Norwegian University of Science and Technology

TPK5115 - Risk Management in Projects

Author:
Jørn Vatn

August 12, 2021

Contents

1 Introduction 5
1.1 About this compendium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Risk Management 8
2.1 Project objectives and criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Risk identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Structuring and modelling of risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Model for project execution time/schedule modelling . . . . . . . . . . . . . . 11
2.3.2 Cost modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.3 Uncertainty in schedule and cost modelling . . . . . . . . . . . . . . . . . . . 11
2.4 Risk elements for follow up: Risk and opportunity register . . . . . . . . . . . . . . . 12
2.5 Correction and control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 Collection and analysis of experience - learning . . . . . . . . . . . . . . . . . . . . . 14

3 Probability theory 22
3.1 Basic probability notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.1 Event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.2 Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.3 Probability and Kolmogorov’s axioms . . . . . . . . . . . . . . . . . . . . . . 23
3.1.4 The law of total probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.5 Bayes theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1.6 Stochastic variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Common probability distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2.1 The normal distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.2.2 The exponential distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.3 The Weibull distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2.4 The gamma distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.5 The inverted gamma distribution . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.2.6 The lognormal distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.7 The binomial distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.8 The Poisson distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2.9 The inverse-Gauss distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2.10 The triangular distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2.11 The PERT distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.3 Assessment of parameters in parametric distributions . . . . . . . . . . . . . . . . . . 36
3.4 Distribution of sums, products and maximum values . . . . . . . . . . . . . . . . . . 37

1
 3.4.1 Distribution of sums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4.2 Distribution of a product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.3 Distribution of maximum values . . . . . . . . . . . . . . . . . . . . . . . . . 38

4 Schedule 41
4.1 Critical Path Method (CPM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2 Program Evaluation and Review Technique (PERT) . . . . . . . . . . . . . . . . . . 43
4.3 Successive schedule planning (SSP) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.4 Monte Carlo simulation (MCS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.5 Penalty for default . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.6 Event uncertainty in the schedule model . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.7 Updating the model as we get more information . . . . . . . . . . . . . . . . . . . . . 49
4.8 Examples of advanced schedule modelling . . . . . . . . . . . . . . . . . . . . . . . . 50

5 Decision under uncertainties 52

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.1.1 Overview of the method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2 Basic concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.2.1 Discrete end consequences vs attribute vector . . . . . . . . . . . . . . . . . . 54
5.2.2 Maximising expected utility . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2.3 Examples with one decision node . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.2.4 Decision trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

6 Life cycle cost and life cycle profit 69

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.2 Net present value calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.2.1 Trend modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.2.2 Example areas of LCC calculations . . . . . . . . . . . . . . . . . . . . . . . . 71

7 Parameter estimation 74
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.2 The MLE principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.3 Method of moments – PERT distribution . . . . . . . . . . . . . . . . . . . . . . . . 75
7.4 The LS principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.5 Bayesian meothds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

8 Expert judgements 81
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.1.1 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.1.2 Extent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.1.3 Use . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
8.2 General theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2.1 History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.2.2 Which types of evaluations/estimates do we consider as expert judgements? . 84
8.2.3 Why is there a need for expert judgements? . . . . . . . . . . . . . . . . . . . 86
8.2.4 Who are the experts? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
8.2.5 How should the expert judgement be carried out? . . . . . . . . . . . . . . . 89
8.2.6 Terms and notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

2
 8.3 Check list for expert judgement exercises . . . . . . . . . . . . . . . . . . . . . . . . . 92
8.3.1 Checklist for phase I: The preparation phase . . . . . . . . . . . . . . . . . . 92
8.3.2 Checklist for phase II: The elicitation phase . . . . . . . . . . . . . . . . . . . 93
8.3.3 Checklist for phase III: The calculation phase . . . . . . . . . . . . . . . . . . 94
8.4 Calculation aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
8.4.1 Criterion for performing calibration . . . . . . . . . . . . . . . . . . . . . . . 95
8.4.2 Method for performing calibration . . . . . . . . . . . . . . . . . . . . . . . . 96
8.4.3 Weighting of experts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
8.4.4 Standard weighting model - Experts only . . . . . . . . . . . . . . . . . . . . 98
8.4.5 Transforming percentiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
8.4.6 Standard weighting model - Experts and data . . . . . . . . . . . . . . . . . . 99
8.5 Worked example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

9 Optimization of turnaround activities 103

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
9.2 Single component considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
9.2.1 Preventive maintenance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
9.2.2 Single activity - Preventive maintenance not included in the turnaround . . . 104
9.2.3 Single activity - Consideration for inclusion in turnaround - Static consideration105
9.2.4 Single activity – dynamic consideration . . . . . . . . . . . . . . . . . . . . . 106
9.2.5 Single activity - “Arctic maintenance” . . . . . . . . . . . . . . . . . . . . . . 106
9.3 Single component - Impact on turnaround duration . . . . . . . . . . . . . . . . . . . 108
9.3.1 Consideration for inclusion in turnaround, τ ∗ >τTA – No cancel possibilities . 108
9.3.2 Consideration for inclusion in turnaround, τ ∗ >τTA – Cancel possibilities . . 108
9.3.3 “Arctic maintenance” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
9.4 Changing the frequency of the turnaround . . . . . . . . . . . . . . . . . . . . . . . . 110
9.4.1 Treating only safety issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
9.4.2 Production related issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9.5 Risk identification for critical activities . . . . . . . . . . . . . . . . . . . . . . . . . . 114

10 Portfolio management 118

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
10.2 Visualization of project portfolios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
10.3 Selection of projects for execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
10.4 Vulnerability of project portfolios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
10.4.1 Dynamic programming - Portfolio selection . . . . . . . . . . . . . . . . . . . 123
10.4.2 Mixed integer programming problem . . . . . . . . . . . . . . . . . . . . . . . 123
10.5 Scheduling of projects within a portfolio . . . . . . . . . . . . . . . . . . . . . . . . . 124

11 Linear, dynamic, non-linear and stochastic programming 128

11.1 Introduction to programming problems . . . . . . . . . . . . . . . . . . . . . . . . . . 128
11.2 Linear programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
11.2.1 Motivating example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
11.2.2 Linear programming problem on standard form . . . . . . . . . . . . . . . . . 129
11.2.3 Solving the linear programming problem by the SIMPLEX method . . . . . . 130
11.2.4 Solving the LP problem by a computer . . . . . . . . . . . . . . . . . . . . . 130
11.2.5 Mixed integer programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
11.3 Dynamic programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

3
 11.3.1 Worked example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
11.4 Stochastic programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
11.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
11.4.2 Discretization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
11.4.3 The Value of the Stochastic Solution . . . . . . . . . . . . . . . . . . . . . . . 141
11.4.4 Expected value of perfect information . . . . . . . . . . . . . . . . . . . . . . 143
11.4.5 Scenario building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
11.4.6 How to perform discretization? . . . . . . . . . . . . . . . . . . . . . . . . . . 144

Bibliography 144

Index 149

4
Chapter 1

Introduction

1.1 About this compendium

This course compendium is to be used in various cources at NTNU. The focus will be on the
following topics:

• Risk identification

• Risk structuring

• Risk modelling in the light of a time schedule and a cost model

• Risk follow up

We will also discuss elements related to decision analysis where risk is involved, use of life cycle
cost and life cycle profit models and methods to assess the numerical values of parameters used in
the risk models.
The course compendium comprises a large number of exercises, and it is recommended to do
most of the exercises in order to get a good understanding of the topics and methods described. A
separate MS Excel program, pRisk.xlsm has been developed in order to assist numerical calculations
and to conduct Monte Carlo simulations.

1.2 Definitions
Aleatory uncertainty
Variation of quantities in a population. We sometimes use the word variability rather than aleatory
uncertainty.

Epistemic uncertainty
Lack of knowledge about the “world”, and observable quantities in particular.

Dependency
The relation between the sequence of the activities in a project.

Observable quantity
A quantity expressing a state of the “world”, i.e., a quantity of the physical reality or nature, that is
unknown at the time of the analysis but will, if the system being analysed is actually implemented,

5
take some value in the future, and possibly become known.

Parameter
We use the term parameter in two ways in this report. The main use of a parameter is that it
is a quantity that is a part of the risk analysis models, and for which we assign numerical values.
The more academic definition of a parameter used in a probability statement about an observable
quantity, X, is that a parameter is a construct where the value of the parameter is the limiting value
where we are not able to saturate our understanding about the observable quantity X whatsoever
new information we could get hold of.

Parameter estimate
The numeric value we assess to a parameter.

Probability
A measure of uncertainty regarding the occurrence of an event.

Risk
Risk is uncertainty regarding occurrence and severity of future events. In project risk management
focus is on undesired events, whereas economists also include opportunities as part of risk. To
quantify risk three elements are introduced: < e, p, S >. p is used as a probability measure of the
occurrence of an event, say e. S represents the severity of the event. Note that S is a multidimen-
sional random quantity, covering several dimensions like personnel safety, environmental impacts,
material damages, project delays, extra costs, etc. Since there is more than one event to treat, i is
used as an index to run through all relevant events. An operational definition of risk is thus the
set of all relevant triplets: R = {< ei , pi , Si >}.

Risk picture
A set of undesired events, the causes and factors that may contribute to the event, the possible
consequences of the event with corresponding influencing factors, and uncertainties related to all
these issues.

Risk acceptance
A decision to accept a risk.

Risk acceptance criterion

A reference by which risk is assessed to be acceptable or unacceptable.

Schedule
A plan which specifies the start and finalisation point of times for the activities in a project.

Stochastic dependency
Two or more stochastic variables are (stochastically) dependent if the expectation of one stochastic
variable depends on the value of one or more of the other stochastic variables.

Stochastic variable
A stochastic variable, or random quantity, is a quantity for which we do not know the value it will
take. However, we could state statistical properties of the variable or make probability statement
about the value of the quantity.

6
Uncertainty
Lack of knowledge about the performance of a system, and observable quantities in particular.

7
Chapter 2

Risk Management

Generally, risk management is defined (IEC 60300-3-9) as a “systematic application of management

policies, procedures and practices to the tasks of analysing, evaluating and controlling risk”. It will
comprise (IEC definitions in parentheses):
• Risk assessment, that is:

– Risk analysis (“Systematic use of available information to identify hazards and to esti-
mate the risk to individuals or populations, property or the environment”), and
– Risk evaluation (“Process in which judgements are made on the tolerability of the risk
on the basis of risk analysis and taking into account factors such as socio-economic and
environmental aspects”)

• Risk reduction/control (Decision making, implementation and risk monitoring).

There exists no common definition of risk, but for instance IEC 60300-3-9 [13] defines risk as a
“combination of the frequency, or probability, of occurrence and the consequence of a specified
hazardous events”. Most definitions comprise the elements of probabilities and consequences. How-
ever, some as Klinke and Renn (2001) [15] suggest a very wide definition, stating: “Risk refers
to the possibility that human actions or events lead to consequences that affect aspects of what
humans value”. So the total risk comprises the possibility of a number (“all”) unwanted/hazardous
events. It is part of the risk analysis to delimit which hazards to include. Further, risk usually
refers to threats in the future, involving a (high) degree of uncertainty. In this presentation risk is
defined as uncertainty regarding occurrence and severity of future events. To make an operational
definition of risk three elements are introduced: < e, p, S >. p is used as a probability measure of
the occurrence of an event, say e. S represents the severity of the event. Note that S is a mul-
tidimensional random quantity, covering several dimensions like personnel safety, environmental
impacts, material damages, project delays, extra costs, etc. Since there is more than one event to
treat, i is used as an index to run through all relevant events. An operational definition of risk is
thus the set of all relevant triplets: R = {< ei , pi , Si >}.
In the following we will present the basic elements of risk management as it is proposed to be
an integral part of project management.

2.1 Project objectives and criteria

In classical risk analysis of industrial systems the use of so-called risk acceptance criteria has played
a central role in the last two or tree decades. Basically use of risk acceptance criteria means that

8
some severe consequences are defined, e.g. accident with fatalities. Then we try to set an upper
limit for the probability of these consequences that could be accepted, i.e., we could not accept
higher probabilities in any situations. Further these probabilities could only be accepted if risk
reduction is not possible, or the cost of risk reduction is very high.
In recent years it has been a discussion in the risk analysis society whether it is fruitful or not
to use risk acceptance criteria according to the principles above. It is argued that very often risk
acceptance criteria are set arbitrary, and these do not necessarily support the overall best solutions.
Therefore, it could be more fruitful to use some kind of risk evaluation criteria, rather than strict
acceptance criteria.
In project risk management we could establish acceptance criteria related to two types of events:

• Events with severe consequences related to health, environment and safety.

• Events with severe consequences related to project costs, project quality, project duration, or
even termination of the project.

In this course we will have main focus on the project costs and the duration of the project. Note
that both project cost and project duration are stochastic variables and not events. Thus it is not
possible to establish acceptance criteria to project cost or duration directly. Basically, there are
three types of numeric values we could introduce in relation to such stochastic variables describing
the project:

1. Target. The target expresses our ambitions in the project. The target shall be something
we are striving at, and it should be possible to reach the target. It is possible to introduce
(internal) bonuses, or other rewards in order to reach the targets in a project.

2. Expectation. The expectations are the value the stochastic variables will achieve in the long
run, or our expectation about the outcome. The expectation is less ambitious than the target.
The expectation will in a realistic way account for hazards, and threats and conditions which
often contribute to the fact that the targets are not met.

3. Commitment. The commitments are values related to the stochastic variables which are
regulated in agreements and contracts. For example it could be stated in the contract that a
new bridge shall be completed within a given date. If we are not able to fulfil the commitments,
this will usually result in economical consequences, for example penalties for defaults, or in
the worst case cancelling of the contract.

Problem 2.1
Discuss targets, expectations and commitments related to a new railway track between two big
cities in Norway. ♢
We sometimes also want to discuss the uncertainty in e.g. the project costs. In Section 2.3.3 we
have discussed the uncertainty concept in relation to project duration and costs.

2.2 Risk identification

In order to establish a risk picture three important questions are put forward:

1. What could go wrong?

2. How likely is it?

9
 3. And if it goes wrong, how serious is it (the consequences)?
With respect to risk identification it is the first question we will answer. Since risk addresses all three
questions, it would have been better to use terms like ‘hazards-’ and ‘threats identification’ rather
than the term ‘risk identification’, but the common practice is to use the term ‘risk identification
when it comes to identification of what could go wrong. The risk identification could be addressed
from different angles:

• A listing of undesired events.

• A listing of scenarios.
• A listing of hazards.
• A listing of threats.

An undesired event is an event which might occur, e.g., a large water leakage in a tunnel. A scenario
is a description of a imagined sequence or chain of events, e.g., we have a water leakage, and we are
not able to stop this leakage with ordinary tightening medium due to the possible environmental
aspects which is not clarified at the moment. Further the green movement is also likely to enter
the scene in this case. A hazard is typically related to energies, poisonous media etc, and if they
are released this will result in an accident or a severe event. A threat is a wider term than hazard,
and we include also aspects as “wrong” method applied, “lack of competence and experience”. The
term threat is also very often used in connection with security problems, e.g., sabotage, terrorism,
and vandalism.

Problem 2.2
List examples of “undesired events”, “scenarios”, “hazards” and “threats” in relation to building a
new railway track between two major cities in Norway. ♢
There exist several methods that could be used in order to identify undesired events and threats,
e.g.:
• Preliminary Hazard Analysis (PHA). PHA is used to establish threats in an early phase of
a project. The method will usually require some project breakdown, e.g Work Breakdown
Structure (WBS) or Cost Breakdown Structure (CBS), project phases or similar. A detailed
project description is usually not available at this moment.
• Task analysis (TA) and Hazard and Operability Study (HAZOP) are used on a more detailed
level where we have knowledge about the various tasks.
• Use of experience data means that we try to identify events and threats based on systematic
analysis of experience from the past, i.e., what have gone wrong in earlier projects.
• Checklists. Checklists exist on different levels, and could either be used in a separate analysis,
or as an aid in another method, e.g. in a PHA. The checklists should, however, be put a way
initially since introducing checklists early will often prevent the process of revealing project
specific conditions. A checklist is primarily a list to be used in the end of the process to
ensure that the “obvious” elements have not been overlooked. In Table 2.1 a such generic list
of risk factors is provided. For each risk factor in the list, also some “cues” are listed which
could be used to assess the significance of the risk factor in a given project. Note that this
list is on a very general level, and not specific to e.g. a construction project, a tunnel project
and so on.

10
2.3 Structuring and modelling of risk
In Section 2.2 we have identified methods to identify events and threats. We now want to relate
these events and threats to the explicit models we have for project costs and project duration.

2.3.1 Model for project execution time/schedule modelling

When analysing the execution time for a project we will have a project plan and typically a Gantt
diagram as a starting point. The Gantt diagram is transformed into a so-called flow network. where
the connections between the activities are explicitly described. Such a flow network also comprises
description of duration of the activities in terms of probability statements. The duration of each
activities are stochastic variables, which we denote Ti for activity i. In a flow network we might also
have uncertain activities which will be carried out only under special conditions. These conditions
could be describe in terms of events, and we need to describe the probability of occurrence of
such events. Thus, there is a set of quantities, i.e., time variables and events in the model. The
objective is now to link the undesired events and threats discussed in Section 2.2 to these time
variables and events. Time variables are described by a probability distribution function. Such a
distribution function comprises parameters that characterise the time variable. Often a parametric
probability distribution is described by the three quantities L (low), M (most likely) and H high.
If an undesired event occur, it is likely that the values of L, M and H will be higher than in case
this event does not occur. A way to include the result from the risk identification process is then
to express the different values of L, M and H depending on whether the critical event occurs or
not. If we in addition are able to assess the probability of occurrence of the critical event, the
knowledge about this critical event has been completely included into the risk model. Based on
such an explicit modelling of the critical event, we could also easily update the model in case of new
information about the critical event is obtained, for example new information could be available
at a later stage in the process and changes of the plan could still be possible in light of the new
information.

2.3.2 Cost modelling

The cost model is usually based on the cost breakdown structure, and the cost elements will again
be functions of labour cost, overtime cost, purchase price, hour cost of renting equipment, material
cost, amount of material etc. The probabilistic modelling of cost is usually easier than for modelling
project execution time. The principle is just to add a lot of cost terms, where each cost term is the
product of the unit price and the number of units. We introduce price and volume as stochastic
variables to describe the unit price and the number of units. The price and volume variables should
also be linked to the undesired events and threats we have identified in Section 2.2. Often it is
necessary to link the cost model to the schedule model. For example in case of delays it might be
necessary to put more effort into the project to catch up with the problems, and these effort could
be very costly. Also, if the project is delayed we may need to pay extra cost to sub-contractors
that have to postpone their support into the project.

2.3.3 Uncertainty in schedule and cost modelling

As indicated above we will establish probabilistic models to describe the duration and cost of a
project. The result of such a probabilistic modelling is that we treat the duration and cost as
stochastic variables. Since duration and costs are stochastic variables, this means that there is
uncertainty regarding the values they will take in the real project we are evaluating. Sometimes

11
we split this uncertainty into three different categories, i) aleatory uncertainty (variability due to
e.g. weather conditions, labour conflicts, breakdown of machines etc.), ii) parameter or epistemic
uncertainty due to lack of knowledge about “true” parameter values, and iii) model uncertainty
due to lack of detailed, or wrong modelling. Under such a thinking, the aleatory uncertainty
could not be reduced, it is believed to be the result of the variability in the world which we
cannot control. Uncertainty in the parameters is, however, believed to be reducible by collecting
more information. Also uncertainty in the models is believed to be reducible by more detailed
modelling, and decomposition of the various elements that go into the model. It is appealing to
have a mental model where the uncertainty could be split into one part which we might not reduce
(variability), and one part which we might reduce by thorough analysis and more investigation
(increased knowledge). If we are able to demonstrate that the part of the uncertainty related
to lack of knowledge and understanding has been reduced to a sufficient degree, we could then
claim high confidence in the analysis. In some situation the owner, or the authorities put forward
requirements which could be interpreted as confidence regarding the quality of the analysis. It is
though not always clear what is meant by such a confidence level. As an example, let E(C) be the
expected cost of a project. A confidence statement could now be formulated as “The probability
that the actual project cost is within an interval E(C)±10% should at least be 70%”. It is, however,
not straight forward to document such a confidence level in a real analysis. The “Successive process
(trinnvisprosessen)” [4] is an attempt to demonstrate how to reduce the “uncertainty” in the result
to a certain level of confidence.
We also mention that Aven [12] has recently questioned such an approach where there exist
model uncertainty and parameter uncertainty, and emphasises that we in the analysis should focus
on the observable quantities which will become evident for us if the project is executed, e.g. the
costs, and that uncertainty in these quantities represent the lack of knowledge about which values
they will take in the future. This discussion is not pursuit any more in this presentation.

Problem 2.3 Discuss different type of uncertainties in a tunnel project. Propose a classification,
and identify uncertainty elements that could be reduced by i) further physical investigation, and
ii) by further analysis. Also list uncertainty elements that could not be reduced before the project
is actually executed. ♢

2.4 Risk elements for follow up: Risk and opportunity register
As risk elements and threats are identified in Section 2.2 these have to be controlled as far as
possible. It is not sufficient to identify these conditions and model them in the schedule and cost
models, we also have to mitigate the risk elements and threats. In order to ensure a systematic
follow up of risk elements and threats it is recommended to establish a so-called threat log. The
terms ‘Risk Register’ and ‘Risk & Opportunity Register’ (R&OR) is sometimes used rather than
the term ‘threat log’.
A R&OR is best managed by a database solution, for example an MS-Access database. Each
row in the database represents one risk element or threat. The fields in such a database could vary,
but the following fields seems reasonable:

• ID. An identifier is required in order to keep track of the threat in relation to the quantitative
risk models, to follow up actions et.

• Description. A description of the threat is necessary in order to understand the content of the
the problem. It could be necessary to state the immediate consequences (e.g. occupational

12
 accident), but also consequences in terms of the main objectives of the project, e.g., time and
costs.

• Likelihood or probability. A judgement regarding how probable it is that the threat or the
risk condition will be released in terms of e.g. undesired or critical events.

• Impact. If possible, give a direct impact on cost and schedule if the event occurs, either by
an expected impact, or by L, M and H values.

• References to cost and schedule. In order to update the schedule and cost models it is
convenient to give an explicit reference from the R&OR into the schedule and cost models.

• Manageability. Here it is descried how the threat could be influenced, either by implementing
measures to eliminate the threat prior to it reveals it self, or measures in order to reduce the
consequences in case of the threat will materialize.

• Alert information. It is important to be aware of information that could indicate the devel-
opment of the threat before it eventually will materialize. If such information is available
we could implement relevant measures if necessary. For example it could be possible to take
ground samples at a certain cost, but utilising the information from such samples could enable
us to choose appropriate methods for tunnel penetration.

• Measures. List of measures that could be implemented to reduce the risk.

• Deadline and responsible. Identification of who is responsible for implementing and follow up
of the measure or threat, and any deadlines.

• Status. Both with respect to the threat and any measure it is valuable to specify the devel-
opment, i.e., did the treat reveal it self into undesired events with unwanted consequences,
did the measure play any positive effect etc.?

Problem 2.4
Consider threats and risk conditions in Problem 2.2 and discuss the possibilities to mitigate the
threats, and if any prior information could be available. ♢

2.5 Correction and control

As the project develops the R&OR is the primary control tool for risk follow up. By following the
status of the various threats, risk elements and measures we could monitor the risk in the project.
This information should of course be linked to the time and cost plans. If a given threat does not
reveal in terms of undesired events, the time and cost estimates could be lowered and this gain
could be utilised in other part of the project, or in other projects. In the opposite situation it
is necessary to increase the time and cost estimates, and we need to consider new measures, and
maybe spend some of the reserves to catch up in case of an expected delay.
During the life cycle of a project it will occur new threats and risk elements which we did not
foresee in the initial risk identification process. Such threats must continuously be entered into the
R&OR, and measures need to be considered.

13
2.6 Collection and analysis of experience - learning
After project execution it is valuable to systematise the information and knowledge we have achieved
during the project. The most important data sources will be

• The risk and opportunity register (R&OR).

• Schedule plans, and the relation between plans and reality.

• Cost plans, and account numbers

Problem 2.5
Identify other sources of information that could be relevant. ♢
It is especially two types of analyses we will conduct:

• Systematising of what went wrong, and which measures that proved to be efficient.

• Estimation of parameters which we could include in later probabilistic schedule and cost
models.

The main elements of the project risk management process is shown in Figure 2.1.

Project objectives,
requirements etc.

Learning
Experience
Risk identification
data

Risk structuring Project

and modelling debriefing

Identify threats and risks Regular update of

to include in R&OR R&OR

Project execution,
correction and controll

Project Project Project Operating

initiation start-up finalisation phase

Figure 2.1: Project risk management

14
 Table 2.1: Risk factors, adapted from State of
Texas: Department of Information Resources,
http://www.dir.state.tx.us/eod/qa/risk/index.htm

Risk Factors Low Risk Cues Medium Risk Cues High Risk Cues
Project Fit to Provider directly supports indirectly impacts does not support or re-
Organization provider organization one or more goals of late to provider organi-
mission and/or goals provider zation mission or goals
Customer Perception customer expects this organization is working project is mismatch
organization to provide on project in area not with prior products
this product expected by customer or services of this
organization
Work Flow little or no change to will change some aspect significantly changes
work flow or have small affect on the work flow or
work flow method of organization
Goals Conflict goals of projects within goals of projects do not goals of projects are in
the program are sup- conflict, but provide lit- conflict, either directly
portive of or compli- tle direct support or indirectly
mentary to each other
Resource Conflict projects within the pro- projects within the pro- projects within the pro-
gram share resources gram schedule resources gram often need the
without any conflict carefully to avoid con- same resources at the
flict same time (or compete
for the same budget)
Customer Conflict multiple customers of multiple customers of multiple customers of
the program have com- the program have dif- the program are trying
mon needs ferent needs, but do not to drive it in very dif-
conflict ferent directions
Leadership program has active pro- program has person or program has no leader,
gram manager who co- team responsible for or program manager
ordinates projects program, but unable to concept is not in use
spend enough time to
lead effectively
Program Manager Ex- program manager has program manager has program manager is
perience deep experience in the some experience in do- new to the domain
domain main, is able to leverage
subject matter experts
Definition of the Pro- program is well-defined, program is well-defined, program is not well-
gram with a scope that is but unlikely to be han- defined or carries con-
manageable by this or- dled by this organiza- flicting objectives in the
ganization tion scope
Continued on next page

15
 Table 2.1 – Continued from previous page
Risk Factors Low Risk Cues Medium Risk Cues High Risk Cues
Political Influences no particular project has several po- project has a variety
politically-driven litically motivated de- of political influences or
choices being made cisions, such as using most decisions are made
a vendor selected for behind closed doors
political reasons, rather
than qualifications
Convenient Date date for delivery has date is being partially date is being totally
been set by reason- driven by need to meet driven by need to meet
able project commit- marketing demo, trade marketing demo, trade
ment process show, or other mandate show, or other man-
not related to technical date; little considera-
estimate tion of project team es-
timates
Use of Attractive Tech- technology selected has project is being done project is being done as
nology been in use for some in a sub-optimal way, a way to show a new
time to leverage the purchase technology or as an ex-
or development of new cuse to bring a new
technology technology into the or-
ganization
Short Term Solution project meets short project is focused on project team has been
term need without short-term solution to a explicitly directed to ig-
serious compromise to problem, with little un- nore the long term out-
long term outlook derstanding of what is look and focus on com-
needed in the long term pleting the short term
deliverable
Organization Stability little or no change in some management management or orga-
management or struc- change or reorganiza- nization structure is
ture expected tion expected continually or rapidly
changing
Organization Roles and individuals throughout individuals understand many in the organiza-
Responsibilities the organization under- their own roles and re- tion are unsure or un-
stand their own roles sponsibilities, but are aware of who is respon-
and responsibilities and unsure who is respon- sible for many of the ac-
those of others sible for work outside tivities of the organiza-
their immediate group tion
Policies and Standards development policies development policies no policies or stan-
and standards are and standards are in dards, or they are ill-
defined and carefully place, but are weak or defined and unused
followed not carefully followed
Management Support strongly committed to some commitment, not little or no support
success of project total
Executive Involvement visible and strong sup- occasional support, no visible support; no
port provides help on issues help on unresolved is-
when asked sues
Continued on next page

16
 Table 2.1 – Continued from previous page
Risk Factors Low Risk Cues Medium Risk Cues High Risk Cues
Project Objectives verifiable project ob- some project objectives, no established project
jectives, reasonable re- measures may be ques- objectives or objectives
quirements tionable are not measurable
User Involvement users highly involved users play minor roles, minimal or no user in-
with project team, pro- moderate impact on volvement; little user
vide significant input system input
User Experience users highly expe- users have experience users have no previous
rienced in similar with similar projects experience with similar
projects; have specific and have needs in mind projects; unsure of how
ideas of how needs can needs can be met
be met
User Acceptance users accept concepts users accept most of users do not accept any
and details of system; concepts and details of concepts or design de-
process is in place for system; process in place tails of system
user approvals for user approvals
User Training Needs user training needs user training needs con- requirements not iden-
considered; training sidered; no training yet tified or not addressed
in progress or plan in or training plan is in de-
place velopment
User Justification user justification com- user justification pro- no satisfactory justifica-
plete, accurate, sound vided, complete with tion for system
some questions about
applicability
Project Size small, non-complex, or medium, moderate large, highly complex,
easily decomposed complexity, decompos- or not decomposable
able
Reusable Components components available components available, components identified,
and compatible with but need some revision need serious modifica-
approach tion for use
Supplied Components components available components work under components known to
and directly usable most circumstances fail in certain cases,
likely to be late, or in-
compatible with parts
of approach
Budget Size sufficient budget allo- questionable budget al- doubtful budget is suffi-
cated located cient
Budget Constraints funds allocated without some questions about allocation in doubt or
constraints availability of funds subject to change with-
out notice
Cost Controls well established, in system in place, weak in system lacking or
place areas nonexistent
Delivery Commitment stable commitment some uncertain com- unstable, fluctuating
dates mitments commitments
Continued on next page

17
 Table 2.1 – Continued from previous page
Risk Factors Low Risk Cues Medium Risk Cues High Risk Cues
Development Schedule team agrees that sched- team finds one phase team agrees that two or
ule is acceptable and of the plan to have a more phases of schedule
can be met schedule that is too ag- are unlikely to be met
gressive
Requirements Stability little or no change ex- some change expected rapidly changing or no
pected to approved set against approved set agreed-upon baseline
(baseline)
Requirements Com- all completely specified some requirements in- some requirements only
pleteness and Clarity and clearly written complete or unclear in the head of the cus-
tomer
Testability product requirements parts of product hard to most of product hard to
easy to test, plans test, or minimal plan- test, or no test plans be-
underway ning being done ing made
Design Difficulty well defined interfaces; unclear how to design, interfaces not well de-
design well understood or aspects of design yet fined or controlled; sub-
to be decided ject to change
Implementation Diffi- content is reasonable content has elements content has components
culty for this team to imple- somewhat difficult for this team will find very
ment this team to implement difficult to implement
System Dependencies clearly defined depen- some elements of the no clear plan or sched-
dencies of the project system are well under- ule for how the whole
and other parts of sys- stood and planned; oth- system will come to-
tem ers are not yet compre- gether
hended
Response or other Per- readily fits boundaries operates occasionally at operates continuously
formance Factors needed; analysis has boundaries at boundary levels
been done
Customer Service Im- requires little change to requires minor changes requires major changes
pact customer service to customer service to customer service ap-
proach or offerings
Data Migration Re- little or no data to mi- much data to migrate, much data to migrate;
quired grate but good descriptions several types of data or
available of structure no good descriptions of
and use what is where
Pilot Approach pilot site (or team) pilot needs to be done only available pilot sites
available and interested with several sites (who are uncooperative or in
in participating are willing) or with one crisis mode already
who needs much help
Alternatives Analysis analysis of alternatives analysis of alternatives analysis not completed,
complete, all consid- complete, some as- not all alternatives con-
ered, assumptions veri- sumptions questionable sidered, or assumptions
fiable or alternatives not fully faulty
considered
Continued on next page

18
 Table 2.1 – Continued from previous page
Risk Factors Low Risk Cues Medium Risk Cues High Risk Cues
Commitment Process changes to com- changes to commit- changes to com-
mitments in scope, ments are communi- mitments are made
content, schedule are cated to all involved without review or
reviewed and approved involvement of the
by all involved team
Quality Assurance Ap- QA system established, procedures established, no QA process or estab-
proach followed, effective but not well followed or lished procedures
effective
Development Docu- correct and available some deficiencies, but nonexistent
mentation available
Use of Defined Develop- development process process established, but no formal process used
ment Process in place, established, not followed or is inef-
effective, followed by fective
team
Early Identification of peer reviews are incor- peer reviews are used team expects to find all
Defects porated throughout sporadically defects with testing
Defect Tracking defect tracking defined, defect tracking process no process in place to
consistent, effective defined, but inconsis- track defects
tently used
Change Control for formal change control change control process no change control pro-
Work Products process in place, fol- in place, not followed or cess used
lowed, effective is ineffective
Physical Facilities little or no modification some modifications major modifications
needed needed; some existent needed, or facilities non
existent
Tools Availability in place, documented, available, validated, unvalidated, pro-
validated some development prietary or major
needed (or minimal development needed;
documentation) no documentation
Vendor Support complete support at adequate support little or no support,
reasonable price and in at contracted price, high cost, and/or poor
needed time frame reasonable response response time
time
Contract Fit contract with customer contract has some open contract has burden-
has good terms, com- issues which could in- some document require-
munication with team terrupt team work ef- ments or causes extra
is good forts work to comply
Disaster Recovery all areas following se- some security measures no security measures in
curity guidelines; data in place; backups done; place; backup lacking;
backed up; disaster re- disaster recovery con- disaster recovery not
covery system in place; sidered, but procedures considered
procedures followed lacking or not followed
Continued on next page

19
 Table 2.1 – Continued from previous page
Risk Factors Low Risk Cues Medium Risk Cues High Risk Cues
PM Approach product and process planning and monitor- weak or non existent
planning and monitor- ing need enhancement planning and monitor-
ing in place ing
PM Experience PM very experienced PM has moderate expe- PM has no experience
with similar projects rience or has experience with this type of project
with different types of or is new to project
projects management
PM Authority has line management or is able to influence has little authority from
official authority that those elsewhere in the location in the organi-
enables project leader- organization, based on zation structure and lit-
ship effectiveness personal relationships tle personal power to in-
fluence decision-making
and resources
Support of the PM complete support by support by most of no visible support;
team and of manage- team, with some reser- manager in name only
ment vations
Team Member Avail- in place, little turnover available, some high turnover, not
ability expected; few inter- turnover expected; available; team spends
rupts for fire fighting some fire fighting most of time fighting
fires
Mix of Team Skills good mix of disciplines some disciplines inade- some disciplines not
quately represented represented at all
Team Communication clearly communicates team communicates rarely communicates
goals and status be- some of the information clearly within team or
tween the team and some of the time to others who need to
rest of organization be informed
Application Experience extensive experience in some experience with little or no experience
team with projects like similar projects with similar projects
this
Expertise with Applica- good background with some experience with no expertise in domain
tion Area (Domain) application domain domain in team or able in team, no availability
within development to call on experts as of experts
team needed
Experience with high experience average experience low experience
Project Tools
Experience with high experience average experience low experience
Project Process
Training of Team training plan in place, training for some areas no training plan or
training ongoing not available or training training not readily
planned for future available
Team Spirit and Atti- strongly committed to willing to do what it little or no commitment
tude success of project; coop- takes to get the job to the project; not a co-
erative done hesive team
Continued on next page

20
 Table 2.1 – Continued from previous page
Risk Factors Low Risk Cues Medium Risk Cues High Risk Cues
Team Productivity all milestones met, de- milestones met, some productivity low, mile-
liverables on time, pro- delays in deliverables, stones not met, delays
ductivity high productivity acceptable in deliverables
Technology Match to technology planned for some of the planned selected technology is a
Project project is good match technology is not well- poor match to the prob-
to customers and prob- suited to the problem or lem or customer
lem customer
Technology Experience good level of experience some experience with no experience with the
of Project Team with technology the technology technology
Availability of Technol- technology experts experts available else- will need to acquire
ogy Expertise readily available where in organization help from outside the
organization
Maturity of Technology technology has been in technology is well un- technology is leading
use in the industry for derstood in the indus- edge, if not ”bleeding
quite some time try edge” in nature
Design Complexity easily maintained certain aspects difficult extremely difficult to
to maintain maintain
Support Personnel in place, experienced, missing some areas of significant discipline or
sufficient in number expertise expertise missing
Vendor Support complete support at adequate support little or no support,
reasonable price and in at contracted price, high cost, and/or poor
needed time frame reasonable response response time
time

21
Chapter 3

Probability theory

3.1 Basic probability notation

In this chapter basic elements of probability theory are reviewed. Readers familiar with probability
theory can skip this chapter. Readers which are very unfamiliar with this topic are advised to read
an introductory textbook in probability theory.

