2, 3, and 5) have large error when the PF=1 forecasting formula is
used. For these projects, application of the selection rules improved the forecasts, on
average, by nearly 13 percent, a significant improvement.

Table 3. Selection Rules Test – Simulated Data

From comparison of the correct and wrong selections at zero and five percent tolerance,
it can be deduced that there are many times when PF=1 and SPI(t) methods compute
nearly the same forecast values. For 5 percent tolerance, a significant increase in the
number correct is seen for six projects (2, 3, 6, 7, 8, and 10).

© 2019 Walt Lipke www.pmworldlibrary.net Page 10 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

Project number 7 is shown to have thirteen out of control periods. As could be expected,
there is a large difference between maximum and minimum lnSPI(t), 0.511. As well, the
variation indicated by lnESp  is incredibly large. The maximum is equal to 1.365, while
the minimum is 0.607. Although 13 of 38 periods were identified to be out of control, the
selection rules forecast was very good at 3.6 percent error.

The real EVM data used for the selection rules testing come from high technology
projects; it is a subset of the data examined in the second paper cited [Lipke, 2017].
Table 4 compiles the testing results in the same format used for table 3. A characteristic
of the data is project performance is highly erratic. The range of values for lnSPI(t) and
lnESp  is large for several projects. As well, four projects (3, 5, 7, and 9) have a large
number of out of control performance periods. Thus, as should be expected, the
forecasting error values are somewhat larger than are those for the simulated data, on
average 3.7 percent higher for the best forecasts.

Eight of the 10 projects have selection forecasts comparing favorably to their

corresponding best forecasts; they are within a difference of five percent. The two
remaining have differences of 7.2 and 7.8 percent, not excessively large values.

Table 4. Selection Rules Test – Real Data

© 2019 Walt Lipke www.pmworldlibrary.net Page 11 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

Just as for the simulated data, several projects (2, 4, 5, 6, 7, 8, and 10) showed that
many of the SPI(t) and PF=1 forecasts are within a difference of five percent. As well,
the review of tables 3 and 4 revealed another characteristic; in general, poor forecasts
for the simulated data were made using PF=1 (projects 1, 2, 3, and 5), whereas for the
real data it was observed for the SPI(t) method (projects 1, 3, and 9).

Another analysis of applying the selection rules is given in tables 5 and 6. These tables
provide succinct evaluation for the following characteristics:

1) S & 1 within 5% – “yes” is shown when SPI(t) and PF=1 methods produce overall
forecast results within five percent, otherwise “no”
2) S or 1 Better? – when the overall forecast using SPI(t) has less error, “S” is
entered, otherwise “1”
3) Select Improve? – “yes” is entered when the value for Err-Selection is less than
either Err-SPI(t) or Err-PF1 methods, otherwise “no”

In comparing the results tabulated for the two tables, two differences stand out. The “S
& 1 within 5%” characteristic occurs only once for the simulated data, whereas for the
real data it is seen for seven projects. For the “S or 1 Better?” evaluation, SPI(t)
forecasting is better for the simulated data, while PF=1 is dominant for real data.
However, it is recognized that among the eight projects having a better forecast from
PF=1, six have forecast values within 5 percent from the SPI(t) values.

Table 5. Compiled Results – Simulation Data

Table 6. Compiled Results – Real Data

© 2019 Walt Lipke www.pmworldlibrary.net Page 12 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

Perhaps, the most significant evaluation characteristic for the proposed selection rules
is “Select Improve?” The idea of this measure is to answer the question: Does
application of the selection rules provide forecasting improvement? For 8 of 10 projects,
for both simulated and real data, the selection rules forecast had lower MAPE than, at
least, one of the two methods. For the two projects in each set, where Err-Selection is
not less than either Err-SPI(t) or Err-PF1, MAPE values are poorer by very small
differences, less than 1.1 percent.

Summary/Conclusion

Two empirical studies reported that the performance factor equal to 1.0 provides better
project duration forecasts than using the ES schedule performance index, SPI(t).
However, for the examined projects, there are instances where SPI(t) was better. Thus,
it was reasoned that there may be performance conditions which cause one method to
forecast more accurately.

The focus of this study was to determine selection criteria for choosing between the two
deterministic forecasting methods, PF=1 and SPI(t). Simulation of project execution,
under various performance conditions, was a key element of the research approach.

The results from the simulation of 25 projects, applying 39 performance scenarios, were
graphed to examine for patterns. The patterns observed for the paired values of lnSPI(t)
and lnESp  provided performance information, leading to criteria for selecting between
the two methods. The proposed rules for method selection were tested using 10
projects each of simulated and real EVM project data. For both data sets, the selection
rules for 8 of 10 projects yielded more accurate forecasts than at least one of the PF=1
and SPI(t) methods.

In final analysis, the two deterministic methods yield very comparable forecasts about
70 percent of the time. The risk of using the PF=1 method is there are instances when
its forecast can be in error by greater than 10 percent. Whereas, the risk of exclusively
using SPI(t) forecasting is it generally has larger error than PF=1, although the
difference is small. The application of the selection rules moderates these risks, and
thereby provides a more reliable forecast.

© 2019 Walt Lipke www.pmworldlibrary.net Page 13 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

Suggested Research

Certainly, the duration forecasting selection rules proposed in this paper should be
further examined before general adoption. It is suggested that those practitioners having
real data apply the rules and report their findings. As well, application of the rules to
simulated EVM data is welcomed, and may yield refinements or a better selection
approach.

For those interested, the forecast method selection spreadsheet is available from the
author upon request.

References

Batselier, J., M. Vanhoucke (2015). “Empirical Evaluation of Earned Value Management

Forecasting Accuracy for Time and Cost,” Journal of Construction Management,
141(11): 05015010, 1-13

Lipke, W. (2017). “Assessing Earned Value Management and Earned Schedule

Forecasting,” PM World Journal (online), August, Vol VI, Issue 8.
https://pmworldlibrary.net/wp-content/uploads/2017/08/pmwj61-Aug2017-Lipke
assessing-evm-and-earned-schedule-forecasting-featured-paper3.pdf

Lipke, W. (2009). Earned Schedule, Raleigh, NC: Lulu Publishing

Lipke, W. (2003). “Schedule is Different,” The Measurable News, Summer: 31-34

© 2019 Walt Lipke www.pmworldlibrary.net Page 14 of 15

PM World Journal Earned Schedule Forecasting Method Selection
Vol. VIII, Issue I – January 2019 by Walt Lipke
www.pmworldjournal.net Featured Paper

About the Author:

Walt Lipke
Oklahoma City, USA

Walt Lipke retired in 2005 as deputy chief of the Software

Division at Tinker Air Force Base, where he led the organization to the 1999 SEI/IEEE
award for Software Process Achievement. He is the creator of the Earned Schedule
technique, which extracts schedule information from earned value data.

Credentials & Honors:

Master of Science Physics
Licensed Professional Engineer
Graduate of DOD Program Management Course
Physics honor society - Sigma Pi Sigma ()
Academic honors - Phi Kappa Phi ()
PMI Metrics SIG Scholar Award (2007)
PMI Eric Jenett Award (2007)
Who’s Who in the World (2010)
EVM Europe Award (2013)
CPM Driessnack Award (2014)
Australian Project Governance and Control Symposium established the annual
Walt Lipke Project Governance and Control Excellence Award (2017)
Albert Nelson Marquis Lifetime Achievement Award 2018

© 2019 Walt Lipke www.pmworldlibrary.net Page 15 of 15

Budownictwo i Architektura 3 (2008) 15-32

Earned value method as a tool for project control

Agata Czarnigowska

Institute of Construction, Faculty of Civil and Sanitary Engineering,

Lublin University of Technology, 20-618 Lublin, Nadbystrzycka 40,
e-mail: [email protected]

Abstract: Earned Value is a well-known project management tool that uses

information on cost, schedule and work performance to establish the current status
of the project. By means of a few simple rates, it allows the manager to extrapolate
current trends to predict their likely final effect. The method is based on a simpli-
fied model of a project, but proved to be useful in practice of cost control. It is
being developed to account better for schedule and time aspects. The paper outlines
the basic principles of the method and its recent extension, the Earned Schedule
method, and, with help of a few examples, investigates into assumptions that affect
their diagnostic and predictive accuracy.
Key words: Earned Value, project control.

1. Purpose of earned value

The Earned Value method has been developed as a tool facilitating project
progress control. It is used for determining a project’s status (is it behind or ahead
of schedule? is it over or under budget?) and the scale of current variances from the
plan. Moreover, it allows a project manager to make inferences on the final effect of
the project in terms of cost and, to some extent, in terms of duration, by extrapolat-
ing current trends [1,2].
The method is simple: it assumes a simplified model of a project, and calcu-
lations require nothing more than four basic arithmetic operations. However, the
method has been recognized as a useful tool by many practitioners and government
agencies and has become a standard in project management. It proved to be versatile
enough to be applied to any type of a project, ranging from defence schemes worth
millions and extending on many years to minor IT projects. The analysis can be
conducted on any level of work breakdown structure and used by both clients and
contractors.
The method, if to be used efficiently, requires a disciplined approach to collec-
tion of data on project cost and progress (on weekly basis) and the findings are to be
processed immediately. The purpose is to detect any deviation as soon as possible,
so that there is enough time to asses if the deviation is dangerous for the project and,
if necessary, to take corrective actions.
16 Agata Czarnigowska

2. The idea of earned value analysis and interpretation

of its results

2.1. Input data

Figure 1 presents the idea of the Earned Value project control. The analysis
requires following inputs [1, 2]:
– BCWS – Budgeted Cost of Works Scheduled – the baseline for the analysis,
cumulated planned costs related to time of their incurrence;
– BCWP – Budgeted Cost of Work Performed – a measure of physical progress of
works expressed by cumulated planned cost of works actually done related to time,
it is also called Earned Value (like the method it is used by);
– ACWP – Actual Cost of Work Performed – cumulated ammount payable for
works done related to time;
– BAC – Budget at Completion – total planned cost of the whole project, it
equals BCWS at the planned finish;
– T – planned duration of the project.

Fig. 1. Earned Value curves; this project is currently (“today”) over budget and behind schedule.

2.2. Project status indicators

PC – Percentage Complete:

BCWP
PC = , (1)
BAC
CV – Cost Variance – a measure of deviation between planned and actual cost of
works done until the date of recording progress in money units. If negative, it indi-
cates that the project is over budget:
CV = ACWP - BCWP . (2)
 Earned value method as a tool for project control 17
To capture the scale of deviation, it is often expressed as a fraction of the budg-
eted cost of works performed:

CV
CV % = × 100% , (3)
BCWP
SV – Schedule Variance – a measure of deviation between the actual progress and
the planned progress. Though it is interpreted as time deviation, it is expressed in
money units. In other words, it is the difference between the planned cost of works
that have been done and planned cost of works that should have been done by the
reporting date. If negative, it indicates a delay:
SV = BCWP - BCWS . (4)
To address any distortion caused by the relative value of activities, it is expressed
as a fraction of BCWS:

SV
SV % = , (5)
BCWS
CPI – Cost Performance Index – compares the planned and actual value of works
done, if less than 1, it indicates that the project has consumed more money than
planned, if greater than 1, there have been savings.

