Print

Menu

Go Back

Next Page

Estimating Project Cost Contingency Beyond the 10%

syndrome
D.Baccarini
Faculty of Built Environment, Art & Design, Curtin Univeristy of Technology, Perth, Western Australia

ABSTRACT
Projects require budgets to set the sponsors financial commitment and provide the
basis for cost control and measurement of cost performance. A key component of a
project budget is cost contingency. This paper reviews the traditional approach of
estimating project cost contingency by means of a percentage addition to the base
estimate and analyses its many conceptual flaws. Over the past decade there has been
a significant increase in interest by both practitioners and academics into more robust
methods for estimating cost contingency. This paper reviews the literature of some of
these methods for estimating project cost contingency.

INTRODUCTION
The cost performance of construction projects is a key success criterion for project sponsors.
Cost contingency is included within a budget estimate so that the budget represents the total
financial commitment for the project sponsor. Therefore the estimation of cost contingency
and its ultimate adequacy is of critical importance to projects. There are three basic types of
contingencies in projects: tolerance in the specification, float in the schedule, and money in
the budget (CIRIA 1996). This paper deals with project cost contingency. Patrascu (1988:115)
observes, "Contingency is probably the most misunderstood, misinterpreted and misapplied
word in project execution". Contingency has been defined as the amount of money or time
needed above the estimate to reduce the risk of overruns of project objectives to a level
acceptable to the organization (PMI 2004)

CONTINGENCY - ESTIMATION
A range of estimating techniques exist for calculating project cost contingency- see Table 1.
Table 1: Contingency - Estimating methods
Contingency Estimating methods
Traditional percentage
Method of Moments
Monte Carlo Simulation
Factor Rating
Individual risks expected value
Range Estimating
Regression Analysis
Artificial Neural Networks
Fuzzy Sets
Influence Diagrams
Theory of Constraints
Analytical Hierarchy Process

References (Examples)
Ahmad 1992, Moselhi 1997
Diekmann 1983; Moselhi, 1997, Yeo 1990
Lorance & Wendling 1999, Clark 2001
Hackney 1985, Oberlander & Trost 2001
Mak, Wong & Picken 1998; 2000
Curran 1989
Merrow & Yarossi 1990; Aibinu & Jagboro 2002
Chen & Hartman 2000; Williams 2003
Paek, Lee, & Ock, 1993
Diekmann & Featherman 1998
Leach 2003
Dey, Tabucanon & Ogunlana 1994

Menu

Next Page

Go Back

Traditional Percentage
Traditionally cost estimates are deterministic i.e. point estimates for each cost element based
on their most likely value (Mak et al 1998). Contingencies are often calculated as an acrossthe-board percentage addition on the base estimate, typically derived from intuition, past
experience and historical data. This estimating method is arbitrary and difficult to justify or
defend (Thompson and Perry 1992). It is an unscientific approach and a reason why so many
projects are over budget (Hartman 2000). A percentage addition results in a single-figure
prediction of estimated cost which implies a degree of certainty that is not justified (Mak et al
1998). The weaknesses of the traditional percentage addition approach for calculating
contingencies has led for a search for a more robust approach as evidenced by the range of
estimating methods set out in Table 1.
Individual Risks - Expected Value
The amount of contingency reserve can be based on the expected value for individual risk
events. Expected value is the mean of a probability distribution of a risk. For example, the
Hong Kong Governments Works Branch introduced Estimating using Risk Analysis (ERA)
for construction projects for determining contingencies using expected value (Mak et al,
1998; Mak & Picken, 2000). Firstly, a risk-free estimate of known scope is produced then risk
events are identified and costed in terms of an average and maximum risk allowance is
calculated. There are two types of risks:

Fixed Risk These are events that will either happen in total or not at all e.g. whether an
additional access road will be required. If it happens, the maximum cost will be incurred;
if not, then no risk will be incurred. The maximum risk allowance will be the cost if the
risk eventuates, whilst the average cost = maximum cost * probability of its occurrence.