3.1.1 Event
In order to define probability, we need to work with events. Let as an example A be the event that
there is an operator error in a control room. This is written:

A = {operator error}

An event may occur, or not. We do not know the outcome in advance prior to the experiment or
a situation in the “real life”. We also use the word event to denote a set of distinct events. For
example the event that we get an even number when tossing a dice.

3.1.2 Probability
When events are defined, the probability that the event occurs is of interest. Probability is denoted
by Pr(·), i.e.

Pr(A) = Probability that A occur

The numeric value of Pr(A) may be found by:

• Studying the sample space.

• Analysing collected data.

• Look up the value in data hand books.

• “Expert judgement” [11].

The sample space defines all possible events. As an example let A = {It is Sunday}, B = {It is
Monday}, .. , G = {It is Saturday}. The sample space is then given by S = {A, B, C, D, E, F, G}.
So-called Venn diagrams are useful when we want to analyse a subset of the sample space S. A
rectangle represents the entire sample space, and closed curves such as a circle are used to represent

22
 A

Figure 3.1: Venn diagram

subsets of the sample space as illustrated in Figure 3.1. In the following we will illustrate frequently
used combinations of events:

Union. We write A ∪ B to denote the union of A and B, i.e., the occurrence of A or B or (A and
B). Let A be the event that tossing a die results in a “six”, and B be the event that we get an odd
number of eyes. We then have A ∪ B = {1, 3, 5, 6}. S AB

Intersection. We write A ∩ B to denote the intersection of A and B, i.e. the occurrence of both
A and B. As an example, let A be the event that a project is not completed in due time, and let
B be the event that the budget limits are exceeded. A ∩ B then represent the situation that the
project is not completed in due time and the budget limits are exceeded. S AB

Disjoint events. A and B are said to be disjoint if they can not occur simultaneously, i.e. A ∩ B
= Ø = the empty set. Let A be the event that tossing a die results in a “six”, and B be the event
that we get an odd number of eyes. A and B are disjoint since they cannot occur simultaneously,
and we have A ∩ B = Ø. SA B

Complementary events. The complement of an event A is all events in the sample space S
except for A. The complement of an event is denoted by AC . Let A be the event that tossing a die A
results in an odd number of eyes. AC is then the event that we get an even number of eyes. S AC

3.1.3 Probability and Kolmogorov’s axioms

Probability is a set function Pr() which maps events A1 , A2, ... in the sample space S to real
numbers. The function Pr(·) can only take values in the interval from 0 to 1, i.e. probabilities are
greater or equal than 0, and less or equal than 1. Kolmogorov established the following axioms

A B

0 Pr(A) Pr(B) 1

Figure 3.2: Mapping of events on the interval [0,1]

which all probability rules could be derived from:

1. 0 ≤ Pr(A)

23
 2. Pr(S) = 1
3. If A1 , A2 , A3 ,... is a sequence of disjoint events we shall then have:
Pr(A1 ∪ A2 ∪ . . .) = Pr(A1 ) + Pr(A2 ) + . . .
The axioms are the basis for establishing calculation rules when dealing with probabilities, but
they do not help us in establishing numerical values for the basic probabilities Pr(A1 ), Pr(A2 ),
etc. Historically two lines of thoughts have been established, the classical (frequentiest) and the
Bayesian approach. In the classical thinking we introduce the concept of a random experiment,
where Pr(Ai ) is the relative frequency with which the event Ai occurs. The probability could then
be interpreted as a property of the experiment, or a property of the world. By letting nature reveal
itself by doing experiments, we could in principle establish all probabilities that are of interest.
Within the Bayesian framework probabilities are interpreted as subjective believe about whether
Ai will occur or not. Probabilities is then not a property of the world, but rather a measure of the
knowledge and understanding we have about a phenomenon.
Before we set up the basic rules for probability theory that we will need, we introduce the
concepts of conditional probability and independent events.

Conditional probability. Pr(A|B) denotes the conditional probability that A will occur given
that B has occurred.

Independent events. A and B are said to be independent if information about whether B has
occurred does not influence the probability that A will occur, i.e., Pr(A|B) = Pr(A).

Basic rules for probability. The following calculation rules for probability apply:
Pr(A ∪ B) = Pr(A) + Pr(B) − Pr(A ∩ B) (3.1)
Pr(A ∩ B) = Pr(A) · Pr(B) if A and B are independent (3.2)
Pr(A ) = Pr(A does not occur) = 1 − Pr(A)
C
(3.3)
Pr(A ∩ B)
Pr(A|B) = (3.4)
Pr(B)
Example 3.1
Let the two events A and B be defined by A = {It is Sunday} and B = {It is between 6 and 8 pm).
First we note that A and B are independent but not disjoint. We will find Pr(A ∩ B), Pr(A ∪ B)
and Pr(A|B)
1 2 1
Pr(A ∩ B) = Pr(A) · Pr(B) =
· =
7 24 84
1 2 1 9
Pr(A ∪ B) = Pr(A) + Pr(B) − Pr(A ∩ B) = + − =
7 24 84 42
Pr(A ∩ B) 1/84 1
Pr(A|B) = = =
Pr(B) 2/24 7
♢

3.1.4 The law of total probability

In many situations it is easier to assess the probability of an event B conditionally on some other
events, say A1 , A2 , . . ., Ar , than unconditionally. The law of total probability could then be used

24
to assess the unconditional probability. Now, we say that A1 , A2 , . . ., Ar is a division of the sample
space if the union of all Ai ’s covers the entire sample space, i.e. A1 ∪ A2 ∪ … ∪ Ar = S and the Ai ’s
are pair wise disjoint, i.e. Ai ∩ Aj = Ø for i ̸= j. An example is shown in Figure 3.3.

A2
A4
A1
A3
S

Figure 3.3: Division of the sample space

Let A1 , A2 , . . ., Ar represent a division of the sample space S, and let B be an arbitrary event in S.
The law of total probability now states:

∑
r
Pr(B) = Pr(Ai ) · Pr(B|Ai ) (3.5)
i=1

Example 3.2
A special component type is ordered from two suppliers A1 and A2 . Experience has shown that
components from supplier A1 has a defect probability of 1%, whereas components from supplier
A2 has a defect probability of 2%. In average 70% of the components are provided by supplier A1 .
Assume that all components are put on a common stock, and we are not able to trace the supplier
for a component in the stock. A component is now fetched from the stock, and we will calculate
the defect probability, Pr(B):

∑
r
Pr(B) = Pr(Ai ) · Pr(B|Ai ) = Pr(A1 ) · Pr(B|A1 ) + Pr(A2 ) · Pr(B|A2 ) =
i=1
0.7 · 0.01 + 0.3 · 0.02 = 1.3%

3.1.5 Bayes theorem

Now consider the example above, and assume that we have got a defect component from the stock
(event B). We will derive the probability that the component originates from supplier A1 . We then
use Bayes formula that states if A1 , A2 , . . ., Ar represent a division of the sample space, and B is
an arbitrary event then:

Pr(B|Aj ) · Pr(Aj )
Pr(Aj |B) = (3.6)
∑
r
Pr(Ai ) · Pr(B|Ai )
i=1

25
Example 3.3
We have
Pr(B|A1 ) · Pr(A1 ) 0.01 · 0.7
Pr(A1 |B) = = = 0.54
∑
r
0.013
Pr(Ai ) · Pr(B|Ai )
i=1

Thus, the probability of A1 is reduced from 0.7 to 0.54 when we know that the component is defect.
The reason for this is that components from supplier A1 are the best ones, and hence when we
know that the component was defect, it is less likely that it was from supplier A1 . ♢

3.1.6 Stochastic variables

Stochastic variables are used to describe quantities which can not be predicted exactly. Note that
the term ‘random quantity’ is often used to denote a stochastic variable.

X is stochastic ⇔ Cannot say precisely the value X has or will take

To be more precise, a stochastic variable X is a real valued function that assigns a quantitative
measure to each event ei in the sample space S. Often the underlying events, ei are of little
interest. We are only interested in the stochastic variable X measured by some means. Examples
of stochastic variables are given below:

• X = Life time of a component (continuous)

• R = Repair time after a failure (continuous)

• T = Duration of a construction project (continuous)

• C = Total cost of a renewal project (continuous)

• M = Number of delayed trains next month (discrete)

• N = Number of customers arriving today (discrete)

• S = Service time for the furst customer arriving today (continious)

• W = Maintenance and operational cost next year (continuous)

Remark: We distinguish between continuous and discrete stochastic variables. Continuous

stochastic variables can take any value among the real numbers, whereas discrete variables can
take only a finite (or countable finite) number of values. ♢

Cumulative distribution function. A stochastic variable X is characterized by it’s cumulative

distribution function

FX (x) = Pr(X ≤ x) (3.7)

We use subscript X to emphasise the relation to the cumulative distribution function of the quantity
X. The argument (lowercase x) states which values the stochastic variable X could take, or is of
our interest. From the expression we observe that FX (x) states the probability that the random
quantity X is less or equal than (the numeric value of) x. A typical distribution function is shown

26
in Figure 3.4. Note that the distribution function is strictly increasing, and 0 ≤ FX (x) ≤ 1. From
FX (x) we can obtain the probability that X will be within a specified interval, [a,b):

Pr(a < X ≤ b) = FX (b) − FX (a) (3.8)

FX (x)
1

x
Figure 3.4: Cumulative distribution function, FX (x)

Example 3.4
Assume that the probability distribution function of X is given by FX (x) = 1 − e−(0.01x) , and we
2

will find the probability that X is in the interval (100,200]. From Equation (3.8) we have:

Pr(100 < X ≤ 200) = FX (200) − FX (100) =

[ ] [ ]
1 − e−(0.01·200) − 1 − e−(0.01·100) = e−1 − e−4 ≈ 0.35
2 2

♢Note that the index X representing

the stochastic variable X often is dropped if it is obvious which stochastic variable we are working
with. Note also the distinction between lowercase and uppercase letters. The uppercase X is used
to denote a stochastic variable, for example number of customers arriving next day. The lowercase
x is just a representation of possible values X can take. For example X = 3.

Probability density function. For a continuous stochastic variable, the probability density
function is given by
d
fX (x) = FX (x) (3.9)
dx
The probability density function expresses how likely the various x-values are. Note that for

fX (x)

x
Figure 3.5: Probability density function, fX (x)

continuous random variables the probability that X will take a specific value vanishes. However,

27
the probability that X will fall into a small interval around a specific value is positive. For each
x-value given in Figure 3.5 fX (x) could be interpreted as the probability that X will fall within a
small interval around x divided by the length of this interval. Especially we have:
∫x
FX (x) = fX (u)du (3.10)
−∞

and
∫b
Pr(a < X ≤ b) = fX (x)dx (3.11)
a

The last expression is illustrated in Figure 3.6.

fX (x)

x
a b
Figure 3.6: The shadded area equals Pr(a < X ≤ b)

Random quantities that take discrete values are said to be discretely distributed. For such
quantities we introduce the point probability for X in the point xj :

p(xj ) = Pr(X = xj ) (3.12)

where x1 , x2 , . . . are possible values X could take.

Expectation. The expectation (mean) of X is given by

 ∫∞

 x · fX (x) dx if X is continuous
E(X) = −∞
∑ (3.13)

 xj · p(xj ) if X is discrete
j

The expectation can be interpreted as the long time run average of X, if an infinite amount of
observations are available.

Median. The median of a distribution is the value m0 of the stochastic variable X such that
Pr(X ≤ m0 ) ≥ 1/2 and Pr(X ≥ m0 ) ≥ 1/2. In other words, the probability at or below m0 is at
least 1/2, and the probability at or above m0 is at least 1/2.

Mode. The mode of a distribution is the value M of the stochastic variable X such that the
probability density function, or point probability at M is higher or equal than for any other value
of the stochastic variable. We sometimes used the term ‘most likely value’ rather than mode.

28
Variance. The variance of a random quantity expresses the variation in the value X will take in
the long run. We denote the variance of X by:
 ∫∞

 [x − E(X)]2 · fX (x) dx if X is continuous
Var(X) = −∞
∑ (3.14)

 [(xj − E(X)]2 · p(xj ) if X is discrete
j

Standard deviation. The standard deviation of X is given by

√
SD(X) = + Var(X) (3.15)

The standard deviation defines an interval which observations are likely to fall into, i.e., if 100
observations are available, we expect that approximate1 67 of these observations fall in the interval
[E(X) − SD(X), E(X) + SD(X)].

1
Precision. The precision, P , is the reciprocate of the variance, i.e. P = Var(X) .

α-percentiles. The upper α-percentile, xα , in a distribution FX (x) is the value satisfying α =

Pr(X > xα ) = 1 − FX (xα ).
We end this section by giving some results regarding expectation and variances. These results
apply when it is easier to express the expectation and variance of one variable if we condition on
the value of another variable.

Result 3.1 Double expectation

Let X and Y be stochastic variables. We then have:

E(X) = E(E(X Y )) (3.16)

Var(X) = E(Var(X Y )) + Var(E(X Y )) (3.17)

♢
It follows easily that

E(X) = E(X B) Pr(B) + E(X B C ) Pr(B C ) (3.18)

Var(X) = Var(X|B) Pr(B) + Var(X|B C ) Pr(B C )

[ ]2
+ [E(X|B) − E(X)]2 Pr(B) + E(X|B C ) − E(X) Pr(B C ) (3.19)

3.2 Common probability distributions

In this section we will present some common probability distributions. We write X ∼ <Name
of distribution>(<parameters>) to express that X belongs to <Name of distribution>, and with
parameters <parameters>. Sometimes we also use an abbreviation for the distribution, for example
we write X ∼ N (3, 4) to express that X is normally distributed with expectation 3 and variance 4.
1
This result is valid for the normal distribution. For other distributions there may be deviation from this result.

29
3.2.1 The normal distribution
X is said to be normally distributed if the probability density function of X is given by:
2
1 1 − (x−µ)
fX (x) = √ e 2σ2 (3.20)
2π σ
where µ and σ are parameters that characterise the distribution. The mean and variance are given
by:

E(X) = µ
Var(X) = σ 2 (3.21)

The distribution function for X could not be written on closed from. Numerical methods are
required to find FX (x). It is convenient to introduce a standardised normal distribution for this
purpose. We say that U is standard normally distributed if it’s probability density function is given
by:
1 u2
fU (u) = ϕ(u) = √ e− 2 (3.22)
2π
We then have
∫u ∫u
1 t2
FU (u) = Φ(u) = ϕ(t)dt = √ e− 2 dt (3.23)
2π
−∞ −∞

and we observe that the distribution function of U does not contain any parameters. We therefore
only need one look-up table or function representing Φ(u). A look-up table is given in Table 3.1.
To calculate probabilities in the non-standardised normal distribution we use the following result:

Result 3.2 If X is normally distributed with parameters µ and σ, then

X −µ
U= (3.24)
σ
is standard normally distributed. ♢
∫a
In many situations we are interested in calculating the “truncated expectation” −∞ xf (x)dx. For
the nomral distribution the following result may be used:

Result 3.3 Let X be normally distributed with parameters µ and σ. We then have:
∫ a ( ) ( )
a−µ a−µ
xf (x)dx = µΦ − σϕ (3.25)
−∞ σ σ

where Φ() and ϕ() are the CDF and PDF for the standard normal distribution respectively. ♢
∫a ∫ (a−µ)/σ
To prove Equation (3.25) first introduce u = (x − µ)/σ yielding −∞ xf (x)dx = −∞ (σu −
µ)ϕ(u)du. The µϕ(u) part of the integral is directly found by the Φ() function whereas for the
√ ∫ (a−µ)2 /2σ2 −z
σuϕ(u) part introduce z = −u2 /2 yielding −σ/ 2π −∞ e dz. The result then follows.

30
Example 3.5 Calculation in the normal distribution
Let X be normally distributed with parameters µ = 5 and σ = 3. We will find Pr(3 < X ≤ 6). We
have:
3−µ X −µ 6−µ 3−5 6−5
Pr(3 < X ≤ 6) = Pr( < ≤ ) = Pr( <U ≤ )
( ) ( ) σ σ σ 3 3
1 −2
=Φ −Φ = Φ(0.33) − (1 − Φ(0.67)) = 0.629 − 1 + 0.749 = 0.378
3 3

Problem 3.1 Consider the example in Example 3.5, and carry out the calculation by means of
the pRisk.xlsm program. ♢

Problem 3.2 Let X be the height of men in a population, and assume X is normally distributed
with parameters µ = 181 and σ = 4. How large percentage of the population is more than 190 cm?
♢

3.2.2 The exponential distribution

X is said to be exponentially distributed if the probability density function of X is given by:

fX (x) = λe−λx (3.26)

The cumulative distribution function is given by:

FX (x) = 1 − e−λx (3.27)

and the mean and variance are given by:

E(X) = 1/λ
Var(X) = 1/λ2 (3.28)

Note that for the exponential distribution, X will always be greater than 0. The parameter λ
is often denoted the intensity in the distribution

Example 3.6
We will obtain the probability that X is greater than it’s expected value. We then have:

Pr(X > E(X)) = 1 − Pr(X ≤ E(X)) = 1 − FX (E(X)) = e−λE(X) = e−1 ≈ 0.37

3.2.3 The Weibull distribution

X is said to be Weibull distributed if the probability density function of X is given by:

fX (x) = αλ(λx)α−1 e−(λx)

α
(3.29)

The cumulative distribution function is given by:

FX (x) = 1 − e−(λx)
α
(3.30)

31
and the mean and variance are given by:
[ )
1 1
E(X) = Γ +1
λ α
( ( ) ( )]
1 2 1
Var(X) = 2 Γ +1 −Γ 2
+1 (3.31)
λ α α
where Γ(·) is the gamma function. Note that in the Weibull distribution X will also always be
positive.

3.2.4 The gamma distribution

X is said to be gamma distributed if the probability density function of X is given by:
λα
fX (x) = (x)α−1 e−λx (3.32)
Γ(α)
α is denoted the shape parameter whereas λ is denoted the scale parameter. For integer values of
α the gamma distribution is often denoted the Erlang distribution. The cumulative distribution
function could then be found on closed form:
∑
α−1
(λx)α −(λx)
FX (x) = 1 − e (3.33)
n!
n=0

For non-integer values of α numerical methods are required to obtain the cumulative distribution
function. The mean and variance are given by:
α
E(X) =
λ
α
Var(X) = 2 (3.34)
λ
If we know the expectation E and the variance V in the gamma distribution, we may obtain the
parameters α and λ by: λ = E/V , and α = λ · E. The gamma distribution is often used as a prior
distribution in a Bayesian approach.
For integer values of α the gamma distribution and in particular the Erlang distribution may
be seen as a distribution for a sum of exponentially distributed stochastic variables:

Result 3.4 Let

∑ Z1 , Z2 , . . . Zk be independent and exponentially distributed with parameter λ. The
variable X = ki=1 Zi is then gamma distributed with shape parameter k and scale parameter λ.

3.2.5 The inverted gamma distribution

X is said to be inverted gamma distributed if the probability density function of X is given by:
( )α+1
λα 1
fX (x) = e−λ/x (3.35)
Γ(α) x
The mean and variance are given by:
E(X) = λ/(α − 1)
Var(X) = λ (α − 1)−2 (α − 2)−1
2
(3.36)
Note that if X is gamma distributed with parameters α and λ, then Y = X −1 has an inverted
gamma distribution with parameters α and 1/λ. If we know the expectation, E and the variance,
V , of an inverted gamma distribution we could obtain α and λ by α = E 2 /V +2, and λ = E ·(α−1).

32
3.2.6 The lognormal distribution
X is said to be lognormal distributed if the probability density function of X is given by:
1 1 1 − 1 2 (log x−ν )2
fX (x) = √ e 2τ (3.37)
2π τ x
We write X ∼ LN(v,τ ). The mean and variance of X is given by
1 2
E(X) = eν+ 2 τ
2 2
Var(X) = e2ν (e2 τ − eτ ) (3.38)

The following result could be utilised:

Result 3.5 If X is lognormally distributed with parameters ν and τ , then Y = ln X is normally

distributed2 with expected value ν and variance τ 2 . ♢

3.2.7 The binomial distribution

Before the binomial distribution is defined, binomial trials are defined. Let A be an event, and
assume that the following holds:

i) n trials are performed, and in each trial we record whether A occurs or not.

ii) The trials are stochastic independent of each other.

iii) For each trial Pr(A) = p

When i)-iii) is satisfied, we say that we have binomial trials. Now let X be the number of times
event A occurs in such a binomial trial. X is then a stochastic variable with a binomial distribution.
This is written X ∼ Bin(n, p).
The probability function is given by
( )
n px
Pr(X = x) = (1 − p )n−x for x = 0, 1, 2, .., n (3.39)
x

The cumulative distribution function Pr(X ≤ x) is given in statistical tables. For the binomial
distribution, expectation and variance are given by:

E(X) = np
Var(X) = np(1 − p) (3.40)

3.2.8 The Poisson distribution

The Poisson distribution is often appropriate in the situation where the stochastic variable may
take the values 0,1,2,. . ., and where the expected number of occurrences is proportional to an
exposure measure such as time or space. For the Poisson distribution we have the following point
distribution:
λx −λ
p(x) = Pr(X = x) = e (3.41)
x!
2
ln(·) is the natural logarithm function

33
For the poison distribution, expectation and variance are given by:

E(X) = λ
Var(X) = λ (3.42)

It can be proved that the Poisson distribution is appropriate if the following situation applies:
Consider the occurrence of a certain event (e.g. a component failure) in an interval (a, b), and
assume the following:

1. A could occur anywhere in (a,b), and the probability that A occurs in (t, t + ∆t) is approxi-
mately equal to λ∆t, and is independent of t (∆t should be small).

2. The probability that A occurs several times in (t, t + ∆t) is approximately 0 for small values
of ∆t.

3. Let I1 og I2 be disjoint intervals in (a, b). The event A occurs within I1 is independent of if
the event A occurs in I2 .

When the criteria above are fulfilled we say we have a Poisson point process with intensity λ. The
number of occurrences (X) of A in (a, b) is then Poisson distributed with parameter λ(b − a):

[λ(b − a)]x −λ(b−a)

p(x) = Pr(X = x) = e (3.43)
x!

Result 3.6 In a Poisson point process with parameter λ the times between the occurrence of the
event A are exponentially distributed with parameter λ.

3.2.9 The inverse-Gauss distribution

The inverse-Gauss distribution is often used when we have an “under laying” deterioration process.
If this deterioration process follows a Wiener process with drift η and diffusion constant δ 2 , the
time3 T , until the first time the process reaches the value ω will be Inverse-Gauss distributed with
parameters µ = ω/η, and λ = ω 2 /δ 2 .
If the failure progression Ω(t) follows a Wiener process it could be proven that Ω(t) - Ω(s) is
normally distributed with expected value η(t − s) and variance δ 2 (t - s). That is η is the average
growth rate in the process, whereas δ 2 is an expression for the variation of the growth around the
average value.
For the inverse-Gauss distribution we have:
(√ ( )) ( √ ( ))
λ t λ t
FT (t) = Φ −1 )+Φ − +1 e2λ/µ (3.44)
t µ t µ

where Φ is the standard normal (standard Gaussian) distribution c.d.f.

and

E(T ) = µ (3.45)
3
Var(T ) = µ /λ (3.46)
3
We use the symbol T rather than the more general symbol X here since this modell is so explicitly linked to the
time.

34
3.2.10 The triangular distribution
The triangular distribution has a probability density function that comprises a triangle. The lover
left corner points out the lowest value (L), the upper right corner points out the highest value
(H). Finally, the x-value of the third corner points out the most probable value, or mode (M ).The
probability density function for the triangular distribution is given by:
{ 2(x−L)
(M −L)(H−L) if L ≤ x ≤ M
fX (x) = 2(H−x) (3.47)
(H−M )(H−L) if M ≤ x ≤ H

The cumulative distribution function is given by:

{ (x−L)2
(M −L)(H−L) if L ≤x≤M
FX (x) = (H−x)2 (3.48)
1 − (H−M )(H−L) if M ≤ x ≤ H

and the mean and variance are given by:

L+M +H
E(X) =
3
L2 + M 2 + H 2 − LM − LH − M H
Var(X) = (3.49)
18
Problem 3.3 Assume that the completion of a project is triangular distributed with parameters
L = 200, M = 240 and H = 350. In our contract we have committed our selves to finish the
project within 220 days. After 220 days we have to pay a penalty of 100 Euro per day in penalty
for default. Find the total expected penalty for default in this project. ♢

Problem 3.4 Consider Problem 3.3 and assume that a special building method could reduce H
from 350 to 300, leaving L and M unchanged. This will cost 2,000 Euro extra. Do a cost benefit
analysis of this option. ♢

3.2.11 The PERT distribution

The PERT distribution has as the triangular distribution three parameters, L (lowest value), M
(most likely value), and H (highest value). To give the probability density function for the triangular
distribution we first define:
4M + H − 5L
α1 =
H −L
5H − 4M − L
α2 =
H −L
x−L
z= (3.50)
H −L
The probability density function is now given by:

(x − L)α1 −1 (H − x)α2 −1
fX (x) = (3.51)
B(α1 , α2 )(H − L)α1 +α2 −1

where B(·, ·) is the beta function. The cumulative distribution function is given by:

35
 Bz (α1 , α2 )
FX (x) = (3.52)
B(α1 , α2 )

where Bz (·, ·) is the incomplete beta function. The mean and variance are given by:

L + 4M + H
E(X) =
6
(E(X) − L)(H − E(X))
Var(X) = (3.53)
7

Problem 3.5 Use the pRisk.xlsm program to find Pr(X ≤ 7) if X ∼ PERT(L = 3, M = 6, H =

10). ♢

Problem 3.6 Consider a situation where the unconditional distribution of the duration of a project
groundwork activity is PERT distributed with parameters L = 0.5, M = 1.5 and H = 3.5 days.
By a detailed analysis into the uncertainty of the situation we recognize that frozen soil is a major
factor to the long duration. Let B represent the event that it is frozen soil. We now make the
following assessment: Given frozen soil, the duration of the activity, T B, is PERT distributed with
parameters L = 2, M = 2.5 and H = 3.5, and if the soil is not frozen the duration of the activity,
T B C , is PERT distributed with parameters L = 0.5, M = 1 and H = 2.5. Find p = Pr(B) such
that the expectation in the conditional situation is the same as in the unconditional situation. Hint:
You may use that E(T ) = E(T B) Pr(B) + E(T B C ) Pr(B C ), see Equation ( 3.18). ♢

Problem 3.7 Make a sketch of the unconditional probability distribution function in the situation
in Problem 3.6 when the consideration of frozen soil is taken into account. ♢

Problem 3.8 Find the unconditional variance of the duration in Problem 3.6. Hint: You may use
equation (3.19). ♢

Problem 3.9 Consider again the situation in Problem 3.6, i.e. we let in the first place T ∼
PERT(L = 0.5, M = 1.5, H = 3). Also Let B represent frozen soil and Pr(B) = 0.2. We now
introduce three factors, fB , fB C and fV that relate the conditional situation to the original situation.
The parameters relevant in the conditional situation are {LB , MB , HB } and {LB C , MB C , HB C } in
the situation where B occurs, and B does not occur respectively. We now let MB = fB · M ,
LB = MB − fV · (M − L), HB = MB + fV · (H − M ), MB C = fB C · M , LB C = MB C − fV · (M − L),
and HB C = MB C + fV · (H − M ). Let fB = 1.5 and fV = 0.5. Find by an iterative procedure the
value of fB C such that the expectation of T is equal to the original expectation. Next find fV by
a similar iterative procedure such that the variance of T is equal to the original variance. ♢

3.3 Assessment of parameters in parametric distributions

We have in the previous section discussed parametric probability distributions. Common for all
these distributions is that they involve parameters. When using a parametric distribution, we
also need to assess the parameters. In this presentation we will not discuss in detail how this
could be done. If we have access to experience data, we could estimate these parameters by e.g.
the maximum likelihood principle. In other situations where we have no, or very little data we

36
would use expert judgement to assess the parameters, see e.g., [11] for further discussion on expert
judgement. In this presentation we will very often assume that the uncertainty in a quantity,
e.g. the duration of an activity could be described by a so-called triple estimate {L, M, H}. We
will then as a general rule assume that the corresponding parametric distribution is the PERT
distribution. We will further assume that the L value is the absolute minimum, and that the H
value is the absolute maximum the quantity could take. It is, however, important to realise that in
other presentation the L and H values are treated as lower and upper quantiles in the distribution,
and often a 90% interval is assumed. This is even the situation for the PERT distribution which is
defined for a finite domain. So if we for a given triple estimate should establish the expected value,
and the standard deviation we should be careful regarding the interpretation of the triple estimate.

3.4 Distribution of sums, products and maximum values

3.4.1 Distribution of sums
If X1 , X2 ,…,Xn are random variables we might obtain the expected value, the variance and the
standard deviation of the sum of the x-es:
(∑n ) ∑n
E(X1 + X2 + . . . + Xn ) = E Xi = E(Xi ) (3.54)
i=1 i=1

(∑ n ) ∑n
Var(X1 + X2 + . . . + Xn ) = Var Xi = Var(Xi ) (3.55)
i=1 i=1

(∑ n ) √
∑n
SD Xi = [SD(Xi )]2 (3.56)
i=1 i=1

Note that Equations (3.55) and (3.56) are only valid if the x-es are stochastically independent. If
there is dependency between the x-es we need to include a covariance term, e.g., if we only have
two variables X1 and X2 we have:

Var(X1 + X2 ) = Var(X1 ) + Var(X2 ) + 2Cov(X1 , X2 ) (3.57)

where Cov(X1 , X2 ) is the covariance between X1 and X2 .

The results above help us in determine the expectation and variance of a sum of stochastic
variables, but the results could not be used to establish the probability distribution of the sum. In
the following we refer some results we could utilise in many situations.

Result 3.7 Sum of normally distributed stochastic variables

Let
∑n X1 , X2 ,…,Xn be independent normally distributed.∑Let Y be the sum of the x-es,
∑n i.e. Y =
n
i=1 Xi . Y is then normally distributed with E(Y ) = i=1 E(Xi ) and Var(Y ) = i=1 Var(Xi ).
♢

Result 3.8 Sum of exponentially distributed stochastic variables

Let X1 , X2 ,…,Xn ∑
independent exponentially distributed with parameter λ. Let Y be the sum of
the x-es, i.e. Y = ni=1 Xi . Y is then gamma distributed with parameters n and λ. ♢

Result 3.9 Sum of gamma distributed stochastic variables

Let X1 , X2 ,…,Xn independent
∑ gamma distributed with parameters α and λ. Let Y be the sum of
the x-es, i.e. Y = ni=1 Xi . Y is then gamma distributed with parameters nα and λ. ♢

37
Result 3.10 Central limit theorem
Let X1 , X2 ,…,Xn be a sequence of identical independent distributed stochastic variables with ex-
pected value µ and standard deviation σ. As n approaches infinity, the average value of the x-es
√
will asymptotically have a normal distribution with expected value µ and standard deviation σ/ n.
Similarly, the sum of the x-es will asymptotically have a normal distribution with expected value
√
nµ and standard deviation σ n. ♢
Several generalizations for finite variance exist which do not require identical distribution but
incorporate some conditions which guarantee that none of the variables exert a much larger influence
than the others. Two such conditions are the Lindeberg condition and the Lyapunov condition.
Now, as n approaches ∑infinity, the sum of the ∑
x-es will asymptotically have a normal distribution
n
with expected value i=1 E(Xi ) and variance ni=1 Var(Xi ).

Problem 3.10 Consider a project consisting of n activities that follow each other in time. Let
each activity have a PERT distribution with parameters L = 3, M = 5 and H = 10. Use the Monte
Carlo simulation procedure in the pRisk.xlsm program to find the cumulative distribution function
for the total duration of the project. Compare the result with using the Central Limit Theorem for
various values of n. How large should n be in order to give a reasonable approximation by using
the normal distribution? ♢

3.4.2 Distribution of a product

If X1 , X2 ,…,Xn are independent stochastic variables we might obtain the expected value, the vari-
ance and the standard deviation of the product of the x-es:
( n )
∏ ∏
n
E(X1 · X2 · . . . · Xn ) = E Xi = E(Xi ) (3.58)
i=1 i=1

The results for the variance and standard deviation is more complicated, and we only present the
results for n=2.