BCWP
CPI = , (6)
ACWP
SPI – Schedule Performance Index – compares the planned cost of works done with
planned cost of works planned; if less than 1, it indicates a delay:

BCWP
SPI = . (7)
BCWS

2.3. Earned Value “forecasting” parameters

EAC – Estimate at Completion – is calculated at the date of reporting progress to
serve as an estimate of the effect of deviations cumulated from the project’s start on
the total project cost, so it informs how much the project is going to be in the end,
if the cost performance index CPI stays the same:

BAC
EAC = . (8)
CPI
It is clear that EAC is a simple linear extrapolation of current tendencies. It
does not allow for any future risks or effects of corrective measures, so it is not
a proper forecast. Nevertheless, EAC indicates the potential scale of cost problems.
As the Earned Value method requires frequent progress checks from the very begin-
ning of a project, an early EAC-based constatation that current tendencies are likely
to double the cost are likely to provide a valuable warning signal and trigger rectify-
ing actions when it is still time.
EAC is not necessarily based on the assumption that future costs are going to
follow the today’s pattern. Other scenarios can be considered but, as the method
18 Agata Czarnigowska

rests upon a simplified model of a project, linear extrapolation is a rule and it proves
to be adequate [3]. The general EAC formula allows for a number of simple scenar-
ios:
BAC - BCWP
EAC = ACWP + , (9)
PF
i.e. EAC is a sum of costs already committed and the reminder of the budget adjusted
by a factor (PF) that reflects the relationship between the project’s future and its
past. This can be project-specific. Scenarios considered most often are as follows:
1. the cost of remaining task is going to be as planned, i.e. future costs are not
related to current costs, PF=1, so:
EAC = BAC + CV ; (10)
2. the cost of remaining tasks is going to stay in proportion to current CPI as
in equation (8); it ignores the real-life time-cost relationship (if a project is
to be accelerated, it usually requires more money);
3. the cost of remaining tasks will be related to current tendencies of both
schedule and cost performance, so the PF is a Critical Ratio (CR), called also
a Schedule Cost Ratio (SCR):
SCI = CPI × SPI . (11)
Another measure used for forecasting (or rather for simple extrapolation) is
TCPI (To Complete Performance Index) – a value of cost performance index that is
to be maintained from now on if the project is to be completed to budget. In other
words, TCPI is a proportion between the remaining work (expressed in terms of
budgeted costs) and the money left from the budget:
BAC - BCWP
TCPI = . (12)
BAC - ACWP

If it is much higher than 1 and the current CPI, it indicates the scale of effort
needed for searching for economies.

3. Application of earned value – example

Figure 2. presents a fragment of a construction project’s schedule – a number
of tasks of the first stages of the project. Table 1. contains input data collected by
the end of the seventh week of the schedule and status indicators calculated on their
basis. Inputs are marked grey:
– a budget for each tasks and for the whole project (BAC),
– percentage complete (PC) of each task as measured or estimated during
inspection (it is represented by black lines over schedule bars and values
above them),
– budgeted costs of works scheduled up to the end of week 7 (BCWS) (e.g.
according to the schedule, the task “Piling” should have been completed, so
its BCWS=BAC, but the task “Slab 1” should be about 66% complete, so its
BCWS=0,66·BAC),
– actual costs of works performed (ACWS).
 Earned value method as a tool for project control 19

Fig. 2. Schedule to serve as basis for calculations – project “Factory”.

Table 1. Earned Value calculations – project “Factory”, end of week 7.

Task BAC PC BCWS BCWP ACWP SV CV CPI SPI EAC TCPI

Factory 5000 8,50% 713,33 425,00 445,00 -288,33 -20,00 96% 60% 5235,29 100,44%

Piling 120 100% 120 120 120 0,00 0,00 100% 100% 120,00 -

Earthworks 50 100% 50 50 60 0,00 -10,00 83% 100% 60,00 -

Pile raft and f. 100 80% 100 80 75 -20,00 5,00 107% 80% 93,75 80,00%

Sub. walls 220 50% 220 110 110 -110,00 0,00 100% 50% 220,00 100,00%

Slab 1 260 25% 173 65 80 -108,33 -15,00 81% 38% 320,00 108,33%

Ground f. walls 150 0% 50 0 0 -50,00 0,00 - 0% - 100,00%

… … 0% 0 0 0 0 0 - - -  

The progress of the whole project (PC) is calculated according to formula (1):

BCWP 425
PC = = = 8, 50% .
BAC 5000
BCWS, BCWP, ACWP and variances SV and CV for the whole project are sums
of values of all tasks. The project’s CPI and SPI are calculated on the basis of its
BCWS, BCWP and ACWP.
SV i SPI indicate that, by the end of week 7, the tasks “Pile raft...”, “Sub. walls”
and “Slab 1” have not been completed to the planned extent, so they are delayed.
Therefore, the whole project is behind schedule in terms of scope of works i.e. less
work has been done than it was planned.
CV i CPI indicate that “Earthworks” and “Slab 1” proved more expensive than
planned. Some savings on “Pile raft...” compensated this additional cost only in
part, so the project as a whole is over budget.
If the project was to be finished on budget, it would have to be continued with
an improved cost performance index (TCPI). If the project was to proceed with
current CPI, its total costs would be EAC.
20 Agata Czarnigowska

4. Practical problems with using earned value

4.1. Determination of percentage complete (PC)
It is possible to determine the exact progress of a task only if its result is meas-
urable. For instance, a percentage complete of a task “brick walls” can be calculated
by dividing the area of walls measured on site and the area that was planned to be
built by the date of progress check; it can be assumed that costs are in proportion to
quantity. Usually, direct measurements are possible only for tasks (or works) of the
lowest (most detailed) levels of work breakdown structure.
The progress more complex tasks (such as “finishings of second floor”) or tasks
that do not have tangible or measurable results (as “quality inspection” or “client
briefing”) has to be estimated. The problem is that such estimates can be biased. The
literature on the subject prompts a number of solutions [1,4], e.g.:
– an assumption that a task is 0% complete until it is done, than it becomes
100% complete (a rather pessimistic and “safe” approach, but it undervalues
partially completed tasks);
– applying a fixed percentage at the beginning of the task (10% or 20% or
50%) and keeping it until the task ends, then it becomes 100% complete;
– “eyeball” assessment i.e. subjective judgement – as it is better to make
a rough estimate of a progress than to give up progress monitoring at all.
All the above approaches cause no distortion to the overall picture of the
project progress only if applied at low level of work breakdown structure (tasks of
duration much lower than the total project duration).

4.2. Accessibility of actual cost data (ACWS) and basis for budget
Decision on what costs are to be included in Earned Value analysis is crucial
for the project control.
Let’s assume that the manager wants to use actual expenditures. There is but
a certain lag between a task and related expenditure: labour is paid monthly or
weekly, terms of payment for materials are various. Another difficulty – how to
treat expenditures on tasks that have not started yet? They could not be included
in ACWP, so additional work would be required to separate them. It would be also
difficult to assess actual costs on the basis of invoices – they can be delayed, and
they are likely to concern money spent on a whole delivery or service that may be
used for a number of tasks. To sum up – it would be counterproductive to process
book-keeping data to get ACWS [6].
Another problem is the scope of costs to be analysed. As a budget is based
usually on a bid, it contains some costs that are assigned to tasks artificially – by a
percentage rate (overheads). It would be advisable to extract them and create sepa-
rate tasks. The cost model would be thus more precise and costs easier to control. It
is also important to remember that the contractor’s BAC, BCWS, BCWP and ACWP
do not contain profit and contingencies – it is cost that is to be analysed, not reve-
nues. If the budget was mechanically copied from the bid (and contained profit),
and ACWP was recorded as cost only, the analysis would lead to false conclusion
that there have been savings.
 Earned value method as a tool for project control 21
From the point of a building contractor it would be convenient to use direct
cost as incurred (so e.g. material costs as materials have been built-in). In practice,
instead of considering full costs or all direct costs, labour cost can be used for creat-
ing the budget for progress control purposes. Certainly, it is advisable only in the
case of labour-intensive projects where all tasks require using it. Labour costs are
directly measurable and the analysis would be simplified. Alternatively, planned
man-hours could be used instead labour costs [1, 6]
For the reasons mentioned above, the Earned Value’s “S-curves” are not the
same as those used for project cash flow analysis. This certainly does not facilitate
combined interpretation of project performance in terms of work progress, receipts
and expenditures.

4.3. Inferences on future schedule development on the basis

of SV i SPI
It is worth remembering that SV ($) and SPI (%) are considered to be measures
of time deviation, but they are in fact the measure of difference between the cost
of work planned and work done. In practice they are only indirectly and approxi-
mately related to the project’s time scale.
The Earned Value model of a project is linear. This does not distort cost vari-
ances and EAC estimates, as total cost is a simple sum of costs of tasks. Regardless of
the task relationships and timing, if each task cost raised by, say, 5%, the total cost
would be also 5% greater than the initial budget. Therefore, equation (8) correct.
As for schedule, the problem is more complex due to precedence relationships
between tasks. As there are usually both critical and non-critical tasks in the sched-
ule, a delay of a single task does not necessarily mean that the project finish date is
going to be affected. The earned value model does not allow for this fact and each
task deviation is reflected in the project’s SV and SPI. Similarly, if some tasks were
behind schedule, and some ahead, the overall SV might equal 0 and indicate that the
project was on schedule, even if the schedule network logic was seriously affected.
Another drawback of the simplified timing matters is related to the follow-
ing fact: completed tasks cease to affect the project’s schedule variance – their
SV=0 (SPI=1) and BCWP becomes equal to BCWS, no matter if the tasks have been
delayed or shifted to earlier of later dates. So, there are two interpretations of a task’s
SV=0 (SPI=1): the task is according to plan or it is fully completed. This quality is
shared by single tasks and the whole project (see Fig. 1: at the end of the project,
BCWP=BCWS=BAC). Therefore, at a certain stage of project development (usually
the last third of its duration) SV and SPI cannot be considered a reliable measure
of schedule deviation – either in terms of time or in terms of completed scope of
works [7].
And yet another shortcoming of the Earned Value’s simplified model: in
general, the graphic representation of a real-life project’s cumulated cost against
time is a complex irregular curve, but the Earned Value model is linear. Thus, the
assumption that the time variance stays in proportion to SV is wrong [8]. This is
illustrated by Figure 3.
22 Agata Czarnigowska

Fig. 3. Non-linear time-cost relationship – no correlation between size of delay and schedule vari-
ance.

An interesting illustration of this fact has been presented by Evensmo and

Karlsen [5]. The problem they presented would be though more approachable if
reversed: let’s assume that a project’s development has been already recorded (so the
actual duration T ‘ and final costs EAC are known). Let’s conduct an ex-post analysis
of accuracy of SPI-based, linear “prediction” of actual duration, by analogy to EAC
estimates (equation 8):

T
T'= . (13)
SPI

4.4. Example – issues of CPI and SPI-based estimates of project’s

actual cost and duration
Let us compare two variants of a project development:
a) BCWS, BCWP and ACWP stay in close relation and are described by a poly-
nomial of 3rd degree representing a symmetrical S-curve [5]:

BAC 2 3
BCWS = 2
(3t 2 - t ) , (14)
T T

BAC 2 3
BCWP = 2
(3t 2 - t ) , (15)
T' T'

EAC 2 3
ACWP = (3t 2 - t ) . (16)
T '2 T'
The parameters are: BAC=$1000 tho., T=10 months (planned duration),
EAC=$1200 tho. (actual total cost), T’=12 months (actual duration) (Fig. 4, Tab. 2).
b) BCWP and ACWP are random, but close to BCWS. BCWS is an S-curve as
in a) (Fig. 5, Tab. 2).
 Earned value method as a tool for project control 23

Fig. 4. Development of project a). Fig. 5. Development of project b).

