Use of Fuzzy Logic For Predicting Design Cost Overruns
Use of Fuzzy Logic For Predicting Design Cost Overruns
Use of Fuzzy Logic For Predicting Design Cost Overruns
on Building Projects
Karla Knight1 and Aminah Robinson Fayek, A.M.ASCE2
Downloaded from ascelibrary.org by UNIVERSIDADE FEDERAL DE PARANA on 07/16/13. Copyright ASCE. For personal use only; all rights reserved.
Abstract: This paper describes a model for use in predicting potential cost overruns on engineering design projects. The output of the
model is useful in assessing the amount of possible risk on a project and the likelihood of making a profit on the job. The model is
intended for use by engineering consultants, i.e., structural, mechanical, and electrical engineers, in the building design industry. This
research uses fuzzy logic to model the relationships between the characteristics of a project and the potential risk events that may occur,
and the associated cost overruns caused by combinations of the project characteristics and risk events. This paper discusses the topic of
scope creep and scope definition, which are significant causes of cost overruns on design projects; identifies the project characteristics and
risk events used in the model; explains the structure of the model and the use of fuzzy logic; and provides recommendations for future
research.
DOI: 10.1061/共ASCE兲0733-9364共2002兲128:6共503兲
CE Database keywords: Cost control; Decision making; Fuzzy sets; Project management; Construction.
Introduction ing can cost almost 10% of the total project cost, according to
Bubshait et al. 共1999兲. Many factors, then, that affect design costs
The structural design, mechanical design, and electrical design of will affect the entire project, and therefore impact the overall
a building are important phases in a project. The engineers and project costs.
designers must produce a design within a given time frame that Research on construction design is limited by the lack of docu-
meets the needs of the building users, is straightforward in its mented data available from previous projects. The model pre-
constructability, contains no design errors, and is within certain sented in this paper attempts to predict design cost overruns or
cost limitations both in creating the design and in constructing the cost savings using data that are realistic to obtain. It identifies the
facility. The designers carry a large responsibility in properly main factors that contribute to escalating costs for the design
managing the project design in order to achieve a successful consultant on construction projects. A list of the main project
project both by their own standards and those of the client. characteristics and risk events that affect design projects has been
A design firm must apply good project management tech- compiled. The model uses fuzzy logic to relate the existence of
niques and practices in order to maintain a well-managed and these project characteristics and the occurrence of these risk
successful project. Poor project management, among other fac- events, to determine the overall cost impact on the design phase
tors, can lead to negative cost impacts for the design consultant of the project.
and the entire project team. Although the design phase represents
a small portion of the overall cost of a construction project, Mc-
George 共1988兲 believes that ‘‘constructability and value engineer- Scope Definition
ing 关in the design phase兴 yield construction cost savings of ten to
twenty times the cost of the extra design input.’’ For example, A lack of scope definition at the onset of a project is one of the
McGeorge claims that an increase in design input of 50% will main causes of cost overruns during design. The Construction
lead to a 10% savings in construction costs. A better design and Industry Institute’s 共CII兲 publication on scope definition and con-
better management of the design process will lead to a more trol 共CII 1986兲 ranks the lack of scope definition during engineer-
organized and controlled project, and therefore a more successful ing as having the second highest impact on cost overruns. Dysert
project. Design deviations account for approximately 60% of con- 共1997兲 claims that ‘‘poor scope definition at the estimate stage
struction project deviations, and poor quality design and engineer- and loss of control of project scope’’ are the most frequent con-
tributors to cost overruns. A poorly defined project is subject to
1
Project Manager, Earth Tech 共Canada兲, Inc., 17203-103 Ave., changes initiated by the client that will require extra work and
Edmonton, AB, T5S 1J4, Canada. effort by the design team to complete.
2
Associate Professor, Dept. of Civil and Environmental Engineering, With a poorly defined scope there is no baseline against which
Univ. of Alberta, Edmonton, AB, T6G 2G7, Canada. changes can be evaluated and monitored to identify those that are
Note. Discussion open until May 1, 2003. Separate discussions must not within the original scope of work. According to Dumont et al.
be submitted for individual papers. To extend the closing date by one
共1997兲, these changes may result in cost overruns and a greater
month, a written request must be filed with the ASCE Managing Editor.