Var(X1 X2 ) = Var(X1 )Var(X2 ) + Var(X1 ) [E(X2 )]2 + Var(X2 ) [E(X1 )]2 (3.59)

√
SD(X1 X2 ) = Var(X1 )Var(X2 ) + Var(X1 )[E(X2 )]2 + Var(X2 )[E(X1 )]2 (3.60)

Problem 3.11 Show that Equation (3.59) is correct by using the fact that Var(X) = E(X 2 ) −
[E(X)]2 . ♢

Problem 3.12 Use the program pRisk.xlsm to simulate the mean and standard deviation of the
product X1 X2 if both X1 and X2 are independent and normally distributed with expected value
10 and standard deviation 2. Compare the result with the exact result. ♢

3.4.3 Distribution of maximum values

Let X1 and X2 be independent stochastic variables, and let Y = max(X1 , X2 ). The cumulative
distribution function of Y is given by:

FY (x) = Pr(Y ≤ x) = Pr(X1 ≤ x ∩ X2 ≤ x)

= Pr(X1 ≤ x) Pr(X2 ≤ x) = FX1 (x)FX2 (x) (3.61)

38
In this situation we could easily obtain the distribution of the maximum of two stochastic variables,
but it is not so easy to obtain the expectation and variance. However, since the probability density
function, fY (x) is the derivative of FY (x) we find:
∫∞ ∫∞
E(Y ) = x · fY (x) dx = x · [fX1 (x)FX2 (x) + fX2 (x)FX1 (x)] dx (3.62)
−∞ −∞

∫∞
Var(Y ) = [x − E(Y )]2 · [fX1 (x)FX2 (x) + fX2 (x)FX1 (x)] dx (3.63)
−∞

Problem 3.13 Find the expectation and standard deviation of Y = max(X1 , X2 ) if X1 and X2 are
independent and normally distributed with µ1 = E(X1 ) = 10, µ2 = E(X2 ) = 7, σ1 = SD(X1 ) = 2,
and σ2 = SD(X2 ) = 3. Hint: You might use the routine for numerical integration implemented in
the pRisk.xlsm program. ♢

Problem 3.14 Consider the problem above, but now find the result by using the Monte Carlo
simulation procedure in the pRisk.xlsm program. ♢

Problem 3.15 Consider the problem above, but now find the result by using the EMax and VarMax
functions in the pRisk.xlsm program. ♢

39
 Table 3.1: The Cumulative Standard Normal Distribution
∫z
1 u2
Φ(z) = Pr(Z ≤ z) = √ e− 2 du
2π
−∞

z .00 .01 .02 .03 .04 .05 .06 .07 .08 .09
0.0 .500 .504 .508 .512 .516 .520 .524 .528 .532 .536
0.1 .540 .544 .548 .552 .556 .560 .564 .567 .571 .575
0.2 .579 .583 .587 .591 .595 .599 .603 .606 .610 .614
0.3 .618 .622 .626 .629 .633 .637 .641 .644 .648 .652
0.4 .655 .659 .663 .666 .670 .674 .677 .681 .684 .688
0.5 .691 .695 .698 .702 .705 .709 .712 .716 .719 .722
0.6 .726 .729 .732 .732 .739 .742 .745 .749 .752 .755
0.7 .758 .761 .764 .767 .770 .773 .776 .779 .782 .785
0.8 .788 .791 .794 .797 .800 .802 .805 .808 .811 .813
0.9 .816 .819 .821 .824 .826 .829 .831 .834 .836 .839
1.0 .841 .844 .846 .849 .851 .853 .855 .858 .860 .862
1.1 .864 .867 .869 .871 .873 .875 .877 .879 .881 .883
1.2 .885 .887 .889 .891 .893 .894 .896 .898 .900 .901
1.3 .903 .905 .907 .908 .910 .911 .913 .915 .916 .918
1.4 .919 .921 .922 .924 .925 .926 .928 .929 .931 .932
1.5 .933 .934 .936 .937 .938 .939 .941 .942 .943 .944
1.6 .945 .946 .947 .948 .949 .951 .952 .953 .954 .954
1.7 .955 .956 .957 .958 .959 .960 .961 .962 .962 .963
1.8 .964 .965 .966 .966 .967 .968 .969 .969 .970 .971
1.9 .971 .972 .973 .973 .974 .974 .975 .976 .976 .977
2.0 .977 .978 .978 .979 .979 .980 .980 .981 .981 .982
2.1 .982 .983 .983 .983 .984 .984 .985 .985 .985 .986
2.2 .986 .986 .987 .987 .987 .988 .988 .988 .989 .989
2.3 .989 .990 .990 .990 .990 .991 .991 .991 .991 .992
2.4 .992 .992 .992 .992 .993 .993 .993 .993 .993 .994
2.5 .994 .994 .994 .994 .994 .995 .995 .995 .995 .995
2.6 .995 .995 .996 .996 .996 .996 .996 .996 .996 .996
2.7 .997 .997 .997 .997 .997 .997 .997 .997 .997 .997
2.8 .997 .998 .998 .998 .998 .998 .998 .998 .998 .998
2.9 .998 .998 .998 .998 .998 .998 .999 .999 .999 .999
3.0 .999 .999 .999 .999 .999 .999 .999 .999 .999 .999

Φ(-z) = 1 - Φ(z)

40
Chapter 4

Schedule

In order to analyse the duration of a project, or a project activity we use flow network models. In
the literature there are two different ways to represent a project in a network model. Activity on
node (AON) networks uses the nodes to represent the project activities whereas the arcs are used
to represent the sequence and dependencies between activities. Activity on arc (AOA) networks
uses the arcs to represent the project activities whereas the nodes are used to represent events
in the project, i.e., starting and finalizing activities. In this presentation we will use the AON
representation of a project.
Visually, a flow network model is similar to a bar chart, or a Gantt diagram. However, we
usually indicate dependencies between activities with arrows, and the y and x axes are usually not
labelled. The symbols used in a flow network used in this presentation are shown in Figure 4.1.
An example flow network diagram is shown in Figure 4.2.

Activity

Uncertain activity

Milestone

S Start point

F End point, finish

Coupling between activities or nodes

Figure 4.1: Symbols used in flow network diagrams

There exist several methods for analysing flow networks. All these models requires that the
flow network is described completely in terms of dependencies between the activities. Further the
duration of the activities should be described by probability distribution functions with numeric
values for the parameters. When analysing such flow network we differentiate between:

• Analytical methods.

• Monte Carlo simulation methods.

41
 D

S A B E F H F

C G
Figure 4.2: Example flow network from [3]

Generally we let T denote the duration of the project we are analysing, or a part of the project,
e.g. a work package. If the project comprises n activities, we often denote these activities Ai , and
the duration of activity Ai is denoted Ti . Sometimes in this presentation we also use the more
simplified notation where each activity is described by a letter, e.g. A, B etc. The main purpose
of the schedule analysis is to establish the cumulative distribution function for the entire project
duration. We might also want to establish the cumulative distribution function for parts of the
project, milestones etc. Another important measure of interest is the probability that an activity
will delay the project, i.e. the criticality index. The methods we will investigate are:

• Critical Path Method (CPM)

• Program Evaluation and Review Technique (PERT)

• Successive schedule planning (SSP)

• Monte Carlo simulation

The example diagram shown in Figure 4.2 will be used to demonstrate the various methods. This
example is adapted from [3]. The parameters to describe the duration of each activity are given
in Table 4.2. Fundamental for all methods is to understand the term ’path’. A path in a flow
network is a set of activities from the the starting point to the end point in the network, where
each activity in the set follows another activity in the set except the first activity that follows the
starting point. This means that all activities in a path have to be executed in order to complete
the project. Usually there are several paths in a flow network. Formally, we also include uncertain
activities in a path, even if they might not be necessary to execute.

Activity L (Lowest) M (Most likely) H (Highest)

A 2 5 9
B 4 6 9
C 7 12 21
D 5 7 10
E 4 7 11
F 2 3 6
G 3 5 9
H 5 7 10

Table 4.1: Data for the schedule demonstration example

42
4.1 Critical Path Method (CPM)
The idea of the CPM method is to find all paths in the flow network. Next, we assume that the
duration of all activities are deterministic, and typically equal to the most likely duration (M ).
The duration of each path is given as the sum of duration of all activities in the path. The path
with the longest duration is denoted a critical path, and the duration of the project is found by the
duration of the critical path (or all critical paths in case of several critical paths). In Figure 4.2 we
have the following paths: P1 = {A,B,D,F,H}, P2 = {A,B,E,F,H} and P3 = {A,C,G,H}. Inserting
the duration of each activity, we get the following durations TP1 = 5 + 6 + 7 + 3 + 7 = 28, TP2
= 5 + 6 + 7 + 3 + 7 = 28 and TP3 = 5 + 12 + 5 + 7 = 29 for P1 , P2 and P3 respectively. Since
P3 has the longest duration, P3 is a critical path, and the project duration is found to be 29. A
disadvantage of the CPM method is that it cannot handle the uncertainty in the duration of each
activity, i.e. it is a deterministic approach.

4.2 Program Evaluation and Review Technique (PERT)

The PERT method is similar to the CPM method when finding the project duration. We also here
find the critical path, i.e. P3 . But rather than using the deterministic value for the duration, we
now treat uncertainties in the activities in the critical path. The expectation and variance of a sum
are given by Equations (3.54) and (3.55). If the expectations and variances are given for the various
activities, we can proceed directly. However, very often the activities are described by a low (L),
most likely (M ), and high value (H) as in Table 4.2. If we now assume that these parameters are
describing the PERT distribution, we recall that the expected value is given by µ = (L+4M +H)/6
and the variance is given by σ 2 = (µ − L)(H − µ)/7.
Note that the PERT method only includes one critical path. In case of more than one critical
path, it will be appropriate to use the path with the highest variance. A weakness in the PERT
method is that the project duration will be underestimated in case of many paths with expected
duration in the same range as the critical path.

4.3 Successive schedule planning (SSP)

The idea behind the SSP-method is that we need to consider more than the critical path. To
motivate for the algorithm we are going to present, consider a project with two activities A and
B executed in parallel. In this situation it is reasonable to assume that the project duration will
be the maximum of the duration of the two activities, and we might apply Equations (3.62) and
(3.63). In this example A and B are both paths in the network, and the two paths have no
common activities. In a general situation, the various paths might have common activities which
complicates the calculation. Now, consider a situation with two activities B and C in parallel,
where the start-up of these two activities will be immediate after the finalisation of activity A. We
might now establish the two paths {A, B} and {A, C}, but we realise that the two paths have a
common activity, A. For each path we could find the expectation and variance similar to what
we did in the PERT method, but since the two paths share a common activity, they will not be
independent, and the result for the maximum of the two paths will not be correct. In order to
overcome this problem, we could in this situation first find the expectation and variance of activity
A, and next add the expectation and variance for the maximum of activity B and C.

Problem 4.1 Consider the situation above with activities B and C in parallel following activity
A. Further let the expectation and variance of the activity durations be given by: µA = 10, µB = 7,

43
µC = 8, σA2 = 22 , σ 2 = 32 and σ 2 = 22 . Find the expectation and duration of the project by first
B C
treating the two paths {A, B} and {A, C} as independent. Next, carry out an exact calculation
and compare the result with the first result. ♢
To structure the analysis we need some definitions. We define a meeting point where two or more
arrows join before or into an activity or the endpoint. For example in Figure 4.2 the activities D
and E join into a meeting point just before activity F . A branching point is a point where one
activity is followed by two or more activities in parallel, i.e. one branch splits into two or more
branches. For example in Figure 4.2 the activities B and C follow in parallel after activity A, and
the branching point is just right to activity A. We also need some numerical routines for solving
the integrals in Equations (3.62) and (3.63). Assume that we have access to the following routines
EMax=EMax(µ1 , σ12 , µ2 , σ22 ) and VarMax=VarMax(µ1 , σ12 , µ2 , σ22 ) for solving Equations (3.62) and (3.63)
respectively. Here µ1 , σ12 , µ2 and σ22 are expectations and variances for the two variables we are
taking the maximum of. We will only consider the situation where EMax and VarMax are implemented
under the assumption of independent and normally distributed variables. See pRisk.xlsm for such
an implementation.

Problem 4.2 Show that EMax(µ1 , σ12 , µ2 , σ22 ) = ∆µ+ EMax(µ1 −∆µ, σ12 , µ2 −∆µ, σ22 ) and VarMax(µ1 , σ12 , µ2 , σ22 ) =
VarMax(µ1 − ∆µ, σ12 , µ2 − ∆µ, σ22 ) ♢
The procedure for successive schedule planning with respect to describing the project duration is
now as follows1 :
1. For each activity i, establish the expectation, µi and variance σi2 for the duration of activity
i.

2. Identify all meeting points, i.e. where one or more branches join into one arrow.

3. Repeat and follow all activities from left to right in the flow network. This process is iterative
since each activity to the left of the current activity has to be processed before we can proceed.

4. For each activity i establish the expected start (EiS ), and expected finalisation (EiF ). The
expected start is equal to the expected finalisation of the preceding activity (or meeting point
in case of branches are joining just before activity i). The expected finalisation is given by
the expected start plus the expected duration of activity i, i.e. EiF = EiS + µi . Note that this
step cannot be executed if one or more of the activities to the left have not been processed.

5. For each activity i establish the accumulative variance, ViF . Here ViF is the accumulative
variance of the activity (or meeting point) preceding activity i plus σi2 , i.e. ViF = VkF + σi2
where k is the activity preceding activity i.

6. If there is a meeting point in the network just before the entry into an activity, we have to
process this meeting point. Note that this means that two or more branches join together
and the succeeding activity cannot start before all the branches, or paths up to this point,
have been finalised (completed). Technically, we now introduce a virtual node at the meeting
point, representing the finalisation of the two (or more) branches going into the meeting point.
The virtual nodes are enumerated V1 , V2 , . . .. If three or more branches join into one meeting
point, we first process two branches into one virtual node, then this virtual node represent
one branch which is then processed together with the third branch into another virtual node
etc.
1
The presentation is slightly different from the original presentation by Lichtenberg (1990).

44
 7. Let Vk be the virtual node we are processing, and assume that it is activities i and j that
are joining into Vk . If one of the activities (or virtual nodes) immediate to the left of Vk
has not been processed, we have to go to the left in the network until we meet processed
activities or nodes. The expectation and variance for the finalisation of activity i are now
given by EiF and ViF respectively. Similar we have EjF and VjF for termination of activity
j. If the two paths up to activity i and j were disjoint, we could easily find the expectation
and variance of the finalisation of the virtual node Vk by Equations (3.62) and (3.63), or
numerically by EMax and VarMax. Typically, the two branches that join together after activity
i and j did split up into two branches from one single branch prior to a branching point. Let
l be the activity at which the branches did split up before joining again at the virtual node
k. When finding the expectation and variance up to the virtual node k we then first find
the expectation and variance up to the branching point l, and then add the expectation and
variance of the maximum of the two branches from the branching point l to the virtual node
k. The accumulated variance along the path from branching point l to the end of activity i
is found by Vi∆F = ViF − VlF . We get similar results for the other branch, i.e. the one with
activity j preceding the virtual node k. The expectation and variance for the finalisation of
the virtual node k is now given by EiF = EMax( EiF , ViF − VlF , EjF , VjF − VlF ) and ViF = VlF +
VarMax( EiF , ViF − VlF , EjF , VjF − VlF ).

8. If there are more branches not processed into the meeting point, repeat until all branches are
processed by creating new virtual nodes.

9. When we reach the end node, we are done.

Example 4.1
We want to demonstrate the calculation process by the flow network shown in Figure 4.2. We
further assume that we have a spread sheet program available. The result of the calculations are
shown in Table 4.2. In addition to the activity row, and the three rows for the low, most likely and
high row, we add four rows for µi , σi2 , EiF and ViF respectively. For each row corresponding to
normal activities we calculate µi = (L+4m+H)/6 and σi2 = (µi −L)(H −µi )/7. Then we calculate
the expected finalisation of each activity, EiF as the expected finalisation of the previous activity (or
virtual node) plus the expected duration of activity i, µi . Similarly the variance of the finalisation
of activity i, ViF is the variance of the finilisation of the previous activity (or virtual node) plus the
variance of activity i, σi2 . For activity A the expected finalisation and variance of the finalisation is
equal to the expectation and variance of the duration of activity A since it is the first activity. We
note that after activity A we have a branching point that meets again after activities F and G. For
Activity B we see that EB F = E F +µ = 5.17+6.17 = 11.3, and V F = V F +σ 2 = 1.73+0.88 = 2.61.
A B B A B
We note that after activity A we have a branching point that meets again after activities F and
G.We proceed similarly with activities C, D and E. We now proceed to the virtual node V1 . It is
convenient to insert a new row in the spread sheet program just before activity F . In order to find
the expectation and variance for this node we take advantage of the functions EMax and VarMax.
The arguments to these functions are the expectation of the finalisation of each of the preceding
activities, and the accumulated variance through the branches from the branching point which
in this case is after activity B. EVF1 = EMax(18.5,3.49-2.61,18.5,4.35-2.61)=19.14. In order to
obtain the variance we use the VarMax function but we have to remember to add the accumulate
variance up to finalisation of activity B, VVF1 = VarMax(18.5,3.49-2.61,18.5,4.35-2.61)+2.61=3.5.
We complete the sheet for the remaining activities, including the virtual node V2 . The expectation
and variance of the duration of the entire project is now given by EH F and V F respectively.
H ♢

45
 D

S A B E V1 F V2 H F

C G
Figure 4.3: Example flow network with virutal nodes, adapted from [3]

Note that we in the SSP-method have used the PERT distribution as a basis. The method could
be used for any distribution for the activities, the essential point is to assess the expectation and
variance of each activity duration.

Problem 4.3 Consider Example 4.1 and carry out the calculations by your self in a spread sheet
program. ♢

Problem 4.4 Find the cumulative distribution function for the entire project duration (T ) based
on the calculation in Problem 4.3, and especially find Pr(T > 35). ♢

4.4 Monte Carlo simulation (MCS)

The analytical models for project duration evaluation is often not flexible enough to capture relevant
aspects of a project. The use of Monte Carlo simulation techniques is a supplement to analytical
methods when we the situation is to complex to be analysed by analytical models. The idea of
Monte Carlo simulation is that we establish a set of stochastic variables and events. Then we
establish deterministic relations between these variables and the events, e.g. the order of which
activities are executed, which activities that could be executed in parallel etc. It is important to
realise that the model that describes these relations is a deterministic model. Such a model could
be implemented in e.g. an MS Excel spreadsheet. The next idea in the Monte Carlo simulation
is to generate the stochastic variables and the events (indicator variables). Most computer codes
or program systems have a function that generate uniform distributed stochastic variables on the
interval from 0 to 1. Given a such function it is also in principle straight forward to generate the

Act. L M H µi σi2 EiF ViF Comment

A 2 5 9 5.17 1.73 5.17 1.73 Branching point for V2
B 4 6 9 6.17 0.88 11.33 2.61 Branching point for V1
C 7 12 21 12.67 7.75 17.83 8.48
D 5 7 10 7.17 0.88 18.50 3.49
E 4 7 11 7.17 1.73 18.50 4.35
V1 19.14 3.50 E & E joins, branches after B
F 2 3 6 3.33 0.51 22.48 4.01
G 3 5 9 5.33 1.22 23.17 9.70
V2 24.13 5.75 G & F joins, branches after A
H 5 7 10 7.17 0.88 31.30 6.63

Table 4.2: Data for the successive schedule planning demonstration in Example 4.1

46
stochastic variables we need. By inserting these stochastic variables into the deterministic model
(e.g. an MS Excel model) we now get one realisation of the system, or more specific the project
duration. Let t1 be the numeric value when this process is done the first time. Now, we might repeat
the process by generating another set of random quantities and insert these into the deterministic
model to yield another value, say t2 . By repeating this process we could think of the generated
values t1 , t2 . . . as realisations of the project, and use the values to obtain statistical properties such
as the mean, the standard deviation, the cumulative distribution function etc.
We will now illustrate how this process could be carried out with the pRisk.xlsm program.

Example 4.2
It will be convenient to establish one row in MS Excel for each activity. The first column (A)
could contain the activity number, the second, third and forth (B, C and D) could then contain the
parameters in the PERT distribution, similar to Table 4.2. Now we introduce three new columns
(E, F and G) to contain the duration, start and finalisation of each activity respectively. We start
to enter the duration of each activity. Assume that activity A is described in row 2 in the Excel
sheet. In cell E2 we now enter the following expression for the duration:
=RndPert(Rand(),B2,C2,D2)
Here the RndPert() function is a pRisk.xlsm specific function, whereas the Rand() function is a
standard Excel function. The procedure is repeated for all activities, and we simply copy the
formula in cell E2 into the cells E3, E4 etc. We will now proceed to the start and finalisation of
each activity. It will be convenient to give the cells containing the start and finalisation names in
Excel. For activity A we give the following names D_A, S_A and F_A for the duration, start and
finalisation respectively. Similarly we give the names S_B, D_B and F_B for the start, duration and
finalisation of activity B respectively, and so on for the remaining activities. By giving name to
the activities, it is easy to access them in formulas in other cells. We now use the convention cell
name = expression where the cell name is the name of the cell we want to assign an expression. By
inspecting the network in Figure 4.2 we easily verify the following statements for the start of the
various activities:
S_A = 0
S_B = F_A
S_C = F_A
S_D = F_B
S_E = F_B
S_F = Max(F_D, F_E)
S_G = F_C
S_H = Max(F_F, F_G)
The finalisation of the activities is given as the start point plus th duration, e.g., for activity A we
enter:
F_A = S_A + D_A
and similar for the other activities.
We have now specified the model and are prepared to simulate several runs. First we note that
each time we press the F9 key Excel updates the model by generating new random numbers since
we used the RAND() function in the cells containing the duration of each activity. Next we switch
to the RunSimul sheet and press the Run button. ♢

Problem 4.5 Consider the example in Figure 4.2. We will now consider an alternative execution

47
method for the last part of the project. Rather than executing activity H as one activity, it is
possible to split this activity into two parallel activities H and I. Each of these activities could be
described by the PERT distribution with parameters L = 3, M = 5 and H = 8. Set up the flow
network for this situation, and use the pRisk.xlsm program to find the expectation and standard
deviation of the project duration by Monte Carlo Simulation. ♢

4.5 Penalty for default

The contracting party might issue penalties for default to ensure that the contractor put necessary
resources and effort into project execution. Penalties for default could be linked to milestones and
finalisation of the entire project. In the following discussion we only consider the situation when
there is defined penalty for default if the project as such is delayed. Let T be the duration of the
project measured from a defined start-up date. Let D be the number of days (from the startup
date) before penalty for default is initiated. Finally, let PD be the size of the penalty per day. The
total penalty for default is then max(0, (T − D)PD). The expected total penalty for default in a
project is thus:
∫ ∞
PDTot = (t − D)PDfT (t) dt (4.1)
D

where fT (t) is the probability density function for the project duration.
In principal we have to perform the integration in Equation 4.1 to find the expected total penalty
for default in a project. In most cases we also need to carry out numerical integration. However, if
we have a Monte Carlo simulation model for the project, we might utilise that for a given project
duration T the total penalty for default is max(0, (T − D)PD), and in e.g. pRisk.xlsm we could
specify in the “Cell to analyse”:
=max(0,PD*( T_End- D_Start )

where D_Start is the name of the cell where we have specified when penalty for default is ini-
tiated, and T_End is the name of the cell where the total project duration could be found.

Problem 4.6 Consider the example in Figure 4.2. Assume that D = 34, and PD = 1, 000 Euro.
Find the total expected penalty for default in this project. ♢
∫a
Problem 4.7 For the normal distribution we have from Equation (3.25) that −∞ xf (x)dx =
( ) ( )
µΦ a−µ
σ − σϕ a−µ
σ . Use this to verify that if the duration, T of a project is normally distributed
with mean µ and standard deviation σ then we have:
∫ ∞
PDTot = (t − D)PDfT (t) dt (4.2)
( [ ])
D
( )
D−µ D−µ
= PD(µ − D) 1 − Φ + PDσϕ (4.3)
σ σ

where Φ() and ϕ() are the cumulative distribution function and probability density function for the
standard normal distribution respectively.

48
4.6 Event uncertainty in the schedule model
When we describe the uncertainty about the duration of an activity by parameters such as L, M
and H these parameters account for all factors and conditions that influence the duration. In some
situations we might want to describe and model some important factors explicit. For example if
the event W denote extremely bad weather conditions, we might describe two set of parameters
{L, M, H}; one if W occurs, and one if W does not occur. That is, the duration of the activity is
described by the parameters {LW , MW , HW } if W occurs, and {LW C , MW C , HW C } if W does not
occur. Next we describe the probability that W occurs by a probability statement, pW = Pr(W ).
We might now include the uncertainty about the weather (event uncertainty) explicit into the
schedule model. The easiest way to include such event uncertainty is to use the Monte Carlo
simulation approach. We have to do the following:
• Describe the event we want to condition on, e.g. W

• Describe the probability of the occurrence of this event, e.g. pW = Pr(W )

• Define conditional parameter statements, e.g.{LW , MW , HW } if W occurs, and {LW C , MW C , HW C }

if W does not occur for each activity duration influenced by W

• Define the event W in the pRisk.xlsm model

• The duration for each activity influenced by W is now entered as one expression if W occurs,
and another if W does not occur.
Now, assume that it is activity A which is influenced by the event W . The following information
could then be specified to pRisk.xlsm:
p_W = 0.1
Event_W = IF(Rand() < p_W,1,0)
D_A = IF(Event_W,RndPert(Rand(),5,8,12) ,RndPert(Rand(),2,5,8))
where the probability of bad weather conditions were set to 0.1, and we defined a cell with cell name
Event_W. Finally we have used the following set of parameter values: {LW , MW , HW } = {5, 8, 12}
and {LW C , MW C , HW C } = {2, 5, 8}.

Problem 4.8 Consider the example in Figure 4.2. Let W be the event {Bad weather}. Assume
that this event influences primarily the activities B and C. Let Pr(W ) = 0.1. The parameters
describing the duration of activity B and C is now similar to the situation in Problem 3.9. The
transformation factors now read fW , fW C and fV . Let fW = 1.5 and find fW C and fV similar to
the procedure in Problem 3.9. Follow this procedure both for activity B and C. Now update the
Monte Carlo simulation model for the example in Figure 4.2 when the event W is introduced in
the model, and compare the simulation results with the original results. ♢

4.7 Updating the model as we get more information

As we proceed with the project execution new information might be available, and we will know
the status of activities that have been completed. At regular intervals we should therefore update
the schedule model in order to optimise the effort we spend on the different activities. To update
the schedule model we take the following into account
• Activities that are completed are replaced with deterministic quantities in the schedule model.

49
 • If new activities were necessary to add to the project, these are added to the schedule model.

• If some activities were cancelled these are removed from the schedule model.

• If the status of events and other risk factors that were included in the schedule model is
known, we replace the probabilistic statements about these with deterministic statements.

• Other parameters (typically L, M and H) are revised in light of the knowledge available at
this moment, e.g. related to resources available.

Problem 4.9 Consider the example in Figure 4.2 and the penalty for default structure as defined in
Problem 4.6. assume that we now are in the project phase, and activity A has just been completed.
Due to special circumstances the duration of activity A was tA = 10 which is even higher than the
most pessimistic assessment. First calculate the expected project duration, and total penalty for
default. Next we will consider alternative production methods to increase the speed in the project.
A major sub activity in activity H is to produce an element on the construction site. It is, however,
possible to have this element prefabricated in advanced. The extra cost of such a prefabrication
is 1,500 Euro. The gain of such a prefabrication is seen in the new distribution of the duration
of activity H, i.e. we now judge TH ∼ PERT(3, 5, 7). Update the pRisk.xlsm model and find the
expected total cost with and without prefabrication. ♢

4.8 Examples of advanced schedule modelling

We will in the following present some situations where we need some more advanced modelling to
capture the situation. This will typically require to use the Monte Carlo simulation approach, and
we will use pRisk.xlsm to model the situation.

Example 4.3 Moving resources between activities

We consider again the example in Figure 4.2 and will investigate the activities F and G. In
expectation activity G will start slightly before activity F , whereas activity G has longer duration
than H in expectation. If the situation in the project is such that activity G is not ready for
start-up when activity F starts, it seems reasonable to move resources from activity F to activity
G. If this is the situation, we will assume that the distributions for F and G are PERT(3,4,8) and
PERT(3,4,8) respectively. To model such an operative measure in pRisk.xlsm we use the following
statements:
D_F = IF(S_F>S_G,RndPert(Rand(),2,3,6),RndPert(Rand(),3,4,8))
D_G = IF(S_F>S_G,RndPert(Rand(),3,5,9),RndPert(Rand(),3,4,8))
When we run this model we get a slightly lower expected value for the project duration. ♢

Problem 4.10 Use pRisk.xlsm to perform the calculations in Example 4.3. ♢

Example 4.4 Time window

We consider again the example in Figure 4.2 and event B which represents an activity that could
only be executed within certain time windows. We will assume that if activity A is not completed
before time t = 8, we could not start up activity B before time t = 20. To model the effect of such
a time window in pRisk.xlsm we use the following statement:
S_B = IF(F_A>8,20,F_A)
♢

50
Problem 4.11 Use pRisk.xlsm to perform the calculations in Example 4.4. ♢

51
Chapter 5

Decision under uncertainties

5.1 Introduction
In this section we will give a general introduction to the field of decision theory where uncertainties
are involved. Examples are related to project risk management. As a motivation consider the
following situation where we have to choose between two or more alternatives:

• Choice of sub contractor.

• Choice of concept for an oil production platform.

• Choice between double track and single track for a new railway line.

• Choice of tunnel trace now, or perform more investigation into the ground.

We could also have decisions related to continuous variables:

• When to make an agreement with one of the sub contractors.

• When to start preparing for a major shut-down.

• Choice of diameter for a gas pipeline towards Skogn.

• Dimension of a critical part in a new construction.

5.1.1 Overview of the method

We will first consider situations where one and only one decision is to be made. We will denote the
decision with the letter d. The decision alternatives will be denoted a1 , a2 , …, am . The decision,
d, could then be which sub contractor to choose, and the alternative ai is the decision that we
choose sub contractor i. The result of our choice will be a set of end consequences Y. The end
consequences could occur at different times in the future, but we will simplify and assume that the
effect will be immediate after the decision is made. In more complex situations we have to consider
that the effect will come some time in the future, and discounting is an issue. Y = [Y1 , Y2 , . . . , Yr ] is
an attribute vector and comprises many dimensions. For example Y1 could be the project duration,
Y2 could be the project costs etc. Further we note that the Yi ’s are stochastic variables and the
values will depend on our decision d.We will seek the decision that gives the “best” value of the
attribute vector Y. It is common to differentiate between the following four situations:

52
 1. Decision under certainty. In this situation all the outcomes are known, and we will know for
sure what the outcome will be for the different decision alternatives.

2. Decision under risk. In this situation all the possible outcomes are known, but we do not
know which outcome will be the result of our decision. We are able to state probabilities for
the various outcomes.

3. Decision under uncertainty. In this situation all the possible outcomes are known, and we
are unsure about the probabilities for the various outcomes.

4. Decision under ignorance. In this situation we do not know all the possible outcomes, and
we are also unsure about the probabilities of those outcomes we know about.

In this presentation we will only consider decisions under risk and uncertainties. We also note that
many authors claim that it is not meaningful to state uncertainty about the probabilities, it is the
outcome which is uncertain, not the probabilities. When it comes to the final outcome, i.e., related
to the attribute vector Y we agree that there is no uncertainty in the probability distribution of
Y, i.e., the probability distribution contains all the uncertainty about Y. We will therefore not
differentiate between the situation of decisions under risk, and decisions under uncertainty. Most
frequently we will use the term ‘decision under uncertainty’.
As indicated above we will seek the decision d that gives the best “value” of the attribute vector
Y, for example the lowest cost and the shortest execution time of a project. There are, however,
some difficulties in this approach:

• We will not be neutral to the risk. Very often we are willing to make a decision that do
not give the maximum expected revenue, but rather choose an option with a lower expected
revenue but with a lower probability of big losses. We are what is called risk averse.

• The attribute vector Y comprises several dimensions, and it is not straight forward how to
weight these dimensions. For example how should we treat a project with low cost, but a
higher risk of accidents during project execution?

In order to treat such decision situations we introduced the concept of utility theory, and utility
functions. We will only briefly mention the major aspects, and refer to [9] for further discussions.
First we will treat situations where we make only one decision, and the end consequences are
assumed to take effect immediately after our decision. In more complex situations the effects will
come on a later stage, and we could make several decision in a sequence with time delays between
each decision. In such decision problems we often use decision trees to help assisting the decision
process.

5.2 Basic concepts

We use the notation d about a decision where it is only one decision to be made, whereas we use the
notation d1 , d2 , . . . in the situation where several decisions have to be made. In this situation we also
need an extra index for the decision alternatives. In the situation where the decision is to choose
a value of a numeric quantity, either discrete or continuous we will notationally not differentiate
between the decision, and the decision variable (d).

53
5.2.1 Discrete end consequences vs attribute vector
Generally we use Y to describe the values of the end consequences resulting from a decision. In
some situations we want to simplify the representation by a set of few end consequences, EC. We
will let the end consequences, ECj , be disjoint. Further we will let pj = Pr(ECj occurs) be the
probability that we get end consequence ECj . It is not always straight forward to determine what is
the most convenient, either to work with the full attribute vector Y, or the set of end consequences
EC1 , EC2 , . . .. This will depend on the level of precision in the risk analysis, the skill of the risk
analyst, or the decision maker etc.
In order to see the difference, consider the occupational safety dimension during project ex-
ecution. If we work with end consequences it could be natural to introduce the following end
consequences:

1. EC0 = No injury

2. EC1 = Minor injury

3. EC2 = Medical treatment

4. EC3 = Serious injury

5. EC4 = 1 fatality

6. EC5 = 2-10 fatalities

7. EC6 = > 10 fatalities

In order to describe the expected result we also specify the corresponding probabilities, p0 , p1 , . . . p6 .
These probabilities will be dependent on the decisions we make. If we want to use a full attribute
vector we could use Y = [Y1 , Y2 , Y3 , Y4 ], where Y1 = number of minor injuries, Y2 = number of
major injuries, Y3 = number of fatalities, and Y4 = is the number of gross accidents, i.e., accidents
with five or more fatalities. In this latter situation we specify the expected outcome in terms of
the joint probability distribution function of Y. We then often introduce parameters that depend
on the decision d we make.

Utility function
The utility function expresses the preferences of the decision maker regarding various attribute
vectors or end consequences. A prerequisite for establishing a utility function is that the decision
maker is able to express preferences between different values of the attribute vector. For example in
a one dimensional situation where we set Y = NPV (net present value) this will be rather obvious
in the first place, it is reasonable that all decision makers will prefer a higher value to a lower value.
Now let y1 and y2 denote two arbitrary values Y may take. The following relations are of interest
between y1 and y2 :
The utility function is now a function that assigns a one-dimensional utility value to each value
of the attribute vector or quantity, u = u(y). For the utility function we require:
y1 ∼ y2 ⇔ u(y1 ) = u(y2 )
y1 ≻ y2 ⇔ u(y1 ) > u(y2 )
y1 ≽ y2 ⇔ u(y1 ) ≥ u(y2 )
y1 ≺ y2 ⇔ u(y1 ) < u(y2 )
y1 ≼ y2 ⇔ u(y1 ) ≤ u(y2 )

54
 Relation Explanation
y1 ∼ y2 y1 and y2 is considered equal
y1 ≻ y2 y1 is preferred over y2
y1 ≽ y2 y1 is as least as preferable as y2
y1 ≺ y2 y2 is preferred over y1
y1 ≼ y2 y2 is as least as preferable as y1

There exists, however, an infinite number of utility functions that satisfy the above criteria and we
therefore want to fix the utility function for some values. In order to be useful, the utility function
should also express how much we prefer for example, y1 over y2 . Further we also want the utility
function to reflect the fact that there will be uncertainty regarding the future value of the attribute
Y . We still consider the one dimensional situation where Y = NPV (net present value). Y will be
a stochastic variable in the decision point. Now, assume that we could choose between a decision
A that for sure gives the net present value Y = y0 and the decision B that gives the net present
value Y = y1 with probability α and the net present value Y = y2 with probability 1 − α. Further
assume that y1 ≺ y0 ≺ y2 . For a given set of values of y0 , y1 and y2 there will exist a value of
α which makes the decision maker indifferent between the two decisions A and B. This will be
reflected in the utility function which must satisfy:

u(y0 ) = αu(y1 ) + (1 − α)u(y2 ) (5.1)

Equation (5.1) could now in principle be used to establish the utility function. In this process we
might restrict our selves to let the utility function take values between 0 and 1, or 0 and 100.

Example 5.1 Private economy

We are asked to do a job in the firm SmartConsult. It will be 100 hours of work, and we are offered
two options for payment:
1. A fixed hour rate of 20 Euro per hour.

2. A baseline hour rate of 10 Euro, and an additional value of 50 Euro which will be paid if the
project reaches the targets that have been set up.
We have been studying the progression in the project so far, and assess the probability that the
extra 50 Euro to be paid is 40%. Simple calculations shows that the second alternative gives the
highest expected hour rate (30 Euro vs 20 Euro). However, we are in a cash position which makes
it very difficult for us if we only receive 10 Euro per hour. After some considerations we have found
out that the two alternatives are equal, i.e., neither of them are prefered over the other.
We will now utilise Equation (5.1) to set up our utility function. Three points could be assessed,
u(1 000), u(2 000), and u(6 000). Since we arbitrary may choose the end points, we let u(1 000) = 0
and u(6 000) = 100. We now have (α = 0.6):

u(2 000) = 0.6u(1 000) + 0.4u(6 000) = 40

The utility function is shown in Figure 5.1 where we have fitted the function u(y) = 55.702 ln(y) −
384.25 and the diamonds represent the assessed values. ♢

Problem 5.1 Consider Example 5.1. What probability would you required for being paid the
additional 50 Euro per hour in order to treat the two alternatives as equal. ♢

55
 u(y)
100

y
1000 2000 3000 4000 5000 6000
Figure 5.1: Utility function for Example 5.1

Problem 5.2 Consider Example 5.1 again, but assume that you were going to work 500 hours.
Make a sketch of your utility function in this situation. ♢
For private economies we are usually risk averse. Risk aversion means a concave utility function as
shown in Figure 5.1. Also smaller enterprises will often be risk averse reflecting that rather than
optimising expected revenue, decisions are taken to minimise the probability of big losses which
could lead to bankruptcy. Larger enterprises will often have an almost linear utility function (in
monetary values) because their economical strength is good, and there is no real possibility for
bankruptcy.