Table 2. Earned Value calculation - example.

project a)
t BCWS BCWP ACWP CPI SPI EAC T’
0 0 0,00 0,00 - - - -
1 28 19,68 23,61 0,83 0,70 1200 14,23
2 104 74,07 88,89 0,83 0,71 1200 14,04
3 216 156,25 187,50 0,83 0,72 1200 13,82
4 352 259,26 311,11 0,83 0,74 1200 13,58
5 500 376,16 451,39 0,83 0,75 1200 13,29
6 648 500,00 600,00 0,83 0,77 1200 12,96
7 784 623,84 748,61 0,83 0,80 1200 12,57
8 896 740,74 888,89 0,83 0,83 1200 12,10
9 972 843,75 1012,50 0,83 0,87 1200 11,52
10 1000 925,93 1111,11 0,83 0,93 1200 10,80
11 1000 980,32 1176,39 0,83 0,98 1200 10,20
12 1000 1000,00 1200,00 0,83 1,00 1200 10,00
project b)
t BCWS BCWP ACWP CPI SPI EAC T’
0 0 0 0 - - - -
1 28 18 20 0,90 0,64 1111,11 15,56
2 104 50 75 0,67 0,48 1500,00 20,80
3 216 155 180 0,86 0,72 1161,29 13,94
4 352 248 450 0,55 0,70 1814,52 14,19
5 500 450 598 0,75 0,90 1328,89 11,11
6 648 788 682 1,16 1,22 865,48 8,22
7 784 876 866 1,01 1,12 988,58 8,95
8 896 899 950 0,95 1,00 1056,73 9,97
9 972 920 1000 0,92 0,95 1086,96 10,57
10 1000 948 1097 0,86 0,95 1157,17 10,55
11 1000 989 1174 0,84 0,99 1187,06 10,11
12 1000 1000 1200 0,83 1,00 1200,00 10,00

The above project models are purely hypothetical and created only to illustrate
the key issues of the method. In fact, any curve would produce similar results. Let’s
assume a) and b) are construction projects and models are created for the needs of
a contractor:
24 Agata Czarnigowska

a) Can be interpreted as follows: after the baseline (budget) had been created,
labour costs rose. At the same time, the workers opposed to working overtime,
which had been assumed while creating the schedule, so the daily output was lower
than planned. However, the manager seemed to reject the obvious facts that the
plan was no longer valid, and while monitoring progress, he kept referring to the
initial baseline.
b) Could be a situation where prices fluctuated (so cost of e.g. materials
changed) and so did labour productivity (e.g. experienced workers came and went),
or wrong decisions were taken to ameliorate variances, or the baseline was wrong.
Analysing CPI and EAC development (Tab. 2, Fig. 6 and 7) one may conclude
that:
– in the case of project a) CPI was stable and EAC estimate was constant and
correct – no surprise as the progress curves reflected the logic of the baseline,
– in the case of project b) cost change tendencies were variable (alternative
savings and cost overruns), so did the consecutive extrapolations of the final cost.
There was little relationship between the project’s future and its past. However,
towards the end of the project the estimate approached the actual cost, so the
Earned Value cost estimate was not entirely wrong.

Fig. 6. CPI changes with the project progress, Fig. 7. Variability of assessment of final project
projects a) and b). cost – projects a) and b).

As for SPI (Fig. 8) and T ‘ estimates (Fig. 9):

– Project a): there was a continuous growth of SPI. According to the classic
Earned Value interpretation, it should indicate improving productivity. However, it
is not logical as the distance between planned (BCWS) and actual progress (ACWP)
graphs along the time axis grew constantly towards the later actual finish. In conse-
quence, the linear extrapolation of total duration decreased constantly – and the
final estimate was the planned duration (T), not the actual duration (T ‘). Therefore,
SPI provides misinformation in terms of project timing.
– Project b) SPI was highly variable (one could expect that as the graphic repre-
sentation of the progress intersected with the plan several times). Accordingly, the
linear extrapolation of total duration fluctuated, and the final estimate was also
the planned duration (T), not the actual duration (T ‘) (the same misleading result
attributable to BCWS=BCWP as the project finishes).
 Earned value method as a tool for project control 25

Fig. 8. SPI changes with the project progress, Fig. 9. Variability of assessment of project dura-
projects a) and b). tion – projects a) and b).

The above proves that using SPI and linear extrapolation to predict total dura-
tion is wrong.

5. Modification of the method to allow for time analysis

5.1. Need for simple tool to asses progress in the terms of time
The literature on the subject argues that Earned Value method cannot (and
has not been intended to) substitute more detailed analyses of project schedules
and costs. However, as it proved useful in cost management, a similar simple tool
for schedule analysis is being looked for. Recent development of Earned Schedule
method [9,10,11] is considered very promising, though it has not joined Earned
Value on the list of classic management tools yet.
Earned Schedule uses the same inputs (BCWS, BCWP, T) as Earned Value, and
just like in the case of the latter method, S-curves are considered to be adequate
(though simplified) models of a project’s time development. However, variances are
to be calculated “horizontally” and expressed in time units, so instead of SV, there
is SV(t) to be calculated (Fig. 10). SV(t) is to inform directly on how many time units
the project is ahead or behind the plan. There had been several attempts to express
schedule variance in the terms of time [7]:
1. By dividing SV by the average planned progress rate (budget divided by
planned duration):

T
SV (t ) = SV × , (17)
BAC
which measures the deviation in time units, but does not account for time deviation
not being proportional to the Earned Value schedule deviation (Fig. 3). SV(t) calcu-
lated this way has the same deficiencies as SV.
2. On the basis of actual time from the project start to the moment of analysis
(t) and the Earned Value’s SPI; deficiencies as above:
SV (t ) = t × (SPI - 1) , (18)
26 Agata Czarnigowska

3. As geometric distance between BCWS i BCWP measured horizontally along

the time scale; as BCWS is seldom described by a mathematic function, it is not
easy to calculate this distance; the following iterative formula has been proposed
by Lipke [9,10]:

BCWP - BCWS N
SV (t ) = N + - t , (19)
BCWS N +1 - BCWS N

where N is a number of units of time since the project starts up to the moment of
analysis which BCWS is lower than BCWP at the moment of the analysis, BCWP is
measured at the moment of analysis (t), BCWSN and BCWSN+1 are budgeted costs of
works scheduled at the Nth and at the (N+1)th unit of time.
A website (www.earnedschedule.com/Calculator.shtml) created by the author
of the third method provides spreadsheets for calculating SV(t) on the basis of BCWS
and BCWP data measured at consecutive units of time.
Having established the time variance SV(t) one can calculate the “earned
schedule” (ES), i.e. the period within that the works actually done should have been
ready according to the plan (see Fig. 10):
ES = t - SV (t ) . (20)

Fig. 10. Time deviation acc. to Earned Schedule method.

The method of “forecasting” the actual duration is similar to Earned Value

approach to extrapolating EAC (equation 8):

T - ES
T'=t + , (21)
PF
so actual duration is the sum of time already consumed (t) and the time that
remained from the planned duration adjusted by a factor (PF) depending on the
assumed relationship between the project’s current and future performance.
The relationships considered most often are [7]:
1. all tasks to follow are going to be according to schedule, PF=1:
T ' = T - SV (t ) , (22)
 Earned value method as a tool for project control 27
2. the rate of progress is going to follow current pattern and stay in proportion
to current SPI(t) (no time-cost relationship considered):

ES
SPI (t ) = , (23)
t

T - ES T
T'=t + = , (24)
SPI (t ) SPI (t )
3. the rate of progress is going to follow current pattern – in proportion to
current schedule cost index SCI(t) as a relationship between actual cost and speed
of works is assumed to exist:
SCI (t ) = CPI × SPI (t ) , (25)

T - ES
T'=t + . (26)
SCI (t )

5.2. Comparison of SPI and SPI(t)-based estimates

of total duration
An extension to the examples a) and b) of chapter 4.4. is presented in Table
3 and Table 4 accordingly. SV(t) has been calculated using equation (19), and T’
according to equation (24). SV is measured in money units and SV(t) in time units,
they are not the same variances measured in different ways.
Table 3 Project a): schedule deviations and duration estimates according to Earned Value and
Earned Schedule).

t BCWS BCWP SV SPI T’ acc. EV SV(t) ES SPI(t) T’ acc. ES

0 0,00 0,00 0,00 - - 0,00 0,00 - -
1 28,00 19,68 -8,32 0,70 14,23 -0,30 0,70 0,70 14,23
2 104,00 74,07 -29,93 0,71 14,04 -0,39 1,61 0,80 12,45
3 216,00 156,25 -59,75 0,72 13,82 -0,53 2,47 0,82 12,16
4 352,00 259,26 -92,74 0,74 13,58 -0,68 3,32 0,83 12,06
5 500,00 376,16 -123,84 0,75 13,29 -0,84 4,16 0,83 12,01
6 648,00 500,00 -148,00 0,77 12,96 -1,00 5,00 0,83 12,00
7 784,00 623,84 -160,16 0,80 12,57 -1,16 5,84 0,83 11,99
8 896,00 740,74 -155,26 0,83 12,10 -1,32 6,68 0,84 11,97
9 972,00 843,75 -128,25 0,87 11,52 -1,47 7,53 0,84 11,95
10 1000,00 925,93 -74,07 0,93 10,80 -1,61 8,39 0,84 11,91
11 1000,00 980,32 -19,68 0,98 10,20 -1,70 9,30 0,85 11,83
12 1000,00 1000,00 0,00 1,00 10,00 -2,00 10,00 0,83 12,00
28 Agata Czarnigowska

Fig. 11. SV ($) and SV(t) (months) compared – project a).

Fig. 12. SPI and SPI(t) compared – project a).

Fig. 13. Extrapolation of actual duration according to Earned Value and Earned Schedule -
project a).

The results of calculations speak in favour of Earned Schedule: its schedule

deviation SV(t) reflects the logic of project development: as graphic representations
of the actual progress is steadily carried away from the plan, the time variance SV(t)
grows and there is no distortion towards the project finish date – quite contrary to
SV development (Fig. 11). The discrepancy between consecutive T ‘ extrapolations
is also much lower and closer to the actual duration (Fig. 13).
Let us repeat calculations for project b) (BCWS being a symmetrical S-curve,
BCWP- an irregular curve):
 Earned value method as a tool for project control 29
Table 4. Project b): schedule deviations and duration estimates according to Earned Value and
Earned Schedule).

t BCWS BCWP SV SPI T’ acc. EV SV(t) ES SPI(t) T’ acc. ES

0 0,00 0,00 0,00 - - 0,00 0,00 - -

1 28,00 18,00 -10,00 0,64 15,56 -0,36 0,64 0,64 15,56
2 104,00 50,00 -54,00 0,48 20,80 -0,71 1,29 0,64 15,51
3 216,00 155,00 -61,00 0,72 13,94 -0,54 2,46 0,82 12,22
4 352,00 248,00 -104,00 0,70 14,19 -0,76 3,24 0,81 12,36
5 500,00 450,00 -50,00 0,90 11,11 -0,34 4,66 0,93 10,72
6 648,00 788,00 140,00 1,22 8,22 1,04 7,04 1,17 8,53
7 784,00 876,00 92,00 1,12 8,95 0,82 7,82 1,12 8,95
8 896,00 899,00 3,00 1,00 9,97 0,04 8,04 1,00 9,95
9 972,00 920,00 -52,00 0,95 10,57 -0,68 8,32 0,92 10,82
10 1000,00 948,00 -52,00 0,95 10,55 -1,32 8,68 0,87 11,52
11 1000,00 989,00 -11,00 0,99 10,11 -1,39 9,61 0,87 11,45
12 1000,00 1000,00 0,00 1,00 10,00 -2,00 10,00 0,83 12,00

Fig. 14. SV ($) and SV(t) (months) compared – project a).

Fig. 15. SPI and SPI(t) compared – project a).

30 Agata Czarnigowska

Fig. 16. Extrapolation of actual duration according to Earned Value and Earned Schedule -
project a).

There seems to be a correlation between SV and SV(t) – the trends of growth

and decrease occur at the same points of time until the last third of project duration,
when SV curve starts to stray away and show wrongly an improving trend. Such
correlation in initial phases of a project has been generally confirmed [7]. Also in
this case, the Earned Schedule fluctuations of consecutive T ‘ extrapolations are also
much lower and closer to the actual duration then those extrapolated by Earned
Value metrics. (Fig. 16).