Variable Risk These are events that will occur but the extent is uncertain (e.g. depth of
piling foundations). The maximum risk allowance, which is assumed to have a 10%
chance of being exceeded, is estimated by the project team based on past experience or
records (e.g. most expensive piling at the maximum length). The average risk allowance is
estimated as the value that has a 50% chance of being exceeded, and may have a
mathematical relationship to the maximum or estimated separately. This 50% level is
chosen on the rationale that the worst values for all risks will not occur but rather there
will be swings and roundabouts effects of the totality of the risk events identified.

The summation of all events average risk allowance becomes the contingency. See Table 1:
Risk

Type

Site Conditions

Variable

Additional Space

Fixed

Probability (Fixed Risks only)

70
CONTINGENCY

Average risk ($)

Max. Risk ($)

525,000

1,000,000

11,760,000

16,800,000

12,285,000

Table 1 (adapted from Mak & Picken, 2000)

The accuracy of contingencies was tested for completed ERA and non-ERA projects (i.e.
using traditional percentage addition). It was found that the ERA-derived contingency was
significantly more accurate.

Menu

Next Page

Go Back

Method of Moments
Each cost item in an estimate is expressed by a probability distribution, reflecting the risk
within the cost item. Each cost item distribution has its expected value and variance. The
expected values and variances for all cost items are added to arrive at the expected value and
standard deviation for the total project cost. Total project cost can be assumed to follow a
normal distribution based on the central limit theorem, but only if the cost items are
independent. Then, using probability tables (z scores) for a normal distribution, a contingency
can be derived from the probability distribution based on a desired confidence level i.e. level
of probability of total project cost not being exceeded. For example using table 2, if a sponsor
wants a baseline budget set at the EV of $116.67 and then add contingency that will have a
90% probability of not being exceeded, then the contingency will need be $67.61
Table 2 Method of Moments - Example
Variable (Cost)

Distribution

Foundation
Walls
Roof
TOTAL PROJECT COST

triangular
triangular
triangular
Normal (CLT)

* EV = (a+b+c)/3;

a
Min.
($)
25
40
10

b
Most
Likely
($)
30
60
15

** V = (a2 + b2 + c2 - ab ac bc)/18;

c
Max
($)
65
80
25

EV*

V**

SD***

40.00
60.00
16.67
$116.67

1415.00
1200.00
175.00
$2790

$52.82

*** SD = V

Monte Carlo Simulation (MCS)

MCS allows us to examine what would happen if we undertook many trials of a project. MCS
is a quantitative technique for analysing risk and provides a structured way of setting the
contingency value in a project cost estimate (Clark, 2001). The output of MCS is a probability
distribution for total cost of the project. An example of its application is provided by
Honeywell Performance Polymers and Chemical, which used MCS in 47 projects ranging
from US$1.4m to US$505m (Clark, 2001). Contingency is be set at 50% probability level
(median), based on the rationale that many projects make up the total annual budget, so cost
variations on one may be offset by another project. This approach often yields a
recommended contingency value of less than 5%, or zero for a well-defined project (Clark,
2001). However, for very large or strategic projects, an 80% or 90% probability level is
chosen for contingency; and at a preliminary stage of a project, 95% is usually required.
Many authors (e.g. Lorance & Wendling 1999) advocate assigning contingency so that the
Base Estimate + Contingency = 50/50 point (median) because this establishes a target for the
project team where they have an equal chance of underrunning or overrunning the estimate.
The US Federal Aviation Administration (FAA) uses the median (50/50) point as its budget
baseline, and contingency is the different between the 50% and 80% confidence level (Fenton
et al, 1999).
Estimate Quality
Oberlander & Trost (2001) developed a quantitative model to predict the amount of cost
contingency required based the quality of the project cost estimate and historical cost data.
The accuracy of an estimate depends on four determinants: who is involved in the estimate