Utility function for quantities other than monetary units

In Example 5.1 we have seen how the utility function could be established for monetary units.
We will now investigate how we could include the safety dimension into the utility function. Two
important questions will be raised:

• What is the benefit, or utility of saving one (statistical) life vs saving 10 statistical lives?

• What is the benefit, or utility of saving one (statistical) life vs the possibility to earn an extra
million Euro?

In the first situation we deal with the question to rank the consequences within the same main
dimension (safety), whereas we in the second situation need to compare benefits or disadvantages
across dimensions. The discussion below will be very short, and we refer to e.g., [10] for further
discussion on this topic.
The first issue we will discuss is the concept ‘value of prevented fatality’ (VPF). The idea behind
this concept is that in any industrial activity, transportation services etc. there will always be a
risk of accidents, and hence possibilities of severe injuries and fatalities. As a decision maker we
have to face this fact. However, we will make effort to reduce this risk, and we are willing to
spend money to achieve such a reduction. The VPF value states then how much we are willing to
spend in order to prevent one statistical fatality. We use the term ‘statistical’ fatality to emphasise
that this willingness to pay is not related to specific persons, but arbitrary persons where it is not
meaningful or possible to identify single persons. In some presentation also the term ‘value of life’
(VOL) has been used. We feel that this term is not appropriate because the term indicates that
the life it self has a value which could be measured in monetary units. This is not our perspective.
The value of life it self could not be measured. What we could assess figures to, is what we are
willing to pay in order to reduce risk, or the probability of fatalities. Hence, the term VPF make
more sense in our understanding.

56
 If we accept that the term VPF make sense, then the next question will be how to assess the value
of VPF. Different approaches exist. One approach is to look into economical considerations from
the society point of view. For example we could calculate the reduction in GNP (Gross National
Product) caused by a fatality. Such calculations have been carried out, and in e.g., Norway this
indicate a value of 3 million Euro for VPF. Another approach has been to ask single persons about
their willingness to pay for risk reduction (see “The change in risk of death” [16]). For example
for buyers of cars, we could ask what they are willing to pay for a given safety system or measure,
for example an improved airbag system. Let assume that the amount one is willing to pay is ∆W ,
and that the assessed risk reduction during the service life of the car is ∆P . It would then be
natural to set VPF = ∆W/∆P . In Norway no such systematic surveys have been conducted, but
more arbitrary surveys at NTNU among ordinary students and continuation students a value of 2.5
million Euro has also been found for VPF. We will emphasise some challenges of such a willingness
to pay approach:
• Different persons have different preferences. For example young people tend to be less willing
to pay for risk reduction compared to older persons with family obligations.
• Individuals are not consistent in their preference statements.
• In real surveys to establish ∆W and ∆P we face the problem that other dimensions than
being killed are involved, e.g., the risk of minor and major injuries. Further one does not only
consider the life of one self, but also the life of family members etc when making decisions
about safety.
• It is not obvious that “what I am willing to pay” is what I want the society to pay for risk
reduction in general, or what I expect my employer to pay for my risk reduction.
It is a tendency to set a lower value for VPF when it comes to the area of public responsibility
compared to industrial activity. For example in the petroleum industry we see VPF values in
the order 10 to 15 million Euro. It is also a tendency to set a higher VPF for multiple fatality
accidents compared to single fatality accidents. This could be interpreted as an aversion against
gross accidents. This aversion should not be confused with risk aversions which would be an aversion
against a high number of fatalities in general, and not the number of fatalities in single accidents.
An another perspective in this discussion is how we should treat injuries in such a framework. One
common approach here is to introduce the concept of ‘equivalent fatality’. For example we could
be willing to pay five times more to prevent a fatality than a severe injury, which corresponds to
an equivalent fatality of 0.2.
In a utility function approach we could now in case of a VPF value 2.5 million Euro let the
utility of one fatality be equivalent to -2.5. If we now extend the situation to include multiple
fatality accidents, and minor and major injuries, we could set up a more general utility function:

u(y1 , y2 , y3 , y4 ) = −0.03y1 − 0.5y2 − 2.5y3 − 7y4 (5.2)

where y1 is the number of minor injuries, y2 is the number of major injuries, y3 is the number
of fatalities in accidents with less than five fatalities, and y4 is the number of fatalities in gross
accidents (five or more fatalities in one accident). It is important to emphasise that the utility
function offered in Equation (5.2) is a function that could be used as a start in a discussion about
value trade-offs and preferences, and should not be considered as the “correct utility function”. Also
note that Equation (5.2) includes an aversion against gross accidents, but there is no risk aversion
in terms of a concave utility function in the attributes. If we also want to include attribute y7 as
the profit in a project measured in million Euro we could extend the utility function:

57
 u(y1 , y2 , y3 , y4 ) = −0.03y1 − 0.5y2 − 2.5y3 − 7y4 + y7 − ae−by7 (5.3)

where a and b are constant. Reasonable values of these constants are a = 0.08 and b = 0.7.
The utility function in Equation (5.3) is an additive utility function. Very often we use additive
utility functions for simplicity. However, arguments could indicate that a situation with one extra
fatality is “worse” if there is a situation with a gross accident than without such a gross accident.
Such discussions will not be pursued any further, and we refer to [9].

5.2.2 Maximising expected utility

In the previous sections we have seen principles for establishing a utility function. The utility
function expresses our preferences and value trade-offs. The utility function is independent of the
given decision situation we are facing and could be viewed as a general function we could use in
many decision situations. We also observe that the utility function is a function of the attributes.
In a given situation these attributes, Y1 , Y2 , …, are stochastic variables which also means that
the utility function will be stochastic, and the idea is to choose the decision that maximises the
expected utility.

Result 5.1 ∫∞
The optimal decision d is the decision that maximises expected utility, E(u(Y)) = −∞ u(y)fY (y)dy
♢
The basic steps in obtaining the optimal decision is then:
1. Establish an explicit expression for the utility function, u = u(y1 , y2 , . . .) which corresponds
to the preferences and value trade-offs of the decision maker.

2. Establish the probability distribution function for the attribute vector Y = [Y1 , Y2 , . . .] for
each decision alternative, or for each value of a decision variable (d).

3. Calculate the expected utility to each decision alternative by integrating the utility function
over the probability distribution of the attribute vector.

4. Find the decision alternative that gives the maximum expected utility.

Problem 5.3 In this problem you shall first make an attempt to construct the utility function
u(y) for a given decision maker. In the problem there is only one dimension, and the attribute
y is measured in thousand Euro by the procedures we have established in the previous sections.
Assume that u(−100) = 0, and u(400) = 1.
a) Why do we have the freedom to assess two points on the utility function, and why is is suitable
to use these two values.
Now, assume that the decision maker makes the following considerations regarding the outcome of
a project:
• An uncertain project which gives -100 with probability 0.50 and +400 with probability 0.50
is considered as equal attractive as receiving the fixed amount +150.

• An uncertain project which gives -100 with probability 0.50 and +150 with probability 0.50
is considered as equal attractive as receiving the fixed amount +100.

58
 • An uncertain project which gives +150 with probability 0.50 and +400 with probability 0.50
is considered as equal attractive as receiving the fixed amount +225.

b) Draw the points on the utility function which you could calculate based on the above infor-
mation, and make a sketch of the utility function in the interval -100 to +400.

c) What does the graph say about the decision makers attitudes to risk?

d) Use the graph to choose the optimum project among the following projects:

A) A project returning -100 with probability 0.2, +150 with probability 0.2 and +350 with
probability 0.6.
B) A project returning 0 with probability 0.4 and +400 with probability 0.6.

e) Which of these two projects would the decision maker choose if he adopts the principle of
maximum expectation. ♢

Problem 5.4 In a tunnel project one could choose between bursting or drilling. Bursting is con-
sidered to be the cheapest alternative, but the risk of personal injuries or fatalities is considered
higher. Assume the utility function given in Equation 5.2 on page 57. Let fi = E(Yi ), i = 1, . . . , 4
be the expected number of minor injuries, serious injures etc. and assume the following numbers:

• Bursting [f1 ,f2 ,…,f4 ] = [10, 1, 0.03,0.008]

• Drilling [f1 ,f2 ,…,f4 ] = [7, 0.2, 0.01,0.001]

How much cheaper need bursting be compared to drilling if these two methods should be equally
valued with respect to utility? ♢

Problem 5.5 A company has established the following utility function: u(y) = y − ae−by , where
a = 1/5 and b = 1/2, to be applied for prioritization of projects. y is given in million NOKs and
represents the profit in the projects.

a. Make a sketch of the utility function, and discuss the risk attitude of the company.

b. The company considers to invest in a project which (i) either gives a profit of 10 (million
NOKs), or (ii) a loss of 10 (million NOKs). What probability of success is required to invest
in the project?

c. The company is going to choose between two projects, A and B. By analysis the following
has been derived: YA ∼ N (6, 42 ) and YB ∼ N (6.1, 52 ), where Y is the profit, and ∼ N (µ, σ 2 )
indicates normally distributed quantities. Calculate the expected utility for each project to
decide which project is best. Discuss the result. Hint: Show that if Y is normally distributed
with expected value µ and standard deviation σ, then the expected utility is found by use of
moment generating function to be: E(u(Y )) = µ − ae−bµ+1/2b σ provided the utility function
2 2

is given by u(y) = u(y) = y − ae−by .

59
5.2.3 Examples with one decision node
In this section we will investigate examples where only one decision is going to be made.

Example 5.2 Maximising expected utility - private economy

In Example 5.1 we established the utility function in a situation with private economy. The function
could be written as: u(y) = 55.702 ln(y) − 384.25. Now, assume that we are offered a job with two
different forms of payment. For both alternatives the possible amounts are:
• Y = YL = 2,000
• Y = YM =5,000
• Y = YH =8,000
However, there is a difference in the probabilities for the two alternative forms of payments. These
are shown under the column p for alternative a1 and a2 in Table 5.1 respectively. Expected utility

Table 5.1: Expected utility and expected values

Alternative a1 Alternative a2
↓ p U V P U V
Amount
2,000 0.1 3.9 200 0.3 11.7 600
5,000 0.8 72.1 4,000 0.4 36.1 2,000
8,000 0.1 11.6 800 0.3 34.9 2,400
Sum→ 1.0 87.7 5,000 1.0 82.7 5,000

is found by equation (5.4).

∑ ∑
E(u(Y )) = u(y) Pr(Y = y) = (55.702 ln(y) − 384.25) Pr(Y = y)
y∈{YL ,YM ,YH } y∈{YL ,YM ,YH }

where each term in the sum is calculated in the column for U in Table 5.1. In the V column we
have similarly calculated the expected monetary value. In the last row the sum is shown, and we
observe that the expected utility for a1 and a2 is 87.7 and 82.7 respectively, and hence alternative
a1 has the largest expected utility. When expectations are considered, the two alternatives are
equivalent.

Problem 5.6 Consider Example 5.2, but now assume that the probability distribution for the
payments are ∼ PERT(4000,5000,6000) and ∼ PERT(2000,6000,8000) respectively. Hint: you
might use the program pRisk.xlsm. ♢

Example 5.3 Tender offer - linear utility function

We are going to prepare a tender for a large road project. We have made a cost estimate, and
found the total cost of executing the project, PC, to be PERT distributed, PC ∼ PERT(L, M, H)
= PERT(10, 30, 80) where all costs are given in million Euro. We also have some knowledge about
our competitors. We have judged the lowest tender from the other contractors, LP, to be PERT
distributed, e.g., LP ∼ PERT(30, 35, 50). The challenge is to set our tender price (TP) such that
it is lower than the lowest of the (serious) competitors , but also not too low as we then will loose
money. The following notation is introduced:

60
 p(TP)
1
.8
.6
.4
.2
TP
30 35 40 45 50
Figure 5.2: Probability p of getting the project as a function of tender price, TP

TP = Our tender price, TP, i.e., the decision variable.

LP = Lowest tender prince among our competitors. We assume a lump sum contract.

PC = Project cost, i.e., a stochastic variable.

Y1 = 1 if we get the contract, 0 otherwise, i.e., Y1 = ILP>TP .

p = The probability that we get the contract, p = p(TP) = E(Y1 ) = Pr(LP > TP) .

Y2 = TP = Our tender price. TP is also an “attribute” because this is our income.

Y3 = PC = project cost.

u = utility function, u(y1 , y2 , y3 ) = y1 · (y2 − y3 ).

Expected utility is given by

E(u(Y1 , Y2 , Y3 |TP)) = E(Y1 ) · E(Y2 − Y3 ) = p(T P ) · (TP − E(PC))

In order to find p = (TP) we could utilise the pRisk.xlsm program, and the function CDFPert. The
syntax to enter in an EXCEL cell where we store the result is:
= 1 - CDFPert(TP,30,35,50)
where TP is a cell reference or a numeric value for the tender price. Figure 5.2 shows the probability
of getting the contract as a function of the tender price. Given that we get the contract, the expected
utility equals TP-E(PC). The expectation in the PERT distribution is given as (L + 4M + H)/6,
i.e., (10 + 4 · 30 + 80)/6 = 35 million Euro, and the expected utility equals:

E(u(Y1 , Y2 , Y3 |TP)) = p(TP) · (TP − 35)

By using the result for p = p(TP) from Figure 5.2 we could easily find the expected utility as shown
in Figure 5.3. The optimum tender price is found to be 39 million Euro. ♢

Example 5.4 Tender offer - concave utility function

We will investigate the situation in Example 5.3 but we will now assume that the decision maker
is risk averse. A possible utility function is:

u(y) = y − a · e−by

61
 E(u)
1
0
-1
-2
-3
-4
-5 TP
30 35 40 45 50
Figure 5.3: Expected utility as a function of the tender price, TP

E(u)
1
0
-1
-2
TP
35 3637 38 39 40 41
Figure 5.4: Expected utility as a function of the tender price, TP

where a and b er parameters. We set a = 0.2, and b = 0.2, and we let money be measured in million
Euro. With these parameters the utility of 10 million Euro is then 9.97 whereas the utility of a loss
of 20 million Euro is - 30.9 and the utility of -30 is -110. In this example we will also add one if
we get the contract. This extra utility unit could represent the value of competence improvement.
The utility function is then:

u(y1 , y2 , y3 ) = y1 · (y2 − y3 ) − 0.2e−0.2·y1 ·(y2 −y3 ) + y1

It will not be easy to maximise expected utility as a function of the tender price, TP. We therefore
use pRisk.xlsm to carry out a Monte Carlo simulation, and the result is shown in Figure 5.4. We
see that the optimum value is slightly increased from 39 million to almost 40 million. The reason
for this is the concave utility function where we want to reduce the probability of the large losses.
♢

Problem 5.7 Make a sketch of the utility function in Example 5.4. Discuss the utility when y
approaches minus infinity and plus infinity. ♢

Problem 5.8 Use pRisk.xlsm to perform the calculations for the example in Example 5.4. Discuss
the influence of the parameter b and discuss the effect of letting b approach zero. ♢

Extra effort in order to reduce penalties for default

We are a part of the project management for a large road development project and realise that it
will be difficult to reach the completion date agreed upon. The following quantities describe the
situation:

CD = 30 = Agreed completion date, in days from now

Y1 =T = Completion date, in days from now. Y1 ∼ PERT(L, M, H).

62
 PD = 10,000 Euro = Penalty for default, i.e the amount to pay each day the project is delayed
(Y1 >CD).

BO = 5,000 Euro = Bonus, i.e the amount BO is paid extra for each day the project is completed
before CD.

Y2 = EE = Extra effort we invest in order to speed up the project.

L0 = 25 = Lowest value of Y1 , if nothing extra is done.

M0 = 35 = Most likely value ofY1 , if nothing extra is done.

H0 = 60 = Highest value of Y1 ,if nothing extra is done.

LY 2 = L0 (0.5 + 0.5e(−Y2 /50000) ) = Lowest value of Y1 , with extra effort Y2 .

MY 2 = M0 (0.5 + 0.5e(−Y2 /50000) ) = Most likely value of Y1 , with extra effort Y2 .

HY 2 = H0 (0.5 + 0.5e(−Y2 /50000) ) = Highest value of Y1 , with extra effort Y2 .

We will use a linear utility function, and we let the utility of one Euro be equal to one. The
following utility function the applies:

u(y1 , y2 ) = PD · (CD − y1 ) · ICD<y1 + BO · (y1 − CD) · ICD>y1 − y2

Also here we utilise the pRisk.xlsm program to calculate expected utility. The following statements
are entered into the Excel sheet:
CD=30
L_0=25
M_0=30
H_0=60
PD=10000
BO=5000
Y_2= 10000
LY_2=L_0*(0.5+0.5*exp(-Y_2/50000))
MY_2=M_0*(0.5+0.5*exp(-Y_2/50000))
HY_2=H_0*(0.5+0.5*exp(-Y_2/50000))
Y_1=RndPERT(RAND(),LY_2, MY_2, HY_2)
We obtain the following values for expected utility:

EE=Y2 0 10 000 20 000 30 000 40 000 50 000

E(u(Y1 ,Y2 )) -44 000 -28 000 -19 000 - 15 000 -15 000 -18 000

The maximum expected utility is obtained by spending between 30 and 40 thousand Euros in order
to reduce the risk of delays in the project.

Problem 5.9 Use the pRisk.xlsm program to perform the simulation as indicated above. Check
the sensitivity in the results as a function of the number of simulation runs. Then find a more
exact result for the optimum value of the extra effort. ♢

63
5.2.4 Decision trees
The use of decision trees is a fruitful approach when we are going to systemise a decision process
where the decisions are made at different point of times. The main reason for postponing a decision
is to follow the development of e.g., a project, and hence make the most appropriate decision when
more information is available. The drawback is that postponing a decision could yield more costly
solutions. Another drawback could be that it is no time to implement necessary measures in due
time if we wait to take action.

Starting point

Decision node

Chance node

Consecutive costs

C End consequences, terminal node

Coupling between nodes and end consequences

Figure 5.5: Symbols used in decision trees

Analysis of decision trees

The following notation is introduced:
CNi = Chance node i
pi,j = probability that chance node i results in outcome j.
DNi = Decision node i.
CINj = Cost of intermediate node j
CTNj = Cost related to terminal node j.
EMV = Expected Monetary Value .
EMVi,j = EMV for branch j into chance node i.
EMVi = EMV for chance node i, or decision node i.
The algorithm for numeric calculating is shown in Figure 5.6.

Example 5.5
Construction Ltd. is the main contractor for a road tunnel project. During the work more water
penetration than expected is discovered. Physically there are three alternatives to choose among:i)
bursting an outlet drain which is very costly but a satisfactorily solution, ii) build a pumping
station to pump away the water which is a cheaper solution but may not be adequate if there is
very much water, and iii) carry out seal work which is even cheaper, but adequate only in case
of very little water. The amount of water is uncertain at the time being. Below we discuss the
decision process:
The first decision is now (DN1 ), and at this decision node we have the following options:

64
Repeat for all end terminals
(*) Move to the node to the left, and bring with the EMV-value in the current node
IF this is a chance node THEN
Calculate EMVi,j = pi,j · EMV
IF EMV has been calculated for all branches into this node THEN
Calculate EMVi = Σj EMVi,j
GoTo (*)
ELSE
GoTo next terminal node
ENDIF
ELSEIF this is a decision node THEN
IF EMV has been calculated for all branches into this node THEN
Let EMVi = Min j ( EMVi,j )
Optimum decision in DNi is the branch with minimum EMVi,j
GoTo (*)
ELSE
GoTo next terminal node
ENDIF
ELSEIF this is a consecutive node THEN
Add EMV of the consecutive node to EMV
GoTo (*)
ELSEIF this is the start node THEN
We are done
ENDIF

Figure 5.6: Algorithm for processing a decision tree

A: Immediate start bursting work

B: Wait half a year until more information about the amount water is available
If we postpone the decision (B) we would have more information about the amount of water in half
a year and a better decision could be made. Two outcomes are foreseen in half a year (CN1 ):
C: It is obviously so much water that bursting the outlet drain is necessary

D: There is still uncertainty regarding the amount of water, and we have a new option in decision
node DN2 :

E: Build a pumping station and hope that this is sufficient, or

F: Wait another half year to obtain even more information

If the pumping station is build at this time (E) this could result in the following outcomes (CN2 ):
G: The pumping station was sufficient

H: The pumping station was not sufficient, and an outlet drain have to be bursted
If we wanted to postpone the decision (F) there are tree possible outcomes (CN3 ):
I: Bursting the outlet drain is required

65
 Drain outlet
A 50 million

DN1 C Drain outlet

60 million
Pump station
B CN1 E CN2 G
20 million

D DN2 H Fiasco
90 million

I Drain outlet
70 million

F CN3 J Seal work

10 million
Pump station
K
25 million
Figure 5.7: Decision tree for tunnel project

J: Sealing work is sufficient

K: A pumping station is sufficient

Table 5.2 shows the associate costs (the letter in parentheses corresponds to the alternative above).

Table 5.2: Cost of the various options

Options Cost now Cost in half a year Cost in one year
Outlet drainage bursting 50 mil. (A) 60 mil. (C) 70 mil. (I,H)
Pumping station 20 mil. (G) 25 mil. (K)
Seal work 10 mil. (J)

At the moment we make the following probability assessments:

P(C|CN1 ) = 30%
P(D|CN1 ) = 70%
P(G|CN2 ) = 90%
P(H|CN2 ) = 10%
P(I|CN3 ) = 10%
P(J|CN3 ) = 40%
P(K|CN3 ) = 50%
Note that in the example we have not used the symbol for consecutive costs. For the calculation
we use the algorithm indicated in Figure 5.6. We start with the upper right terminal node, and
“collect” the EMV = 50 mill. into the decision node to the left, e.g., DN1 . In this decision node we
observe that not all branches (from the right) into node DN1 have been processed, and we therefore
need to go back to a new terminal node. We go back to the next upper terminal node and collect
EMV = 60 mill. which is multiplied with the branch probability (30%) such that we get EMV =
0.3 · 60 mill. = 18 mill. into chance node CN1 . Here, the second branch into the chance node has
not been processed and we again have to go back to the next non-processed terminal node. Here we

66
 50 Drain outlet
50 million

DN1 18 Drain outlet

30% 60 million
90% Pump station
34.45 CN1 27 CN2 18
20 million
70% 10% Fiasco
16.45 DN2 9
90 million

7 Drain outlet
10% 70 million
40% Seal work
23.5 CN3 4
10 million
50% Pump station
12.5
25 million
Figure 5.8: Decision tree for tunnel project with calculations

collected EMV = 20 mil which is multiplied with 90% gives EMV = 18 mill. into chacne node CN2 .
Similarily we get EMV = 90 mill. · 10% = 9 mill. for the second branch into chance noe CN2 . We
may now complete the processing of chance node CN2 by adding the EMV values entering the node
from the right, yilding an EMV of 27 mill. This number now goes into decision node DN2 . Now the
remaining end nodes are processed, and we get the EMV to collect from CN3 equal to 23.5 which
again will be the second EMV into decision node DN2 . In decision node DN2 we shall choose the
branch having the lowest EMV value, i.e., branch F with an EMV of 23.5. In decision node DN2 it is
most beneficial to postpone the decision for another half year. We complete the tree similarly, and
find that in decision node DN1 the optimal decision is to postpone any physical activity. We remain
then with an EMV equal to 34.45. The number from these calculations are shown in Figure 5.8.
Also note that we have not taken the discounting aspects into account, something that also would
have been an argument for postponing the decision. ♢

Problem 5.10 An oil company has the rights for a given oil field, and have the options between:
• Drill a well (D)

• Sell our rights (S)

The decision will depend on the amount of oil that might reside in the field. There are two options:
• Profitable oil pool (P)

• Non profitable oil pool (NP)

Before the final decision is made, our oil company could conduct an expensive seismic investigation
which might give information regarding the probability that the field contains a profitable oil pool.
The result from such an investigation will be one of the following statements:
• No structure (NS)

• Open structure (OS)

• Closed structure (CS)

67
In this problem you should establish a decision tree for the situation. The decision node following
the start node should be:

• Perform a seismic investigation (SEI)

• Do not perform a seismic investigation (NINV)

The cost involved in the decision tree is as follows:

Cost of seismic inveestigation: 10

Cost of drilling a well: 100

Net profit (after drilling) if oil: 700

Selling price without seismic investigation: 30

Selling price with seismic investigation:
- No structure 0
- Open structure 50
- Closed structure 200

The probability that shall go into the decision tree is as follows:

Probabilities in relation to a seismic investigation (e.g., based on experience figures from geol-
ogists):

No structure: 0.60
Open structure: 0.30
Closed structure: 0.10

The probabilities for finding a profitable oil pool, given the result of the seismic investigation:

No structure: 0.10
Open structure: 0.25
Closed structure: 0.70

The probability for finding oil when no seismic investigation has been performed: 0.20. Find the
optimal decision in each decision node, and formulate the conclusions from your analysis. ♢

68
Chapter 6

Life cycle cost and life cycle profit

6.1 Introduction
In this chapter we will give a short introduction to life cycle cost (LCC) modelling and anaylysis
in connection with project management. The term LCC is defined in IEC 60300: “LCC is the
cumulative cost of a product over its life cycle”. The LCC concept was first introduced in the
US Army and the idea was to establish the cost of development, production and use (operation
and maintenance) of military equipment. In the original use the revenues was not included in the
modelling. However, in order to get a complete picture we will usually also include the possible profit
of a new system or product. Hence the term ‘Life Cycle Profit’ has been introduced. Kawacuchi
and Rausand [8] suggest a process for LCC analysis comprising the following steps:

1. Problem definition

2. Cost element definition

3. System modelling

4. Data collection

5. Cost profile development

6. Evaluation

7. Reporting

In this presentation we will focus on the cost modelling aspects, i.e. mainly step 3 in the above
procedure.

6.2 Net present value calculation

The formulas for LCC calculation are based on standard formulas used in net present value (NPV)
calculations. In the following we will summarise the most frequent used formulas. The basic idea
in NPV calculation is that money received in the future will be less valued than the same amount
of money today. To treat this formally all future amounts are discounted to the present time, i.e.
present values. We will only consider discrete time, i.e. all amounts will occur at the end of each
year, or now (beginning of year one). The cash flow is illustrated in Figure 6.1.

69
 (now)
X1 X2 X3 X4 X5 XT −1 XT
time
0 1 2 3 4 5 T −1 T
Figure 6.1: Visualisation of the cach flow, Xt

The net present value of an amount Xt that occurs at the end of year t is:

NPV = Xt (1 + r)−t (6.1)

where r is the discount rate. Similarly, we find the net present value of a cash flow X0 , X1 . . . . , XT :

∑
T
NPV = Xt (1 + r)−t (6.2)
t=0

where X0 represents in or outgoing cash now, and T is the number of years to consider.
Sometimes we want to establish the net present value of a constant yearly (nominal) amount
∑n i.e.i the samen amount each year. By utilising the formula for the sum of a geometric series,
XA ,
i=1 q = q(1 − q )/(1 − q) we obtain:
[ ]
1 − (1 + r)−T
NPV = XA (6.3)
r

Note that NPV approaches XA /r as T approaches infinity.

Now, consider a situation with a fixed increasing yearly value, were the first in or outgoing
amount is XA,v (at the end of the first year), and where the amount is increasing by a factor (1 + v)
each year. The net present value for T years is then found to be:
 ( )T 
 1 − 1+r 
1+v
NPV =   XA,v (6.4)
r−v

IF r = v in Equation (6.4) we use NPV = XA,v T /(1 + r) obtained by l’Hopitals rule.

The expression in Equation (6.3) assumes that the amount XA,v occurs every year. In some
situations we want to consider an amount XA,v which occurs every k year, where k > 1. The net
present value is now given by (assuming first amount now (t = 0)):
∞
∑ XA
NPV = XA (1 + r)ki = (6.5)
1 − (1 + r)−k
i=0

If the first amount occurs at the end of year l we obtain:

XA (1 + r)−l
NPV = (6.6)
1 − (1 + r)−k

6.2.1 Trend modelling

When modelling trend it is important to find a simple mathematical expression for the time de-
velopment. Further note that the change in the yearly amount is due to at least the following
factors:

70
 • The monetary value increases due to general conditions, such as inflation.

• The monetary value increases due to increased operating costs, e.g. physical deterioration
and hence more maintenance is required.
Increased operating costs due to deterioration could usually be reset by a renewal of the system we
are considering, whereas external conditions like inflation is not affected by e.g. a system renewal.
In the modelling we will assume a fixed inflation rate, even if we in a more advanced model also
could let the inflation rate vary. This inflation rate will be denoted v, and we could use Equation 6.4
to calculate the net present value of a amount that changes due to inflation. When we want to
model increased operating cost due to deterioration, we need to introduce a local age parameter.
We will let a denote the age of the system, or the age of the system since the last system renewal.
When we consider degradation, we introduce the degradation rate d where we assume that the
yearly increase due to deterioration equals (1 + d). This corresponds to an exponential growth
which very often is found realistic if we have degradation mechanisms that drive the costs. Now,
let c0 be the yearly cost of operation, maintenance etc now (i.e. at time t = 0). We then have the
yearly cost in year t (occurring at the end of year t):

ct = c0 (1 + d)t (6.7)

In order to obtain the degradation rate d we usually need data about the costs as a function of
time. A very simple approach if we know that c(t) has increased by a growth factor (GF) during
a period of T years. We then have:

d = eln(GF)/T − 1 (6.8)

6.2.2 Example areas of LCC calculations

In the following we give examples of areas where LCC analysis and calculation could be used. We
differentiate between situations were decisions are related to project execution, and the progression
of one project, or a portfolio of projects, and the situation where we consider which project are
profitable, or how the profitability could be maximised. Examples related to project execution:
• Invest in equipment to increase efficiency in project execution, e.g. a new excavator.

• Choice between construction method A and B.

• Outsourcing of truck-maintenance.

• Lease equipment rather than by our selves.

Examples related to project profitability:
• Development of one or more oil fields.

• Construction of a new passing loop.

• Renewal of ballast in a railway track.

• Point wise refill of ballast in order to postpone the need for a full renewal (ballast cleaning).

• Grinding of rails.

• Invest in a new production line.

71
There are several aspects to consider when conducting an LCC analysis, for example:

• Visualise the cost picture, enabling the possibility to work actively with eliminating the main
cost drivers, or the effect of these.

• Use the LCC model as a decision support when making decision about the profitability of
projects or measures, and when to conduct or implement these.

• Use the LCC model as a basis for contractual follow-up, e.g. LCC contracts.

Example 6.1 We will consider a railway system where the quality of the ballast has deteriorated
during the last years, and in order to compensate for this it is proposed to do a point wise replace-
ment of the ballast on the line. The age of the ballast is 35 years, and without this point wise refill
of ballast it is expected that a full renewal (ballast cleaning) is necessary within five years. If we
conduct the project we could postpone the ballast cleaning with another five year. The length of
the line we are considering is 10 km. The quantities to include in the LCC model is as follows:

RC = 2.5 million Euro = Renewal cost = 250 Euro per meter for ballast cleaning.

IC = 400,000 Euro= Improvement cost, e.g. cost of point wise ballast refill.

LT = 40 years = Life length of ballast = period between ballast cleaning.

a = ballast age, i.e. effective age relative to the implemented measures. Without point wise
refill of ballast a = 35 years, and with point wise refill of ballast a = 30 year. For a track
that has just being renewed a = 0.

c0 = 25,000 Euro = yearly cost of maintenance and operation of the track, for a new track, i.e.
just being renewed.

c40 = 250,000 Euro = yearly cost of maintenance and operation of the track, for a track that has
reached it’s service life, e.g. 40 years.

d = eln(250000/25000)/40 − 1 = 0.05925

ct = c0 (1 + d)t+a = 25000(1 + 0.05925)t+a = total maintenance and operation cost in year t

(from now), and a is the effective age of the track.

r = 6% = interest rent.

We start by calculating the various LCC-terms (in million Euros) if the improvement project (point
wise refill of ballast) is not executed. The total renewal cost if found by Equation (6.6):

RC(1 + r)−5
LCCRC = = 2.069
1 − (1 − r)−40

The variable cost the next five years (up to the next renewal) is found from Equation (6.4)
 ( )5 
 1 − 1+r 
1+d
35
LCCVC,1 =   c0 (1 + d) = 0.883
r−d

72
After the renewal in five year the variable costs will be reset to v0 , and then start increasing again.
The net present value in one cycle is:
 ( )40 
 1 − 1+d
1+r 
LCCVC,0 =   c0 (1 + d) = 0.986
r−d

The amount LCCVC,0 will then be repeated every 40 year, and the first time will be in five years:

LCCVC,0 (1 + r)−5
LCCVC,∞ = = 0.816
1 − (1 + r)−40
Finally we have the total contribution from variable costs:

LCCVC = LCCVC,1 + LCCVC,∞ = 1.699

If we execute the improvement project, the calculations are similar. We start with the total renewal
cost (first renewal after 10 years):
RC(1 + r)−10
LCCRC = = 1.546
1 − (1 + r)−40
The variable cost the next ten years (up to the next renewal) noting that the effective age after the
improvement project is a = 30:
 ( )10 
 1 − 1+d
1+r  30
LCCVC,1 =   c0 (1 + d) = 1.322
r−d

After the renewal in ten year the variable costs will be reset to v0 , and then start increasing again.
The net present value in one cycle, LCCVC,0 , is the same as without the improvement project, but
the first cycle will start in ten years:
LCCVC,0 (1 + r)−10
LCCVC,∞ = = 0.610
1 − (1 + r)−40
Finally we have the total contribution from variable costs:

LCCVC = LCCVC,1 + LCCVC,∞ = 1.932

In this last situation we also need to include the investment cost:

LCCIC = 0.4

Summing up all LCC contributions we find that implementing the improvement project gives a
total LCC of 3.878 million versus not implementing the project gives a total cost of 3.768. Thus
the improvement project is not profitable. ♢

Problem 6.1 Consider Example 6.1. Find the value of the discount r that makes the two alter-
natives equal from a LCC point of view. Why is this value of r higher than the initial one?

Problem 6.2 Consider Example 6.1 and investigate if ballast cleaning after 40 years is optimal.
If not, find the optimal period for ballast cleaning, i.e., the optimal renewal period.

73
Chapter 7

Parameter estimation

7.1 Introduction
In this chapter we will briefly describe principles for parameter estimation. A parameter in this
context is a quantity in the risk analysis for which we assign numerical values. There are two
principles for establishing numerical values (parameter estimates):

• Statistical analysis of historical data

• Use of expert judgements

If we have access to relevant data we will usually use these data to estimate the parameters.
Often we have little relevant data, and we then have to rely on expert judgements. In some
situations we combine historical data with expert judgements by use of Bayesian methods.

7.2 The MLE principle

The basic idea behind the Maximum Likelihood Estimation (MLE) principle is to choose the nu-
merical values of the parameters that are the most likely ones in light of the data. The procedure
goes as follows:

• Assume that we know the probability density function of the observations for which we have
data. Let this distribution be denoted f (x; θ).

• The involved parameters, are unknown, and are generally denoted θ.

• We have n independent observations (data points) that we denote X1 , X2 , …Xn . When we

refer to the actual numerical values we have observed, we use the notation x1 , x2 , …xn .