5.3. Implementing Earned Schedule

The Earned Schedule approach, though better suited for assessing schedule
status than Earned Value, rests upon the same simple model of a project: a time-cost
S-curve. The method’s calculations can be done on any level of a work breakdown
structure (from simplest activities to project stages) in exactly the same way. For the
sake of accurate progress estimates (PC), it would be advisable to analyse the project
on activity level (see chapter 4.1.)
However, an S-curve is a result of schedule logic. On the level of activities,
projects tend to have many paths, some of them critical, other not. A tiny (if
compared to total project duration) schedule variance of an activity can become a
cause for substantial rearrangements in the project network, it can change critical
path or order of activities, enforce a total reorganisation of works. This cannot be
captured by either Earned Value or Earned Schedule calculations - potential prob-
lems might be masked by compensating positive and negative schedule deviations.
If the project was “zoomed out” so the objects of analysis were not activities,
but larger complex tasks, the model might be reduced to one path and thus simpli-
fied. Within a complex task, schedule problems should be assessed by scheduling
tools, and Earned Schedule method might be used to capture a more general picture
of the project.

6. Conclusions
Earned Value, despite its being based on a much simplified model of a project,
has been widely used in practice. An analysis of numerous projects the method was
used at led to interesting conclusions: there occur to be statistically confirmed rules
of project development [12]:
 Earned value method as a tool for project control 31
– the final cost variance (VAC=EAC-BAC) will be greater than cost variance
at 20% of project development, so early negative tendencies are practically
irreversible;
– after the project becomes about 20% complete, CPI stabilises: usually, it
does not change more than by 10%, and in most cases its changes are for
worse;
– the CPI-based EAC proves to be a reliable lower estimate of final cost over-
run.
Nevertheless, the above rules have been established on the basis of US Depart-
ment of Defence projects that have been assessed and managed according to specific,
formalised and uniformed procedures.
Many organizations worldwide adopted Earned Value as a standard manage-
ment tool (e.g. US Department of Defence [13], an Australian standard [14]). It is
described in practically all management handbooks and incorporated into manage-
ment software. However, if to be implemented, the method should be used accord-
ing to its purpose: it is not a tool for forecasting; instead, it facilitates progress
monitoring, determination of project status (on time? to budget?), identification of
potentially negative occurrences and a rough estimate of their combined effect on
the project’s outcome. If the project is to be managed consciously, these occurrences
should be then investigated into by means of more accurate methods.

References
[1] Burke R., Project Management. Planning and Control Techniques, John Wiley & Sons,
2006.
[2] Dałkowski B.T., W trosce o publiczne pieniądze. zarządzanie projektami metodą
Earned value, Materiały III Konferencji Project Management, www.spmp.org.pl/
files/3dalkowski1.pdf.
[3] Christensen D.S., Using performance indices to evaluate the estimate at completion, Jour-
nal of Cost Analysis and Management, 1994 (Spring):17-24.
[4] Ruskin A. M., Two issues concerning the use of Earned Value Measurements, Engineering
Management Journal, 2004: 16(3) s. 26-30.
[5] Evensmo J., Karlsen J.T., Earned Value Based Forecasts – Some Pitfalls, 2006 AACE Inter-
national Transactions.
[6] Cass D.J., Earned Value Programs for US Dept. of Energy Projects, Cost Engineering
Vol 42 No 2 2000, s. 24-43.
[7] Vandevoorde S., Vanhoucke M., A comparison of different project duration forecasting
methods using earned value metrics, International Journal of Project Management, 2006:
24, s. 289-302.
[8] Corovic R., Why EVM Is Not Good for Schedule Performance Analyses (and how it could
be…), The Measurable News, Winter 2006-2007, www.earnedschedule.com/papers.
[9] Lipke W., Henderson K., Earned Schedule - an emerging enhancement to EVM, www.
pmicos.org/topics/EVMDEC07.pdf.
[10] Henderson K., Earned Schedule A Breakthrough, Extension to Earned Value Management,
Proceedings of PMI Global Congress Asia Pacific, January 2007 www.earnedschedule.
com/papers.
[11] Van De Velde R., Time Is Up: Assessing Schedule Performance with Earned Value, PM
World Today - October 2007 (Vol. IX, Issue X). www.earnedschedule.com/papers.
32 Agata Czarnigowska

[12] Christensen D., Templin C., EAC evaluation methods: do they still work?, Acquisition
Review Quarterly 9 (2002) s. 105-116.
[13] DOD (1997), Earned Value Management Implementation Guide, Washington: United
States of America Department of Defense.
[14] AS 4817-2006, Project performance measurement using Earned Value.
 The Measurable News
The Magazine of the College of Performance Management Issue 3, 2012

Speculations on Project Duration Forecasting

Walt Lipke
PMI® Oklahoma City Chapter

ABSTRACT – Project duration forecasting has been enhanced with the introduction and application of the techniques
derived from Earned Schedule (ES). The computed forecast results from ES have been shown to be better than any
other EVM-based method using both real and simulated performance data. However, research has shown that as
the topology of the network schedule becomes more parallel, the accuracy of the ES forecast worsens. This paper
examines a possible approach for overcoming the dilemma to further improve the effectiveness of ES forecasting.

Introduction
Earned Schedule is a measure of time duration indicating how much of the Earned Value
EVM World 2013
TM

Management performance baseline has been completed. Having the measure, allows for the
creation of schedule performance efficiency. The ES schedule performance index, SPI(t) is equal THE PROJECT PERFORMANCE MANAGEMENT CONFERENCE
to ES divided by AT, the duration from the project start to the status point [Lipke, 2003]. The
concept is illustrated in figure 1 (page 4). Join us for the 29th
Annual International
The derived schedule efficiency, SPI(t), is utilized to forecast project duration though the simple Conference with
formula [Henderson, 2004]: Keynote Speaker

IEAC(t) = PD / SPI(t) Howard Dean

where IEAC(t) = Independent Estimate at Completion (time units) May 29-31, 2013
PD = Planned Duration Naples, FL
The forecasting capability of the formula has been shown to be reasonably good. It has been
verified by simulation and application to real project data. A comprehensive examination of the
capability of two Earned Value Management (EVM) based methods and ES was made by the CONTENTS
research team of Vanhoucke and Vandevoorde, who applied schedule simulation techniques
Speculations on Project Duration
for assessing project duration forecasting performance [Vanhoucke & Vandevoorde, 2007]. The
Forecasting
conclusion from their work indicated, “The results …confirm…that the Earned Schedule method By Walt Lipke ........................................ cover
outperforms, on average, the other forecasting methods.”
Accepted Standards and Emerging
The following year, real data from 16 projects were used for comparing the EVM time conver- Trends in Over Target Baseline (OTB)
sion forecasting methods to ES [Lipke, 2008]. The comparison was constructed to determine Contracts
whether the four EVM methods, as an aggregate, produce better forecasts than does ES. The By Simon Dekker ....................................... 10
analysis strategy segregated the project data into seven ranges of percent complete in order to
Earned Value Management: Adapted for
isolate possible forecasting characteristics or tendencies among the methods. Conclusively, ES
use in Underground Mining Operations
was shown to be the best method of forecasting project duration. By Ridwan Wibiksana ................................ 16

This evidence is compelling for applying ES forecasting when EVM is employed for project con- Customize your Independent Estimate at
trol. However, recent research has demonstrated that the topology of the schedule has impact Completion (IEAC) Formula
on the “goodness” of the forecast. The ES forecasts are more accurate for schedules which are By Bill Mendelsohn .................................... 20
more serial and less so when parallel [Vanhoucke, 2009].
Vendors & Services ................................. 24
In turn, this deficiency has lead practitioners and researchers to seek additional techniques
for schedule control when schedule topology is predominantly parallel. The approach recently
examined combines two techniques, ES forecasting and Schedule Risk Analysis (SRA) [Van-
houcke, 2012]. The combination has shown promise in the testing performed on both simulated
and real data. However, it does complicate the analysis and significantly add work to the project
control process. R.E.P. , and the Registered Education Provider logo are
CONTINUED ON PAGE 4
registered marks of the Project Management Institute, Inc.

College of Performance Management • 101 South Whiting Street, Suite 320 • Alexandria, VA 22304 • 703.370.7885 • 703.370.1757• www.mycpm.org
4
The Measurable News 2012, Issue 3

Speculations on Project Duration Forecasting - CONTINUED FROM COVER

Figure 1. Earned Schedule Concept

Possibly, there is a second approach to the analysis and project control Methodology
dilemma. If the ES deficiency for parallel topology schedules can be Using notional data, duration forecasts are computed for the various
overcome, more reliable and significantly improved forecasting can be serial paths to project completion. The forecasts are compared to the
expected. Achieving this, the project control process discussed previ- corresponding forecast for the total project. The results from the com-
ously may be simplified, as well. parisons are used to assess the validity of the following statements:

The idea for resolving the problem is to apply ES forecasting to the 1. The forecasts using the current LP are improved from those for
current longest path of the schedule. The remainder of the paper the total project
describes the theory and explores its application using notional data.
2. The total project forecast is the “lower bound” when LP > TP.
Theory
The fundamental idea for utilizing the longest path (LP) is that the ES 3. The current LP forecast overcomes the negative effect of parallel
forecast is from a schedule topology that is completely serial. As dis- schedule topology
cussed earlier, research has shown the best ES forecast is for a serial
schedule. Thus, the ES forecast should be improved if the LP, as it To perform the forecasting calculations the ES (special case) calculator
evolves in the project execution, reflects the likely duration outcome. is used.c The calculator takes into account periods for which no work is
planned. This capability is needed because the tasks begin at various
The concept of the current LP is an extension of the planned critical times during the project execution.
path (CP). The current LP is the longest duration from among the paths
remaining to be executed from the present status point. The longest Project Data
duration is determined by applying ES forecasting to each remaining The notional data used for the analysis is displayed in figure 2. There
serial path. This methodology has been described in the literature for are ten tasks in the project. For each task the periodic values for PV
comparison of planned CP performance to the total project [Lipke, and EV are shown. It is assumed that the precedence relationships
2006]. Fundamentally, the remaining Planned Value (PV) for the serial between the tasks are finish to start.
path examined is used as the Performance Measurement Baseline
(PMB). The Earned Value (EV) accrued for the tasks on, say, path m is Knowing the precedence, the reader should be able to identify various
used with its PMB to calculate the path earned schedule, ESm, and the serial paths to completion. For example, the planned CP includes tasks
associated duration forecast.a 1, 4, 8 and 10, symbolized by 1-4-8-10. It should be noted that task 9
does not feed into task 10. Thus, Task 9 is another outcome which has
The longest duration forecast from the remaining executable paths the potential of becoming the LP.
may be longer than the forecast made for the total project (TP). It can
be deduced that when LP > TP there must be shorter paths included
in the total project. Assuming this is true, it answers a question posed
not long after the creation of ES, “Is the duration forecast from ES, the
“lower bound?”b From deduction it is apparent, the duration forecast
for the total project is always optimistic when the forecast has this a) For a description of EVM and its terminology reference [Project Management Institute, 2011]
relationship, LP > TP. b) The idea of a “lower bound” comes from the research studies of the forecast of final
cost using EVM [Christensen, et. al., 2002].
c) The ES (special case) calculator (ES calculator vs1) is available from the Earned
The theory proposed is the LP forecast at each status point resolves
Schedule website (http://www.earnedschedule.com/Calculator.shtml). For more
the described limitations of the ES forecasts, thereby providing better information reference [Lipke, 2011].
and more direct information for project control.
 2012, Issue 3
The Measurable News 5