The manuscript for this paper was submitted for review and possible potential for disputes. Dumont et al. also claim that these changes
publication on March 28, 2001; approved on August 21, 2001. This paper may ‘‘delay the project schedule, cause rework, disrupt project
is part of the Journal of Construction Engineering and Management, rhythm, and lower the productivity and morale of the workforce’’;
Vol. 128, No. 6, December 1, 2002. ©ASCE, ISSN 0733-9364/2002/6- but an increased level of scope definition will ‘‘improve the ac-
503–512/$8.00⫹$.50 per page. curacy of cost and schedule estimates as well as the probability of
everyone is working toward the same goal. Misunderstandings predict cost overruns and underruns, as a percentage of the con-
and misinterpretations between parties can cause problems such tract fee, based on the degree of existence of certain project char-
as rework and extra work, and will invariably cause disputes. acteristics and the degree of occurrence of certain risk/
Good communication, good organization and control, and proper opportunity events. The model uses 13 project characteristics and
scope definition are therefore important elements of a successful eight risk events that are rated by the user in reference to a spe-
project. cific project. Standard strengths within the model represent the
relationships between the project characteristics and risk events,
and the potential cost overrun/underrun that combinations of
Previous Studies on Design Cost Overruns these factors may cause. Based on the user input ratings of the
project characteristics and risk events, the model calculates the
Literature on design cost control and cost overruns is very lim- predicted cost overrun or underrun above or below the consult-
ited. A study by Cox et al. 共1999兲 reviews the causes and effects ant’s fixed engineering fee. A cost overrun, as defined in the
of design changes on a construction project, and Glavinich 共1995兲 model, occurs when the actual cost of the work 共including profit
reports on the issue of constructability and its effect on the con- and overheads兲 to the design firm is above the fee paid to the
struction process. Love et al. 共2000兲 created a model that simu- design firm by the client to complete the work. It is assumed that
lates the causes and effects of design errors. The CII has a pub- a project fee is estimated at the onset of the project and this fee is
lication entitled ‘‘Scope definition and control’’ 共CII 1986兲 that set as the maximum upset price for the design work.
discusses the effects of a poorly defined scope. The CII has also The data used to build the model are based on previous
prepared a ‘‘project definition rating index’’ 共PDRI兲 for industrial projects and information obtained from project managers within a
projects 共CII 1996; Dumont et al. 1997兲, and a PDRI for commer- local leading design firm. The model is applicable to the work of
cial building projects 共CII 1999兲. The PDRI for building projects structural, mechanical, and electrical design engineers performing
has a checklist of scope definition elements and a score sheet that work in the commercial building construction sector. The engi-
allows the user to rate each element to determine the level of neering design firm should be either the consultant 共i.e., a con-
scope definition on the project. The risks to a consultant due to sultant to the prime consultant; typically the design firm兲 or the
the client have also been researched and modeled by Kometa prime consultant 共i.e., a consultant directly to the client; typically
et al. 共1995, 1996兲. the architect兲, and the contract a guaranteed maximum price con-
Many of the factors that affect design costs have been identi- tract where the fee is estimated by the project manager or deter-
fied and studied, but no tools are currently in place for predicting mined as a percentage of the total project cost. The project size
design cost overruns. Project management of the design phase is can vary, but the model is most applicable to projects with an
underresearched in comparison to the amount of research that has engineering fee in the $100,000–$500,000 range.
been done on project tracking, monitoring, and control for the The model is intended to be used during the proposal stage of
construction phase of a project. Factors for the proposed model the project 共i.e., as the consultant is preparing the fee estimate兲 or
were chosen partly through literature that reviewed the issues that at the onset of the consultant’s work on the project. The model is
are important to design work, and partly through discussion with intended for use in conducting a risk analysis or for project
project engineers and other personnel working with a design firm. awareness, and not as a guide for estimating. The consultant’s
estimate of the fee should be complete at this point, and the model
will identify the sensitivity of the project to certain characteristics
and risk events.
Fuzzy Logic in Construction
Fuzzy logic is the modeling technique used in the model to pre- Framework of Model
dict design cost overruns. Fuzzy logic has become an increasingly
popular modeling tool in construction research. Fuzzy logic and The structure of the model is shown in Fig. 1, and each of its
fuzzy expert systems are used increasingly in situations where components is discussed next.
little deterministic data are available. ‘‘Fuzzy set theory was origi-
nally devised to model uncertainty associated with human percep-
Project Characteristics
tion or subjective probability judgments’’ 共Nguyen 1985兲.