The MLE principle now tells us to estimate θ by the value which is most likely given the observed
data. To define “likelihood” we use the probability density function. The simultaneous probability
density for X1 , X2 , …Xn is given by:

∏
n
f (x1 ; θ)f (x2 ; θ) . . . f (xn ; θ) = f (xi ; θ) (7.1)
i=1

74
This density express how likely a given combination of the x-values are, given the value of θ.
However, in our situation the x-values are given, whereas θ is unknown. We therefore interchange
the arguments, and consider the expression as a function of θ:
∏
n
L(θ; x1 , x1 . . . xn ) = f (xi ; θ) (7.2)
i=1

where L(θ; x1 , x1 . . . xn ) in equation (7.2) denotes the likelihood function. The MLE principle will
now be formulated as to choose the θ-value that maximizes the likelihood function. To denote the
MLE estimator we write a “hat” over θ, θ̂. Generally, θ will be a function of the observations:

θ̂ = θ̂(X1 , X2 , . . . Xn ) (7.3)

When we insert numerical values for the x’s we denote the result as the parameter estimate.

Example 7.1 Estimation in the exponential distribution

We consider the situation where we have observed n failure times, and we will estimate the failure
rate, λ, under the assumption of exponentially distributed failure times. The observed failure times
are denoted t1 , t2 , …, tn . Equation (7.2) gives:
∏n
L(λ; t1 , t2 , . . . , tn ) = λe−λti
i=1

Note that the parameter is denoted λ, whereas we generally use θ. Further we denote the observa-
tions with t because we here have failure times. The probability density function in the exponential
distribution is given by f (t) = λe−λt . A common “trick” when maximising the likelihood function
is to take the logarithm. Because the logarithm (ln) function is monotonically increasing, ln L will
also be maximised for the same value as for which L is maximised. We could then find:
∑n
l(λ; t1 , t2 , . . . , tn ) = ln L(λ; t1 , t2 , . . . , tn ) = n ln λ − λti
i=1

By taking the derivative wrt λ and set this expression to zero, we easily obtain:
/∑n
λ̂ = n ti
i=1

Problem 7.1 Find the MLE for µ and σ in the normal distribution. ♢

7.3 Method of moments – PERT distribution

The maximum likelihood principle is not numerically stable for the PERT distribution. We will
therefore apply another principle, i.e., the methods of moments. The method of moments is a
method of estimation of population parameters by equating sample moments with unobservable
population moments and then solving those equations for the quantities to be estimated.
In the PERT distribution we have the following moments:
L + 4M + H
E(X) = (7.4)
6
(E(X) − L)(H − E(X))
Var(X) = (7.5)
7

75
 ∑
From (7.4) we easily obtain M = (6E(X) − L − H)/4. Now let x̄ = i xi denote the sample mean
(first order moment), and we may estimate M by:
M̂ = (6x̄ − L̂ − Ĥ)/4 (7.6)
The challenge now is to find L̂ ∑
and Ĥ. By rearranging equation (7.5) and inserting x̄ for E(X) and
2
i (xi − x̄) for Var(X) we have:
1
the sample variance S 2 = n−1

7S 2 = (x̄ − L)(H − x̄) (7.7)

where L and H are the two unknowns to find by one equation. To overcome the problem of under-
determination it seems reasonable to require (x̄ − L)/(H − x̄) = (x̄ − xMin )/(xMax − x̄) where xMin
and xMax are the sample minimum and maximum respectively. Hence we have:
x̄ − xMin
7S 2 = (H − x̄)2 (7.8)
xMax − x̄
yielding
√
xMax − x̄
Ĥ = x̄ + S 7 (7.9)
x̄ − xMin
and
(Ĥ − x̄)(x̄ − xMin )
L̂ = x̄ − (7.10)
xMax − x̄
thus, the final estimates in the PERT distribution are given by equations (7.6), (7.9) and (7.10).

Problem 7.2 Assume that we have observed the following durations for a typical activity in
projects: 9.3, 10.5, 9.4, 9.0, 9.4, 8.6, 10.1, 10.7, 12.0, 10.6, 9.8, 13.1, 12.0, 8.6 and 10.9. Estimate
the parameters if you assume that the durations are PERT distributed. Hint: Use the Average()
and STDEV() functions in MS Excel to find x̄ and S. ♢

7.4 The LS principle

The least squares (LS) principle for estimation is used when we have observations that do not come
from the same distribution, but we know the expectation of each variable as a function of a set
of parameters θ, and a set of explanatory variables. Previously we denoted the observations by
the letter ‘X’, but we will now change the notation to let ‘Y ’ denote the observations, whereas we
reserve the letter ‘X’ for explanatory variables. We now let ϕi (θ) denote the expectation of Yi (the
i’th observation), where the functions ϕi are all known, but the parameter vector θ is unknown
and shall be estimated. The LS principle now states that we may estimate θ by the value that
minimises the square sum of the deviations between the observed and expected values, i.e.:
∑
n
Q(θ) = [yi − ϕi (θ)]2 (7.11)
i=1

Equation (7.11) is the starting point for estimating the parameters in so-called regression models.
The most simple formula is given by:
E(Yi ) = β0 + β1 xi (7.12)
In this model x is denoted the independent variable, whereas Y is denoted the dependent variable
because it depends on the independent variable, x.

76
Problem∑7.3 Prove that the estimators for β0 and β1 in equation (7.12) is given by:
(x −x̄)y
β̂1 = ∑i (xi −x̄)2i and β̂0 = ȳ − β̂1 x̄.
i i
♢
The model in equation (7.12) could be extended to cover more independent variables. These
are denoted regression variables , or explanatory variables. To extend the model we introduce an
extra index for each x. We write xij , where index i denotes the i′ th data point, whereas index j
denotes the j’th explanatory variable. The model then reads:

E(Yi ) = β0 + β1 xi,1 + β2 xi,2 + · · · + βr xi,r (7.13)

To obtain the LS estimators in this situation, we introduce matrix notation. Let y = [y1 ,y2 , …,
yn ]T be a column vector containing the dependent variables, and let X be the design matrix given
by:
 
1 x11 ...... x1r
 1 x21 x1r 
X=  :

 (7.14)
xij
1 xi1 ....... xnr

It may be shown that the LS estimator for β = [β0 , β1 , β2 , · · · , βr ]T is given as the solution of the
following matrix equation:

XT y = XT Xβ (7.15)

If the design matrix has full rank, XT X will be non-singular, and the solution is given by:

β̂ = (XT X)−1 XT y (7.16)

If one has access to a tool for matrix calculus, we easily obtain the LS estimates. We could also
use commercial available statistical programs, or the “analysis” module of MS Excel.

Example 7.2 Estimation of the effects of regression variables

We will consider a situation where we have observed the duration of construction the foundation
wall of houses. The different values are shown in the Y-column below. The variable x1 denotes
the base in square meters, whereas x2 is an indicator of ground frost. A value is given as 1 if there
is ground frost, 0 otherwise. We have also introduced the variable x3 that denotes the walking
distance from the workmen’s hut to the building site:
Y x1 x2 x3
8.4 100 1 100
7.8 150 1 50
11.4 250 1 50
6.1 80 0 75
6.1 100 0 200
8.3 90 1 30
7.5 180 0 25
7.2 200 0 50
6.0 110 0 75

77
From MS Excel we obtain the following parameters: β̂0 = 4.211, β̂1 = 0.0167, β̂2 = 2.196, and β̂3 =
0.0011
♢

Problem 7.4 Use MS Excel to verify the above results. ♢

Note that we in equation (7.13) have written the expected value of Yi . Generally we write:

Yi = β0 + β1 xi,1 + β2 xi,2 + · · · + βr xi,r + εi (7.17)

where εi is an error-term. Very often we assume εi to be normally distributed, but we might

also assume that εi is PERT distributed. To estimate the parameters in an underlying PERT
distribution we calculate the predicted values:

yˆi = βˆ0 + βˆ1 xi,1 + βˆ2 xi,2 + · · · + βˆr xi,r (7.18)

then we estimate the error-terms by the residuals:

εˆi = yi − yˆi (7.19)

The residuals εˆi may now be used as input in the method of moments to estimate PERT parameters.
Below se summarize the procedure to get L, M and H values for an activity, cost element etc. in
a specific project.

1. Estimate regression parameters from data from similar projects: β̂ = (XT X)−1 XT y

2. Calculate the predicted values: yˆi = βˆ0 + βˆ1 xi,1 + βˆ2 xi,2 + · · · + βˆr xi,r

3. Estimate the error-terms by the residuals: εˆi = yi − yˆi

4. Use the estimates εˆi as basis for estimation of L, M and H, i.e., find L̂, M̂ and Ĥ by the
method of moments

5. For the new activity, cost element etc., find the corresponding x-vector, and denote it x =
[x1 , x2 , . . . , xr ]

6. Find the prediction of the new observation by y0 = βˆ0 + βˆ1 x1 + βˆ2 x2 + · · · + βˆr xr

7. The PERT parameters to use in the analysis are given by L = L̂ + y0 , M = M̂ + y0 and

H = Ĥ + y0

Problem 7.5 Calculate the residuals in equation (7.19) with the data in Example 7.2 and estimate
the parameters by assuming the residuals are PERT distributed. Find the probability that the
duration for observation number 5 is shorter than the observed value of 6.1. From the observations
it seems that duration number 3 is rather long. Conclude on this by applying the regression model.
♢

78
7.5 Bayesian meothds
In Bayesian estimation procedures we utilise prior information about the relevant parameters. The
procedure could briefly be described as follows:

1. Specify a prior uncertainty distribution of the parameter vector, π(θ).

2. Structure reliability data information into a likelihood function, L(θ; x), see equation (7.2).

3. Calculate the posterior uncertainty distribution of the para,eter parameter vector, π(θ|x).
The posterior distribution is found by π(θ|x) ∝ L(θ; x)π(θ), and the proportionality constant
is found by requiring the posterior to integrate to one.

4. The Bayes estimate for the parameter vector is given by the posterior mean, which in principle
could be found by integration.

Example 7.3 Exponential distribution

In the following we give an example showing the main elements of the procedure. In the example
we will estimate the failure rate in the constant failure rate situation. Assume that we express our
prior believe1 about the failure rate λ of a certain component (gas detector used on an oil and gas
platform), in terms of the mean value E = 0.7 · 10−6 (failures / hour), and the standard deviation
S = 0.3 · 10−6 . For mathematical convenience, it is common to choose a gamma distribution2 with
parameters α and ξ for the prior distribution. The expected value (E) and the variance (V ) in the
gamma distribution are given by α/ξ and α/ξ 2 respectively, and we obtain the following expressions
for α and ξ:

ξ = E/V = E/S 2 = (0.7 · 10−6 )/(0.3 · 10−6 )2 = 7.78 · 106

α = ξE = (7.78 · 106 ) · (0.7 · 10−6 ) = 5.44

To establish the likelihood function, we look at the data. In this example we assume that we
have observed identical units for a total time in service, t, equal to 525 600 hours (e.g. 60 detector
years). In this period we have observed n = 1 failure. If we assume exponentially distributed failure
times, we know that the number of failures in a period of length t, N (t), is Poisson distributed with
parameter λ · t. The probability of observing n failures is thus given by:

L(λ; n, t) = Pr(N (t) = n) ∝ λn e−λ·t

and we have an expression for the likelihood function L(λ;n, t).

The posterior distribution is found by multiplying the prior distribution with the likelihood
function:

π(λ|n) ∝ L(λ; n, t) · π(λ) ∝ λn e−λ·t · λα−1 e−ξλ ∝ λ(α+n)−1 e−(ξ+t)λ

and we recognize the posterior distribution as a gamma distribution with new parameters α′ = α+n,
and ξ ′ = ξ + t. The Bayes estimate is given by the mean in this distribution:
α+n 5.44 + 1
λ̂ = = = 0.78 · 10−6
ξ+t 7.78 · 106 + 525600
1
This could be based on statements from experts, see Øien et.al (1998), or by analysis of similar components
(empirical Bayesian analysis).
2
π(λ) ∝ λα−1 e−ξλ for the gamma distribution.

79
 We note that the maximum likelihood estimator gives a much higher failure rate estimate
(1.9 · 10−6 ), but the “weighing procedure” favours the prior distribution in our example. Generally
we could interpret α and ξ here as “number of failures” and “time in service” respectively for the
“prior information”. ♢

80
Chapter 8

Expert judgements

Note that this chapter is widely based on: Øien, K. & Hokstad, P.1998 ”Handbook for performing
Expert Judgement”. SINTEF report STF38 A98419 (ISBN 82-14-00449-7). Whereas Øien &
Hokstad have a general approach, we will here focus on estimation where the objective is to assess
two or more parameters describing the uncertainty distribution of durations or costs.

8.1 Introduction
When statistical data do not exist or are not available, the alternative is to obtain such information
from experts/resource persons. To handle this in the best possible way, this should be carried out as
a structured and systematic process, both during planning, elicitation and calculation, as proposed
in this handbook. Such a structured collection of information is what we call “expert judgements”.
Thus, expert judgements are carried out to provide necessary input data for our analyses.
ESA [17] gives the following definition of expert judgement data:
“Expert Judgement data are estimates of unknown values about a system made by specialists
who have system-related knowledge.”

8.1.1 Purpose
The purpose of this chapter is primarily to give a simple and complete “recipe” of how expert
judgements may be carried out. Note that the referred handbook by Øyen and Hokstad gives a
significant more detailed picture than this chapter. At the end of the chapter combining expert
judgement findings with statistical data is also discussed.

8.1.2 Extent
Figure 8.1 shows tasks that are covered for performing expert judgements. This chapter covers all
steps of all 3 phases.

8.1.3 Use
It has been an overall objective to make this chapter flexible, that is, also to be applicable for
expert judgements carried out with little resources, and thereby in a simple way. In addition, there
are guidelines of how to perform more comprehensive expert judgements.
Therefore, it is not the intention that all the procedure items are to be carried out every time.
Which items are to be carried out depend on which resources are available, the problem to solve,

81
 Problem
description
Evaluation of
available resources
Choosing
experts
Evaluation of
experts
Choosing method
of calculation
Developing a
PHASE I: THE PREPARATION PHASE
questionnaire

Information
to the experts
Accomplishment
of elicitation
Establishing
3-point estimates
Establishing a basis
for giving weights
Final
discussion
Possible adjustments
PHASE II: THE ELICITATION PHASE
of weights

Calibration
of experts
Weighting
of experts
Calculation of
combined estimates

Documentation

Communication/
Presentation
Supplementary
PHASE III: THE CALCULATION PHASE
work
Figure 8.1: The steps in an
82 expert judgement process
and the choices made during the work/process, (among others: choosing method of calculation).
All items should therefore be looked into, in order to evaluate whether they are relevant. The short
version in Section 3 gives a suggestion for what to include for very simple expert judgements.
This chapter represents an alternative to an unstructured and undocumented “engineering”
judgements.

8.2 General theory

8.2.1 History
Speculations, brainstorming and guessing made by experts and used as basis in structured decision
processes, is of relatively new date. It all started with the establishment of RAND Corporation in
the USA after the second World War. At first, two methodologies were dominating. These were
the scenario method and the Delphi method. Both were developed at RAND.

Scenario analyses
Herman Kahn developed this method as a kind of system analyses method, where hypothetical
sequences of events are constructed for the purpose of focusing on cause processes and decision
points. This gives the answers to two kinds of questions:

1. Precisely how might some hypothetical situation come about, step by step?

2. What alternatives exist, for each actor, at each step, for preventing, diverting, or facilitating
the process?

This type of scenario analysis must not be confused with accident scenarios, and are only used to
examine the main trends, which are extrapolated into the future. This is done without evaluating
or using the probabilities of the scenarios to occur.
When the trends are extrapolated into the future, any theoretical or experience based knowledge
that may influence the extrapolation is considered.

The Delphi method

This method is also developed at RAND [30] and is the most well-known method to obtain and
treat the experts’ opinions/judgements.
The Delphi method can be summarised into 8 steps:

1. An “observation team” (the analysts) defines the purposes and choose the respondents (ex-
perts). Normally, the respondents are anonymous to each other, and the answers are anony-
mous.

2. The observation team prepares a temporary questionnaire, which is sent to the respondents
for comments.

3. The answers are reviewed, and a final questionnaire is established.

4. The respondents answer the questionnaires.

5. The observation team analyses the answers and calculates median values and the interquartile
range (the 25% and the 75% estimates).

83
 6. The results are returned to the respondents, who are asked whether they want to adjust their
answers. Those who still are outside the interquartile range, are asked to give arguments for
their prediction.

7. The revised predictions are processed in the same way as the first responses, and arguments
for “outliers” are summarised. This information is sent back to the respondents, and the
whole process is iterated (3 - 4 times).

8. The median values on the final round are taken as the best predictions. Generally, the spread
in the last round is smaller than in the first round, which is taken to indicate a degree of
consensus.

This method was very popular in the 1960’s and 1970’s, but some later evaluation studies which have
seen carefully and critically on the Delphi method, have concluded that it violates methodological
rules for common experimental science. Comparisons with other methods have also shown that the
Delphi method was the poorest.
One major criticism is directed to the “reward” given to the expert for changing his estimates
towards the median value through group interaction. It has been shown that this does not increase
the relative frequency of correct estimates. The Delphi method seems to have lost its popularity.

8.2.2 Which types of evaluations/estimates do we consider as expert judge-

ments?
In practice you can be an expert in any field or any topic, and in this respect one could regard any
information or knowledge about this as expertise, and treat it as “expert judgements”. However,
in this chapter we will limit ourselves to a specific type of expert judgement, as we ask the experts
to give their knowledge/experience on a subject.
There is no clear definition in the literature about what type of information that may be
seen as expert judgements. However, in most of the literature, expert judgements are performed
to “characterise” the collection of quantitative data in lack of statistical data, (e.g. subjective
probabilities). Most likely duration of an activity is an example of such data.
Figure 8.2 illustrates some of the information that is needed in a project (for instance, a re-
liability analysis), to show what we mean by “expert judgements” compared to other types of
information.
In the system definition phase, we have a need for information about the system, how it works,
is supposed to be used, maintained, etc. This is information that we may obtain from the personnel
who have this experience, and may be viewed as “experts”, but this type of information (no matter
how important it is) is not what we usually mean by expert judgements.
In the analytical phase, the analysts have to choose the methods that will be used, based on
the type of problem they are facing. Here, the analysts may be viewed as “experts” of the method,
or may consult senior personnel who are perceived as experts of analytic methods, so that these
will choose the best methods to the relevant problem. This is also important information, however,
neither this is what we usually mean by expert judgements.
When the method is chosen, the problem must be adjusted to the methods. That is, modelling
the problem by making simplifications and assumptions. Such a modelling (conversion) of the
problem should be evaluated by a responsible person within the clients organisation (the “problem
owner”) who has basic understanding of the methods that will be used, and who may evaluate
whether the simplifications and assumptions are valid and acceptable. Such a verification of the

84
 Information from per-
SYSTEM sonnel who have knowl-
DEFINITION edge/experience about the
system

Usually chosen by the ana-

CHOICE OF
lyst. May also consult other
METHODS
“senior”-analysts.

Evaluation of the modelling

ANALYTICAL
MODELLING performed by client/problem
PHASE
owner

Data from databases, etc.

INPUT Expert judgements
DATA performed by
experienced personnel

Evaluation of the steering

RESULTS/CONCLUSIONS committee in relation to the
project goal

Figure 8.2: Expert judgements of input data compared to other type of information

85
modelling is of great importance (and may easily be disregarded or handed over to the analysts
only). This type of information/evaluation is neither what we usually view as an expert judgement.
Input data to our models may either be statistical data (objective /”hard” data) or subjective
(“soft”) data provided by personnel who have the necessary knowledge and experience to provide
this information. Such subjective data estimates is what we here view as expert judgements. Also,
we have restricted this to quantitative data. However, in general also qualitative estimations, as
for instance ranking, may be viewed as expert judgements.
Final results and conclusions are evaluated by a steering committee/client against the project
targets. Such an “expert judgement” of the results is also important, but is not what we usually
mean by the term expert judgements.
Evaluations/quality assurance performed by the person(-s) responsible of the quality assurance
are not illustrated in the figure, but should be performed in all phases of the project. Neither such
an evaluation is what we usually mean by expert judgements.
To summarise, we may say that the type of evaluations that are viewed as expert judgements
in this case is:

Quantitative estimates of input data to models

8.2.3 Why is there a need for expert judgements?

Expert judgements are primarily of interest when data is lacking, or when we have insufficient
statistical data. They are not meant as an alternative to statistical data. Ideally, we should
preferably establish databases to obtain statistical data. However, in anticipation of databases
being established and events to occur, we have to use the knowledge and experience that are in the
possession of persons who work with the system. Otherwise, the problem is not solved. Besides,
for new systems and for very rare events, a sufficient base of statistical data will not exist.
It may be claimed that experts have always been utilised in order to give estimates for quan-
titative values of different kinds, because statistical data have not existed, or have not been easily
available. However, this has been done in an informal way. It could be questioned whether this
is actually satisfactory. Why formalise this too much ? Why do we not just continue using “engi-
neering judgements” when quantitative values in our models/calculations are needed?
This is not necessarily a black and white situation. There are many reasons not to perform
an expert judgement in a too formalistic and extensive way. On the other hand, there are good
reasons to be sceptical to an informal handling, such as using the so called “engineering judgements”.
Advantages and disadvantages of both “extremes” are discussed in Table 8.1.
Some of the differences are discussed in the following.

Documentation
By using a formal expert judgement, all the steps in the process are well documented. Normally,
by using engineering judgements, very little is documented. A more structured method, as expert
judgements, also promotes documentation of the results. Further, the information that is collected
often is less precisely defined by engineering judgements. Neither, important assumptions that
make a basis for the judgements are not necessarily stated.

Extent of information (Completeness)

The extent of the information collected via formal expert judgements are often limited by only
being a response to well defined questions. It does not intercept any “extra” information than the

86
Table 8.1: Advantages and disadvantages of strictly formalised expert judgements vs. informal
engineering judgements.
Factors Formal expert judge- Engineering judge-
ment ment
Structure Systematic and struc- Unsystematic and
tured method/process. unstructured process.
“Discussion across the
table”.
Specification of Well specified. Only in- Imprecise. Assump-
information formation given as an- tions are not specified.
swers of well defined
questions.
Documentation All steps of the pro- Poor or no documenta-
cedure are well docu- tion.
mented.
Extent of col- Limited. Only that ob- Wide. May cover many
lected informa- tained through prede- aspects of the subject,
tion fined questions. also through follow-up
questions.
Evaluation of ex- “Objective” rules for The confidence in a spe-
perts evaluation and possible cific expert is judged
weighting of the ex- subjectively by the an-
perts. alyst.
Simplicity Extensive and expen- Very simple. Performed
sive. without preparations.

87
answers to the questions, thus making demands to the questionnaire being worked out. There is
little room for follow-up questions and improvisation, something that may impair the completeness
of the information gathered about the subject. In engineering judgements, one may more easily
gain a general insight in addition to precise quantification of a given subject.

Objective evaluation of the experts

In formal expert judgements, unequal weighting is not allowed if there are no rational reasons for
this (as for example use of control questions). Therefore, this method promote a objective handling
of the experts, while for engineering judgements, it is often a matter of subjective opinion in who
the analyst chooses to have the greatest confidence in.

Simplicity
With its lack of structure and preparation, engineering judgements are often too simple to become
credible. Formal expert judgements have a tendency to become quite complex and extensive.
Thus, it is a problem that too extensive expert judgements hardly make a relevant alternative to
engineering judgements, due to high costs.
Thus, a conclusion is that nowadays’ engineering judgements have a big problem in respect
to documentation and objectivity. As for how complete the information gets, the picture is more
mixed. The advantage of having a well structured list of specified questions is acknowledged.
However, this can lead to a non-flexible elicitation, and is not necessarily the best in all situations.

8.2.4 Who are the experts?

The experts are persons with knowledge and experience about the system we want information
about. This could for instance be maintenance personnel, so one does not necessarily have to have
higher education (e.g. University Degree) to be viewed as an “expert”.
In the literature this question of “who is the expert” is considered to a very limited extent.
Svenson [28] discusses this based on Shanteu, where strong criteria are set to which experience and
qualifications a person must have in order to be viewed as an expert. These are:

1. Experience in performing judgements and making decisions

2. More than 10 years of experience within the current subject

3. Inherent qualities like self-confidence and adaptability

A person that satisfies the first or the two first criteria, but not the last, is evaluated by Shanteau
to be a novice and not an expert. In his evaluation, such a person will not be able to train to
satisfy item iii). Thus, he will stay a novice, even though he sees himself as an expert, and also if
the people around him do so.
To us, this is a too rigid judgement of what is required to characterise a person as an expert.
We do not ask for “experts” meeting some fixed criteria, but rather the persons having as much
knowledge/experience on the subject as possible. Thus, we have quite a pragmatic view to what
we mean by the notion “expert”.

88
8.2.5 How should the expert judgement be carried out?
The method of this chapter is based on some basic requirements some of which were also discussed
in Section 8.2.3, were we performed an evaluation of expert judgements vs. engineering judgements.
The basic requirements are:

1. Documentation

2. Objectivity

3. Empirical control (The estimates should be able to be empirically controlled)

4. Completeness

5. Simplicity

The requirements 1-4 are necessary to obtain scientific credibility. Requirement number five about
simplicity is included to secure a practical method that leads to a widespread use.

8.2.6 Terms and notions

In this section we define some terms and notions related to expert judgements. Many of these
describe qualities or attributes of the experts or the estimates they provide.
In the presentation below we will always assume that there exist some “true” values for the
quantities of interest. In order to make this meaningful we then have to distinguish the random
nature of the problem at hand, and the underlying (true) construct of the problem. Let X be
the quantity in the real problem which is of interest, e.g., the duration of a critical activity. Due
to a set of factors it is not possible to state the actual value of X, and we therefore treat X
as a random quantity (stochastic variable) to represent the variability of the problem. In many
cases it might make sense to consider the problem as a repetitive problem where X could be seen as
realisation of an underlying probability distribution. The parameters in this underlying distribution
are, however, unknown. The objective of expert judgements is then to reveal by an elicit process
these parameters. These underlying parameters are considered to exist, in the sense that we talk
about true underlying parameters. For example, if X is PERT distributed, the quantities L, M
and H are considered to be such true underlying parameters of interest.

Unbiasedness
Unbiasedness may be defined as the degree of “accuracy” of the assessments, and describes to what
extent the assessments show a systematic deviation from the true value. Systematic deviations may
be caused by psychological factors, or by that the experts’ experience somehow is not representative
for the system in question (e.g. having different operational conditions than the expert is familiar
with).
A definition of biasedness may be:
Biasedness = The degree of systematic deviation from the true value
A common measure of the experts’ biasedness is:
Bias = mean of the estimated values - true value

89
Calibration
Calibration may have two meanings. Calibrate (as a verb) means to correct the estimates of
an expert that shows systematic deviations (giving several estimates), that is provide unbiased
estimates. This corrected (calibrated) estimate will then be the input to the analysis.
Calibration (as an adjective) is used by e.g., Cooke [18] to describe the accuracy of one specific
estimate (independent of systematic or random errors). Cooke defines calibration as:
Calibration = to what extent the estimated probability agree with the observed relative frequency
Thus, calibration (as an adjective) is a characteristic of one estimate (the distance between the
expert’s best estimate and the true value).Unlike unbiasedness (which is applied having more than
one estimate), poor calibration is not necessarily due to biasedness (it might be unsystematic). In
this memo we use the term ‘calibration’ as a verb.

Over- and underestimation

In relation to biasedness, notions like over- and underestimation are often used. Overestimation
means that the experts in too many cases give an estimate that is higher than the true value.
Underestimation means that the expert too often gives an estimate that is too low compared to
the true value. If this is done systematically, there is a bias. However, if we calibrate the experts’
estimates, this will give very important information. The more systematic the deviations are, the
better will the calibrated value become.

Informativeness
Even though the experts do not show any systematic deviations from the correct values, this
does not mean that he always gives the correct values. He may even have random/unsystematic
deviations, which result in spread or variation of the estimates round the true values. Here we use
the notion informativeness to describe the experts’ unsystematic (unpredictable/random) deviation
from the correct values. This is also referred to as the degree of precision.
Informativeness = the degree of unsystematic variation/spread of the estimates around the true
values. (High informativeness = low variation).
An expert that has great variations in his estimates will, even though he sometimes strikes the
target, be perceived to have little informativeness (is not to “be trusted”), and should be given
small weight when his estimates are weighted with the other experts’ estimates.
One way to measure the informativeness is to use the sum of squares of the deviations between
the experts’ estimates and the true values.
NOTE! Some authors (e.g., [18]) use informativeness about the experts’ own evaluation of how
certain he is of his estimate, that is, the confidence interval he puts around his best estimate. This
we will denote “subjective informativeness”, see below.

Subjective informativeness
Subjective informativeness = the informativeness (degree of precision) of the estimate, as assessed
by the expert himself
This subjective informativeness is given by the confidence interval he puts around his best
estimate. For example in case of PERT distributed random quantities in the model, we ask first
the expert to assess the values of L, M , and H. Then for each of these, we ask the expert to assess
the informativeness regarding these values. For example for the parameter M , the expert gives

90
MEV (expected value for M ), MLV (low value for M ), and MHV (high value for M ). MHV – MLV
is an expression of how confident the expert is in his own estimate.
Note that it has been shown not to be a good correlation between an expert’s calibration
(“accuracy”) and his subjective informativeness (self-confidence). Thus, there is no guarantee
whatsoever that an expert who gives narrow confidence intervals gives better estimates (more
accurate) than one with wide confidence intervals.

Over- and underconfidence

Over- and underconfidence are characteristics of the expert’s confidence in his own judgements.
Overconfidence means that the expert has a too great confidence in his own estimates, by giving
too narrow confidence intervals around his best estimate. Underconfidence means that the expert
gives too wide confidence intervals around his best estimate.

Dependence
Dependence is often related to biasedness. If the expert gives biased estimates and this biasedness
shows special patterns, we talk about dependencies. For instance, the biasedness may increase
when the correct values increase. This represents a positive dependence (correlation) between the
correct values and the bias. (Dependence may be seen as a special type of biasedness.)

Resolution
Resolution = the ability to separate the probability of a specific event from the average probability
for the total set of events
An example of this is the ability to estimate the most likely duration of the given project with
it’s characteristics compared to a general project.

Consistency
Consistency = the assessment is independent of the method/approach that is used and also of when
the evaluation is carried out
Consistency means that the judgements are reproducible, that is, we get the same result inde-
pendent of time and method. For each expert this means that he is consistent if e.g. he provides
the same result/estimate, even though the question is phrased in different ways, and that he gives
the same answer if the assessment is repeated at a later stage of the elicitation. If the expert is
inconsistent, he should be given a lower weight than the other experts, when the overall estimate
is calculated.

Coherence
Coherence = conformity with the laws of the theory of probability
This may also be referred to as “logical consistency”.
For example, according to probability theory, Pr(A ∪ B) = Pr(A) + Pr(B) − Pr(A ∩ B), and
Pr(AC ) = 1 − Pr(A). If the expert does not provide answers that conform with these rules, he is
said to be incoherent. (If the experts are not coherent, the analyst should avoid using questions
that include probabilities.)

91
Reproducibility or Inter-expert reproducibility
Reproducibility = the extent to which different experts provide similar results to a given question
(when using the same method)
Lack of reproducibility should indicate that at least some of the experts do not give satisfying
estimates.
Reproducibility may be measured through the spread.

8.3 Check list for expert judgement exercises

In the following we present a checklist for expert judgement exercises where the most important
recommendations/warnings are given.
This may be used as a check-list for relatively simple expert judgements, where individual
weighting of the experts is not to be used. It is not recommended to carry out calibration and
weighting of the experts (in phase 3) during simple expert judgements.
To get a further explanation of each step in the procedure, and the use of calibration and
weighting of experts, we refer to the next chapters.

8.3.1 Checklist for phase I: The preparation phase

1. Describe the problem, what is to be estimated

2. Avoid too narrow delimitation of the problem

3. State the reason for the need for experts judgements in relation to the existing problem

4. Estimate the available budget for executing the expert judgement

5. Appoint a time for when the expert judgement must be executed

6. Indicate how much time each expert must set aside

7. Determine the number of experts to be used

8. Possible statistical data should be exhibited to the experts

9. Choose the experts with most knowledge/experience about the problem of interest

10. Choose the experts who are available and motivated

11. Avoid having too many experts with a great interest in the project (interest of a specific
result)

12. Avoid particular dominating persons of high status

13. The client should control the list of experts

14. Gather information about the experts’ background and experience

15. Decide whether individual estimating or a group process is to be used

16. Formulate an estimation method/ a group technique, respectively

17. Give an attractive format for the questionnaire

92
 18. Formulate clear and simple questions

19. Explain to the experts the difference between uncertainty (variability) in the quantity to
assess (e.g., activity duration), and the uncertainty in the parameter estimates

20. The questions must be logically correct and understandable

21. Use graphics to indicate uncertainty (high/low estimates)

22. Decide on suitable level of decomposition of the problem

23. Carry out a test of the questionnaire on colleagues

8.3.2 Checklist for phase II: The elicitation phase

1. Explain to the experts the problem and what is to be estimated

2. Explain to the experts how the expert judgements is to be executed

3. Explain how the results will be treated

4. Go through the forms the experts shall fill in

5. Emphasise the importance of the experts being sincere and honest when giving the estimates

6. Call on the experts to not be affected/dominated by the other experts or by the ana-
lyst/process leader

7. Explain to the experts some of the most important phenomenon which causes biased/incorrect
estimates

8. The experts must be treated in a fair and correct way

9. The analyst/process leader must remain neutral and not actively take part in the evaluations
so that the estimates are affected

10. The analyst/process leader must be present during the elicitation to show interest, control,
and to answer questions

11. Estimate the duration of the questioning. (By individual estimation, max. 1 hour before
having a break).

12. The analyst/process leader must see to that not one (or a few) expert dominates or affects
the others

13. If only one expert is used, the estimates must be well-founded

14. For each parameter of interest, be as specific as possible to describe the cost element, the
activity etc for which parameters are required

15. Ask the experts to first indicate their lowest and highest estimate. Thereafter, their best
estimate

93
 16. If the variable of interest is PERT distributed, it is recommended to ask the experts to
estimate the most likely value (M ), the P10 and the P90 percentiles1 .

17. To assess the subjective informativeness ask the expert to give low and high values for each
of the three estimates in a typical case (base case)

18. For subsequent quantities subjective informativeness is assessed by L, E and H where L

means less subjective informativeness (more uncertain than in the base case), E means equal
subjective informativeness, and H means higher subjective informativeness (more certain than
in the base case).

19. Equal weights of the experts should be used when there is no strong argument for weighting

20. Weighting is not recommended for simple expert judgements

21. Finally, ask the experts if there are any questions that are not clear, so that these may be
clarified before ending the elicitation

22. In a group processes (after ind. proc.) the experts should affirm for that the result is complete,
before finishing the process.

8.3.3 Checklist for phase III: The calculation phase

1. Calibration of the experts is not recommended for simple expert judgements

2. Unequal weighting is not recommended for simple expert judgements

3. Establish a common estimate based on the experts’ estimates

4. The analyst must document all steps in the process

5. All assumptions (made by both analyst and experts) should be gathered in a separate form

6. The reason why each expert is chosen, plus his strengths and knowledge of the current subject,
must be documented.

7. The “raw-estimates” made by the expert must remain available.

8. Any form for manipulation/weighting of the estimates must not be carried out without in-
cluding argumentation/documentation

9. All calculations must be reproducible

10. The experts’ names should be stated, but each single estimate does not need to be allocated
to a specific expert, if anonymity is considered to encourage objectivity

11. Present the results to the client; how results are achieved, and who the experts are.

12. Graphical presentation of the results may be advantageous, especially to show the uncertainty.

13. The results, including the handling of each single estimate are presented to the experts (may
be in writing)
1
The Px percentile is defined by Pr(X ≤ P x) = x%. For example Pr(X ≤ P 10) = 10%

94
 14. Evaluate whether the results seem reasonable and accurate.