Figure 2. Notional Project Data

Figure 3. Path Performance

Analysis
The identification of various executable paths with their associated Two paths, 2-5-9 and 6-9 indicate completion two periods past the
task aggregated PV and EV is shown in figure 3. The cumulative values, planned duration of 10 periods. Thus, we know from simple inspec-
PVc and EVc, are the data used with the special case calculator to tion of the figure that execution of the planned CP (1-4-8-10) did not
compute ES for each of the paths at each status point. complete the project and that the longest path must have changed
during project execution.
To complete the understanding of figure 3 a brief explanation of the
symbol “XX” is needed. When used in the PV row, the XX indicates no Figure 4 contains the computed forecasts for all of the paths and the
work was planned for the period. In the EV row, the interpretation is total project. For the various paths the longest duration forecast for
the execution was delayed for that period. For example, performance each status period is identified in the chart by the lime color. It is clearly
was not planned to begin for path 2-4-8-10 until period 3, as shown in seen that the current LP was identical to the planned CP for only one
the PVp row. For performance path 2-5-9, it is observed that although performance period, period two. Path 7-10 indicated the current LP for
execution was planned to begin in period 3, it did not commence until periods 4 through 7, while from period 8 through project completion,
period 4. This is shown with XX in the EVp row for periods 1 through 3. period 12, the longest duration forecasts were for path 6-9.
6
The Measurable News 2012, Issue 3

Figure 4. Forecast Comparison

Figure 5. Standard Deviation Comparison

Figure 6. Longest Path vs Total Project Forecasts

 2012, Issue 3
The Measurable News 7

Results References:
A significant observation from figure 4 is that for every period the LP Christensen, D. S., D. A. Rees. “Is the CPI-Based EAC A Lower
forecast is greater than the forecast for the total project. This result Bound to the Final Cost of Post A-12 Contracts?,” Journal
supports the expectation postulated earlier in the theory section. Thus of Cost Analysis and Management, Winter 2002: 55-65.
it provides credence for the idea that the forecast for the total project is
the lower bound; i.e., it is consistently optimistic when LP > TP. Henderson, K. “Further Developments in Earned Schedule,” The
Measurable News, Spring 2004: 15-22
Figure 5 indicates variation from the actual duration using the standard
deviation. As shown, excluding periods 2 and 3, the standard deviation
Lipke, W. “Schedule is Different,” The Measurable News, Summer
for the current LP is fairly constant with respect to the mean value of
2003: 31-34
0.446. The standard deviation for the total project behaves differently.
It is significantly larger than the value for LP and, in general, improves
from a beginning value of approximately 1.60 to a project comple- Lipke, W. “Applying Earned Schedule to Critical Path Analysis and
tion value of 1.16. This comparison strongly suggests, forecasting is More,” The Measurable News, Fall 2006: 26-30
improved through the use of the current LP.
Lipke, W. “Project Duration Forecasting:Comparing Earned Value
Figure 6 provides a good visual supporting the improvement in ES Management Methods to Earned Schedule, CrossTalk, December
forecasting provided by using the current LP. As can be observed, the 2008: 10-15
variation of the LP forecast is reasonably uniform around the actual
duration, whereas the total project forecast has much more variation in Lipke, W. “Earned Schedule Application to Small Projects,” PM World
converging to the actual duration. Today, April 2011: Vol XIII, Issue IV

As discussed earlier in this section, two of the three statements out- Project Management Institute. Practice Standard for Earned Value
lined in the preceding methodology section have been demonstrated. Management, Newtown Square, PA.: PMI, 2011
However, the third (LP forecasting removes the effect of parallel
schedule topology) has not been shown through the exercise with the
Vanhoucke, M., S. Vandevoorde. “A Simulation and Evaluation of
notional data; performance was not examined for varying topologies.
Earned Value Metrics to Forecast Project Duration,” Journal of the
Even so, logically it is plausible because the LP forecasting is applied
Operations Research Society, October 2007, Vol 58: 1361-1374
to completely serial networks.
Vanhoucke, M. Measuring Time – Improving Project Performance
Summary and Conclusion
Using Earned Value Management, London: Springer 2009
The point of this paper is to demonstrate that ES forecasting is improved
by using the current LP. The results from application to the notional
data indicate achievement of the objective. Figures 5 and 6 illustrate Vanhoucke, M. “Measuring the Efficiency of Project Control Using
the improvement theorized, while figure 4 indicates the forecast for the Fictitious and Empirical Project Data,” International Journal of
total project may be considered the lower bound when the forecast for Project Management, February 2012, Vol 30, Issue 2: 252-26
LP is the greater of the two.
About the Author:
Although it is more complex than the normal application of ES, the Walt Lipke retired in 2005 as deputy chief of the Soft-
LP methodology is reasonably straightforward. The added complexity ware Division at Tinker Air Force Base. He has over
is thought to be a good trade-off for gaining the reliable information 35 years of experience in the development, mainte-
needed for project control. nance, and management of software for automated
testing of avionics. During his tenure, the division
A secondary improvement is that the LP forecasting may reduce the achieved several software process improvement
effort for the project manager and the EVM analysis staff. The recently milestones, including the coveted SEI/IEEE award
proposed combined approach of SRA with ES forecasting is indicated for Software Process Achievement. Mr. Lipke has published several
to be labor intensive. The two project control methods appear to articles and presented at conferences, internationally, on the benefits
require a significant amount of analysis and threshold establishment to of software process improvement and the application of earned value
successfully apply the combined methodologies. management and statistical methods to software projects. He is the
creator of the technique Earned Schedule, which extracts schedule
The results from the notional data example are compelling. However, information from earned value data. Mr. Lipke is a graduate of the USA
they are insufficient to say LP forecasting should be adopted and DoD course for Program Managers. He is a professional engineer with
employed without further examination and testing. It is recommended a master’s degree in physics, and is a member of the physics honor
that those with EVM data experiment using the methods described in society, Sigma Pi Sigma (SPS). Lipke achieved distinguished academic
this paper and report their results. For those researchers that have the honors with the selection to Phi Kappa Phi (FKF). During 2007 Mr.
capability to create schedules of various topology characteristics and Lipke received the PMI Metrics Specific Interest Group Scholar Award.
simulate task performance, you are challenged likewise to examine the Also in 2007, he received the PMI Eric Jenett Award for Project Man-
LP approach to forecasting. agement Excellence for his leadership role and contribution to proj-
ect management resulting from his creation of the Earned Schedule
Should LP forecasting become a topic of research and application, it is method. Mr. Lipke was selected for the 2010 Who’s Who in the World.
proposed that the method be referenced as “ES-LP.” The terminology
creates common language necessary for understanding.
PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

Assessing Earned Value Management

and Earned Schedule Forecasting

Walt Lipke
PMI Oklahoma City Chapter

Abstract
Recent research indicates cost and schedule forecasting from EVM data is improved
when the performance factor, PF = 1, is used. This paper uses a small set of real data
to examine the research finding, to either confirm or refute. As well, the application of
PF = 1 is employed in statistical forecasting; results are tested and compared to the
index method. Observations from the research and this study are made referencing
historical studies. Further research is encouraged on these topics, but with some
precaution when real data is used.

Introduction
The 2015 paper, “Empirical Evaluation of Earned Value Management Forecasting
Accuracy for Time and Cost” authored by Batselier and Vanhoucke, is the inspiration for
this article [Batselier et al, 2015]. Their paper is an impressively comprehensive
examination of forecasting from the use of Earned Value Management (EVM) data
taken from 51 projects, predominantly construction.

In the history of EVM and Earned Schedule (ES) research, covering 25 years for cost
and 15 years for schedule, one type of forecasting formula, incredibly, has been
ignored. Included in these past studies are several published by Christensen1,
Vanhoucke2, Crumrine3, and Lipke4. Uniquely, Batselier and Vanhoucke (B&V) examine
several methods of forecasting. B&V demonstrate overwhelmingly in their analysis this
ignored formula yields forecasts more often better than the ones most frequently
employed by EVM and ES practitioners.

This article, using a smaller set of data than that used by B&V, attempts to corroborate
their finding. The primary objective, however, is to implement the improvement shown
1
See references: Christensen et al, 2002(-1,-2), 1995, 1993(-1,-2)).
2
See references: Vanhoucke et al, 2007, 2006.
3
See reference: Crumrine et al, 2013.
4
See references: Lipke, 2017, 2016, 2015, 2014, 2012, 2011, 2009-1, 2006.

© 2017 Walt Lipke www.pmworldlibrary.net Page 1 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

for deterministic forecasting into statistical forecasting. The focus is to assess whether
the improved nominal forecast translates to better statistical forecasts. As well, the
investigation may reveal logical reason for the B&V results.

The subsections following, EVM & ES Forecasting, and Statistical Forecasting, provide
background for understanding the remainder of the article.

EVM & ES Forecasting

EVM and ES forecasting formulas are very similar. They each have the same basic
construct; i.e., the forecast is equal to the current value plus the remainder yet to
accomplish divided by a selected performance factor.

Before discussing the formulas, the following EVM and ES terminology is introduced in
table 1. It is assumed the reader has a fundamental understanding of EVM and ES. If a
more complete description is needed, please reference the following: Practice Standard
for Earned Value Management [PMI, 2011], and Earned Schedule [Lipke, 2009-2].

Table 1. EVM and ESM Terminology and Formulas

In the B&V paper several performance factors (PF) are examined for EVM and ES
forecasting. For this paper, only four are used. As depicted in table 1, cost applies 1 and
CPI, while schedule uses 1 and SPI(t). The reason only these are studied is to
corroborate the B&V finding that use of PF = 1 provides in most instances a better
forecast of the actual outcome than does the most often used cumulative value for the
performance indexes.

© 2017 Walt Lipke www.pmworldlibrary.net Page 2 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

For many years, it has generally been conceded that overall the performance indexes
provide the most reliable of the possible EVM and ES forecasting methods. Some
rationale for reliance on the cost index comes from Dr. Chistensen’s conclusion that CPI
tends to worsen as the project progresses toward completion [Christensen, 1993]. As
well, Christensen determined that forecasts using CPI are optimistic, which he termed
the “low bound” [Christensen et al, 2002-1). His research indicates that the forecast
using CPI will be better than PF = 1; i.e., the CPI forecast will be optimistic, but PF = 1
will be even more so. This comparative deduction may be the reason PF = 1 had not
been seriously examined prior to the B&V paper.

Statistical Forecasting

The use of statistical methods for inferring outcomes is a longstanding proven

mathematical approach. The statistical forecasting method for duration has been
demonstrated to perform reasonably well [Lipke et al, 2009-3].

The current statistical method of duration forecasting is derived from the ES equation,
IEAC(t) = PD / SPI(t), where using the cumulative value of SPI(t) yields the nominal
deterministic forecast. The associated high and low Confidence Limits5 (CL) are
computed from the variation of ln SPI(t)P, i.e., the logarithm of the periodic index values.
As well, the statistical forecasting method for duration is equally applicable to cost by
using the appropriate indexes.

Because B&V have shown PF = 1 to provide better forecasts, curiosity is raised

concerning its use in statistical forecasting. Thus, It is desired to adapt PF = 1
forecasting such that comparison can be made to the present method. The adaptation is
not difficult, but does need some explanation.

Let’s begin with the PFS = 1 duration forecasting expression:

IEAC(t) = AT + PD – ES

First, multiply and divide the ES term by AT. Then by arranging terms, the formula is
transformed to:

IEAC(t) = AT + PD – AT  (ES/AT)

5
Information about Confidence Limits may be found in [Crowe et al, 1960]. Confidence Limits are
sometimes misunderstood to be thresholds for management action. The limits, instead, describe the
region containing the “true” value of the parameter at the prescribed probability, i.e. Confidence Level.