Due to the imprecise nature of many factors that affect con- There are many project characteristics that can potentially affect
struction projects, and a general lack of data for proper quantifi- the costs of a project, both in a negative way and in a positive
cation of factors, fuzzy logic lends itself well to many construc- way. The project characteristics that are most recognizable in their
tion applications. The measurement of construction factors is effect on project costs, and the project characteristics that poten-
often subjective and uncertain, where actual data are not available tially have the greatest impact on project costs have been identi-
or when the data come from the experience and judgment of those fied in consultation with the collaborating design firm. The
in the industry. For this reason, fuzzy logic is being used more project characteristics ( P i ) used in this model are the following:
C5 0 0 Zero
C6 0 0 Zero
C7 0–5 0–20 Low
C8 5–10 20–50 Medium
C9 10–20 50– 80 High
Fig. 1. Structure of model for predicting design cost overruns
C 10 20–100 80–100 Very high
1. Willingness of the prime consultant to approach the client 7. Inadequate design team resources, and
for extra fees, 8. Adequacy of the general contractor and subcontractors.
2. Time taken by the client/prime consultant/architect/ As with the project characteristics, each of these risk events is
engineer to make decisions, given a rating by the user between 0 and 10. This rating, wR k ,
3. Knowledge base of the client, represents the extent to which the risk event, R k , is likely to
4. Level of project scope definition between the consultant occur and the extent of its impact on the specific project. A rating
and the client/prime consultant at the proposal stage, of 0 indicates the worst case, and a rating of 10 is the best case. A
5. Definition of scope duties passed on by the consultant’s rating of 5 indicates that the risk event is likely to have a neutral
project manager to the design team, effect on project costs.
6. Experience of the consultant’s project manager, The model proposed here does not take into account the prob-
7. Experience of the prime consultant’s project lead or project ability of the risk event occurring given that the project charac-
manager, teristics exist to a certain degree, but rather the extent of the
8. Skill set of the consultant’s design team, combined impact of the existence of project characteristics and
9. Skill set of the prime consultant’s design team, the occurrence of risk events. The project characteristics and risk
10. Experience of the project team with similar projects, events are considered independent factors in the model, with each
11. Project complexity, combination of project characteristic and risk event having a po-
12. Timeline for design and construction, and tential impact on design costs.
13. Project location.
Each of these project characteristics is given a rating between Percent Cost IncreaseÕDecrease
0 and 10 by the user. This rating, w P i , represents the degree of
existence of the project characteristic, P i , on a specific project. A Each combination of project characteristic and risk event results
rating of 0 indicates poor, and a rating of 10 is an excellent rating. in a potential cost increase or decrease (C m ) above or below the
A rating of 5 indicates that the project characteristic has a neutral contract fee, expressed as a percentage of the contract fee. This
effect on the project, or that the characteristic exists in a state that percentage is the expected cost overrun, above the contract fee,
is typical or expected on a project. given that the project characteristic exists to the worst degree
possible and the risk event occurs to the worst degree. Or, for a
best-case scenario, the percentage is the expected cost underrun,
Risk Events
below the contract fee, given that the project characteristic exists
Risk events are undesirable or fortuitous events that may occur to the best degree possible and the risk event occurs to the best
during the design and construction phases of a project, potentially degree. Each cost increase/decrease, or C m , is a percentage range
leading to a cost overrun or savings, respectively. Risk should that has been identified through expert opinion, as shown in Table
always be taken into account on a project, but due to the com- 1, and is hard-coded in the model. The ranges each have a lin-
petitive nature of the industry not every possible risk event can be guistic descriptor ranging from very high negative to very high
accounted for in the project fee. This puts the onus on the con- positive. The expert opinions were solicited by means of a panel
sultant and the contractor to deal with, and possibly cover the cost discussion with five project managers within the collaborating
of, both negative and positive risk events. Eight risk events have design firm.