15. Evaluate whether there is need for further work (more data)

16. File the case/project with all documentation.

17. Update possible databases/bases of knowledge

18. Make a “debriefing” of the expert judgement process (how did it work?)

19. Correct the items in the procedure for carrying out the expert judgement, when weaknesses
are experienced.

8.4 Calculation aspects

8.4.1 Criterion for performing calibration
Calibration is carried out if there are distinct indications of the expert systematically over/under-
estimating the correct value, such that valuable information is lost if he is not calibrated. (Unless he
is calibrated, he will be assigned a very low weight, representing a loss of information.). Calibration
requires the use of control questions (seed variables). Generally there should be strong evidence
for calibration, which requires several control questions. Now, introduce:
n = The number of control questions
xi = the correct value (control question no. i), known by the analyst
Yi = the expert’s estimate
Z = The number of Yi − xi (control question no. i) that are > 0
We will calibrate the expert when Z is either near to 0 or near to n. The following simple rules
are suggested:
(1) For n ≥ 5, a calibration is done when
√ √
Z < n/2 − n or Z > n/2 + n
Example: For n = 5 we calibrate if Z = 0, or Z = 5
For n = 10 we calibrate if Z = 0, 1, 9, 10
(2) For 2 ≤ n ≤ 4 and Z = 0 or Z = n, a calibration is done if, in addition, Yi /xi generally are
“large” (»1) or “small”∑(«1). That is:
(i) Z = 0 and 1/n ∑i (Yi /xi ) < 1/(6 − n), or
(ii) Z = n and 1/n i (Yi /xi ) > (6 − n)

Example 8.1 Example cost estimate

Assume n = 3 control questions xi , (i = 1, . . . , 3) for cost estimation of the activities A, B and C.
Assume the estimates and correct values shown in Table 8.2:

Table 8.2: Estimates and correct values from 3 activities, one expert
Activity xi (true, known value) Yi (estimate)
A 50 70
B 60 90
C 70 140

∑
All Yi in this case are larger than xi , that is, Z = 3, but: 1/n i (Yi /xi ) = 1/3 (7/5 + 9/6 +
14/7) = 1.63 <(6 - n) = 6 - 3 = 3. That is, here it is chosen not to calibrate. ♢

95
8.4.2 Method for performing calibration
In the following we describe the principle for calibration if the criterion for calibration is satisfied.
We assume that there is a linear relation between the true values (xi ’s) and the estimates (Yi ’s):

xi = β0 + β1 Yi + error term (8.1)

Note that here we write x as a function of Y and not the other way as we usually do. To estimate
β0 and β1 we apply standard LS methods, i.e.:
∑
(Yi − Ȳ )xi
β̂1 = ∑i ( )2 (8.2)
i Yi − Ȳ

β̂0 = x̄ − β̂1 Ȳ (8.3)

For a given value of Y , say y, we now estimate the parameter of interest by:

x̂ = β̂0 + β̂1 y (8.4)

Note that if Y is less than Yi,min = mini Yi it is recommended to use a regression line through the
origin, i.e.,
( )
β̂0
x̂ = + β̂1 y (8.5)
xi,min

Example 8.2 With the data given above we find (although calibration was not recommended):

β̂0 = 33.077 and β̂1 = 0.27

Assume now that we ask the expert to estimate a new value for duration of a new activity D (M =
Most likely value). He assigns the value y = 100. The estimate is now given by:

x̂ = β̂0 + β̂1 y = 33.077 + 0.27 · 100 ≈ 60 (8.6)

8.4.3 Weighting of experts

Various principles for weighting of experts exist. In the following we describe some of these.

Equal weighting
Give the experts a weight 1/m, where m = the number of experts. Equal weighting of the experts
is used when there is no basis for performing differentiation of weighting (e.g. control questions,
mutual weighting, knowledge profile.)

96
Based on control questions
Give the experts a weight related to the estimated random error (variance) obtained from evaluation
of the control questions. This expresses the experts’ “true informativeness”. Note that we now write
the relation between x and Y in the normal way:

Yi = α0 + α1 xi + error term (8.7)

where the objective is to estimate the variance of the error term. To find the variance of the error
term we estimate the regression line, let SS be the square sum of the residuals for expert k:
∑
SS k = (yi − αˆ0 − α̂1 xi )2 (8.8)
i

i.e., the differences between the observation (yi ’s) and the estimated line. An estimate for the
variance is given by:

Sk2 = SS k /(n − 2) (8.9)

where n is the number of control questions given to expert k. If we use e.g., MS Excel to estimate
the parameters we directly find S as the “Standard Error”. If we have m experts the weight of
expert k is then

S −2
wk = ∑ k −2 (8.10)
j Sj

where Sk is the estimated standard deviation of expert k.

Based on mutual evaluation

Give the experts a weight related to how they where evaluated by the other experts. All the experts
evaluate the other experts by giving them points (p) from 0-10 (10 is maximum).
The weight for each single expert is then the sum of scores which the expert achieves divided
by the total score given by all the experts. (Thus, the sum of the weights equals 1.0). The setup
is shown in Figure 8.3.
Weight for expert k is given by:
∑
j pj,k
wk = ∑ ∑ (8.11)
j i̸=j pj,i

Based on knowledge profile

Give the experts a weight based on their knowledge. For example, the number of years as project
manager, the experience with similar type of projects etc.

Based on arguments/documentation used in the discussion

Give the experts a weight related to their argumentation for their estimates. In the argumentation,
personal experiences and references to sources, among other things, may be included. (Such an
“expert judgement” of the experts performed by the analyst, must be substantiated and documented
if it is to be used.)

97
 Expert no. (who is evauated)
1 2 3 ... m
...

Expert no. (who evauates)

1 p1,2 p1,3 p1,m

2 p2,1

3 p3,1

...

...
m pm,1

Figure 8.3: Mutual evaluation of experts

8.4.4 Standard weighting model - Experts only

Assume we have m experts that have made individual statement, x̂j regarding the parameter of
interest. Further assume that weights wi are established for each expert according to procedures
described in Section 8.4.3. The combined estimate of the parameter to estimate (x) is now given
by:
∑
x̂ = wj x̂j (8.12)
j=1:m

8.4.5 Transforming percentiles

It is recommended to ask the expert to assign percentiles in the PERT distribution rather than
the extreme values L and H.Assume that the weighting model in equation (8.12) is applied for
all parameters, and the results are denoted P̂10 , M̂ and P̂90 . We now seek L̂and Ĥ. If we have
implemented the CDFPert() function we should then require:

CDFPert(P̂10 , L̂, M̂ , Ĥ) = 10% (8.13)

and

CDFPert(P̂90 , L̂, M̂ , Ĥ) = 90% (8.14)

And try to solve these two equations wrt L̂ and Ĥ.

Example 8.3 Assume that P̂10 = 6, M̂ = 10 and P̂90 = 14. By applying the pRisk.xls program
we may now generate one cell containing CDFPert(P̂10 , L̂ , M̂ , Ĥ), and one cell containing
CDFPert(P̂90 , L̂ , M̂ , Ĥ). Then we set a third cell equal to the sum of these two cells. By using
the solver requiring the sum to be one, and for example the cell with CDFPert(P̂10 , L̂ , M̂ , Ĥ)
to be equal to 0.1 in the “Constraints” editor, we find L̂ = 2.11 and Ĥ = 17.89. Observe that the
CDFPert(P̂90 , L̂ , M̂ , Ĥ) cell evaluates to 90%. ♢

98
8.4.6 Standard weighting model - Experts and data
In some situations we both have data and experts judgements. In principle we may use Bayesian
updating strategies to find the final estimates to apply. In the following we propose a slightly simpler
approach where we weight the expert judgement results with the estimates found by statistical
analysis of data.

Weight of experts
Equation (8.12) is used to find the combined estimate of the parameter of interest. One way to
weight this value with the data is to calculate the “variance” of the estimate in equation (8.12).
An unbiased estimator for the variance based on the variation of the expert statements is:

1 ∑
m
2
SVE = ∑m wj (x̂j − x̂)2 (8.15)
1− 2
j=1 wj j=1

If there are few experts this measure is not very reliable. A better approach may then to use the
“Subjective informativeness” assessed by the experts. Now, assume that each expert, k, has stated
an estimate, x̂k , but also low (x̂k,L ) and high (x̂k,H ) values for the estimate. Assume that the
low and high values corresponds to the P10 and P90 percentiles. If we assume that the underlying
uncertainty distribution of the expert is PERT distributed, we may apply the approach shown in
the previous example to find the L, M and H values of the uncertainty distribution, and hence the
variance by standard formulas. A more pragmatic approach would be to claim that the standard
deviation is proportional to the distance between the low and high value, and then try to assess
the proportional constant. Following the example we then assess the “self evaluated standard
deviation” by :

Ŝk = 0.37(x̂k,H − x̂k,L ) (8.16)

If Ŝk−2 is used as basis for the weighting of expert k, and the Ŝk−2 ’s are considered as true variances,
it can be shown that the variance of the weighted means equals the reciprocal of the sum of these
variances, and hence a reasonable estimate for the variance of the weighted estimate based on all
experts is found by:
2 1
SSE =∑ −2
(8.17)
j=1:m Ŝj

If there are few experts it is recommended to apply equation (8.17). If there are a medium number
of experts, say 3 to 5 one may calculate the variance by both equation (8.15) and equation (8.17)
and use the maximum of these two values. Generally we denote the estimate of the variance of the
expert judgement estimator by SE2 .

Weight of data
In some cases it is easy to find the variance of the estimator used when statistical data is available.
For example the standard deviation of the estimator for the mean value in the normal distribution
is found by the sample standard deviation divided by the square root of the number of observations.
In more complicated situations we may use the principle of bootstrapping to find the variance of the
estimator. The procedure is as follows: Let z1 , z2 , ... , zn be observations from the distribution for
which we are seeking estimates. Apply an estimation procedure to estimate the parameter vector.
Denote the result by θ̂. Now repeat, k = 1, 2,…

99
 1. Generate n pseudo random numbers from the actual probability distribution with parameter
vector θ̂.

2. Apply the estimator again to find a new estimator for the pseudo generated numbers, and let
the result be denoted θ̂k for the k th iteration.

The sample variance of the j th element of the θ̂k -sequence may now be used as an estimate for the
variance of the estimator based on data. Let SD 2 denote the variance based on the data.

Combining it
If we let x̂E and x̂D denote the estimates from combined experts and data respectively, and further
SE2 and SD 2 the corresponding estimated for variances, we find the final weighted estimate of experts

and data by:

SE−2 x̂E + SD
−2
x̂D
x̂ = −2 −2 (8.18)
SE + SD

8.5 Worked example

Assume that we have three experts. They are all calibrated with three control questions shown in
Table 8.3:

Table 8.3: True values and estimates from three experts

Activity (Control) xi (true, known value) Yi (Expert 1) Yi (Expert 2) Yi (Expert 3)
A 50 70 40 60
B 60 90 80 50
C 70 140 60 90

All experts are asked to give the most likely value (M ) for a new activity D, with corresponding
uncertainties in the estimate shown in Table 8.4:

Table 8.4: New values and self assessed uncertainty by the experts
Activity Y (M in PERT) Low Y (P10 ) High Y (P90 )
Expert 1 100 75 150
Expert 2 80 60 95
Expert 3 60 55 70

Observation from previous project shows the following costs: 53.2, 60.4, 52.4, 66.8, 56.2, 72.2,
71.7 and 73.2.
Find a combined estimate from the experts, and an estimate from the data. Then combine
these two estimates into a final estimate.
We decide to calibrate the expert independent of the “calibration test”. Regression analysis
from MS Excel is shown in Table 8.5.
Predictions of a new value for a new activity, D, by the experts are shown in Table 8.5. Low
and high values, specified by the expert are given in Table 8.6.

100
 Parameter estimate Expert 1 Expert 2 Expert 3
β̂0 = x̄∑− β̂1 Ȳ 33.08 45.00 36.92
(Yi −Ȳ )xi
β̂1 = ∑
i
2 0.27 0.25 0.35
i (Yi −Ȳ )

Table 8.5: New values predicted by the experts with calibrated values
Quantity Expert 1 Expert 2 Expert 3
Y = Prediction by expert 100 80 60
x̂ = β̂0 + β̂1 y 60.00 65.00 57.69

Table 8.6: Low and high values specified by the expert and derived results
Quantity Expert 1 Expert 2 Expert 3
Low Y by expert =P10 75 60 55
High Y by expert =P90 150 95 70
Calibrated low, (P10 = x̂k,L ) 53.27 60.00 55.96
Calibrated high, (P90 = 73.46 68.75 61.15
x̂k,H )
Ŝk = 0.37(x̂k,H − x̂k,L ) 7.471 3.238 1.921
−2
Ŝk 0.018 0.095 0.271
Ŝ −2
wk = ∑m k −2 0.047 0.248 0.705
j=1 Ŝj

101
 Weighted calibrated prediction based on self evaluated weights:

∑
m
x̂ = wj x̂j = 59.61
j=1

Corresponding variance of this weighted average:

2 1
SSE =∑ −2
= 2.60
j=1:m Ŝj

Problem 8.1 Assume that we have three experts. They are all calibrated with three control
questions shown in Table 8.3 above. All experts are asked to give a high value (H) for a new
activity D, with corresponding uncertainties in the estimate as shown in Table 8.7. Observation
from previous project shows the following costs: 53.2, 60.4, 52.4, 66.8, 56.2, 72.2, 71.7 and 73.2 (all
in thousands).

a. Check whether there are reasons to calibrate the experts

b. In the remaining of this you should calibrate the experts independent of the result from (a).
Calibrate the experts.

c. Find an estimate of the variance of each expert (i.e., around the regression line).

d. Use the variance of each expert as a basis for weights to each expert. Then find a weighted
estimate for the high value.

e. Estimate the parameters in the distribution by assuming a PERT distribution based on the
data given.

f. Perform ”Bootstrapping” to find variances for the estimates for L, M and H.

g. Combine the estimates from data and experts to find a final estimate for the high value (H).

Table 8.7: New values and self assessed uncertainty by the experts for the high value in the
distribution
Activity Y (H in PERT) Low Y (P10 ) High Y (P90 )
Expert 1 140 105 210
Expert 2 96 72 114
Expert 3 72 66 84

102
Chapter 9

Optimization of turnaround activities

9.1 Introduction
Turnarounds are scheduled events wherein an entire process unit of an industrial plant such as an
oil & gas production platform is taken shut-down for an extended period for modification and/or
renewal. Sometimes the term ‘revision stop’ is used to denote a turnaround.
Turnarounds are expensive - both in terms of lost production while the process unit is offline
and in terms of direct costs for the labour, tools, heavy equipment and materials used to execute
the project. Turnarounds have unique project management challenges due to the large number of
activities to execute, logistics for labour and spare parts, and safety issues due to many activities
in parallel.
In this chapter we mainly discuss how maintenance activities either could be conducted individ-
ually, or if they should be in cooperated in a turnaround to save cost since the production already
is down.

9.2 Single component considerations

In this section we consider the situation where only one maintenance activity is considered. The
question to be answered is whether a maintenance activity should be carried out at its “optimal
execution time”, or if it should be carried out as a part of a turnaround (revision stop).

9.2.1 Preventive maintenance

Due to wear and tear of components their failure probability often increases with component age.
In order to reduce the likelihood of failure components are often preventively maintained. For
example a switch machine in a railway system is in Norway tentatively overhauled every six year.
Such an overhaul activity includes replacement of components exposed to wear, lubrication, cleaning
etc. In maintenance theory it is made a distinction between age or calendar based activities, and
condition based activities. Historically, preventive maintenance was carried out based on the age
of a component. The only trigger for maintenance was age, running hour, mileage run for a
car etc. Since the correlation between age and failure is rather vague, one seeks to find more
precise indicators that could be correlated to the actual failures of the component. This has led
to increased used of condition monitoring techniques where the objective is to utilize the condition
of a component as an indicator whether a failure will occur in the near future. An example of a
condition based activity is replacement of a rail when cracks of critical length are found by ultrasonic

103
inspection. The main rationale for preventive maintenance is that it is cheaper to prevent a failure
from occurring by a preventive maintenance task than it will be if a failure occurs. A range of
optimization models have been derived in order to determine the appropriate level of preventive
maintenance. The reason why maintenance is addressed within project risk management is that
turnarounds essential is a huge package of preventive maintenance activities organized as a project.

9.2.2 Single activity - Preventive maintenance not included in the turnaround

Consider a component where a preventive maintenance (PM) action is conducted at predetermined
intervals due to an increasing hazard rate1 , z(t). In order to find an optimal interval for a main-
tenance action we may establish the average cost per time unit as a function of the maintenance
interval, say τ :

C(τ ) = CPM /τ + λE (τ ) [CCM + CEP + CES ] (9.1)

where CPM is the cost of a preventive maintenance action (to prevent failures), CCM is the cost
of a corrective maintenance (CM) action (given that a failure did occur), λE (τ ) is the effective
failure rate, i.e., the expected number of failures per time unit when the component is preventively
maintained every τ time unit, CEP is the expected production losses upon a component failure,
and finally CES is the expected safety cost upon a component failure, including material damages
and environmental losses.
In the following we let CU = CCM + CEP + CES denote the expected unplanned cost upon a
failure. The effective failure rate depends on the life time distribution of the component. The
Weibull distribution is a widely used distribution for aging components. In the case of Weibull
distributed life times we may find approximation formulas for the effective failure rate. If we know
the mean time to failure , MTTF (without maintenance), and the ageing parameter (α) of the
lifetime distribution of the component, the effective failure rate may be approximated by:
( )
Γ(1 + 1/α) α α−1
λE (τ ) = τ (9.2)
MTTF

where Γ(·) is the gamma function. The approximation is good when the maintenance interval is
small compared to the MTTF. If the maintenance interval is approaching the MTTF value, the
approximation in equation (9.2) is not very accurate, and we might use the following improved
approximation:
( )
Γ(1 + 1/α) α α−1 [ ]
λE (τ ) = τ 1 − 0.1α(τ /MTTF)2 + (0.09α − 0.2)τ /MTTF (9.3)
MTTF

In MS Excel the gamma function could be found by Gamma(x). In the following we will always
assume that the approximation in equation (9.2) is sufficient for our purpose. By setting the
derivative of C(τ ) in equation (9.1) equal to zero, we find the optimal interval to be:
( )1/α
MTTF CPM
τ∗ = (9.4)
Γ(1 + 1/α) CU (α − 1)
1 fX (x)
The hazard rate is given by hX (x) = 1−F X (x)
, where fT (t) is the probability density function of the lifetime
T of the unit, and FT (t) is the cumulative distribution function. The hazard rate is also to be understood as the
conditional probability of a failure in (t, t + ∆t) given that the unit has survived up to time t. In some textbooks
the hazard rate is also dented the failure rate function.

104
9.2.3 Single activity - Consideration for inclusion in turnaround - Static con-
sideration
A turnaround is usually conducted every year, every two years, or every three years. There is an
aim to conduct the turnaround in a period where working conditions are good (summer period),
and where also production losses are as low as possible which historically also has been in the
summer period since energy prices usually reduces when the demand goes down. It is therefore a
common practice to conduct the turnarounds in the summer. The interval between turnarounds is
denoted τTA .
The formula for the average cost C(τ ) in equation (9.1) needs to be modified if maintenance is
considered. Let CTA be the preventive maintenance cost if the maintenance of the component is
included as part of the turnaround. Usually we will have that CTA <CPM . We will now investigate
whether the activity should be included in the turnaround. The criterion for inclusion in the
turnaround is that the average cost is reduced. An example is given to demonstrate the steps. The
relevant parameters are given in Table 9.1 where the time unit is months, and cost unit is thousand
NOKs.

Table 9.1: Parameters for the decision regarding inclusion in turnaround, τ ∗ >τT A
Parameter Value
α 2.5
MTTF 48
CPM 5
CTA 4
CU 50
τTA 12

We first apply equation (9.4) and get τ ∗ ≈ 18.3 months. The total average cost in the situation
where the PM activity is not included in the turnaround is thus ≈ 0.455 per month by inserting
τ = τ ∗ in equation (9.1). If the PM is included in the turnaround, the average cost is found by
inserting τ = τTA and CPM = CTA in equation (9.1). This gives a monthly cost of ≈ 0.43 <0.455.
It is thus in average cheaper to include the PM into the turnaround. We observe that τTA <τ ∗
<2τTA . We might therefore also consider to insert τ = 2τTA and CPM = CTA in equation (9.1).
This gives a monthly cost of ≈ 0.44 which is still better than the situation without including the
PM in the turnaround, but less favourable comparing by including the PM in every turnaround.
The optimal strategy is thus found to be to include the PM activity in every turnaround with the
given assumptions.

Problem 9.1 We will consider a preventive maintenance action (PM) for inclusion in a turnaround.
For detailed description of the situation, see the introductory example above and use the data in
Table 9.1. Find the optimum interval for the maintenance interval if the PM is conducted outside
the turnaround by using the Excel Solver. ♢

Problem 9.2 Since the maintenance interval is rather close the MTTF the suggested approxima-
tion formula for the effective failure rate is not very accurate. Find the optimum interval if the
improved approximation in equation 9.3 is used. ♢

Problem 9.3 In the example above it was shown that the long run average cost will be lower both
in the case we include the PM in every turnaround, and in the case where we include it in every

105
second turnaround. Since the individual optimal interval is close to 18 months, it seems reasonable
to include the PM in every third turnaround. This means that average PM cost will be reduced.
Find the total cost for this strategy. ♢

9.2.4 Single activity – dynamic consideration

In Section 9.2.3 we considered a situation where a static regime was established with respect to
finding the appropriate maintenance interval of the component. By “static” we here understand
that the same interval applies for the component during the entire “life” of the installation. This
is appropriate if we are able to synchronise all maintenance activity from the start-up of the
production, and when nothing “changes”, in terms of major modifications, increase in reliability
etc. In practice, however, things will change all the time, and it is usually required to be able to
change strategies “on the fly”. This means that we need a dynamic approach. We now consider
a situation where we are planning the next turnaround. For the time, we let t = 0 represent the
time we are going to conduct the next turnaround. Let x be the “age” of the component we are
considering. This means that the component was preventively maintained x time units ago. To
proceed we continue the example provided in Section 9.2.3, but we now assume that x = 8 months.
The “due” point of time of PM if we did not include the activity would then be approximately in
18-8 = 10 months. In this case, it is therefore reasonable to postpone the maintenance until the
next PM. If we include the PM in the coming turnaround the total cost in the next period (i.e.,
from now until the PM next year) is:

CFirst TA = CTA + λE (τTA ) · CU · τTA (9.5)

If we choose to skip the PM activity for the coming turnaround, and wait till the next we get
the following total cost:

CSecond TA = λE (τTA + x) · CU · (τTA + x) − λE (x) · CU · x (9.6)

where λE (τTA + x) · CU · (τTA + x) is the expected “unplanned” cost from the last PM until the
next turnaround, and λE (x) · CU · x is the expected “unplanned” cost from the last PM up to now
(i.e., what already has been “paid”). With the example data we get:

CFirst TA = CTA + λE (τTA ) · CU · τTA ≈ 5.16

and

CSecond TA = λE (τTA + x) · CU · (τTA + x) − λE (x) · CU · x ≈ 3.73

Hence, the PM activity should not be included in the coming turnaround but in the turnaround
coming next year. Note that we might consider to execute the PM in between the two coming
turnarounds, but the example in Section 9.2.3 indicates that the gain by synchronizing with a
turnaround is always the best alternative.

9.2.5 Single activity - “Arctic maintenance”

In some situations it is not possible to conduct a corrective maintenance action in the period
between the major shutdowns (turnarounds) of the plant. For example in the arctic it is considered
almost impossible to conduct a major repair in wintertime. If a failure occurs in the period between
two turnarounds, there will thus be a production loss from that failure until the next turnaround.

106
In the modelling we still assume Weibull distributed lifetimes. For the Weibull distribution2 we
have:

α
fT (t) = αβ(βt)α−1 e−(βt) (9.7)

α
R (t) = Pr (T > t) = 1 − FT (t) = e−(βt) (9.8)

Γ(1/α + 1)
β= (9.9)
MTTF
In the modelling we now we still assume that the cost of an unplanned failure is CU . But in addition
to the immediate cost of a failure we assume that for the remaining time until the next opportunity
for maintenance there will be a cost of CW per time unit until the component could be repaired.
If we include the PM in the coming turnaround the total cost in the next period (i.e., from now
until the PM next year) is:
∫ τTA
CFirst TA = CTA + [1 − R(τTA )] CU + fT (t) (τTA − t) CW dt (9.10)
t=0

If we choose to skip the PM activity for the coming turnaround, and wait till the next we get the
following total cost in between the two turnarounds:

∫ τTA
CSecond TA = [1 − R(τTA + x)/R(x)] CU + fT (t + x) (τTA − t) CW dt/R(x) (9.11)
t=0

We proceed by the same data as in the previous examples but we now also let CW = 10. We
then get:
∫ τTA
CFirst TA = CTA + [1 − R(τTA )] CU + fT (t) (τTA − t) CW dt ≈ 5.93
t=0

and
∫ τTA
CSecond TA = [1 − R(τTA + x)/R(x)] CU + fT (t + x) (τTA − t) CW dt/R(x)
t=0
≈ 7.06

Note that we here have assumed that even if we include the PM activity in the first turnaround,
it will also be included in the second turnaround. In a more general setting, it might be that if
we include now, we might skip the activity in the next turnaround. The analysis then becomes
more complicated because to assess the costs that follow one needs to take into account different
strategies after the second turnaround.
2
Note that various parametrization exist for the Weibull distribution, and also note that the the symbol for the
scale parameter often is λ rather than β as used here.

107
9.3 Single component - Impact on turnaround duration
Up to now we have considered that the cost of including the PM activity in the turnaround could be
described by a single number, CTA . Since the inclusion of any activity in a turnaround will influence
the duration and complexity of the turnaround, it might be an oversimplification to consider a fixed
cost incurred by the PM activity we are considering. At the end, we might still be able to calculate
an expected cost for CTA , but we will now investigate the situation in some detail.
Assume that there are several activities that are to be included in the turnaround, and that the
PM activity we are considering is the last activity to include. For the time being, we assume that
the duration of the turnaround is fixed, say DTA . This is to say, that the planned duration is DTA .
The actual duration is a random quantity which we denote TTA . Now, assume that we have an
oversight over the remaining activities to include. The question is whether we should include the
PM activity or not. Let ATA,0 denote the set of activities and their relations already considered
to include in the turnaround. Further let ATA,1 be the activities to consider if we include the PM
activity we are considering.

9.3.1 Consideration for inclusion in turnaround, τ ∗ >τTA – No cancel possibili-

ties
We now consider the same situation as in Section 9.2.3. The only difference is that CTA is not
known. We will assess the expected value of CTA . To do this, we both consider the extra cost of
conducting the PM activity (spare parts, man hour costs) etc. In addition we need to consider the
influence wrt increasing the probability that we are not able to complete the turnaround in due
time. Let CTA,F demote the fixed cost of the PM activity in terms of spare parts and man hour
costs. Next let CTA,R denote the random cost related to the possibility of delaying the turnaround.
The expected value of CTA,R may be found as the difference in expected production losses with and
without the PM activity included. If the production loss per day is CPL we have:
∫ ∞
E(CT A,R ) = CPL [fTI (t) − fTO (t)] (t − DTA ) dt (9.12)
t=DTA

where fTI (t) and fTO (t) are the probability density functions for the project duration with and
without the PM activity included respectively (I = Included, O = Outside the turnaround). DTA is
the due date for the turnaround, where there will be no production loss if project duration T is
less than DTA , and the production loss is T -DTA if T is greater than DTA . The situation is now
identical to the situation in Section 9.2.3 if we set:

CTA = CTA,F + E(CTA,R ) (9.13)

Thus the total cost of the PM activity execution within the turnaround is the fixed cost plus
the random cost, E(CTA,R ), caused by the increased risk of exceeding the due date. Note that a
first approximation is presented here. In reality the proses of select activities for inclusion is more
challenging than considering one and one activity individually. Some kind of dynamic programming
to select the most appropriate PM activities out of a huge list is required.

9.3.2 Consideration for inclusion in turnaround, τ ∗ >τTA – Cancel possibilities

In Section 9.3.1 we considered a situation where the question was to determine whether to include
the PM activity in the turnaround or not. If production losses are huge if the turnaround is delayed,
we might in some situations choose another strategy where it is possible to cancel the PM activity

108
if the turnaround for some reasons is delayed. Assume there is a critical milestone upfront of the
execution of the PM activity. If a critical time is passed, one need to consider to skip the PM, and
rather execute it at its optional time (τ ∗ ). Let Let pD denote the probability that the turnaround
is delayed and one needs to consider to skip the PM activity as part of the turnaround. If the PM
activity is skipped, it is assumed to be conducted at its optimal time, i.e., at τ ∗ -τTA time units
after the turnaround. The decision diagram is shown in Figure 9.1.

The PM is planned outside the TA, execute at τ ∗PM outside

TA

Skip PM, ->τ ∗ PM planned in

TA, but cancelled
TA delayed
DN1 DN2

PM in TA
& problems
Keep inside TA
CN
The PM is planned 1
inside the TA, (τTA ) PM in TA without
No problems any problems

Figure 9.1: Decision tree, artic maintenance

The following quantities are now introduced:

CTA,P : The preparation cost of executing the PM activity as part of the turnaround. This cost is
to be paid independent of if the activity is executed as part of the turnaround or not.
CTA,E : The execution cost of the PM activity if executed during the turnaround.
CPM : Cost of executing the PM outside the turnaround, i.e., if executed at τ ∗
pD : the probability that the turnaround is delayed and one needs to consider to skip the PM
activity as part of the turnaround
fTN (t): The probability density function of the duration of the turnaround if no delays at the
critical milestone
fTI (t): The probability density function of the duration of the turnaround if delays at the critical
milestone and the PM is kept within the turnaround.
fTO (t): The probability density function of the duration of the turnaround if delays at the critical
milestone and the PM is taken out of the turnaround.
C ∗ : Average cost per unit time if the PM is executed at its optimal time, i.e., at τ ∗ .
The remaining parameters required have been defined in previous sections.
In the calculation there are some challenges because the various end consequences in the decision
three in Figure 9.1 has different time horizons. If the PM is executed within the turnaround, the
next due time for PM is earlier than if the PM is executed at its optimal time, i.e., at τ ∗ . In order
to have the same time horizon we assume that if the PM is executed during the coming turnaround
it will also be maintained at the next turnaround. If the PM is not executed as part of the PM,
but at its optimal interval at τ ∗ , the cost for the remaining time until the second turnaround is C ∗
per time unit. For the calculations we now assume that current time, t0 , correspond to the time

109
when the coming turnaround is executed. The duration of the turnaround is so short that we may
ignore this duration in comparison to τTA . Further we assume that the previous execution of the
PM took place at time t0 − τTA , i.e., at the previous turnaround. The following cost equations may
now be found for each end consequence, cf Figure 9.1:

PM outside TA

C1 = CPM + CU [λE (τ ∗ )τ ∗ − λE (τTA )τTA ] + C ∗ · (2τTA − τ ∗ ) (9.14)

PM planned in TA, but cancelled

C2 = CPM + CTA,P + CU [λE (τ ∗ )τ ∗ − λE (τTA )τTA ] + C ∗ · (2τTA − τ ∗ )

∫ ∞
+CPL fTO (t)(t − DTA )dt (9.15)
t=DTA

PM in TA & problems
∫ ∞
C3 = CTA,E + CTA,P + CU λE (τTA )τTA + CPL fTI (t)(t − DTA )dt (9.16)
t=DTA

PM in TA without any problems

∫ ∞
C4 = CTA,E + CTA,P + CU λE (τTA )τTA + CPL fTN (t)(t − DTA )dt (9.17)
t=DTA

With these cost figures the optimal decision is found by processing the decision tree in Figure 9.1.

9.3.3 “Arctic maintenance”

The situation to consider here is similar to the previous section, but we are now not allowed to
execute the PM between two turnarounds. Also, if a failure occurs, we need to wait with the
repair until the next opportunity, which here is set to the next turnaround. The situation is now
illustrated in in Figure 9.2.

9.4 Changing the frequency of the turnaround

Up to now the interval between turnarounds, τTA , has been considered fixed. In this Chapter we
will establish models for extending the intervals. The following aspects will be taken into account:

• Currently there exist tasks that from a safety point of view need to be conducted at intervals
of length τTA,0 , where τTA,0 is the . In order to extend the turnaround interval it is therefore
necessary to improve the reliability of these components, or propose more extensively condi-
tion monitoring in the period between turnarounds. In the following we will only consider
the possibilities to increase component reliability.

• Currently there exist tasks where the “optimum” interval is close to τTA,0 , and for these
components we will also consider to increase the reliability by conducting upgrade projects
prolonging the MTTF.

110
 The PM is planned for the second turnaround PM in second
TA

Skip PM, ->2τTAPM planned in

TA, but cancelled
TA delayed
DN1 DN2

PM in TA
& problems
Keep inside TA
CN1
The PM is included in
the coming turnaround PM in TA without
No problems any problems

Figure 9.2: Decision tree, Arctic maintenance

• τTA,0 = 2 years in the example, and the question to be raised is whether it pays off to extend
the interval, i.e., τTA = 3.

9.4.1 Treating only safety issues

One of the limitation with respect to extending the period between turnarounds is safety critical
maintenance activities that need to be carried out at the current turnaround interval, τTA,0 . We
now introduce the following quantities: {SC}: List of safety critical activity types where the current
maximum interval is close to τTA,0.
{NC}: List of activity types that are not safety critical.
ni : Number of identical components representing activity type i.
CUG,i : Cost of upgrading component i.
CPM,i : Cost of executing preventive maintaining activity ioutside the turnaround. This might be
an option if it is too costly to upgrade.
CTA,i : Cost of maintaining a component of maintenance activity type i as an integral part of the
turnaround.
CTA,B : Total cost of executing the turnaround treating only activities in the set {NC}, i.e., exclusive
activities in {SC}.
MTTFi : Mean time to failure for components with activity type i ∈ {NC}.
αi : Ageing parameter for components with activity type i ∈ {NC}.
CU,i : Unavailability cost upon a failure for components with activity type i ∈ {NC}.
λE,i = λE,i (τTA ; αi , MTTFi ): effective failure rate of components with activity type i ∈ {NC}.
In addition we have for the example:CTA,B = 50 (thousand NOKs).
We start by treating the average cost per year as a function of τTA assuming that the upgrading
of safety critical components takes place:
∑ ∑
C(τTA ) = CTA,B /τTA + ni CTA,i /τTA + ni CU,i λE,i (τTA ) (9.18)
i∈ {SC} i∈ {NC}

111
 Table 9.2: Reliability and cost data treating safety issues only
i ni CUG,i CPM,i CTA,i MTTFi αι CU,i i ∈ {SC}
1 4 8 4 1 Yes
2 3 2 1 0.5 Yes
3 4 3 2 1 Yes
4 4 2 5 4 20 No
5 8 2 6 3 30 No
6 6 3 6 2.5 25 No

where we have not taken discounting into account. By calculating C(τTA ) for τTA = 2 and
τTA = 3 in equation (9.18) we may calculate the yearly gain or loss by changing the interval of the
turnaround. If there is a gain, we also need to consider the (yearly) cost of the upgrading project:
∑
CUG = ni CUG,i /T (9.19)
i∈{SC}

where T is the number of years to consider, e.g., T = 10 if the installation will be disposed after
10 years. CU G should not exceed the gain by increasing τTA from 2 to 3 years.