Please leave footer empty

Menu

Go Back

Next Page

(16% influence); how the estimate is prepared (23%); what is known about the project (39%);
and other factors considered whilst preparing the estimate (22%). These four determinants
were decomposed into 45 elements used to measure the quality of an estimate. The model was
based on detailed analysis of 67 completed capital projects (US$5.6bn) in the process industry
For the estimate, each element is rated, from 1 (best) to 5 (worst). Examples of these elements
are: relevant experience of estimating team; time allowed for preparing estimate; what is
known about technology.
The score for each element is entered into the model and an overall score for estimate quality
is automatically derived. An ordinary least-square (OLS) fit through the data of the 67
projects provides the basis for predicting of the accuracy of an estimate. The prediction model
is y = mx + b, where y represents the percentage contingency and x represents the estimate
score, m is the slope and b is the intercept. This score then predicts the accuracy of the
estimate; the higher the score, the greater the inaccuracy and therefore the need for more
contingency for a chosen confidence level. For example, a score of 41.8 has a 90% chance of
underrunning, if 34.9% contingency is added to the base estimate. This information can be
used to check the amount of contingency determined by other methods, as well as a method of
predicting its won contingency
Regression Analysis
Regression models have been used since the 1970s for estimating cost (Kim et al, 2004). The
purpose of linear regression is to use the linear relationship between a dependent variable (e.g.
estimated final cost) and independent variables (e.g. location, size) to predict or explain the
behaviour of the dependent variable. Multiple regression analysis is generally represented in
the form: y = a + b1x1 + b2x2 +bnxn . Where y is the total estimated cost; x1, x2, etc are the
measures of variables that may help estimate y; and b1, b2 etc are the coefficients estimated by
regression analysis. The regression equation can then be used to predict the value of a
dependent variable once the values of the independent variables are inserted. For example
Merrow & Yarossi (1990) have y as the estimated cost/actual cost and x variables such as
level of scope definition, and level of unproven technology.
Artificial Neural Networks (ANNs)
ANNs are an information processing technique that simulates the biological brain and its
interconnected neurons (Chen & Hartman, 2000). The structure of ANNs mimics the nervous
system by allowing signals to travel thorough a network of simple processing elements (akin
to neurons) by means of interconnections among these elements. These processing elements
are organised in a sequence of layers consisting of an input layer, followed by one or more
hidden layers and culminating in an output layer. The input processors accept the data into the
ANN (e.g. variables that have a relationship to the amount of cost overrun in projects), the
hidden processors represent relationships in the data and the output layer produces the
required result (e.g. predicted amount of cost overrun). ANNs employ a mechanism to learn
and acquire problem-solving capabilities from training examples by detecting hidden
relationships among data and generalising solutions to new problems. ANNs are suitable for
non-linear modelling of data, which contrasts with the linear approaches using regression
Over the past decade the use of ANNs for cost estimating has grown. ANN can be used to
predict project cost overruns and thereby assist management in developing an appropriate

Menu

Go Back

Next Page

contingency (Chen & Hartman, 2000). Examples of the application of ANNs to predict the
level of cost overrun/underrun include:
Chen & Hartman (2000) used ANN to predict the final cost of completed oil and gas
projects from one organisation using 19 risk factors as the input data. It was found that
75% of the predicted final cost aligned with the actual variance i.e. where the ANN model
predicted an overrun/underrun, an overrun/underrun actually occurred. The prediction
accuracy of ANN outperformed multiple linear regression
Chau et al (1997) used 8 key project management factors to predict the final cost of
construction projects. It was found that more than 90% of the examples did not differ by
more than one degree of deviation from the expected
Gunaydin & Dogan (2004) used 8 design parameters to estimate the square metre cost of
reinforced concrete structure systems in low-rise residential buildings and found that the
ANN provided an average cost estimation accuracy of 93%
The research on the application of ANN to predict cost performance often compares the
accuracy of ANN with multiple linear regression and in most cases ANN produce more
accurate predictions (e.g. Chen & Hartman, 2000; Sonmez, 2004; Kim at al, 2004)
SUMMARY
Traditionally, contingencies are often calculated as an across-the-board percentage addition
on the base estimate, typically derived from intuition, past experience and historical data. This
estimating method is serious flaws. It is usually illogically arrived at and may not be
appropriate for the proposed project. This judgmental and arbitrary method of contingency
calculation is difficult for the estimator to justify or defend. A percentage addition results in a
single-figure prediction of estimated cost, which implies a degree of certainty that is simply
not justified. It does not encourage creativity in estimating practice, promoting a routine and
mundane administrative approach requiring little investigation and decision making.
This paper briefly reviews more robust and justifiable approaches to estimating project cost
contingency. In particular the application of Monte Carlo simulation is growing as project
cost estimators become more aware of its improved effectiveness over the traditional
approach.

