Chemical Engineering December 2014 - 42

Cover Story
Table 5. TyPIcal SySTemIc conTIngency allowanceS baSed on ProjecT aTTrIbuTeS
(owner PerSPecTIve)
Systemic contingency as percentage of unexpended base estimate
Scope definition
Complexity
class 3
Low
Technology
Medium
High
Impurities
Bias
Teams/system
Fixed
Best (p10)
Worst (p90)
Low
3
6
15
Medium
8
11
20
High
12
15
24
Low
10
13
22
class 4
Medium High Low
20
15
18
27
23
32
adjustments to the above (select from within the ranges):
Feedstock or intermediate streams have solids, contaminants, corrosives, and so on: +5 to +10
Aggressive base/target pricing: +10; Conservative base/small plant-based/punitive: -10
Immature/weak/indecisive/understaffed: +10; Mature/strong/decisive/full staff: -10
Separate out costs based on firm-priced quote and use 50% of above for that portion
After making adjustments above, the best case for sensitivities (p10) is about -0.5×
After making adjustments above, the worst case for sensitivities (p90) is about 3×
comes. With a specific distribution
and base estimate value in mind,
contingency is simply the cost that
must be added to get a desired probability
of underrunning the total, as
shown in Figure 4. Management
will decide on their desired probability.
Probability of underrun
levels from p50 to p70 are common,
depending on how risk-tolerant or
risk-averse the company's management
is. Note that contingency has
no effect on the project's cost distribution
or absolute cost range. Those
are fixed attributes of the project
and its uncertainty profile as it
stands at that point in time. From
the distribution, a worst-case value
can also be obtained for sensitivity
analysis (for instance, p90).
So how can engineers come up
with a probabilistic cost distribution
like the one shown in Figure
4? Since the 1980s, when the first
Monte-Carlo simulation (MCS)
add-on for spreadsheets was introduced,
the traditional answer was
a method called line-item ranging
(LIR). This method involves assigning
low and high range values
(ends of triangular distributions) to
one's cost-estimate line items, and
running the MCS software. However,
LIR has been discredited by
research. It fails because it violates
a first principle of risk analysis; it
does not " quantify the risks. " LIR
does not make a risk-to-impact
link. Teams cannot intuitively, and
certainly not item by item, make
the risk-to-impact leap. In terms of
statistical significance, LIR tends
to produce almost the same curve
every time it is used - the red curve
of Figure 1. Research published by
IPA Inc. shows that LIR results in
9% contingency with ±4% standard
deviation, regardless of actual risks
[4]. The problem is not MCS itself;
the problem is that the LIR model
only addresses " estimating process
quality " risk. However, like a broken
watch, it does get the number right
occasionally, such as when there are
no other project risks.
Without using LIR, what is an
engineer to do? Rather than fall
back on a rule-of-thumb percentage,
AACE recommends several riskquantification
methods that apply
first principles in RP-40R-08 [5]. Two
methods are readily available, and
can be applied without specialized
software and training, although, on
major projects, using experts for risk
analysis is highly recommended.
Additionally, these two methods
can provide an average contingency
value without probabilistic modeling,
if desired. These AACE methods
are covered in the following RPs:
* 42R-08: Risk Analysis and Contingency
Determination Using
Parametric Estimating
* 44R-08: Risk Analysis and Contingency
Determination Using
Expected Value
Two methods are needed because
the two very different types of risk,
systemic and project-specific, must
be addressed. The Parametric Estimating
method is used to quantify
systemic risks, because their impacts
are only knowable through
analysis of historical experience.
The Expected Value method is used
to quantify project-specific risks, because
their occurrence and impacts
are readily estimable by the team.
Quantification: point value only
Parametric estimating uses a predetermined
equation in which the
predicted average contingency percentage
(to apply to the base estimate)
is the equation outcome (the
dependent variable), and numeric
ratings of the systemic risks are the
inputs (the independent variables),
as shown in Equation (1).
MSC, % = Constant + A × (SDR) + B
× (CR) + C × (EBR) + (OR)
(1)
Equation (1) follows a typical
parametric model or output of a regression
analysis in the following
form: Value = constant + coefficient
A × parameter A + coefficient B × parameter
B, and so on. In this case, to
obtain the mean systemic-risk contingency
(MSC) percentage value,
the parameters are quantified values
of scope definition rating (SDR),
complexity rating (CR), estimate
bias rating (EBR) and the product of
any other coefficients and systemic
ChemiCal engineering www.Chemengonline.Com DeCemBer 2014 41
19
22
32
class 5
Medium High
24
27
37
29
32
42

http://www.Chemengonline.Com

Chemical Engineering December 2014

Table of Contents for the Digital Edition of Chemical Engineering December 2014