Problem 9.4 Consider the example data above, and calculate the yearly cost for τTA = 2 and
τTA = 3 in order to determine if the turnaround interval could be increased. ♢
It is a simplification to ignore the discounting of the cost elements in this situation. We will now
discuss how to discount the variable cost related to failures. λE (τTA ) represents the yearly expected
number of failures if the time unit is year as we assume in this example. It is then reasonable to
calculate the expected number of failures each year j, j = 1,.., τTA and discount the cost to present
values. We assume that all costs are incurred at the end of the year, although it could also been
argued that it is more reasonable to assume that a failure occur in the middle of a year. We now
introduce ΛE,i as a measure of “discounted number of failures” in a period between two turnarounds
where the turnaround interval is τTA :
∑
τTA
ΛE,i (τTA , r) = (jλE,i (j) − (j − 1)λE,i (j − 1)) (1 + r)−j (9.20)
j=1

where r is the interest rate. Further we let n(T ,τTA ) be the number of turnarounds for an installation
that is disposed in year T (no turnaround the last year) when the turnaround interval is τTA .
Table 9.3 shows n(T, τTA ) as a function of τTA and T .

Table 9.3: n = n(T, τTA ) for various combinations of τTA and T

τTA \T 8 10 12 15 20
1 8 10 12 15 20
2 4 5 6 8 10
3 3 4 4 5 7
4 2 3 3 4 5

112
 We always assume that the turnaround is executed at time t = 0, then after time τTA , 2τTA
and so on. We still assume that the upgrade of all safety critical components takes place. The
turnaround related cost up to time of disposal is given by:

C A (τTA ) =
∑n(T,τTA ) ( ∑ ∑ )
−(j−1)τTA
(1 + r) CTA,B + ni CTA,i + ni CU,i ΛE,i (τTA , r) +
j=1 i∈{SC} i∈{NC}
∑
∆CFLP + ni CUG,i (9.21)
i∈{SC}

where ∆CFLP is a correction term to account for failures the last period, i.e., the difference between
ΛE,i (τTA ,r) calculated after the last turnaround, and the actual discounted number of failures
multiplied with the failure cost and summed over all activities that are not safety critical. There
might be a difference because either we skip a turnaround the last year and hence there is an extra
failure cost for that year not accounted for, or because we accumulate failure costs in the calculation
formula for the period after T . If T is large, i.e., in the order of magnitude 10 years, we may ignore
∆CFLP due to discounting effects.
The accumulated turnaround related cost for the remaining life of the installation may now be
compared by applying equation (9.21) for τTA = 2 and τTA = 3 respectively, where we for τTA = 2
do not include upgrade costs, CUG,i .
If upgrade cost, CUG,i , is high for some components we may also consider to exclude the corre-
sponding activities as part of the turnaround. This means to remove the corresponding terms for
ni CTA,i and ni CUG,i in equation (refeq:turnaroundLCC1), and similarly add the cost of preventive
∑n(T,τTA,0 )
maintenance: j=1 (1 + r)−(j−1)τTA,0 ni CPM,i to the total cost for τTA equal the extended
turnaround interval (i.e., τTA = 3) and τ TA,0 is the original turnaround interval.

Problem 9.5 Consider the example data above, and calculate the total accumulated cost for the
remaining period up to year T for τTA = 2 and τTA = 3 in order to determine if the turnaround
interval could be increased from a life cycle cost perspective. ♢
Note that the cost terms CUG,i might contain some implicit interrelations. Since the upgrade is
part of the next turnaround, it is reasonable that adding more and more upgrade projects totally
increases the cost more than the individual activities requires due to the increase of complexity,
and impact on the shutdown duration. This is not pursued further in this presentation.

9.4.2 Production related issues

For production related activities we also need to consider whether an upgrade of the components
could have an impact on the reliability. We here assume that if we upgrade a component we could
increase the corresponding MTTF. In the following we assume that the upgrade cost per production
related component is CUG,i = 2, which will increase MTTF with 25%. In the optimization we
therefore need to consider the following options:

1. Keep the activity in the turnaround, do not upgrade

2. Keep the activity in the turnaround, upgrade in order to prolong the MTTF
3. Take the activity out of the turnaround, and pay a higher cost CPM,i but with the flexibility
this gives to apply the optimal individual maintenance interval. We assume that CPM,i =
1.5CTA,i .

113
 If there are no interrelations between upgrading cost of the production related activities we may
perform individual analysis for each of the activity types. If there are interrelations the situation
becomes much more complicated. A dynamic programming approach might be required, or at least
we need to consider the activities pair wise.
In the following we assume that we may treat each activity individually.

Keep the activity in the turnaround, do not upgrade

In this situation we apply equation (9.21) as it is written. By extending the turnaround inter-
val there will be fewer turnarounds to be executed, but the term ΛE,i (τTA , r) will become more
significant since the effective failure rate increases at the end of the period between turnarounds.

Keep the activity in the turnaround, upgrade in order to prolong the MTTF
We apply equation (9.21) but need to add the cost of upgrading, i.e.,ni CUG,i . Since MTTF is
prologned the term ΛE,i (τTA , r) will also be reduced.

Exclude the activity from the turnaround

We apply equation (9.21) but need to reduce the term CTA,B . As a first approximation we subtract
the cost of executing the preventive maintenance activity as part of the turnaround, i.e., CTA,B =
CTA,B − ni CTA,i . We also need to remove the corresponding failure cost terms in equation (9.21).
But since we have to execute the PM activity i outside the turnaround we now add:

n(T,τi∗ )
∑ ∗
(1 + r)−(j−1)τi (ni CPM,i + ni CU,i ΛE,i (τi∗ , r)) (9.22)
j=1

Note that τi∗ is not an integer so we may need to rewrite the n() and Λ() functions.

Problem 9.6 Assume that we increaseτTA to 3 years. For each i ∈ {NC}find out whether it pays
off to pay the upgrading cost. ♢

9.5 Risk identification for critical activities

In this section we will consider critical activities in the turnaround. We will only treat those
activities that relate to an upgrade of the flare which is a main part of the turnaround. The
activities are listed in Table 9.4:
The relation between the activities is as follows:

• 0 – Shutdown is conducted first, then follows in parallel:

– A - Flare gas meters/B - Change flare tip (following each other)

– C - LP Flare drum
– D - NF HP Flare drum
– E - HP Flare drum

• Activity: F - Install new Flare tip system, may start when all the other activities are com-
pleted.

114
 Table 9.4: Activities to include and durations (normal conditions)
ID Activity L M H
0 Shutdown 1 2 3
A Flare gas meters 1 2 3
B Change flare tip 1 2 3
C LP Flare drum 4 5 6
D NF HP Flare drum 3 4 5
E HP Flare drum 3 4 5
F Install new Flare tip system 5 6 8

In Table 9.5 we have listed some threats related to activity: B - Change flare tip. We will consider
only those activities where a schedule impact is explicitly listed. The turnaround is scheduled for
14 days. It cost 10 million NOKs each day the turnaround is delayed. Costs in Table 9.5 are given
in million NOKs.

Problem 9.7 We will consider the impact of including the PM into the turnaround. So far we
have included the following activities (PERT distributed durations) as given in Table 9.6
First all A-activities are conducted in serial, then all the B-activities also follows in serial, but
in parallel with all the C-activities. The C-activities also follows each other in serial. The duration
of the shutdown is 14 days. After 14 days there is a penalty of 50 (thousand NOKs) per day.
Assume the fixed cost of the PM activity (i.e., the cost contribution not related to the duration)
is 2 (thousand NOKs). Find the total expected cost of including the PM activity after activity C3
(there is some expected slack in the “C-branch” of the model). Assume the duration of the PM is
∼ PERT(1.5,2,4)-distributed. ♢

Problem 9.8 Conduct a cost benefit analysis of the measure where one wants a more powerful
helicopter taking the risk elements of Table 9.5 into account. ♢

115
 Table 9.5: Threats related to activity: B - Change flare tip
Threat Probability Schedule impact Comment/Measure Cost
element
Technical 0.05 1 Technician on type (TLM) 0.05
problem be available for repair.
with heli-
copter or
equipment
Failure in 0.01 Accepted and tested strap
clearance of arrangement for the flare tip,
lifting specified by OM.
Wind force 0.08 1 per windy day Given wind force over 30 1
over 30 knots knots (critical limit) the day
of execution, the probabil-
ity of the wind problem re-
peats the subsequent days is
50% for each day. Proposed
measure: More powerful he-
licopter that may operate in
stronger wind. To model the
impact, we assume that all
the probabilities are reduced
by 30%
Failure in ra- 0.02 Stop all operations until
dio commu- the problem is solved/re-
nication established.
Loosing 0.01 Insist take off and delivering
object fly area. Cancel the take-off and
around rotor postpone operation until the
wind area is safe. This has to be a
continuous process.
Failure with 0.03 Use of certificated and
lifting strap approved lifting equipment.
Prohibited zone for flying
with hanging load has to be
defined.
Failure with 0.02 All equipment are to be func-
lifting hook tion test before operation.
Prohibited zone for flying
with hanging load has to be
defined.
Unexpected 0.05 Use of certificated and ap-
rotation on proved straps and lifting
load equipment. In case the pilot
can’t stop rotation by chang-
ing of the speed, the load has
to be return back to helideck
and defined again.
Failure with 0.01 Personnel shall not stay un-
116
motor or der the loading zone where
loss of motor the load is lowering.
craft
Table 9.6: Activity durations
Activity L M H
A1 0.5 1 2
A2 0.2 1.5 3
A3 2 2.5 4
B1 1 2 4
B2 1 3 5
B3 1 2 3
C1 0.5 1 2
C2 1 3 5
C3 0.2 0.5 4

117
Chapter 10

Portfolio management

10.1 Introduction
A project portfolio is a group of projects to be carried out under the sponsorship of a particular
organization. The portfolio may be seen from various stakeholders. For example it might be a
set of oil fields to be developed for a major oil company. Further, it might be a set of installation
project for a sub-sea equipment manufacturer. A third stakeholder will be a financial institute
(banks etc.) supporting the project financially in a limited time period. In this chapter we will
discuss the following:

• Visualization of project portfolios

• Selection of projects for execution

• Vulnerability of project portfolios

• Scheduling of projects within a portfolio

10.2 Visualization of project portfolios

There are various ways to visualize the projects within a potential portfolio. Such a visualization
is valuable to get an oversight of a huge number of projects and get a preliminary oversight of
these with respect to which ones to go for. We will now, and in the following assume that for the
individual projects key numbers or characteristics have been derived. Examples of such numbers
are:

• Expected net present value (NPV) of the project

• Standard deviation of the NPV

• Probability that NPV <0

• Probability that actual cost is larger than a factor, f , times the budget asked for, for example
f =1.5, and f = 2

• Probability of major failure of technology development

• Return factor, i.e., return (in Euro) per Euro invested

118
 Table 10.1: Project key numbers
Project Name NPV Pr(Cost overrun) Return ratio
A 10 0.3 3
B 5 0.2 2
C 8 0.1 2
D 4 0.45 5
E 7 0.3 4
F 4 0.2 3

Bubble diagrams are convenient to visualise tree of the dimensions at a time. To cover more
dimensions several bubble diagrams are required. We will consider the data shown in Table 10.1:
The net present value (NPV) represents the total discounted cost and income in the project,
either for a period of years, or an infinite time horizon if relevant. The corresponding bubble
diagram is shown in Figure 10.1.
Pr(Cost overrun)

0.6
0.5
D
0.4
0.3 E A
0.2 F B
0.1 C
0
2 4 6 8 10 12
Net present value

Figure 10.1: Example of bubble diagram where the size of the bubbles represents the return ratio

10.3 Selection of projects for execution

A first approach to select projects will be to combine the key numbers describing each project.
Now, let xij denote key number j for project i. Further, assume that each key number has been
given a weight, wj . In order to make the presentation as simple as possible we assume that the
weights are positive if a high value of xij is preferred, and the weights are negative if a low value
of xij is preferred. A score for project i may be achieved by

Si = Σj wj · xij (10.1)

We return to the example presented in Table 10.1, where the following weights has been introduced:
w1 = 4 = Weight of (expected) NPV
w2 = -2 = Weight of probability of overrun (e.g., more than a factor of 1.5)
w3 = 3 = Weight of return ratio
The calculated costs are shown in Table 10.2. The most promising projects according to the
applied weights are A, E and C respectively. Projects might be chosen according to the calculated

119
scores, where budget constraints might limit the number of projects. Assume that the budget is 8
units.

Table 10.2: Example data with calculated scores

Project Name NPV Pr(Cost overrun) Return ratio Project cost Score
A 10 0.3 3 3.3 48.4
B 5 0.2 2 2.5 25.6
C 8 0.1 2 4.0 37.8
D 4 0.45 5 0.8 30.1
E 7 0.3 4 1.8 39.4
F 4 0.2 3 1.3 24.6

Starting from the top of the list sorted on the score gives projects A, D, E and F where we
always fulfil the budget constraints. We then spend 7.2 cost units, and the total score of the selected
projects is 142.5. In general the problem is a dynamic programming problem. In order to set the
weights often the Analytic Hierarchy Process (AHP) is applied, see e.g., [31].
Note that the score of each project is a weighted sum of various criteria. It is, however, not
obvious that such a weighted score make sense in selecting a range of project. If there are budget
constraints as indicated above, the best strategy would obvious be to chose project with a low cost
and a reasonable high score. An alternative approach would be to derive a “utility” based score.
Ideally we could find a utility function and then search for a portfolio of projects that maximizes
expected utility under the budget constraints. By introducing a utility function the probability of
cost overrun could be discarded from the utility function since the uncertainty related to overrun
is reflected in the concave utility function. The return ratio would also be of less importance.
Obvious it seems reasonable to consider the return ratio, but it might be argued that how much we
get back of the money we invest is not the most appropriate measure, the most interesting measure
is the total profit at the end of the day. A pragmatic approach is now to use the NPV value as the
baseline for the score of a project, but introduce a penalty for projects with high probability of cost
overrun, and projects with high cost, i.e., those projects that lock capital. A proposed measure is:

Si∗ = NPV (1 − 0.3 Pr (CostOverrun)) − 0.2 · ProjectCost (10.2)

where the weights 0.3 and 0.2 represent how negative overruns and lack of capital are considered
respectively.
Table 10.3 is prepared for using the Excel solver to optimize which projects to include. In the
column “Included” a binary variable is introduced, where 1 means that the project is included, and
0 means that the project is not included. In the Excel solver all the “Included” variables are defined
as binary variables in the “Constraints” (Underlagt begrensningene) in Figure 10.2. Further the
sum of the project cost (cell J8) is limited to 8 units. The target cell is here set to cell I8 for the
alternative score, S∗ . The optimization in this particular situation gives the same result as for the
original score. The reason for this is that project C locks much capital even though the score is
high (S∗ = 4.8 which is the second highest) .

10.4 Vulnerability of project portfolios

The figures in Table 10.2 represent expected values. These numbers have most likely been assessed
by various risk analyses, hence we often have uncertainty distributions over the key figures. In

120
 Table 10.3: Project summary, alternative score S∗
ProjectNPV Pr(Cost Return Project Score IncludedScore Score* Pr.Cost
Name over- ratio cost |Incl |Incl |Incl
run)
A 10 0.3 3 3.3 48.4 1 48.40 6.33 3.33
B 5 0.2 2 2.5 25.6 0 0.00 0.00 0.00
C 8 0.1 2 4.0 37.8 0 0.00 0.00 0.00
D 4 0.45 5 0.8 30.1 1 30.10 2.64 0.80
E 7 0.3 4 1.8 39.4 1 39.40 4.55 1.75
F 4 0.2 3 1.3 24.6 1 24.60 2.53 1.33
Σ= 142.50 16.06 7.22

Figure 10.2: Setup for optimization in Excel

121
portfolio analysis we distinguish between random variation and systematic variation. Random
variation represents variation within one project independent of the other projects. Systematic
variation represents variation due to one or more factor possible affecting two or more projects. In
portfolio analysis it is the systematic variation which usually is of concern. The random variation
will be wiped out due to the “law of large numbers”. We will therefore in the following discuss
systematic variation. We now propose a 8 step method for selecting projects taking systematic
variation into account.

1. Calculate key numbers for relevant projects as described in Section 10.3

2. Choose the most promising projects based on a preliminary score (Si ), but include more
projects than the total budget constraints

3. Identify one or more vulnerability factor affecting two or more projects

4. For each project the vulnerability factors are considered explicitly. Further for each combi-
nation of the vulnerability factors recalculate the expected values of the key numbers. Let
{V}be the set of all combination of the vulnerability factors.

5. Present the corresponding projects key figures by bubble diagrams for the corresponding
values of {V}in order to visualise the impact of the vulnerabilities.

6. Recalculate the expected values of the weighted sum for various subsets of projects using
the law of total probability, where pv is the probability for element v in {V}. This means
that we for each v in {V}first calculate the sum of scores, Σi Si |v, for projects included by
conditioning on the value v, then we sum these weighted scores over all v in {V}in a new
weighted sum using the pv -values.

7. Calculate also the variance of the weighted sum.

8. Make a final selection of projects based on the vulnerability calculations.

9. Highlight the vulnerability factors as an important input to the overall portfolio management
(e.g., risk registers for each project, and global enterprise risk registers).

Table 10.4 shows the situation where only one vulnerability factor is considered. Two values are
considered, © = Good, § = Bad.

Table 10.4: Key project figures with vulnerability included

Project NPV © NPV § Project Project Return Return
Name cost © cost § ratio © ratio §
A 11 3 3 7 3.7 0.4
B 6 4 2.5 4 2.4 1.0
C 8.5 2 3 7 2.8 0.3
D 4.5 3 0.6 1.5 7.5 2.0
E 8 4 1.5 4 5.3 1.0
F 4.5 3 1.2 2 3.8 1.5

We assume that Pr(©) = 0.8, and Pr(§) = 0.2.

122
10.4.1 Dynamic programming - Portfolio selection
The situation described inTable 10.4 is of a dynamic programming type. The knapsack problem is
one of the standard example used to illustrate the knapsack problem:

• The knapsack problem is a problem in combinatorial optimization

• Given a set of items, each with a weight and a value

• Determine the number of each item to include in a collection so that the total weight is less
than or equal to a given limit and the total value is as large as possible

• To solve the knapsack problem we may

– Formulate it as a dynamic programming problem

– Formulate it as a mixed integer programming problem, and solve with Excel

An introduction to dynamic programming is given in Section 11.3. Note that for dynamic pro-
gramming we need to do the programming, whereas if we use mixed integer programming we use
Excel without any “programming” (coding).

10.4.2 Mixed integer programming problem

The knapsack problem and the problem above could be solved by mixed integer programming. An
introduction to mixed integer programming is given in Section ??. We can use the Solver in Excel
to solve the portfolio selection problem.

• Define the objective function to be the sum of score for included projects. The objective
function is the function to maximize (Set Objective in Excel.

• Each project is described by a binary variable. If we arrange the projects as a table in Excel,
we introduce a column named: Included.

• The set of binary variables are the decision variables for maximizing the objective function.
In excel we name the range of all “included” cells, i.e., we give them the name Included

• In Excel, we use the Subject to the Constraints: to ensure our binary variables to be
binary, i.e., we use the bin option in the constraints editor.

• We also need to add the constraints regarding maximum cost, maximum vulnerability etc in
the Subject to the Constraints:-window.

Note that we typically define a cell in Excel to hold the objective function. This cell could then
be SUMPRODUCT(Scores,Included) where Scores is the named range of the scores of all projects,
and Included is the named range of our binary variables.

Problem 10.1 Discuss steps 1-8. Maximise expected NPV under the constraints that the total
spending given the bad value (§) of the vulnerability factor should not exceed 12. You may either
use mixed integer programming or dynamic programming. ♢

123
10.5 Scheduling of projects within a portfolio
Given a selected portfolio of projects we need to schedule them in time. In some situations the
scheduling might be unproblematic, but in other situations we need to take resource constraints
into account. We will in the following only consider one critical resource (e.g., labour force of a
critical discipline). We assume that we have done a preliminary assessment of resources required in
each project in the various project phases. We will not give a complete approach to the scheduling
problem, but just illustrate some aspect of the scheduling problem.

• Given a selected portfolio of projects we need to schedule them in time

• Several project may “fight” for the same resources

• In the following we assume

– Each project has a due date with penalty of default defined

– We know the relation between progress in the projects, and resources we allocate
– We do not discriminate between different disciplines
– All quantities are deterministic, a probabilistic approach is much more

Table 10.5 shows the example data we will use. The variables are discussed in the following:
• DDi = Due date project i

• PDi = Penalty of default per time unit, project i

• µRUD,i = Progress per time unit under ramp up/down project i

• µFS,i =Progress per time unit under full speed project i

• RRUD,i = Resources required in ramp up/down period for project i

• RFS,i = Resources required in full speed period for project i

• AR =Available resources per time unit. Here we consider AR = 100 = fixed value for the
entire planning horizon.
We assume that all projects run through three phases, “ramp up”, “full speed” and “ramp
down”. From Table 10.5 we see that progress per time unit is less for ramp up and ramp down
compared to full speed. But we also see that the demand for resources is lower in the ramp up and
ramp down phases.

Table 10.5: Key figures for the projects

Project DD PD µRUD µFS RRUD RFS
1 120 10000 0.5 0.8 33.33 50
2 140 20000 0.5 0.8 33.33 50
3 150 10000 0.5 0.8 33.33 50

We now introduce the following decision variables

• xRU,i = Starting ramp up project i

124
 • xFS,i = Starting full speed project i

• xRD,i = Starting ramp down project i

• xi = End of project i

Figure 10.3 illustrates the resources needed in the various phases, and we also indicate the progress
µ per unit time. We will simplify and make some assumptions:

Resources

mRUD mFS mRUD

xRU xFS xRD x Time

Figure 10.3: Scheduling of a project within a portfolio

• We assume that we only have 3 projects

• We assume that project 1 is the one that always start first

• We can not run all 3 projects on full speed simultaneously due to constraint limitations

• A project is completed with it reaches a progress of 100%

∑
• Progress is given by j µj tj where we sum over the three project phases, and tj is the time
spent in each phase

• The cost of using resources equals the resources used, i.e., measuring on the same scale as
penalties. In general we need some scaling factors.

To solve the problem we need to introduce some constraints:

• All projects need to achieve 100% progress

• xRU,i ≤ xFS,i ≤ xRD,i ≤ x,i , i.e., the chronological order of the phases, cf. Figure 10.3

To simplify the constraint specification in Excel we use x ≤ y ⇐⇒ x − y ≤ 0.

The objective function is defined in a cell, say Z:

Z=MAX(X_1-DD_1,0)*PD_1
+MAX(X_2-DD_2,0)*PD_2
+MAX(X_3-DD_3,0)*PD_3
+(XFS_1-XRU_1)*RRUD_1+(X_1-XRD_1)*RRUD_1+(XRD_1-XFS_1)*RFS_1
+(XFS_2-XRU_2)*RRUD_2+(X_2-XRD_2)*RRUD_2+(XRD_2-XFS_2)*RFS_2
+(XFS_3-XRU_3)*RRUD_3+(X_3-XRD_3)*RRUD_3+(XRD_3-XFS_3)*RFS_3

125
The contribution from project 1 is:

• MAX(X_1-DD_1,0)*PD_1 = number of days exceeding the due date multiplied with penalty
per day

• (XFS_1-XRU_1)*RRUD_1+(X_1-XRD_1)*RRUD_1+(XRD_1-XFS_1)*RFS_1 = number of days for

each phase multiplied with the resources per day for this phase, and then summed over all
phases

There are different types of constraints. First, all projects need to be completed, i.e., progress
should be 100. For project 1 we define the following cell:

progress\_1=muRUD\_1*(XFS\_1-XRU\_1)+muFS\_1*(XRD\_1-XFS\_1)+muRUD\_1*(X\_1-XRD\_1)

and similarly for project 2 and 3. Then in the constraint editor we requires this cell to be greater
or equal to 100.
Next we need to ensure the chronological order of the projects, again for project 1 we need cells
defined by:

=XRU_1-XFS_1
=XFS_1-XRD_1
=XRD_1-X_1

In the constraints window these cells should then be less or equal to 0. We repeat for project 2 and
3.
To ensure that project 1 is the first one to start, we will require the following cell to be less or
equal to 0:

=XRU_1-MIN(XRU_2,XRU_3)

Finally, we need to ensure that not all three projects run on full speed simultaneously, this means
that we either need to ramp down project 1 or 2 before we give “full speed” for project 3, i.e.,

MIN(XRD_1,XRD_2)<=XFS_3}

or we ned to ramp down project 1 or 3 before project 2 is given “full speed”, i.e.,

MIN(XRD_1,XRD_3)<=XFS_2}

One of these inequalities need to be true. To formulate this constraint into a cell to be non-positive,
we use:

=IF(OR(MIN(XRD_1,XRD_2)<=XFS_3,MIN(XRD_1,XRD_3)<=XFS_2),0,1)

i.e., the cell is 0 only if one of the inequalities are true.

The solution is shown in Table 10.6. Note that the problem formulation is very similar to those
given for linear programming. However, note that the MAX()-function in the objective function
cause the model to be non-linear. Also the constraint on full speed causes difficulties. Therefore we
cannot solve the problem with the SIMPLEX algorithm, and we have to use non-linear optimization
techniques, i.e., the GRC Nonlinear option in the solver window. It takes several minutes to run
the model.

126
 Table 10.6: Optimal scheduling
Project RampUp FullSpeed RampDown EndOfProject
1 0 0 125 125
2 0 0 100 140
3 0 100 162.5 162.5

127
Chapter 11

Linear, dynamic, non-linear and

stochastic programming

11.1 Introduction to programming problems

Programming problems deals with the use or allocation of resources in the best possible manner
in terms of minimizing cost or maximizing profit. Resources are materials, labour, machines,
transportation capacity, energy, capital and so forth. In general resources are limited. In this
chapter three types of programming are introduced:

• Linear programming

• Dynamic programming

• Non-linear programming

• Stochastic programming

The learning objective of this chapter is to understand these type of programming and obtain skills
to apply the programming techniques. This means that after the end of this chapter the student
shall be able to formulate problems and solve problems by appropriate tools such as MS Excel.
The deeper understanding of the mathematical ideas behind these methods are left for courses in
operations research.
This chapter was originally developed for the course TPK4161 - Supply chain analytics, and we
have simplified to meet the requirements for TPK5115.

11.2 Linear programming

Linear programming deals with problems defined by the following conditions:

1. The decision variables are non-negative

2. The objective function is given as a linear function of the decision variables. The linear
assumption implies that only the first powers of the decision variables are included, and no
cross terms are allowed.

3. Constraints in terms of limitation of resources can be expressed as a set of linear equations

or linear inequalities.

128
11.2.1 Motivating example
As an introduction example consider the shipment of goods from Trondheim to Oslo. A truck has
a capacity to take 100 units. There are four products types that could be transported. The profit
per unit depends on the the product type and are given by c1 = 10, c2 = 5, c3 = 8 and c4 = 4 for
product type 1, 2, . . . , 4. Let xi denote the number of units to take of product type i. The objective
function is then given by:
Z(x1 , x2 , x3 , x4 ) = 10x1 + 5x2 + 8x3 + 4x4 = c1 x1 + c2 x2 + c3 x3 + c4 x4 (11.1)
There are some constraints in addition to the truck capacity. The customer in Oslo has storage
restrictions implying that 2x1 + 3x2 cannot exceed 50. Further there is a maximum of 50 units
available in Trondheim of product type 3. The constrains can thus be formulated by:
x1 + x2 + x3 + x4 ≤ 100 = b1
2x1 + 3x2 ≤ 50 = b2 (11.2)
x3 ≤ 50 = b3
In this problem it is quite obvious that the number of units cannot be negative, hence x1 ≥ 0, x2 ≥
0, . . . , x4 ≥ 0.
The LinearProgrammingIntroExample.xlsx file provides a solution to this transportation
problem. The optimal solution is given by x1 = 25, x2 = 0, x3 = 50 and x4 = 25 giving a
final profit equal to Z = 750.

11.2.2 Linear programming problem on standard form

A linear programming problem generally comprises an objective function to maximize or minimize,
and a set of constraints. The objective function, Z() is always a function of the decision variables
xj , j = 1, . . . , n but usually we do not explicitly state this relation on the left hand side of the
equation. Further we assume m restrictions regarding linear combinations of the decision variables.
Finally all decision variables and some parameters have to be non-negative. This is written:
Maximize: Z = c1 x1 + c2 x2 + · · · + cn xn (11.3)

Subject to: a11 x1 + a12 x2 + · · · + a1n xn = b1

a21 x1 + a22 x2 + · · · + a2n xn = b2
..
. (11.4)
am1 x1 + am2 x2 + · · · + amn xn = bm
x1 ≥ 0, x2 ≥ 0, . . . , xn ≥ 0
b1 ≥ 0, b2 ≥ 0, . . . , bm ≥ 0
Note that the LP problem formulation given by Equation (11.3) and the following constraints could
be written in matrix form:
Maximize: Z = cx

Subject to: Ax = b
x≥0
b≥0

129
A linear programming problem can be solved by the SIMPLEX method which we will come back to.
The motivating example was not exactly on standard form because there were some inequalities in
the constraints, i.e., · · · ≤ bi rather than · · · = bi . In other cases there are other types of restrictions
that do not completely match the standard format of the problem. The following steps may be
used to transform a linear programming problem into standard form:
• If the problem is to minimize Z convert the problem to a problem that maximizes −Z.
• If there is a less or equal inequality in one or more rows of the constraints, i.e., ai1 x1 + ai2 x2 +
. . . ain xn ≤ bi , convert it into an equality constraint by adding a nonnegative slack variable
si yielding the resulting constraint: ai1 x1 + ai2 x2 + . . . ain xn + si = bi , where si ≥ 0.
• If there is a greater or equal inequality in one or more rows of the constraints, i.e., ai1 x1 +
ai2 x2 + . . . ain xn ≥ bi , convert it into an equality constraint by subtracting a nonnegative
surplus variable si yielding the resulting constraint: ai1 x1 + ai2 x2 + . . . ain xn − si = bi , where
si ≥ 0.
• If some of the bi ’s are negative, multiply that constraint equation by -1.
• If one of the decision variables, say xj is unrestricted in sign, replace it by xaj − xbj everywhere
in the problem formulation and add to the constraints: xaj ≥ 0 and xbj ≥ 0.

Note that si , xaj and xbj will be renamed to maintain a set of n decision variables, xj , j = 1 . . . n.

11.2.3 Solving the linear programming problem by the SIMPLEX method

To maximize the objective function in Equation (11.3) and the related constraints we typically use
the SIMPLEX method. It is beyond the scope of this course to understand the SIMPLEX method,
so we rather stick to Excel in order to solve such problems. The SIMPLEX algorithm in EXCEL is
usually sufficient for our problems of limited size in example cases. For real problems, we often need
smart ways to solve such problems, i.e., smart heuristics taking advantages of specific structures in
the optimization problem. In particular when some variables are of integer type.

11.2.4 Solving the LP problem by a computer

For practical purposes we need a computer to solve a LP problem. The method presented here is
the basis for most computer codes, but refinement of the algorithm has made it much more efficient.
We will not discuss the content of such refinements but refer to standard text books in operations
research.
For very many users MS Excel will be the computer tool to solve a large range of LP prob-
lems. In order to utilize MS Excel the ”Solver” need to be installed. The ”Solver” if found under
->File->Options->Add-ins. When the ”Solver” is installed it is found under ->Data. The solver
needs 5 types of input:
1. The cell to optimize. This cell should contain a cell that contains the objective function.
Although we have used the name Z for the objective function any cell may be used. (Set
objective:)
2. The type of optimization. A choice could be set to either minimization or maximization, or
even target the objective function to a given value. In MS Excel we therefore do not need
to convert a minimization problem to a maximization problem by multiplying the objective
function by -1 on both sides as we do in the manual procedure. (To:)

130
 3. Define the decision variables, i.e., those variables we could change in order to find the best
solution. (By Changing Variable Cells:)

4. Define the constraints, i.e., specify the equations that define the constraints. Note that MS
Excel will accept inequalities in the constraints, hence we do not need to add slack variables.
(Subject to the Constraints:)

5. Define the solving method. For linear models we always choose the Simplex LP method. For
nonlinear models the Simplex LP method does not work. (Select a solving method:)

Figure 11.1: Final tableau

Figure 11.1 shows the specification screen for the MS Excel Solver. Note the following:

• The cell containing the objective function has been named Z. Further, to calculate Z we used
the =SUMPRODUCT() function. This function accepts two arguments, the first argument is the
x-values, and the second argument is the objective function coefficients.

• The cells containing the x-variables has been given the name xValues.

• To simplify the specification of the constraints it is convenient to move the ∑ constant term to
the left-hand side of the inequality sign, i.e., we can create cells containing ai,j xj −bi . These
cells are then used in the specification of constraints. Also here the =SUMPRODUCT() function is
useful for calculating the sum of products. The three cells containing the constraint equations
have been given the name Constraints. When the ≤ comparison is specified it applies to all
the constraint equations.

Example 11.1 Production scheduling

This example is from [36] where we are scheduling the next 4 weeks of production. The weekly
demands are 300, 700, 900 and 800 items for the next 4 weeks. Production cost is $5 for week
1 and 2, and $10 for week 3 and 4. Normal production cannot exceed 700 units per week. It is

131
possible to use overtime to produce up to 200 extra units per week in week 2 and 3 but then at
an additional item cost of $5. Excess production can be stored at a cost of $3 per item (from one
month to the other). Table 11.1 shows the decision variables to use and other important quantities
for the problem.

Table 11.1: Variables and quantities in Example 11.1

Variable/Quantity Explanation
xi , i = 1 : 4 Normal production week i
di , i = 1 : 4 Demand week i, = {300,700,900,800}
x5 , x 6 Overtime in week 2 and 3 respectively
xi , i = 7 : 10 Stock level at end of week i − 6

Minimize: Z = 5x1 + 5x2 + 10x3 + 10x4 + 10x5 + 15x6 + 3x7 + 3x8 + 3x9 + 3x10

Subject to: x1 ≤ 700

x2 ≤ 700
x3 ≤ 700
x4 ≤ 700
x5 ≤ 200
x6 ≤ 200
x1 − x7 = 300
x2 + x5 + x7 − x8 = 700
x3 + x6 + x8 − x9 = 900
x4 + x9 − x10 = 800

where it is implicitly assumed that all decision variables are non-negative. ∑

An MS Excel solution is shown in Figure 11.2. The constraints equations given in the aij − bi
column is named constraintsLE for the ≤ constraints and constraintsEQ for the = constraints
respectively for easy specification in MS Excel.

11.2.5 Mixed integer programming

Many LP problem formulations put additional logical constraints on the decision variables. Typi-
cally some of the decision variables are restricted to integer values or even binary variables. If this
is the case the problem is said to be a mixed integer programming problem. A naive approach to
mixed integer programming is to ignore the integer constraint and solve the problem as if it was
an ordinary LP problem and then take the nearest integer from the continuous solution for those
variables. This solution is however not particular good. To really find an optimal solution there is
no general approach that guarantees that we can succeed because all possible combinations of the
decision variables have to be investigated, a problem that generally can not be solved.
The branch and bound (B&B) algorithm is maybe the most widely method used for solving
mixed integer problems in commercial computer codes. The B&B algorithm is just an efficient
enumeration procedure for examining all possible integer feasible solutions. The idea of the branch

132
 Figure 11.2: MS Excel solution of Example 11.1

and bound algorithm is presented below but technical details regarding the implementation is left
out and the reader is referred to a operations research text book for further reading. A mixed
integer programming problem (MIP) is generally given on the following form:

Maximize: Z = cx (11.5)

Subject to: Ax = b
x≥0
b≥0
xj is an integer forj ∈ I

where I is the set of all integer variables.