© 2017 Walt Lipke www.pmworldlibrary.net Page 3 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

= AT + PD – AT  SPI(t)

This expression facilitates the statistical use of the cumulative and periodic index
values; the identical values used in the current statistical method. Thus, the forecast
confidence limits are computed using the index limits in the current method.6 That is, for
example, the high forecast limit becomes:

IEAC(t)H = AT + PD – AT  e^(CLL)

where e = the base number for natural logarithms

subscript H denotes the high confidence limit for the forecast
subscript L denotes the low confidence limit for the logarithm of the index

Analogously, the PFC = 1 formula for IEAC is transformed for statistical forecasting of
cost:

IEAC = AC + BAC – AC  CPI

Methodology

EVM data from 16 projects are included in the study. The project data comes from three
sources and is highly varied: two projects are information technology; twelve come from
high technology product development; two are construction type projects. The projects
range in duration from a few months to several years. There is no indication in the data
of reserves for cost or duration. Although it cannot be verified with certainty, it is
believed the projects have not undergone re-planning. The use of projects void of re-
planning and other anomalies such as stop work and planned down time, enables a
cleaner, less encumbered evaluation of the study results. Disturbances such as these
impact the computations and the subsequent analysis.

Utilizing the PF = 1 formulas derived for IEAC(t) and IEAC, the nominal and confidence
limit forecasts are computed for each project. The forecasts are then analyzed utilizing
four hypothesis tests, two each for schedule and cost forecasts. The hypothesis test
applied is the Sign Test [NIST, 2017]. The test is made for the null hypothesis, identified
as Ho. When there is insufficient statistical evidence to support Ho, the test result is the
alternate hypothesis, Ha.

6
For a more complete description of Confidence Limit calculations using EVM and ES data consult the
following reference [Lipke, 2016].

© 2017 Walt Lipke www.pmworldlibrary.net Page 4 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

The four hypothesis tests for evaluating the forecast confidence limits, expressed in the
form of the alternate hypothesis, are defined below:

1) H1: Final Cost is less than IEACH

2) H2: Final Cost is greater than IEACL
3) H3: Final Duration is less than IEAC(t)H
4) H4: Final Duration is greater than IEAC(t)L

It should be clear from the test definitions that the testing determines the likelihood that
the outcome value (final cost or final duration, as appropriate) resides between the
computed forecast confidence limits. Should the testing indicate the final value is likely
outside of the confidence limits, the statistical forecast is not considered reliable.

For each of the four tests, the test statistic is computed and compared to a significance
level (α) equal to 0.05.7 When the test statistic value is less than or equal to 0.05, there
is enough evidence to reject the null hypothesis. The test statistic for the Sign Test is
computed using the binomial distribution with each trial having a success probability of
0.5.

Results/Analysis

To verify that duration forecasting formula PFS = 1 produces, generally, better results
than PFS = SPI(t), Mean Absolute Percentage Error (MAPE)8 calculations were made.
As observed in table 2, of the 16 projects, the deterministic forecasts using formula PFS
= 1 had lower error for 12.

Recognizing that the PFS = 1 forecast is not always better, a limited investigation was
made to see if a combination of the two methods would yield results having less error.
For this, attention was shifted to cost forecasting.

To begin, recall Dr. Chistensen’s observation, cited earlier, that CPI generally
decreases as the project progresses. If this is true then it follows that, at some point in
project completion, index forecasting should converge to the final outcome faster than
PF = 1. As well, when CPI or SPI(t) forecasting is used, it is commonly observed that
computed results are volatile early in project execution. In fact, many analysts discount
the first 15-20 percent of the execution because they believe the EVM and ES indicators

7
Complete descriptions of the terms “test statistic” and “significance level” are available in mathematics
books of statistics [Crowe, et al, 1960].
8
MAPE = 1/n  ( |AD – Forecast(i)|/AD), where Forecast(i) is one of the n forecasts.

© 2017 Walt Lipke www.pmworldlibrary.net Page 5 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

are not reliable enough for making management decisions. Thus, it is reasonable to
believe PF = 1 forecasting should be superior in the early stages of the project.

Table 2. Comparison of Forecasting Accuracy

With these two thoughts, consideration was given to creating a composite forecast
using both forecasting formulas: PF = 1 for the initial two-thirds of project performance,
with PF = index for the final third. After several trials, the composite approach did not
produce improved forecasts. Although not nearly as comprehensive as the B&V study,
the investigation corroborated their finding of PF = 1 performing well in every partition of
project completion.

Having established for the 16 projects that PFS = 1 generally provides the superior
forecast it was thought that statistical forecasting may, likewise, show improvement in
comparison to the index method. The tabulation of the hypothesis test results for CLs
computed at 90 percent confidence level are presented in table 3. To assist with
interpreting the results, recall the general meaning of Ho and Ha:

1) Ha indicates the confidence limit encapsulates the final outcome

2) Ho indicates the outcome lies outside of the confidence limit

Examining the table, it is readily seen: the low CL encapsulates the final outcome for
both cost and schedule, whereas high CL generally does not. The low CL for cost had
the test result Ha for all 16 projects, while for schedule the low CL was observed for 14
projects. For the cost high CL, 15 of the 16 projects indicate the test result Ho. For
schedule, 9 of 16 have Ho results

© 2017 Walt Lipke www.pmworldlibrary.net Page 6 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

Table 3. Hypothesis Test Results

These results are tabulated as probabilities and shown in table 4. The numbers in the
table indicate the probability that the confidence limit encapsulates the final value. The
results shown for PF = CPI and SPI(t) come from a previous study [Lipke et al, 2009-3].
As readily seen, statistical forecasting using the indexes produces considerably more
reliable CLs.

Table 4. Comparison of Confidence Limit Probability

In the previous study (PF = index), 98 percent confidence level was examined with the
following resulting probabilities: Cost High CL = 0.927, Low CL = 1.000; Schedule High
CL = 1.000, Low CL = 0.997 [Lipke et al, 2009-3]. The consistency of the probability
values indicates the CLs are very reliable. For the present study (PF = 1), increasing
confidence level did not cause appreciable increase in probability. Thus, it is reasoned
the PF = 1 statistical forecasting is unreliable.

© 2017 Walt Lipke www.pmworldlibrary.net Page 7 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

Figure 1. Cost Forecast, PFC = CPI

Figure 2. Cost Forecast, PFC = 1

From these results it appears the CLs from PF = 1 forecasting are optimistically biased.
Visually, this can be deduced from graphs for cost and schedule, comparing the
statistical forecasts from each computation method. Figures 1 and 2 clearly illustrate the
optimistic bias of forecasting using PFC = 1, as well as showing PFC = CPI forecasting
yields more reliable CLs.

© 2017 Walt Lipke www.pmworldlibrary.net Page 8 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

Figure 3. Schedule Forecast, PFS = SPI(t)

Figure 4. Schedule Forecast, PFS = 1

Similar results are obtained for duration forecasting, using the same EVM data as was
used for the cost graphs, Figure 3 illustrates PFS = SPI(t) statistical forecasting, while
figure 4 shows PFS = 1. In general, both graphs have plots portraying optimistic
forecasts. However, the high CL for the PFS = SPI(t) does satisfy the hypothesis test
and graphically shows a feature seen consistently. There is a graphical component to
interpreting the statistical forecasts produced from PF = index. It is observed in both the
cost and schedule graphs; the most horizontal plot is generally a very good forecast of
the actual outcome. For the index graphs, figures 1 and 3, the most horizontal plots are
the nominal forecast for cost and high CL for schedule.

© 2017 Walt Lipke www.pmworldlibrary.net Page 9 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

Observations

Some exploration was made with notional data. The objective was to see if there is
something generally true about the two forecasting methods, PF = 1 and PF = index. If
a characteristic could be discovered, then possibly project managers would have
information as to when a particular forecasting formula should be applied.

The exploration was not very structured. Nevertheless, it did show that when the index
is constant, this method of forecasting is superior to PF = 1. However, as the variation in
performance increased, PF = 1 became the more accurate. Possibly, this is an area for
future study. There may be a variation value which demarcates regions for which each
forecasting PF produces its best forecasts.

This observation about variation led to reflection on how organizations handle EVM
data. Non recognition of re-plans, stop work, and down time can inflate index variation,
thereby causing index forecasting to appear worse than it should.

Table 5. Notional Data

© 2017 Walt Lipke www.pmworldlibrary.net Page 10 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

To illustrate this problem, a notional set of data was created. It is shown in table 5. The
PV, EV, and ES data have five highlighted entries. Each of these entries is a repeat of
the entry just prior. If all of the highlighted entries were removed, the planned duration
would be 10 periods with project completion occurring in period 20. It is fairly easy to
deduce with the yellow entries removed that SPI(t) = 0.5 and has no variation. In this
circumstance the index forecast of final duration is better than the PF = 1 forecast.

Now, let’s consider what these entries might be. Possibly each is a re-plan. Or, it could
be that each of the yellow PV entries is planned down time. Then, when the down time
occurred, conditions were such that it was not possible to accomplish work and, thus,
EV did not progress. When EV does not increase, neither does ES. For the remainder
of the discussion, let’s assume the entries describe down time and stop work.

Figure 5. Notional Data Forecasting Comparison

In the table, there are three deterministic duration forecasts: PD/SPI(t)C, AT + (PD –
ES), and IEAC(t)sp. As each method is discussed it may be helpful to view figure 5. The
figure graphically portrays the performance of the three methods.

The PD/SPI(t)C forecast is made by simply using the data strings of PV and EV without
regard to seeing a need for further review of the highlighted entries. The consequence
is the forecast values are erratic, yet the calculation converges to the actual duration.

© 2017 Walt Lipke www.pmworldlibrary.net Page 11 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

As well, the forecast method, AT + (PD – ES), does not examine the highlighted entries
and uses the ES calculated values to make forecasts. It, too, converges to the actual
duration. One observation is these forecasts are consistently optimistic.

Lastly, the IEAC(t)sp forecasts, a modified form of the index method, perfectly align with
the final duration. These forecasts are made using the ES Calculator (Special Cases)9.
This calculator takes into account down time and stop work. It filters through the
interruptions to make a better forecast. For this example, the special cases calculator
provided forecasting perfection; in general, improvement is expected when the
conditions of down time and stop work exist, but not perfection.

The take-away from this exercise is that real EVM data used in testing forecasting
methods needs close examination. If at all possible, data having re-plans should be
avoided. For projects having down time and stop work, the places in the data where
they occur need identification so that they can be handled appropriately.

Summary/Conclusion

Forecasts of project duration were made using real data from 16 projects. The forecasts
using performance factors, SPI(t) and PFS = 1, were compared using MAPE values; 12
of the 16 forecasts made with PFS = 1 were observed to be have less error with respect
to the final duration. This result is in agreement with the finding stated by B&V [Bastelier
et al, 2015]; i.e., forecasts using PFS = 1 are generally better. As well, a very limited
examination confirmed the B&V finding that PF = 1 performs well throughout the project.

With confirmation that PF = 1 forecasts generally produce more accurate results, its use
in statistical forecasting was explored. The examination revealed that the associated
confidence limits are unreliable for both cost and schedule. The CLs are optimistically
skewed. Thus, statistical forecasting with PF = 1 is not recommended.

Duration forecasting comparison was made using notional data which included down
time and stop work. Three methods were compared; two ignoring the conditions and
one recognizing them. The index method, PD/SPI(t), provided highly volatile pessimistic
forecasts. The PFS = 1 method was less volatile and consistently optimistic. The method
recognizing the conditions, IEAC(t)sp, yielded an accurate forecast. The significant point
derived from the exercise is real data needs to be closely examined and used

9
The Earned Schedule Calculator (Special Cases) is available from the Earned Schedule website
(www.earnedschedule.com).

© 2017 Walt Lipke www.pmworldlibrary.net Page 12 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

appropriately when performing forecasting studies. Otherwise, the study results are
suspect.

Suggested Research

In the limited investigations of this study it was observed, when the index is reasonably
constant, the deterministic forecasts were better than those made with PFS = 1. Thus,
there may be a demarcation value for the variation of ln SPI(t)P identifying which
forecasting method should be applied; i.e., below a specific value of variation the index
method is used and above it, PFS = 1 is preferred. It is suggested to researchers that
this area be investigated.

At present, the application of Earned Schedule-Longest Path (ES-LP) forecasting has

not been sufficiently tested. Possibly, the various forecasting formulas could be used
with ES-LP to explore further improvements to forecasting.