been selected in consultation with the design firm as being the
most common or having the largest impacts on projects. The cho- Standard Strengths
sen risk events (R k ) are
1. Accuracy of site investigation, There are two sets of standard strengths in the model—both of
2. Design errors or omissions, which were identified by the same panel of experts described
3. Design/scope changes by the client, consultants, or architect, earlier. The standard strengths are hard-coded in the model. The
4. Communication among the project team, first, SS 1 , relates the project characteristics and risk events; the
5. Overengineering, second, SS 2 , relates the risk events and cost ranges. The standard
6. Constructability issues, strength SS 1 represents the sensitivity of the impact of the risk
strength of 1.0. The neighboring cost ranges receive strengths Once these two binary relations are formed, they are combined
decreasing outwardly in increments of 0.2. These incremental val- to produce a third binary relation using the composition opera-
ues give the decreasing likelihood of the cost ranges occurring. tion. This operation, shown in Eq. 共2兲, yields the fuzzy binary
These standard strengths account for the fact that the chosen cost relation Q( P,C), which relates the project characteristics to the
range may be the most likely to occur, but the other ranges may percent overrun/underrun through their respective relationship to
also occur but with less likelihood. the risk events
Q 共 P,C 兲 ⫽SoF 共 P,C 兲 ⫽S 共 P,R 兲 oF 共 R,C 兲 (2)
Use of Fuzzy Set Theory in Model Two types of composition operations are used and compared—
The model presented in this paper uses fuzzy binary relations to the maximum-minimum 共max-min兲 composition operation and
approximate the relationships between the project characteristics the cumulative-minimum 共cum-min兲 composition operation. The
and risk events in order to predict the expected cost overrun or max-min composition operation, shown in Eq. 共3兲, determines the
underrun. Binary relations are a fuzzy logic technique that in- most likely solution based on the strongest indicator or piece of
volves approximating the relationship between two data sets evidence. The cum-min composition operation, shown in Eq. 共4兲,
given the degree of association between the sets. The advantage takes into account each factor that points to the output value and
of using binary relations is that they do not rely on membership increases the strength with which that output value is recom-
functions, which can require substantial data sets formed on ex- mended based on all supporting evidence. The accuracy of each
pert opinions. This is unfortunately also the biggest weakness of of these two composition operations is compared during the
binary relations, because the user must provide ratings instead, model testing and sensitivity analysis
which are often subjective and relative to the user’s context. The SoF 共 P,C 兲 ⫽max-min关 S 共 P,R 兲 ,F 共 R,C 兲兴 for all R (3)
method used to overcome this weakness was to solicit opinions
from a panel of experts, as described, and to base the values used SoF 共 P,C 兲 ⫽sum-min关 S 共 P,R 兲 ,F 共 R,C 兲兴 for all R (4)
on a consensus of opinion. A fuzzy binary relation can be repre- After the composition operation is used to combine the
sented by a matrix, with the values of the matrix representing the S( P,R) and the F(R,C) relations to obtain the Q( P,C) relation,
values of the membership grades between sets. The fuzzy relation the total strength of each percentage range is calculated using Eq.
allows for partial membership, as opposed to the crisp binary 共5兲. This equation takes the sum of the strengths for each percent-
relation, which only allows for values of 0 or 1, or in other words, age range, and divides the sum by the total of the project charac-
either the presence or absence of association 共Klir and Yuan teristic ratings. This operation for combining the strengths is
1995兲. analogous to the statistical concept of weighted means. The
The first binary relation formed is between the project charac- ‘‘strengths’’ are the membership values of the Q( P,C) relation
teristics and risk events, called the S( P,R) relation. This relation for each percentage range. The highest strength points to the most
takes into account the user ratings of the project characteristics likely range of percent cost overrun/underrun, based on the stan-
and the risk events for the specific project under analysis, and the dard strengths within the model and the user input ratings for the
preset standard strengths relating project characteristics and risk specific project under analysis
events. The S( P,R) relation is calculated using Eq. 共1兲
兺 in⫽1 Q 共 P n ,C m 兲
S 共 P,R 兲 ⫽w P i * wR k * SS 1ik (1) Q 共 P,C m 兲 ⫽ (5)
兺 in⫽1 w P n
where S( P,R)⫽fuzzy binary relation between the project charac-
teristics and risk events, from 0.0 to 1.0; w P i ⫽rating of project where Q( P,C m )⫽total strength with which cost range C m is ex-
characteristic i from 0 to 10, input by the user and scaled by a pected to occur; 兺 Q( P n ,C m )⫽sum of the membership values
factor of 10; wR k ⫽rating of risk event k from 0 to 10, input by for each C m element from the Q( P,C) relation; and 兺 w P n ⫽sum
the user and scaled by a factor of 10; and SS 1ik ⫽standard strength of the user input project characteristic ratings.