The B&B algorithm starts with maximizing the objective function in Equation (11.5) ignoring
the integer restrictions. This problem is denoted LP-1 and will give a maximum objective function,
say Z1 . Figure 11.3 is used as an illustrative example where we assume x1 and x2 are restricted
to integer values. Since the integer restrictions were ignored in LP-1 it is likely that the solution
for some integer variables contain fractional values. The idea now is to split the LP-1 problem
into two new problems, say LP-2 and LP-3. These problems are identical to the LP-1 problem
but a new restriction is introduced for each of LP-2 and LP-3. The restrictions relate to one of
the integer variables that contains a fractional value. Various principle for selecting this branching
variable exist in the literature but will not be discussed here. Assume that variable xj is taken as
the branching variable and that the LP-1 solution gave an optimal value of xj = βj . Since xj = βj
is not a feasible solution due to the integer restriction we add xj ≤ int(βj ) as a new restriction
to the LP-2 problem, and similarly xj ≥ int(βj ) + 1 as a new restriction to the LP-3 problem.
For example if LP-1 has an optimal solution with x2 = 1.5 Figure 11.3 shows as an example that
we have split the problem into LP-2 with an additional restriction x2 ≤ 1 and into LP-3 with an
additional restriction x2 ≥ 2.
The B&B algorithm now proceeds by solving the LP-2 and LP-3 problems, again as if they were
standard LP problems, i.e., ignoring the integer restrictions. Figure 11.3 shows as an example that

133
 LP-1 Z1 = 9

x2 ≤ 1 x2 ≥ 2

LP-2 Z2 = 8 LP-3 Z3 = 8.5

x1 ≤ 4 x1 ≥ 5

LP-4 Z4 = 7 LP-5 Infeasible

Figure 11.3: B&B example

LP-2 has an optimal solution of Z2 = 8 and LP-3 has an optimal solution of Z2 = 8.5. Assume
that LP-3 has a feasible solution, i.e., that integer restrictions are fulfilled.
We proceed with LP-3 and split into two new problems, say LP-4 and LP-5 and so on. Here
the integer variable x1 is used to put new restrictions on the LP-problems. Occasionally the new
restrictions put on one the integer variables will give integer value for all the integer variables in
the optimal value in the solution, and we are closer to the solution. If this is the case we have
found a solution for that particular LP-problem. It is no guarantee that it is optimal, but it is at
least a lower bound for the optimal solution. This also means all LP-problems with a Z-value lower
than the best feasible solution up to now can be ignored for further investigation. In the example
in Figure 11.3 LP-4 has a solution less favourable compared to the best feasible solution in LP-2,
and since LP-5 gave no feasible solution the best solution found is the solution found in LP-2.
There is no guarantee that an optimal solution is found, but we are able to squeeze the solution
into limits which might be good enough. When one LP-problem is branched into two new problems
it is not evident which one to proceed with. Some principles are discussed in the literature, and
again we refer to an operation research textbook for further reading.
In MS Excel we may specify that some decision variables are integer or binary variables. This
is done when specifying the constraint comparison symbol where we can use ”int” or ”bin”.

11.3 Dynamic programming

Dynamic programming (DP) is a method for finding an optimal strategy to apply to a dynamical
system that evolves over time. Let N be the number of stages a system runs through. Further the
following quantities are defined for stage number n:

• Sn = input state, i.e., the system state when the system enters state n

• dn = decision at stage n, i.e., what we can do to influence the system performance

• S̃n = output state, i.e., the system state when the system leaves state n

• rn (Sn , dn ) = return function, typically a cost or profit function that depends on the system
state and the decision made

Figure 11.4 shows the main elements of an N -stage multistage system. The stage numbering is
typically in an opposite direction compared to the flow of information.
Usually we solve the problem computationally by starting at stage 1 and proceed from right
to left to stage N . Such an approach is denoted backward analysis or backward recursion. In a

134
 rN rN −1 rn r1

SN S̃N SN −1 S̃N −1 Sn S̃n S1 S̃1

N N -1 n 1

dN dN −1 dn d1
Figure 11.4: An N -stage multistage system

backward analysis we always have that Sn−1 = S̃n . The idea is that for the rightmost stage it is
rather easy to find the optimal solution for a given input, i.e., for a given S1 we easily find the
best decision d1 . When proceeding left to stage 2 we then know all the stage 1 optimal decisions
and corresponding return functions r1 for each of the output values S̃2 . Then adding r1 and r2 for
possible d2 decisions makes it possible to find optimal decisions at stage 2 for a given input S2 . We
then proceed till stage N is reached.
The change in system state in a given stage is governed by some transition function, i.e.,:

S̃n = gn (Sn , dn ) (11.6)

The return function rn (Sn , dn ) specifies what is the cost or profit related to what happens in stage
n. The accumulated total return calculated over n stages, i.e., from right to left or numerically
from 1 to n is denoted fn (Sn , dn ). Further fn∗ (Sn ) is the accumulated optimal returns up to stage
n.
Mathematically it is a challenge to be explicit on the formulation of fn (Sn , dn ) and fn∗ (Sn ).
Typically these functions are formulated recursively.

Elements of an optimal decision policy

The idea of dynamic programming is to start at the rightmost stage, i.e., at stage 1. If the input to
stage 1 is known, i.e., we know the value of S1 , then it is rather easy to obtain the optimal decision,
say d∗1 at stage 1, since what has happened previously should not affect the best thing to do at this
final stage. Note that d∗1 = d∗1 (S1 ) and obviously depends on the return function, r1 (S1 , d1 ).
Then at stage 2 we may in a similar way ignore what has happened up to stage 2 as long as
we know S2 . To find an optimal decision, say d∗2 at stage 2 we need to consider both the return
function, r2 (S2 , d2 ) but also all possible scenarios for stage 1. Since S1 = S̃2 this means that we
may use d∗1 = d∗1 (S1 ) = d∗1 (S̃2 ) to find the optimal return function contribution from stage 1 when
optimizing at stage 2.
More generally, at stage n we ignore what has happened previously, i.e., for stages left to
n (higher values than n), when finding the optimal decision at stage n. We have to consider
accumulated contribution to the return function to the right of stage n. That is, in the optimization
at stage n we utilize fn−1∗ (S̃ ) which in principle is known for each output state of stage n.
n
Finally, at stageN the final optimization is performed, i.e., the calculation steps performed at
stage n = N yields the optimal policy.
To solve such a problem by a computer code we need some functions. Let g(n,S,d) be a function
that implements the transition that takes place in stage n given that decision d is implemented,
see Equation (11.6). Further let r(n,S,d) be a function that implements the return in stage n
given that decision d is implemented. In most cases the accumulated returns are found by adding
the return function contributions. In such situations we could then write a recursive function, f(),
implementing fn∗ (Sn ) on the form:
f(n, S) = opt{r(n, S, d) + f(n-1, g(n,S,d))}

135
where opt is the optimization with respect to decision dn at stage n. Now, assume that S_N is the
initial state of the system, the optimal strategy is found by calling f(N, S_N).
Note that f() needs to be implemented in a programming language that allows recursive calls
because this function calls itself. Further note that the all the functions need to be carefully
designed to take care off all boundary conditions, for example to stop when state 0 is invoked!
Further, since the function calls are recursive, the number of calls may explode in magnitude.

11.3.1 Worked example

This example is adopted from [36]. A shoe store sells rubber shoes for the winter season. The sales
division has forecasted the demand for the next season given in Table 11.2. We assume a perfect
forecast, i.e., a deterministic model will be applied.

Table 11.2: Montly demand

Month Demand
October 40
November 20
December 30
January 40
February 30
March 20

The store purchases pairs of shoes from a manufacturer at a purchasing cost of 4 $ per pair
(1976 prices). The supplier only sells in lots of 10, 20, 30, 40 or 50 pair. A discount is achieved for
large orders as given in Table 11.3.

Table 11.3: Discount depending on order quantity

Lot Size Discount
0 0%
10 5%
20 5%
30 10%
40 20%
50 25%

For each order there is a fixed cost of 10 $ to account for transportation, insurance, packaging,
and so on. The shop has limited capacity and can not carry an inventory level higher than 40 at
the end of each month. There is a holding cost of 0.3 $ based on the end-of-month inventory level.
The season starts with an empty inventory, and should also end with an empty inventory. Table
11.4 gives summarized information.
The cost structure favours large orders, but the downside is a holding cost at the end of each
month. Also large order could save the fixed cost if we could escape from ordering one month. It
is assumed that the pair of shoes arrives the first day of the month.
The backward analysis means that the system runs through 6 stages, where October corresponds
to n = 6, November to n = 5 and so on to March where n = 1. The decision variables, dn is the
number of pairs to order for each month. The state variable Sn is the number of pairs in the

136
 Table 11.4: Additional parameters
Parameter Value Description
cP 4 Purchasing cost per unit
cF 10 Fixed cost per order
cH 0.2 Holding cost per unit at end of month
SL 40 Storage limitation at end of month
S6 0 Inventory level at the beginning of the season
S0 = S̃1 0 Inventory level at the end of the season

beginning of the month when the order from the manufacturer has arrived. From Table 11.2 we
can read the demand, Dn each month, again remembering that October corresponds to n = 6 and
so on.
The transition function must relate the state variable at stage n to that of stage n − 1:

Sn−1 = Sn + dn − Dn (11.7)

The return function comprises the cost of an order and holding cost at the end of the month:

rn (Sn , dn ) = ϕ(dn ) + cH (Sn + dn − Dn ) (11.8)

where ϕ(dn ) is the cost of an order comprising the fixed cost and the discounted variable cost per
pair of shoes. It is assumed that ϕ(0) = 0. Now, assume that the transition function in Equation
(11.7) and the return function in Equation (11.8) are implemented by the functions g(n,S,d) and
r(n,S,d) respectively. Then we need to write a function for the accumulated return. Below is
given the main structure of the computer code we need. The programming language is VBA, i.e.,
the built in Visual Basic for Microsoft applications.

Function f(n As Integer, S As Integer)

If n = 0 Then
f = 0
Exit Function
End If
fOptimum = infty
For Each d In lotSizes
isFeasible = True
If (g(n, S, d) > 40 Or g(n, S, d) < 0) Then isFeasible = False
If (n = 1 And g(n, S, d) <> 0) Then isFeasible = False
If isFeasible Then
fTest = r(n, S, d) + f(n - 1, g(n, S, d))
If fTest < fOptimum Then
fOptimum = fTest
End If
End If
Next
f = fOptimum
End Function

137
 Table 11.5: Intermediate results for n = 1
S1 d∗1 f1∗
0 20 86
10 10 48
20 0 0

The recursive formulation causes the f(n,S) function to call it self recursively starting from
n = N = 6. At the lowest level, i.e., n = 1 there are only 3 feasible values of S1 to consider, i.e., 0,
10 and 20. It is easy to verify that the optimal decisions are given in Table 11.5
These results for n = 1 are needed several times when the function is called from stage n = 2.
Therefore it would be a good idea to save the optimal values. This applies also for all other sates. In
fact, for this example we need to calculate fn () more than 1 100 times, which reduces to around 30
calculations by clever saving of intermediate results. In computing, memoization is an optimization
technique used primarily to speed up computer programs by storing the results of expensive function
calls and returning the cached result when the same inputs occur again. Memoization here means
to save calculated values for f ∗ for (S, n) combinations.
The intermediate results for the other (Sn , n) combinations, i.e., n > 1 is not given in this
presentation.
Note that since the function f(n,S) only returns the accumulated cost, but we have not kept
track of the optimal decision at the various levels. A technical solutions to this challenge is left for
the reader as an exercise.
Table 11.6 summarizes the final results for the DP analysis.

Table 11.6: Final results

Month fn∗ Sn S̃n Dd d∗n
October 606 0 0 40 40
November 468 0 30 20 50
December 302 30 0 30 0
January 302 0 0 40 40
February 164 0 20 30 50
March 0 20 0 20 0

11.4 Stochastic programming

11.4.1 Introduction
In the standard linear programming problem discussed in section 11.2 we assumed all quantities to
be know in advance. That is, the LP problem formulation was given by Equation (11.3), and to
repeat in matrix form we have:
Maximize: Z = cx (11.9)

Subject to: Ax = b
x≥0 (11.10)
b≥0

138
The notation used is that x is a decision vector which is controlled by the decision maker, c is
the profit/cost vector, A is the coefficient matrix and b is the requirement vector. Up to now we
have assumed that c, A and b are all known at the time of the decision to be made, that is we
have a deterministic optimization problem at hand. In real life, we often have to make decisions
when some of these quantities are not known. Typically the requirement vector could be uncertain,
and we need to treat this vector as a stochastic variable. The requirement vector often represents
resource constraints which is only or partly known at a later stage in the decision process.
Stochastic programming is a technique to solve such challenges. In this presentation we limit
the presentation of stochastic programming only to deal with linear programming situations.
To motivate for the formalism to be introduced, consider a fish processing facility where we have
to plan tomorrow’s production before we actually know the catch, i.e., the amount and quality of
fish available for production. Fish of high quality can be used for production of fresh fillet. Frozen
fish can be produced from both high and low quality fish. It is assumed that the selling price of
fresh fillet is higher than of frozen fish, but production of fresh fillet requires more workload. Today
we have to decided upon how many workers, x to allocate for the production. x is often referred
to as the here an now decision. The production profile, y can be determined at a later stage,
i.e., when we know the catch. The objective function is now split into two objectives function,
Z1 and Z2 . The maximization (minimization) of Z1 = cx subject to some Ax = b is denoted
the first stage optimization. In the example this corresponds to finding the number of workers to
allocate for tomorrows production. This is obvious not straight forward since we need to take into
account the catch during the night. Now, assume that we can describe tomorrows situation by
some stochastic vector U. Given that we know the value of this vector, say U = u there is a second
stage optimization problem defined by:

Maximize: Z2 = q(u)y (11.11)

Subject to: B(u)x + C(u)y = d(u)

y≥0 (11.12)
d(u) ≥ 0

Since the decision to make tomorrow is influenced by the decision we make today through the
constraints B(u)x we are facing the classical two stage linear stochastic programming problem
were the decision the two days are interconnected. In the first stage we need to take into account
what will be the result of the second stage decision but this again depends on the unknown vector
U. The idea now is to add the expected value of the best solution tomorrow to the today’s objective
function:

Maximize: Z1,2 = cx + EU [Q(x, U)] (11.13)

Subject to: Ax = b
x≥0 (11.14)
b≥0

where the expectation is taken with respect to the stochastic vector U and Q(x, u) is the solution
to:

Maximize: Z2 = q(u)y (11.15)

139
 Subject to: B(u)x + C(u)y = d(u)
y≥0 (11.16)
d(u) ≥ 0

In the second stage problem the profit/cost vector q(u) and the requirement vector d(u) both depend
on the value, u, the stochastic vector U will take. Further the coefficient matrix in B(u)x + C(u)y
is built up by two terms: B(u) relates to the first stage decision vector x and C(u) relates to the
second stage decision vector y. Both matrices depend on the value the stochastic vector U will
take.
In some presentations the second stage problem is considered as a recourse action where the
term C(u) compensates for a possible inconsistency of the system B(u)x = d(u) and q(u)y is the
cost of this recourse action.
In the example −cx represents the cost of allocating x workers for tomorrows production. q(u)y
is the profit of the tomorrows production. B(u)x + C(u)y = d(u) gives the constraints taking the
number of workers and catch into account.

11.4.2 Discretization
To solve the two-stage stochastic problem numerically it is common to perform a discretization
where the stochastic vector U can only take a finite number of possible realizations. These re-
alizations are often denoted scenarios, say u1 , u2 . . . , uk , with corresponding probabilities pi =
Pr(U = ui ). The expectation in the second-stage problem is then given by:

∑
k
EU [Q(x, U)] = pi Q(x, uk ) (11.17)
i=1

It is now possible to write the stochastic problem on a so-called extensive form, that is a determin-
istic equivalent problem:

∑
k
Maximize: Z1,2 = cx + pi q(ui )yi (11.18)
i=1

Subject to: Ax = b
B(ui )x + C(ui )yi = d(ui ), i = 1, 2, . . . , k
x≥0
(11.19)
b≥0
yi ≥ 0
d(ui ) ≥ 0, i = 1, 2, . . . , k

For the scenarios i = 1, 2, . . . , k we have to specify B(ui ), C(ui ) and d(ui ) one by one based on the
understanding of the problem at hand.

Example 11.2 Fish processing facility

We consider a fish processing facility where tomorrow’s supply is uncertain. A discretization into
3 different supply scenarios is performed as shown in Table 11.7. Table 11.8 shows additional
parameters required for the optimization.

140
 Table 11.7: Supply scenarios
Scenario, i pi High quality Low quality
1 0.15 16 000 12 000
2 0.5 8 000 6 000
3 0.35 6 000 3 000

Table 11.8: Parameters

Parameter Value Description
ρFi 0.6 Utilization ratio fresh fillet
ρFz 0.7 Utilization ratio frozen fillet
pFi 80 Selling price, fresh fillet (NOKs/kg)
pFz 30 Selling price, frozen fillet (NOKs/kg)
µFi 70 Processing capacity, kg fresh fillet /man hour
µFz 300 Processing capacity, kg frozen fillet /man hour
cMHR 500 Man hour cost (NOKs)

For fish of high quality the most beneficial product is fresh fillet despite the higher workload
required. This means that if we knew the supply we will always allocate work-force to process
all high quality fish to fresh fillet, and the low quality to frozen fillet. However, at the decision
point for the work-force we do not know the supply. If we allocate sufficient work-force for the best
scenario, i.e., 250 man-hours (some 30 persons), and actually get less raw material, we have to pay
for man-hours we are not utilizing.
The following decision variables are introduced:

• x1 = Number of man-hours allocated

• y1,i = Number of kg processed for fresh fish fillet, scenario i

• y2,i = Number of kg processed for frozen fish fillet, scenario i

In the objective function the profit coefficient corresponding to x1 is −cMHR , the profit coefficients
corresponding to y1,i are pi ρFi pFi and the profit coefficients corresponding to y2,i are pi ρFz pFz .
In the constraint matrix for the deterministic equivalent problem we have for each scenario to
make sure that y1,i not exceeds supply of high quality fish, and that y1,i + y2,i not exceeds total
supply of fish. Further for each scenario we have to make sure that y1,i /µFi + y2,i /µFz ≤ x1 .
Solving the deterministic equivalent problem gives x1 ≈ 134. If scenario 1 occurs, 3 739 kg of
high quality fish is used for fresh fillet, and 24 260 kg (= remaining high quality fish + low quality
fish ) is used to produce frozen fish fillet. The Excel model used is shown in Figure 11.5.

11.4.3 The Value of the Stochastic Solution

The Value of the Stochastic Solution (VSS) is introduced to measure the benefit of solving the
stochastic problem rather than treating the problem as a deterministic problem. To be more
precise on what is meant by treating the problem as a deterministic problem we mean that the
stochastic variables are replaced by their expected values where uncertainties affecting the second
stage decision are ignored. This is referred to as an expected value (EV) approach. This means

141
 Figure 11.5: Specification of the stochastic programming example in Excel

that to solve the two stage problem the decision maker anticipate the value of U by u = E(U),
and optimize:

Maximize: ZEV = cx + q(u)y (11.20)

Subject to: Ax = b
B(u)x + C(u)y = d(u)
x≥0
(11.21)
b≥0
y≥0
d(u) ≥ 0
The first stage solution is denoted xEV . The first stage solution is optimal if the stochastic variable
U takes the value u. However, since the problem is stochastic in nature, other values could also be
the case. Therefore, the decision maker needs to modify the second stage decision when the value
of the stochastic variable U is revealed. The expected value of the expected value solution is now
the value of the following optimization problem:

∑
k
Maximize: ZEEV = cxEV + pi q(ui )yi (11.22)
i=1

Subject to: B(ui )xEV + C(ui )yi = d(ui ), i = 1, 2, . . . , k

yi ≥ 0 (11.23)
d(ui ) ≥ 0, i = 1, 2, . . . , k
Here the solution for each scenario i could be found individually. The difference compared to the
optimization problem in Equation (11.18) is that xEV is fixed. This also mean that the constraint
B(ui )xEV + C(ui )yi = d(ui ) could be very costly for some scenarios since these constraints were
only “tested” for u.
Let the value of the optimization problem in Equation (11.22) be denoted ZEEV∗ and the value
∗
of the optimization problem in Equation (11.18) be denoted Z1,2 . The difference between these two
is denoted the value of the stochastic solution, i.e.,
∗ ∗
VSS = Z1,2 − ZEEV (11.24)

142
11.4.4 Expected value of perfect information
In decision theory, the expected value of perfect information (EVPI) is the price that one would
be willing to pay in order to gain access to perfect information. In our context this means the
difference between the expected value of the stochastic solution and expected value of the solution
if we knew u at the time of the first stage decision. For the latter case we often say that this
correspond to a wait and see situation were we imagine that we can wait and observe u before we
need to decide upon x.
The optimization problem for the wait and see situation given that we observe u = ui is given
by:

Maximize: Zi = cxi + q(ui )yi (11.25)

Subject to: Axi = b

B(ui )xi + C(ui )yi = d(ui )
x≥0
(11.26)
b≥0
yi ≥ 0
d(ui ) ≥ 0

Let Zi∗ be the result of the optimization problem

∑ in Equation (11.25). The expected profit when
averaging over all scenarios are given by i pi Zi∗ . The expected value of perfect information is
thus:

∑
k
EVPI = pi Zi∗ − Z1,2
∗
(11.27)
i=1

There are several reasons why we would like to calculate the EVPI. For example if we know EVPI
we can decide upon how much effort we should pay to reduce uncertainty. For example in Example
11.2 we could imagine that we are able to establish a better prediction model for the catch. Such a
model could take weather forecast, type of catch tools and so on into account. A better prediction
model will not reduce variability in the catch, but will reduce the epistemic uncertainty, and hence
we are able to take better decisions.

11.4.5 Scenario building

In our two-stage model the uncertainty is defined through the stochastic vector U. In Example
11.2 we assumed that we only could have three different values. In general U could have many
elements, where each element could take many different values, or even be continuous variables.
This will lead to a very large number of scenarios which is a challenge both from a modelling and
from a computational point of view. There are many approaches to overcome the challenges. The
sample average approximation (SAA) is one approach that is often used. The idea is to use Monte
Carlo techniques to generate a limited sample of the stochastic vector U. That is, we realize N
vectors, say ui , with the same distribution as U. The optimization problem to solve is then:

1 ∑
N
Maximize: Z1,2 = cx + q(ui )yi (11.28)
N
i=1

143
subject to the same constraints as in Section 11.4.2.
Another approach is to investigate element by element in the stochastic vector U. That is, for
each element Ui we first calculate the VSS where we use 3 values, say low, medium and high for
the discretization. If SSV is relatively small, we may be tempted to ignore uncertainty of element
Ui .
However, if SSV is large the uncertainty of element Ui should be taken into account. The next
question is then if a discretization using only 3 values is sufficient. If Ui is a continuous variable we
would expect to get a better result if we approximate Ui with a 5-point discrete variable compared
to a 3-point discrete variable. Again, we calculate the expected value of the stochastic solution
(SSV). The situation is now more challenging. We have to compare the 5-point solution with the 3-
point solution. If we assume that the 5-point discretization of Ui is “perfect”, we can then calculate
the expected cost of assuming a 3-point discretization for the first stage decision, and then face
the 5-point discretization for the second stage decision. The procedure for calculating the expected
profit for the 3-point discretization is then:
1. Formulate the problem as a two stage optimization problem on extended form where a 3-point
discretization is used
2. Find the solution for both x and yi , i = 1, 2, 3. Denote the solution for x by x3p
3. Switch to the 5-point discretization
4. For each of the 5 scenarios, solve the second stage problem assuming that first stage solution
is bounded to x3p
5. For each of the 5 scenarios, calculate the expected profit
6. Calculate an average weighted profit by using corresponding pi ’s for each scenario
The calculated weighted average is then subtracted from the expected value where a 5-point dis-
cretization is used in the ordinary manner.
If the difference is significant, we could compare the 5-point discretization with the situation
where a 7-point discretization is assumed to be “perfect”. We proceed in this manner until the
value of a higher point discretization is relatively low. Here we need to be pragmatic, so that the
total number of combinations taking all elements of U into account is manageable.

11.4.6 How to perform discretization?

[33] propose a standard approach for approximating a continuous distribution by a discrete distribu-
tion where (i) the outcome region is divided into intervals, then (ii) for each interval a representing
point is chosen, and (iii) a probability, pi is assigned to each point. Usually the intervals are found
by dividing the outcome region into k equally probable intervals, where the representative point is
the mean of the corresponding interval, and the assigned probability is 1/k. When generating a
limited number of discrete outcomes, some statistical properties should be specified. It is common
to include the first four central moments as properties, i.e., the moments for the discrete stochastic
variable should be as close as possible to the moments of the variable we are approximating.
[34] have used such an approach if the stochastic variable is normally distributed with mean value
µ and standard deviation σ. Odd numbers of points are used, and Table 11.9 presents standardized
distances from the mid point. The midpoint is always given by µ. If a k point approximation is
required, the two nearest points to the mid point are given by the (left and right) distance dk,1 σ
from the midpoint at µ, the second nearest points to the mid point are given by the distance dk,2 σ
from the midpoint at µ and so forth.

144
 Table 11.9: Standardized distances from mid point
k = # of scenarios dk,1 dk,2 dk,3 dk,4 dk,5
3 1.22474
5 0.87889 1.31436
7 0.16787 0.49042 1.79758
9 0.21902 0.60872 0.67431 1.90442
11 0.26459 0.43883 0.70498 0.89051 1.98681

145
Bibliography

[1] Fischhoff, B., S. Lichtenstein, P. Slovic, S.L. Derby, and R.L. Keeney. Acceptable Risk. Cam-
bridge University Press, New York, 1981.

[2] Austeng, K. og Hugsted, R.: Trinnvis kalkulasjon, BATEK, 1995.

[3] Klakegg, O.J.: Tidsplanlegging under usikkerheit, BATEK, 1994.

[4] Klakegg, O.J.: Trinnvis-prosessen, Institutt for bygg- og anleggsteknikk NTH, 1993.

[5] Chapman, C. & Ward, S., 1997. Project Risk Management; Processes, Techniques and Insights.
John Wiley & Sons, England.

[6] Hokstad, P. Life Cycle Cost Analysis in Railway Systems. SINTEF Report STF38 A98424.
ISBN 82-14-00450-0. 1988.

[7] IEC 60300. International Standard, IEC 60300-3-3, Dependability management - Part 3: Ap-
plication Guide - Section 3: Life Cycle Costing

[8] Kawauchi, Y. and Rausand, M. Life Cycle Cost (LCC) analysis in oil and chemical process
industries. NTNU report . 1999.

[9] Keeney, R. L. and H. Raiffa. Decisions with Multiple Objectives: Preference and Value Trade-
offs. New York: Wiley. 1976.

[10] Vatn, J. 1998. A discussion of the acceptable risk problem. Reliability Engineering and System
Safety, 61(1-2):11-19, 1998.

[11] Øien, K, P.R. Hokstad, and R. Rosness. 1998. Handbook for performing Expert Judgement.
SINTEF report STF38 A98419. Could be ordered from http://www.sintef.no.

[12] Aven, T. Foundations of risk analysis. Wiley, West Sussex, 2003.

[13] IEC 60300-3-9. International Electrotechnical Vocabulary (IEV) - Dependability management

- Part 3: Application guide - Section 9: Risk analysis of technological systems. International
Electrotechnical Commission, Geneva, 1995.

[14] Kaplan, S (1997): The Words of Risk Analysis. Risk Analysis 17(4), 407-417.

[15] Klinke, A. and Renn, O. (2001): Precautionary principle and discursive strategies: classifying
and managing risk. Journal of Risk Research 4 (2), 159-173.

[16] Needleman, K. Methods of valuing life. In Technological Risk. University of Waterloo, 1982

146
[17] ESA. 1991 European Space Agency - ESA, Expert Judgment, Requirements and Methods,
PSS-01-405, Issue 1 Draft 1-Nov. 1991, Noordwijk, the Netherlands.

[18] Cooke, R., Experts in Uncertainty - Opinion and Subjective Probability in Science, Oxford
University Press, 1991.

[19] Gossens, L.H.J., Cooke, R.M., van Steen, J.F.J., Expert Opinions in Safety Studies, Vol. 1:
Final Report, TUDelft, 1989.

[20] van Steen, J.F.J., Oortman Gerlings, P.D., Expert Opinions in Safety Studies, Vol. 2: Litera-
ture Survey Report, TUDelft, 1989.

[21] Hokstad, P. and Øien, K., Expert Judgment - For Assessing Input Data to a Small/medium
Size Reliability Study, Presentation at the Growth Point Centre Group Meeting, Trondheim
1994-08-25, STF75 S94018.

[22] Hokstad, P., Reinertsen, R. and Øien, K., Recommendations on the use of expert judgment
in safety and reliability engineering studies. Recommendations on the use of expert judgment
in safety and reliability engineering studies. Two offshore case studies Reliabiliity Engineering
and System Safety 61 (1998) 65-76. case studies. To appear in Reliability Engineering and
System Safety, 1997.

[23] Klakegg, O. J., The stepwise process (In Norwegian), Institutt for bygg- og anleggsteknikk,
NTH, 1993.

[24] Hauge, S. and Tor Onshus, T (2004). Reliability Data for Safety Instrumented Systems – PDS
Data Handbook, 2010 Edition. SINTEF report SINTEF A13502. Trondheim 1994.

[25] Austeng, K. og Hugsted, R., Stepwise calculation (In Norwegian), Institutt for bygg- og an-
leggsteknikk, NTH, 1995.

[26] The PS 2000 project team (Hokstad, P., Klakegg, O.J., Rosness, R. and Øien, K.)

[27] Kahneman, D., Slovic, P. and Tversky, A., Judgment under Uncertainty. Heuristics and Biases.
Cambridge, Cambridge University Press, 1982.

[28] Svenson, O., On expert judgments in safety analysis in the process industries. Reliability
Engineering and Systems Safety, 25, 219-256, 1989.

[29] Arentz, C., Bakken, J., Kilde, H.S., Klakegg, O.J., Krogh, J., Competence as a steering
parameter - The basis for development (in Norwegian), NTH Rapport NTH 95010, PS2000,
1995.

[30] Norman Dalkey, Olaf Helmer (1963) An Experimental Application of the Delphi Method to
the use of experts. Management Science, 9(3), Apr 1963, pp 458-467

[31] Saaty, T.L. 1980 The Analytic Hierarchy Process: Planning, Priority Setting, Resource Allo-
cation, ISBN 0-07-054371-2, McGraw-Hill

[32] Archer, N. and Chasemzadeh, F. (2004) Project Portfolio Selection and Management. In The
Wiley Guide to Managing Projects Morris, P and Pinto, J.K. (Editors). Willey. ISBN: 978-0-
471-23302-2

147
[33] Høyland, K. and Wallace, SW. (2001) Generating Scenario Trees for Multistage Decision Prob-
lems. Management Science, 47(2), pp 295 - 307

[34] Bjerke, MM. and Bykvist, HR. (2018) Managing Supply Uncertainty in the Operational Pro-
duction Planning in a Whitefish Value Chain - A Stochastic Programming Approach. Master
thesis. NTNU - Norwegian University of Science and Technology.

[35] Birge, JR. and Louveaux, F. (2011). Introduction to Stochastic Programming. Springer Inter-
national Publishing, second edition.

[36] Phillips, DT., Ravindran, A. and Solberg, J. (1976) Operations research: Principles and prac-
tice. John Wiley&Sons. New York.

148
Index

α-percentiles, 29 Dependence, 91
Dependency, 5
Activity on arc, 41 Stochastic, 6
Activity on node, 41 Design matrix, 77
Ageing parameter, 104 Deterministic equivalent problem, 140
Arctic maintenance, 110 Discount rate, 70
Decision tree, 111 Disjoint events, 23
Single activity, 106 Distribution
Maximum values, 38
Bayes theorem, 25
Product, 38
Bayesian methods, 79
Sums, 37
Biasedness, 89
Double expectation, 29
Overestimation, 90
Dynamic programming, 134
Underestimation, 90
Binary variables, 132 Effective failure rate, 104
Binomial distribution, 33 EMax function, 44
Bootstrapping, 99 Engineering judgement, 86
Branch and bound algorithm, 132 Erlang distribution, 32
Bubble diagram, 119 Event, 22
Event uncertainty, 49
Calibration, 90
Expectation, 28
Method for, 96
Project expectation, 9
Central limit theorem, 38
Expert judgement, 37, 81
Checklists, 10
Data, 81
Coherence, 91
Explanaotry variable, 77
Commitment, 9
Exponential distribution, 31
Complementary event, 23
Extensive form, 140
Completeness, 86
Conditional probability, 24 Failure rate function, see Hazard rate
Confidence level, 12
Consistency, 91 Gamma distribution, 32
Cost modelling, 11
Critical path method, 43 Hazard, 10
Cumulative distribution function, 26 Hazard and Operability Study, 10
Hazard rate, 104
Decision trees, 64
Decision variable, 128 Independent events, 24
Degradation rate, 71 Informativeness, 90
Delphi method, 83 Subjective, 90

149
Input state, 134 Conditional, 24
Intersection, 23 Density function, 27
Inverse-Gauss distribution, 34 Total, 24
Inverted gamma distribution, 32 Program Evaluation and Review Technique
(PERT), 43
Life cycle cost, 69 Project portfolio, 118
Life cycle profit, 69 Project selection, 119
Likelihood function, 75 Scheudling, 124
Linear programming, 128 Visualization, 118
Solving by MS Excel, 130 Vulnerability, 120
standard form, 129
Lognormal distribution, 33 Random variable, see Stochastic variable
LP, see Linear programming Regression variable, 77
LS principle, 76 Reproducibility, 92
Residual, 78
Maximum likelihood principle, 36 Resolution, 91
Mean time to failure, see MTTF Return function, 134
Median, 28 Risk, 6
Memoization, 138 Acceptance criterion, 6
Method of moments, 75 Identification, 9
Mixed integer programming, 132 Management, 8
MLE, 74 Picture, 6
Mode, 28 Risk and opportunity reg, 12
Monte Carlo simulation, 46
MTTF, 104 Scenario, 10
Scenario analyse, 83
Net present value, 70 Schedule, 6, 41
Normal distribution, 30 Model, 11
SIMPLEX method, 130
Objective function, 128 Slack variable, 130
Observable quantity, 5 Stakeholder, 118
Output state, 134 Standard deviation, 29
Overconfidence, 91 Stochastic programming, 138
discretization, 140
Parameter, 6 expected value of perfect information,
Estimate, 6 143
Estimation, 74 scenario building, 143
Penalty for default, 48 value of the stochastic solution, 141
PERT Stochastic variable, 6, 26
Distribution, 35 Successive schedule planning, 43
Method, 43 Surplus variable, 130
Poisson distribution, 33
Posterior distribution, 79 Target, 9
Precision, 29 Task analysis, 10
Preliminary hazard analysis, 10 Threat, 10
Preventive maintenance, 103 Total probability, 24
Priror distribution, 79 Triangular distribution, 35
pRisk, 5 Triple estimate, 37
Probability, 6, 22 Turnaround, 103

150
 Changing the frequency, 110 Utility
Exclude activities, 114 Maximising expected, 58
Two stage stochastic programming Utility function, 54
formulation, 139
Variance, 29
VarMax function, 44
Unbiasedness, 89
Venn diagram, 22
Uncertainty
Aleatory, 5, 7 Weibull distribution, 31
Cost and schedule model, 11 Used in maintenance models, 104
Decision under, 52 Weight of data, 99
Epistemic, 5 Weighting of experts, 96
Event, 49 Based on control questions, 97
Underconfidence, 91 Based on knowledge profile, 97
Undesired event, 10 Based on mutual evaluation, 97
Union, 23 Equal weighting, 96

151