REFERENCES
Ahmad I (1992) Contingency allocation: a computer-aided approach AACE Transactions, 28
June - 1 July, Orlando, F.4.1-7.
Aibinu A A and Jagboro G.O (2002). The effects of construction delays on project delivery in
Nigerian construction industry. International Journal of Project Management, 20, 593-599.
Chen D and Hartman F T (2000) A neural network approach to risk assessment and
contingency allocation. AACE Transactions, 24-27th June, Risk.07.01-6
Chua, D.K.H., Kog, Y.C., Loh, P.K., & Jaselskis, E.J. (1997). Model for construction budget
performance neutral network approach, Journal of Construction Engineering and
Management, 214-222
CIRIA (Construction Industry Research and Information Association) (1996) Control of risk:
a guide to the systematic management of risk from construction. London: CIRIA
Clark, D.E. (2001). Monte Carlo Analysis: ten years of experience. Cost Engineering, 43(6),
40-45.
Please leave footer empty

Menu

Go Back

Curran M W (1989) Range Estimating, Cost Engineering, 31(3), 18-26.

Dey, P., Tabucanon, M.T., & Ogunlana, S.O. (1994). Planning for project control through risk
analysis; a petroleum pipelaying project. International Journal of Project Management,
12(1), 23-33.
Diekmann, J.E. & Featherman, W.D (1998). Assessing cost uncertainty: lessons from
environmental restoration projects. Journal of Construction Engineering and Management.
124(6), 445-451
Diekmann, J.E. (1983). Probabilistic estimating: mathematics and applications. Journal of
Construction Engineering and Management, 109(3), 297-308.
Fenton, R.E, Cox, R.A. & Carlock, P. (1999). Incorporating contingency risk into project cost
and benefits baselines; a way to enhance realism. INCOSE Conference
Gunaydin, H.M. & Dogan, S.Z (2004). A neural network approach for early cost estimation of
structural systems of buildings, International Journal of Project Management, 22, 595-602
Kim G.H, An S.N. Kang, K.I (2004). Comparison of construction cost estimating models
based on regression analysis, neutral networks, and case-based reasoning, Building and
Environment, 39, 1235-1242
Hackney J W (1985) Applied contingency analysis. AACE Transactions, B.1-4
Hartman, F.T (2000). Dont park your brain outside. Project Management Institute. Upper
Darby, PA.
Leach L P (2003) Schedule and cost buffer sizing: how to account for the bias between
project performance and your model. Project Management Journal, 34(2), 34-47.
Lorance, R.B. & Wendling, R.V. (1999). Basic techniques for analysing and presenting cost
risk analysis. AACE Transactions, Risk.01.1-7
Mak, S, & Picken, D (2000). Using risk analysis to determine construction project
contingencies. Journal of Construction Engineering and Management, 126(2), 130-136
Mak, S, Wong, J, & Picken, D (1998). The effect on contingency allowances of using risk
analysis in capital cost estimating: a Hong Kong case study. Construction Management and
Economics, 16, 615-619.
Merrow E W and Tarossi, M.E (1990) Assessing project cost and schedule risk, AACE
Transactions, H.6.2-7
Moselhi, O (1997). Risk assessment and contingency estimating in AACE Transactions
Dallas, 13-16th July, D&RM/A.06.1-6
Oberlender, G.D. & Trost, S.M. (2001). Predicting accuracy of early cost estimates based on
estimate quality. Journal of Construction Engineering and Management. 127(3), 173-182
Patrascu A (1988) Construction cost engineering handbook. New York: M. Dekker.
PMI [Project Management Institute] (2004) A guide to the project management body of
knowledge. 3rd Edition, Newtown Square: PMI.
Sonmez, R (2004). Conceptual cost estimation of building projects with regression analysis
and neural networks, Canadian Journal of Civil Engineering, 31, 677-683.
Thompson P A and Perry J G (1992). Engineering construction risks. London: Thomas
Telford.
Williams T P (2003) Predicting final cost for competitively bid construction projects using
regression models, International Journal of Project Management, 21, 593-599
Yeo, K.T. (1990). Risks, classification of estimates, and contingency management. Journal of
Management in Engineering. 6(4), 458-470.