between project characteristic i and risk event k, from 0.0 to 1.0. The output of the model can now be defuzzified to obtain a
The second binary relation, the F(R,C) relation, relates the single 共crisp兲 value, C * . Defuzzification methods use the strength
project characteristics and risk events to percentage cost overruns/ for each percentage range to recommend the most likely cost
underruns, and is equal to values ranging from 0.0 to 1.0 derived percentage, either by calculating the area under the curve using
based on SS 2 , the preset standard strengths relating risk events the center of area 共COA兲 method, or simply by examining the
and cost ranges. For each combination of project characteristic percentage range with the highest strength 关the mean of maxi-
and risk event, a most likely range of percentage cost overrun and mum, largest of maximum 共LOM兲, smallest of maximum 共SOM兲,
underrun has been identified through expert opinion. The ranges and middle of maximum 共MOM兲 methods兴. The center of area
of cost overruns and underruns used are shown in Table 1. The method, shown in Eq. 共6兲, takes the centroid, or midpoint, of each
membership values, or standard strengths, for this matrix are val- range and multiplies it by its membership value, then divides by
ues decreasing outwardly by increments of 0.2, starting with a the sum of the membership values. The values for LOM, SOM,
max-min LOM 10 4 2 4
cum-min SOM 8 5 3 2
max-min SOM 9 4 2 4
共b兲 Small projects 共⭐$55,000兲
cum-min COA 14 3 1 4
max-min COA 15 2 2 4
cum-min MOM 9 4 3 1
max-min MOM 10 3 4 1
cum-mina LOMa 9a 7a 1a 0a
max-min LOM 8 7 1 0
cum-min SOM 13 4 4 0
max-min SOM 13 4 4 0
Note: COA⫽center of area, MOM⫽middle of maximum, LOM⫽largest of maximum, and SOM⫽smallest of maximum.
a
Best prediction results.
scenario were therefore quite low; all except two of them were tion; the defuzzification methods tested were the COA, MOM,
below 5, limiting the influence of the underruns on the project LOM, and SOM, as discussed earlier. Table 2 shows the results of
output. The first calibration to the model was to normalize the model testing for each combination of composition operation and
standard strengths of the best-case model. Two of the existing defuzzification method, after calibration of the model. The cum-
standard strengths were 8, which became 10, and the remaining min composition operation and the LOM defuzzification method
strengths at 5 and below were doubled. This calibration increased provide the best prediction results 共footnoted items in Table 2兲.
the influence of the best-case model and enabled the model to For large projects, the average absolute difference between the
predict cost underruns without changing the relationships between actual and the predicted cost overrun/underrun was 9%; for five
the factors in the best-case model. of these projects, the model predicted the same linguistic descrip-
The next problem with the model was that it was underpredict- tor for the magnitude of the overrun/underrun as the actual term;
ing certain overruns and underruns. It was observed that the ma- for two of the remaining five projects, the linguistic term predic-
jority of the projects with the high cost overruns and underruns tion was only one term off from the actual. For small projects, the
were smaller projects. This is logical, since the percent overrun/ average absolute numerical difference was 9%; seven of the
underrun is relative to the size of the project. For example, on a project overruns/underruns were accurately predicted in linguistic
small project of $10,000, if the project is $2,000 overbudget this terms, and one project was one term off.
represents a 20% cost overrun. However, on a larger project of
$100,000, this would only be a 2% overrun, which is considered
quite low. Therefore, it was necessary to separate the large and the Sample Project
small projects, and adjust the ranges used on the small projects.