References:

Batselier, J., M. Vanhoucke (2015). “Empirical Evaluation of Earned Value Management

Forecasting Accuracy for Time and Cost,” Journal of Construction Management,
141(11): 05015010, 1-13

Christensen, D. S., D. A. Rees (2002-1). “Is the CPI-Based EAC A Lower Bound to the Final Cost
of Post A-12 Contracts?,” Journal of Cost Analysis and Management, Winter: 55-65.

Christensen, D. S., C. Templin (2002-2). “EAC Evaluation Methods: Do They Still Work?,”
Acquisition Review Quarterly, Spring: 105-116

Christensen, D. S., R. C. Antolini, J. W. McKinney (1995). “A Review of Estimate At

Completion Research,” Journal of Cost Analysis and Management, Spring: 41-62.

Christensen, David S (1993-1). “The Estimate At Completion Problem: A Review of Three

Studies.” Project Management Journal, Vol 24 March: 37-42.

Christensen, D. S., S. R. Heise (1993-2). “Cost Performance Index Stability,” National

Contract Management Journal, Vol 25: 7-15.

Crowe, E., F. Davis, M. Maxfield (1960). Statistics Manual, New York, NY: Dover Publications

Crumrine, K., J. Ritschel (2013). “A Comparison of Earned Value Management as Schedule

Predictors on DOD ACAT 1 Programs,” The Measurable News, Issue 2: 37-44.

Lipke, W (2017). “Forecasting Schedule Variance using Earned Schedule,” PM World

Journal (online), February, Vol VI, Issue 2

© 2017 Walt Lipke www.pmworldlibrary.net Page 13 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

Lipke, W. (2016). “Earned Value Management and Earned Schedule Performance Indexes,”
Wiley StatsRef: Statistics Reference Online, August,
DOI:10.1002/9781118445112.stat07891

Lipke, W. (2015). “Applying Statistical Forecasting of Project Duration to Earned Schedule-

Longest Path,” The Measurable News, Issue 2: 31-38

Lipke, W. (2014). “Testing Earned Schedule Forecasting Reliability,” PM World Journal (online),
July, Vol III, Issue 7

Lipke, W. (2012). “Speculations on Project Duration Forecasting,” The Measurable News, Issue
3:1, 4-7

Lipke, W. (2011). “Earned Schedule Application to Small Projects,” The Measurable News,
Issue 2: 25-31

Lipke, W. (2009-1). “Project Duration Forecasting …a comparison of Earned Value

Management methods to Earned Schedule,” The Measurable News, Issue 2: 24-31.

Lipke, W. (2009-2). Earned Schedule, Raleigh, NC: Lulu Publishing

Lipke, W., O. Zwikael, K. Henderson, F. Anbari (2009-3). “Prediction of Project Outcome:The

Application of Statistical Methods to Earned Value Management and Earned Schedule
Performance Indexes,” International Journal of Project Management, May Vol 27: 400-
407

Lipke, W. (2006). “Applying Earned Schedule to Critical Path Analysis and More,” The
Measurable News, Fall: 26-30.

National Institute of Standards and Technology. Sign Test. (2017)

http://www.itl.nist.gov/div898/software/dataplot/refman1/auxillar/signtest.htm

Project Management Institute. Practice Standard for Earned Value Management, Newtown
Square, PA.: PMI 2011

Vanhoucke, M., S. Vandevoorde (2007). “A Simulation and Evaluation of Earned Value Metrics
to Forecast Project Duration,” Journal of Operational Research Society, Issue 10 Vol 58:
1361-1374.

Vanhoucke, M., S. Vandevoorde (2006). “A Comparison of Different Project Duration

Forecasting Methods Using Earned Value Metrics,” International Journal of Project
Management, Issue 4 Vol 24: 289-302.

© 2017 Walt Lipke www.pmworldlibrary.net Page 14 of 15

PM World Journal Assessing EVM and Earned Schedule Forecasting
Vol. VI, Issue VIII – August 2017 by Walt Lipke
www.pmworldjournal.net Featured Paper

About the Author

Walt Lipke
Oklahoma, USA

Walt Lipke retired in 2005 as deputy chief of the Software Division at Tinker Air Force
Base, where he led the organization to the 1999 SEI/IEEE award for Software Process
Achievement. He is the creator of the Earned Schedule technique, which extracts
schedule information from earned value data.

Credentials & Honors:

 Master of Science - Physics
 Licensed Professional Engineer
 Graduate of DOD Program Management Course
 Physics honor society - Sigma Pi Sigma ()
 Academic honors - Phi Kappa Phi ()
 PMI Metrics SIG Scholar Award (2007)
 PMI Eric Jenett Award (2007)
 EVM Europe Award (2013)
 CPM Driessnack Award (2014)
 Australian Project Governance and Control Excellence Symposium established
the annual Walt Lipke Project Governance and Control Award (2017)

Mr. Lipke can be contacted at [email protected]

© 2017 Walt Lipke www.pmworldlibrary.net Page 15 of 15

 Why EVM Is Not Good for Schedule Performance Analyses
(and how it could be…)

By Radenko Corovic, MBA

Abstract
Project performance researchers are unanimous that the conventional EVM (Earned Value
Management) is a very good tool to calculate project cost performance. On the other hand, they
have much more divergent opinions about the EVM capacity to provide reliable schedule
performance indicators. Nevertheless, even though they recognise that the schedule
performance indicator – SPI is useless after the planned end of the project, most of them don’t
question its use and reliability over the first two thirds of the project. My researches show that
the schedule performance indicators are not reliable and are essentially erroneous over entire
project life cycle for most commercial projects with a non-linear cumulative cost curve.
But, getting the good performance indicators is not enough. In analyzing what to do with EVM
results, I propose a new performance indicator which, in my opinion, is very significant for project
managers.

The main objective of project performance analysis is to tell as how the project is performing
related to three principal project factors: scope (size), cost and time. It has to indicate how our
process of product development is performing, are we delivering our product (scope) on time
and within planned costs. We are practically measuring the effectiveness of our process by
measuring its attributes, efforts (cost) and time [1]. We are measuring how are we delivering the
product (instead in product scope, which is often very difficult to measure when the process is in
progress, a product is expressed in its other attribute, its value, either planned or earned)
comparing to the attributes of process – costs (cost performance analysis) and time (schedule
performance analysis)[1].
The fundamentals of EVM are based on this concept. EVM is supposed to indicate how our
process (project) is performing by measuring and comparing the costs (actual costs-AC) and
the time spent in order to produce a certain amount of a product (earned value-EV). While the
EVM works well when calculating cost performance, it is much more different when we talk
about schedule performance. Because of that, the cost performance is not a subject of this
analysis and the following article is focusing principally on time or schedule performance.

EVM - Schedule Performance

The EVM concept is graphically presented on two-dimension diagram. The planned as well as
the earned value of the product is presented as a curve, for which each point is determined by
the corresponding points on a vertical (cost) axis and on a horizontal (time) axis.
Cost performance indicators are based on the difference between earned value and actual costs
at a given time. All this is projected on a vertical (cost) axis. Following this logic, time (schedule)
performance should be calculated by measuring the value of product delivered at a given time
(earned value) and by comparing this time to the time when this value was supposed (planned)
to be delivered. The difference (called schedule variance) should be presented on a horizontal
(time) axis.
But, for some reason, the conventional EVM calculates schedule performance by comparing EV
delivered at a given time to the value planned to be delivered at the same time. Instead of
calculating and expressing the results as the amount of time and presenting them on a
 horizontal (time) axis, the conventional EVM expresses the result as value and presents it on a
vertical (cost) axis as we see in Figure 1.
Budget (cost)

Figure 1 – Conventional EVM Approach

This fundamental mistake is not just semantic. It has important implications not just on the unit of
measure in which results are expressed (in measure of value instead of measure of time), but
more importantly, it influences the schedule performance indicators in a way that make them
practically useless and often even misleading.
The first EVM indicator, the schedule variance (SV), is expressed in product value and it is not of
a great use for project managers. If you say to a project manager that his project is $500 000
late, you can expect a question: Yes, but how many months or days is it? Classical EVM cannot
provide a valid response.
The second EVM indicator, the schedule performance indicator (SPI), is a ratio between EV and
PV and at the end of the project, if the project has delivered all what was planned, EV and PV
must be equal. This characteristic makes SPI useless after the planned end of the project
because it tends to be 1 near to the project end. After the planned end of the project, planned
value remains constant while the earned value is supposed to grow until the real end of the
project. If your project is planned to be finished in 10 months, for example, and your EV after 10
months is 70%, you are obviously 30 % behind schedule. But, when you continue your project
(you have to finish it eventually), and for some reason you do nothing for three months, your
schedule performance will remain the same! At the end of the project your EV will be equal to
your PV which, according to EVM, means that you have finished your project on time even
though you are couple of months late. Great, isn’t it? But I don’t think that the client will share
your enthusiasm.

EVM Schedule Indicators Are Wrong

This weakness of EVM is already very well explained by many authors (Fleming and Koppelman
[2], Lipke[4], Henderson[3], Vandevoorde&Vanhoucke[8]) and even included in PMI Practice
Standard of EVM[6]. Most of them argue that the EVM is not a reliable predictor of project
duration and even recommend that EVM, relating to project schedule performance, should be
used just as warning mechanism [2] and not as a real tool to analyze how the project is
performing in time. Certain researchers maintain that SPI is good and reliable in first two thirds
of the project [3] [4] [8]) and it starts to be defective over the final third of a project's life cycle. All
this could give the impression that the classical EVM, even though imperfect, could be used to
calculate schedule performance over the first two thirds of a project.
However, my research shows that the conventional EVM schedule indicators are unreliable and
deficient even in the early project stages for projects in which a cumulative cost curve is not
linear (the vast majority of commercial projects are characterized by non-linear cumulative cost
curves).
To illustrate this, we will take one IT project for which cumulative costs are presented in Figure 2.
The planned project duration is 10 months and the budget at completion (BAC) is $1M. I
calculated the project schedule performance with conventional EVM indicators and I have
obtained the following results:

A) Calculated at June1st

EV 1= $233,000 Schedule variance (SV1) = EV- PV = $-450,000

PV 1= $683,000 Schedule Performance Indicator (SPI 1) = EV/PV= 0.34

Real delay 1 = 1.6 months1

B) Calculated at September 1st

EV 2= $ 775,000 Schedule variance (SV2) = EV- PV = $-183,000

PV 2= $ 958,000 Schedule Performance Indicator (SPI 2) = EV/PV= 0.81

Real delay 2 = 2.7 months

Cost

Figure 2 – EPM versus Earned Schedule

1
Real delay is calculated with Earned Schedule (ES) methodology which will be explained later.
We can see that in the first case (A), where real delay is 1.6 months, SV and SPI are showing
worst performance than in the second case (B) where real delay is 2.7 months. How this can be
explained?
It can be explained with the slope of the cumulative costs curve portion determined by the point
representing the planned value at a given time and the point on the PV curve when the actual
earned value should have been done. In Figure 2 those points are respectively PV1 and EV1’
(Case A) and PV2 and EV2’ (Case B). The slopes of these portions are determined by the
tangents lines (S1 and S2). If the slope of this portion is greater than the average slope of the
cumulative cost curve, the EVM schedule performance indicators will show the results which are
worse than they actually are. In the opposite case (slope of the measured portion smaller than
the average cost curve slope), the EVM schedule performance results will indicate that the delay
is smaller than in reality.
Figure 3 shows an example of a real IT project whose schedule performance was measured on
five consecutive occasions. Those measurements were performed with a tool capable of
automatically calculating EVM indicators by identifying the value on a PV curve corresponding to
the earned value at a given time and the time when this EV should have been .
600 000