This calibration addressed the problem of underprediction in the The following sample project illustrates the calculations per-
majority of cases. formed in the model and the use of the model. The use of the
Upon examination of the data, $55,000 共design fee兲 was cho- model is illustrated by screens captured from the software imple-
sen as the cutoff between large and small projects. This put eight mentation of the model using Microsoft Visual Basic 6.0 共Mi-
projects into the small project category, and 10 into the large crosoft Corp., Redmond, Wash.兲; Fig. 2 illustrates the required
project category. The cost ranges for the small projects were then user input, and Fig. 3 illustrates the output of the model. For the
amended to attempt to more accurately predict their cost overruns sample project calculations, the full range of factors is not used in
and underruns. The new ranges chosen, based on the given data, order to simplify the calculations and reduce the size of the ma-
are shown in Table 1. These values were chosen to best fit the trices. The output of the sample calculations therefore does not
existing data, and were not chosen by experts, as with the original match the output of the model shown in Fig. 3. The sample
cost ranges. project is the construction of a clubhouse for the local cross-
country ski group in a small remote town in northern Alberta,
Canada, with a design fee of $84,300. It is therefore analyzed
Validation Results and Sensitivity Analysis
using the model for large projects. Three project characteristics,
Two composition operations and four defuzzification methods time taken to make decisions ( P 1 ), project location ( P 2 ), and
were tested during the model validation. The composition opera- experience of the consultant’s project manager ( P 3 ), and three
tions tested were the cum-min operation and the max-min opera- risk events, accuracy of the site investigation (R A ), adequacy of
the general contractor and subcontractors (R B ), and inadequate verts this to a degree of existence (w P 1 ) for the best-case sce-
design team resources (R C ), are used for illustration purposes. nario of 0.2, and 0.8 for the worst-case scenario. P 2 , project
Sample calculations are provided at each step. location, receives a rating of 0 because the project is in a small
town 6 h north of the consultant’s office (w P 2 ⫽0 for the best-
Step 1 case scenario, and 1.0 for the worst case兲. P 3 , the experience of
the consultant’s project manager, receives a rating of 6, as the
Identify the user input ratings for the project characteristics and engineer is an intermediate engineer with two years of experience
risk events. The user rates each project characteristic and risk as a project manager (w P 3 ⫽0.6 for the best-case scenario, and
event on a scale from 0 共worst case兲 to 10 共best case兲. The model 0.4 for the worst case兲.
scales the input ratings by a factor of 10, giving numbers between Next, the ratings for the risk events are determined. R A , the
0 and 1.0 for the best-case scenario, and then calculates the accuracy of the site investigation, receives a rating of 6 because
complement of these numbers for the worst-case scenario. typically soils in this area have been slightly unpredictable. The
For the sample project, P 1 , the time taken to make decisions, degree of occurrence for the best-case scenario, wR A , is 0.6, and
receives a rating of 2 because the owners are a volunteer organi- 0.4 for the worst-case scenario. R B , the adequacy of the general
zation with 12 members on the project committee and are very contractor and subcontractors, is given a rating of 1 because the
slow at making decisions. As explained before, the model con- contractors in this remote location are unknown and all contrac-
tors are presently busy with other work, so it will be difficult to
get a good contractor for this project (wR B ⫽0.1 for the best-case
scenario, and 0.9 for the worst-case scenario兲. The final risk
event, R C , inadequate design team resources, receives a rating of
5, as the design firm is also presently busy, so its designers’ com-
mitment to the project may be low at times, although they have
enough resources to complete the project (wR C ⫽0.5 for both the
best-case scenario and the worst-case scenario兲.
Step 2
2C 0.9 C 10 3A 0.6 0.8 1.0 0.8 0.6 1.0 0.8 0.6 0.4 0.2
3A 0.3 C6 3B 0.4 0.6 0.8 1.0 0.8 0.6 0.8 1.0 0.8 0.6
3B 0.6 C8 3C 0.2 0.4 0.6 0.8 1.0 0.8 1.0 0.8 0.6 0.4
3C 0.7 C7
共b兲 Best case Step 5
1A 0.6 C2 Apply the maximum-minimum composition operation and the
1B 0.5 C3 cumulative-minimum composition operation to obtain the
1C 0.5 C3 Q( P,C) relation 共Table 6兲.
2A 0.8 C1
2B 0.3 C4 Sample Calculation Using max-min
2C 0.4 C2
3A 0.2 C3 Q 共 P i ,C m 兲 ⫽max-min关 S 共 P i ,R k 兲 ,F 共 R k ,C m 兲兴
3B 0.1 C4 Q 共 P 1 ,C 6 兲 ⫽max-min关共 0.32,0.40兲 , 共 0.58,0.40兲 , 共 0.36,0.60兲兴
3C 0.1 C5
⫽max共 0.32,0.40,0.36兲 ⫽0.40
Calculate the values of the S( P,R) relation 共Table 4兲. Sample Calculation Using cum-min
Step 4 Step 6
Calculate the values of the F(R,C) relation 共the most likely cost Combine the best-case and the worst-case models by combining
overrun/underrun receives a value of 1.0, and the surrounding the strengths of each cost range calculated in the Q( P,C) relation
costs receive values decreasing outwardly by increments of 0.2兲
共Table 5兲.
Table 6. Membership Values of Q( P,C) Relation for Sample Project
Best Case Worst Case
Q( P,C) C1 C2 C3 C4 C5 C6 C7 C8 C9 C 10
Table 4. Membership Values of S( P,R) Relation for Sample Project
共a兲 max-min
S( P,R) A B C
1 0.07 0.07 0.07 0.07 0.07 0.40 0.58 0.58 0.58 0.58
共a兲 Worst case 2 0.00 0.00 0.00 0.00 0.00 0.45 0.45 0.45 0.45 0.45
1 0.32 0.58 0.36 3 0.07 0.07 0.07 0.07 0.07 0.22 0.22 0.22 0.22 0.22
2 0.16 0.45 0.45 Sum/ 兺 w P 0.18 0.18 0.18 0.18 0.18 0.49 0.57 0.57 0.57 0.57
3 0.05 0.22 0.14 共b兲 cum-min
共b兲 Best case 1 0.13 0.13 0.13 0.13 0.13 1.08 1.26 1.26 1.26 1.26
1 0.07 0.01 0.05 2 0.00 0.00 0.00 0.00 0.00 0.81 1.01 1.06 1.06 1.01
2 0.00 0.00 0.00 3 0.11 0.11 0.11 0.11 0.11 0.40 0.40 0.40 0.40 0.40
3 0.07 0.01 0.03 Sum/ 兺 w P 0.30 0.30 0.30 0.30 0.30 1.04 1.21 1.24 1.24 1.21
0.24a 0.24a 0.24a 0.24a 0.24a 0.84a 0.98a 1.00a 1.00a 0.98a suit either new buildings or renovations, for example. The size of
a
Normalized. the projects applicable to the model should be further defined, and
possibly the engineering discipline 共e.g., mechanical, electrical,
structural, and so on兲. An alternative to predicting a percent cost
共for both the max-min and the cum-min composition operations兲, overrun/underrun would be to predict an actual dollar value over-
and normalize the values of the matrix. The values are normalized run or underrun.
to prevent having recommendations greater than 1.0 and to ensure The model would benefit from future research by expanding it
that at least one recommendation equals 1.0 共Table 7兲. to include more project characteristics and risk events, and by
refining the standard strengths used in the model. A more elabo-
rate survey may provide more accurate standard strengths that are
Step 7
based on real projects instead of a consensus of opinions. There
Apply the center of area method of defuzzification. are also other factors that could be taken into account, such as the
bidding conditions, whether the owner of the project was public
Sample Calculation from max-min Composition Operation or private, and the project delivery method—all of which could be
†Eq. „6…‡ used to further classify projects.
冒兺
A key factor in the accuracy of the model is the accuracy of
C *⫽ 兺 m *C m C m ⫽ 关共 ⫺35% * 0.32兲 ⫹ 共 ⫺15% * 0.32兲 the estimated contract fee that is used to calculate the actual per-
cent cost overrun/underrun. Further research should examine
⫹ 共 ⫺7.5% * 0.32兲 ⫹ 共 ⫺2.5% * 0.32兲 ⫹ 共 0% * 0.32兲 methods of making the fee estimates more rigorous and accurate.