PV
500 000

5
400 000
4
EV
3
300 000

2
200 000

100 000

0
Jan-05 Mar- May- Jul-05 Sep- Nov-05 Jan-06
05 05 05

Figure 3- Example of a real IT project measurement

The schedule performance indicators were calculated in conventional EVM as well as with
Earned Schedule (ES) methodology. The schedule variance (real delay) was also calculated
and the results were compared as shown in Table 1.
 Number of performance measures
1 2 3 4 5
SVt (Real Delay) in days 31 31 31 31 31
Earned Value (EV) ($) 96 000 145 000 195 000 331 000 382 000
SPI (EVM) 0.66 0.75 0.59 0.86 0.88
SPI (ES) 0.79 0.83 0.85 0.87 0.90

Table 1 – Comparative results of an IT project

In a case where the schedule variance is constant, as in our example, SPI should be gradually
growing and the increments should be smaller near to the project end. But, as we can see, even
though the schedule variance in all measurements is the same (31 days), we can see that SPI
calculated with conventional EVM was behaving in a non-regular way. Sometimes it increases
(0.75) and after that, for some reason, it declines (0.59).
Its behaviour is not random because, as explained above, it depends on the slope of the PV
curve portion between planned value at a given time and the point on PV curve where the actual
EV should have been done.
On the other hand, the SPI calculated with ES methodology has relatively consistent behaviour.
When the schedule variance - SV is constant for a certain time (as shown in Figure 3), SPI is
supposed to grow slightly, but steadily and with the smallest increments toward the project end
(that is due to the fact that, even though schedule variance remains the same in absolute
numbers, the earned schedule grows faster than the actual time).
In order to confirm these findings, I’ve made the simulations with a project whose cumulative
cost curve is linear. For projects with linear cumulative cost curves, schedule performance
indicators calculated with conventional EVM are identical to those calculated with Earned
Schedule and both have consistent behaviour. This corroborates the finding that the non-
linearity of cumulative cost curves is a principal cause of distortion of EVM schedule indicators.
All these examples clearly confirm that the schedule performance indicators, as presented
and standardised in conventional EVM methodology (PMI Practice Standard of EVM, EIA
Standard ANSI-EIA-748), are basically wrong for the projects with non-linear cumulative
costs curve2.
How can we explain that the EVM schedule indicators, even though clearly deficient, were used
for decades without being seriously questioned and eventually changed? The possible
explanation has to take into account that EVM was used mostly to analyze project cost
performance which works very well. Fleming and Koppelmann, probably the most cited EVM
authors, advocate the use of earned value management primarily for cost management. They
recognise that EVM is not reliable for schedule performance analyses, especially to predict the
end of a project. Instead of EVM, they recommend the use of the critical path method – CPM as
the most reliable way to forecast project duration.[2]
Another explanation is related to the fact that the researches about EVM do not address
explicitly the non-linearity of a cumulative cost curve. It is interesting because, in those research
works, cumulative cost curve is often presented as an S curve, so non-linear, but analyzed as if
it were a straight line.

2
As stated before, the non-linear curve is a common curve for the majority of commercial projects. It is especially
true for a project in IT and construction where the costs start to grow exponentially when, after the phases of
feasibility analysis and architecture, the project enters into a construction stage.
Earned Schedule Is Better
However, all that doesn’t mean that the EVM is useless in calculating how a project is
performing in time. As I mentioned in the beginning, EVM is very useful when calculated as the
difference on the horizontal (time) axis. This concept is already explained and described as
Earned Schedule (ES) approach. It is now included in PMI Practice Standard of EVM, but just as
an emerging practice and not as the standardized and recommended methodology to calculate
schedule performance.
The concept of Earned Schedule corrects the fundamental weakness of the classical EVM
concept and proposes the schedule performance indicators which are time and not cost-based.
The ES is presented in Figure 4.

Planned
value (PV)
Budget (cost)

Earned
Value (EV)

Remaining
Earned Schedule SV Time
Actual Time
Project Duration

Figure 4 – Earned Value Concept

Schedule Variance (t): SV(t) = ES – AT

Schedule Performance Index: SPI(t) = ES / AT
Estimate at Completion: EAC(t) = PD / SPI(t)

As we can see, Earned Schedule indicators are analogous to the EVM cost performance
indicators. The first two ES indicators (SV and SPI) are logical and should be reliable if correctly
calculated. They basically say whether the project is behind or ahead of planned schedule and
they quantify it.

Project Duration Estimate

The third EVM indicator, EAC, should reveal when the project is supposed to be completed
considering its current status. Most of the EVM authors (PMI Standard proposes the same) think
that the future project performance (cost or schedule) is, to some extent, influenced by its past
performance.
In my opinion, this way to estimate the time needed to complete the project is rather simplistic,
because the time is influenced by many factors and has to be more carefully calculated. To
know if and how the past performance is likely to influence a future project performance, the
project manager has to make an analysis of the factors which have caused the bad (or good,
which is less frequent) performance.
The project, for example, may be late a couple of months in the beginning because of missing
infrastructure. Or, the needed resources may not have been available on time. In these cases,
once the causes of delay are resolved, the project is likely to continue according to the planned
dynamic. If you apply the proposed formula for EAC that is driven by past performance, you will
probably overestimate the schedule. Therefore, the best way to predict the time at completion in
this case is to simply add the recorded delay to the remaining planned time.
On the other hand, if the causes of the project delay are more structural (unrealistically
compressed planned schedule or the lack of expertise), they will probably, to some extent,
influence the future project performance. In this case using the past performance to forecast the
end of the project could be justified.
Consequently, the estimated time at completion should be calculated as follows:

EAC(t) = Actual Time + (Planned Project Duration – Earned Schedule) *Cf

Cf is an adjustment factor representing the degree by which the past performance is supposed
to affect project performance in the remaining time. It should be determined by the project
manager and depends on the level with which the causes of a bad performance are likely to
influence the rest of the project. It is expressed as 1 plus percentage by which the causes of a
bad performance are likely to affect the remaining time. If the schedule compression, for
example, is expected to augment the remaining time by 15 %, Cf is 1.15.
If the causes of a project delay are structural and expected to continue to influence project
performance, Cf should be higher than 1, but it shouldn’t exceed the value of 1/SPI.
If the factors responsible for the bad performance are rather temporary and not likely to influence
the rest of the project, Cf should be 1.
I recommend that, even in the case where causes of bad project performance are of a temporary
nature, Cf should be >1. That is because the accumulated delay, even though not likely to
influence directly the rest of the project, will probably change a project dynamic, which could
result in certain delay.
For the reasons cited above, I believe that the adjustment factor should be calculated as well
while estimating the budget at completion. Nevertheless, we must remember to not apply
automatically the same factors for the budget and the schedule estimate. The factors likely to
affect the schedule and budget estimate may not be the same, or they may not influence both
with the same intensity. For example, if you fail to include in your project plan an important task,
executing the plan will result in delay and an increase in the project costs. Yet, this is not likely to
influence too much the rest of the project. On the other hand, if your team is less experienced
than expected, your schedule as well as your cost will probably be affected for the whole project
and the adjustment factor will be much higher than 1.

What to Do With EVM Data

It is certainly very important to know a project's status. EVM is a very good tool to indicate if a
project is within planned parameters or not. EVM indicators, if correctly calculated, should give a
reasonably accurate picture of current project performance.
However, knowing that is not enough. The prime objective of a project manager is to assure that
his project is going according to the baseline. If it’s not the case, he has either to take the
corrective actions to finish the project according to the planned schedule or he must establish a
new baseline.
It is clear that the EVM indicators, in addition to providing an early warning about the project
performance, should also give an unambiguous lead to the project manager on what is the most
appropriate action. Because of that the quantitative value of EVM indicators is very important.
To make an analogy, if your cholesterol is high, it is an early warning that you have to do
something. However, what you have to do depends principally on the cholesterol level. If it’s not
too high, you can consider changing your lifestyle, doing more exercise, etc. On the other hand,
if the cholesterol level is too high, your doctor will probably recommend more drastic measures,
like medication, for example.
It’s the same with a project. But, as your doctor, before prescribing a treatment, will probably
perform additional exams, so the project manager has to analyze other factors to decide what
the most appropriate action is.
One of the important factors he has to take into account is schedule compression. If the
schedule is already compressed, it could be very dangerous to try to recover the lost time by
adding new resources. The best way may be to reschedule the project and to establish a new
baseline. On the contrary, if the project schedule is comfortable (planned duration is much
longer than the critical path), adding people could solve the problem.
Another important factor which project manager has to consider is the point of a project's life
cycle in which the performance is measured. If the project is late 4 months and the remaining
time is also 4 months, putting additional resources and trying to finish the project within the
planned time is probably not the best solution. Sometimes adding people to a late project can
even make it more late (especially for IT projects).
On the other hand, if the project is late 4 months and the remaining time is 10 months, adding
people or/and putting some activities in parallel could be the reasonable solution in order to
recover the lost time.
For these reasons I recommend another ES indicator (I named it Schedule Variance Criticality –
SVC).

Schedule Variance Criticality: CVC = SV / RT

As we can see, CVC is expressed as a ratio between schedule variance (SV) and remaining
time (RT). It could be very useful indicator for a project manager when he has to decide what to
do with a late project. In any event, this indicator is much more useful than the SPI needed to
finish the project on time. If you say to a project manager that he must have SPI equal to 1.3
until the end of the project if he wants to finish the project on time, it doesn’t have any practical
meaning to him. But, if you say that the delay represents 80 % of the remaining time in one
project and 20 % in another, it is certain that he will not take the same actions in both projects.

Conclusion
The majority of recent research suggests that the results concerning the schedule performance
obtained with Earned Schedule are better than results calculated with conventional EVM.
Nevertheless, many of them claim that EVM shows good results in the first two thirds of a
project's life cycle, which may be understood as the conventional EVM still provides usable
results.
My research shows that schedule performance indicators calculated with conventional EVM are
deficient for the majority of commercial projects and, as such, they shouldn’t be used in schedule
performance management. They are not just less precise, they are fundamentally wrong.
On the other hand, the alternative practice called Earned Schedule works very well. It expresses
schedule delay in time units and, if applied well, it’s reasonably reliable. In my opinion, there is
no doubt that the EVM rules and standards related to schedule performance should be changed
and established according to the proposed Earned Schedule approach. It is much more
important if we know that through the efforts of the Office of Management and Budget (OMB),
the application of EVM is now required for all major acquisitions throughout the U.S. Federal
Government.

References:

[1] Fenton N. and Pfleeger Sl., Software Metrics : A Rigorous and Practical Approach,
International Thomson Computer Press, 1996

[2 Fleming, Q. W. and Koppelman J. M., Earned Value Project Management, 3rd edition.
Newton Square, PA: Project Management Institute, 2005

[3] Henderson, K., Further Developments in Earned Schedule, The Measurable News, 2004,
Spring

[4] Lipke, W. H., Schedule is Different, The Measurable News, 2003, Summer

[5] Lipke, W. H., Independent Estimate at Completion - Another Method, Proceedings of the
Society of Cost Estimating and Analysis, 2005

[6] Project Management Institute,.Practice Standard for Earned Value Management, Newtown
Square, Pennsylvania, Project Management Institute, 2005

[7] Stratton, R., Not Your Father's Earned Value, Projects@Work, 2005

[8] Vandevoorde, S. and Vanhoucke M., A Comparison of Different Project Duration Forecasting
Methods Using Earned Value Metrics, International Journal of Project Management, 2006

About the Author

Radenko Corovic has over 20 years experience in public and private sectors. He is specializing
in the areas of software measurement, IT performance, business process improvement and IT
strategy. His interest focuses on IT management, particularly on business aspect of IT and IT
contribution in organizational productivity.
Mr. Corovic holds a degree in administration from the University of Sarajevo, a postgraduate
diploma in management and an MBA with specialization in IT management from the Laval
University of Quebec. He is currently a manager of a PMO in one Agency of Quebec
Government.

Contact Information

[email protected]
Phone: 418.528.4981