Another solution would be to include a factor within the model
⫹ 共 0% * 0.86兲 ⫹ 共 2.5% * 1.00兲 ⫹ 共 7.5% * 1.00兲 ⫹ 共 15% * 1.00兲 that identifies that amount of time and preparation that was put
into the estimate of the fee, the project manager’s confidence in
⫹ 共 60% * 1.00兲兴 / 共 0.32⫹0.32⫹0.32⫹0.32⫹0.32⫹0.86 his or her estimate, or a rating of the project manager based on the
accuracy of his or her previous estimates, to assess the likely
⫹1.00⫹1.00⫹1.00⫹1.00兲 ⫽10.2% accuracy of the estimated fee. The estimates from previous
The output of this sample project predicts that the conditions of projects could be compared to the actual costs of the projects to
this project will cause a 10.2% cost overrun above the estimated assess the project manager’s historical estimating accuracy.
contract fee, using the max-min composition operation, or an One of the possible inaccuracies in the data used in the model
11.6% cost overrun above the estimated contract fee using the is that the project ratings are not always evaluated on the same
cum-min composition operation. scale, because it is subjective rather than objective. Increased uni-
formity and objectivity of the project ratings could reduce this
inaccuracy. Categories could be developed for the rating of each
Conclusions and Future Development factor, and the user need only choose the appropriate category
into which his or her project factors fall. For example, the expe-
The use of fuzzy logic to predict cost overruns/underruns yields a rience of the consultant’s project manager and the prime consult-
more realistic model in the design context. It enables the assess- ant’s project lead could be categorized into the number of years of
ment of the project characteristics and risk events to be made experience in the industry and the number of projects managed.
subjectively, which is usually the case in practice. The output of The skill set of the design teams could be categorized by the
the model is presented both numerically and linguistically, pro- number of junior and senior designers, the number of years of
viding the decision maker with a useful and realistic guide to the experience, and a rating of the designers’ abilities 共e.g., poor,
likely cost overrun or underrun for the project. This feedback is average, or excellent兲.
useful to the user in setting an appropriate fee for the project The design industry could greatly benefit from future research
based on its likely performance. The user may also choose to in the area of controlling and managing project costs during the
modify the existing conditions surrounding the project that he or design phase. This research was a first attempt at addressing the
she can control in order to change the likely outcome of the effects of combinations of factors on design project costs. It has
project. The modifications to the project characteristics and/or provided a realistic preliminary model for forecasting potential
risk events can then be rerun through the model to assess the design cost variations. The results of this research have helped to
impact on cost due to changes in project conditions. identify some of the issues that need to be addressed in conduct-
The model can also be used as a risk assessment tool. The user ing future research in managing design projects.
may decide that a project is too risky to pursue or find ways of
transferring the risk back to the client or to other parties. By Acknowledgments
rating each of the factors, it also makes the user more aware of The writers would like to acknowledge the support of the partici-
potential problems on the project that he or she should address pating design firm in conducting this research. This research was
Bubshait, A. A., Farooq, G., Jannadi, M. O., and Assaf, S. A. 共1999兲. Kometa, S. T., Olomolaiye, P. O., and Harris, F. C. 共1995兲. ‘‘Quantifying
‘‘Quality practices in design organizations.’’ Constr. Manage. client-generated risk by project consultants.’’ Constr. Manage.
Econom., 17, 799– 809. Econom., 13, 137–147.
Construction Industry Institute 共CII兲. 共1986兲. ‘‘Scope definition and con- Kometa, S. T., Olomolaiye, P. O., and Harris, F. C. 共1996兲. ‘‘Validation of
trol.’’ Publication 6-2, Cost/Schedule Controls Task Force, Austin, the model for evaluating client-generated risk by project consultants.’’
Tex. Constr. Manage. Econom., 14, 131–145.
Construction Industry Institute 共CII兲. 共1996兲. ‘‘PDRI: Project definition Love, P. E. D., Mandal, P., Smith, J., and Li, H. 共2000兲. ‘‘Modelling the
rating index—Industrial projects.’’ Implementation Resource, 113-2, dynamics of design error induced rework in construction.’’ Constr.
Front End Planning Research Team, Austin, Tex. Manage. Econom., 18, 567–574.
Construction Industry Institute 共CII兲. 共1999兲. ‘‘PDRI: Project definition McGeorge, J. F. 共1988兲. ‘‘Design productivity: Quality problem.’’ J. Man-
rating index—Building projects.’’ Implementation Resource, 155-2, age. Eng., 4共4兲, 350–362.
PDRI for Building Projects Research Team, Austin, Tex. Nguyen, V. U. 共1985兲. ‘‘Tender evaluation by fuzzy sets.’’ J. Constr. Eng.
Cox, I. D., Morris, J. P., Rogerson, J. H., and Jared, G. E. 共1999兲. ‘‘A Manage., 111共3兲, 231–243.