Sunteți pe pagina 1din 12

Pergamon International Journal of Project Management Vol. 16, No. 3, pp.

153-164, 1998
© 1998 ElsevierScienceLtd and IPMA. All rights reserved
Printed in Great Britain
0263-7863/98 $19.00 + 0.00

Plh S0263-7863(97)00035-5

Which contractor selection


methodology?

Gary D Holt
School of Engineering and The Built Environment, University of Wolverhampton, Wulfruna Street,
Wolverhampton, West Midlands WV1 1SB, UK

Industrial and academic awareness of the need for judicious construction contractor selection is
increasing. Earlier investigations of this client procurement decision have more recently been
expanded upon; particularly, by work conducted in the USA and UK. Other countries are now
also addressing this issue and procurement research abounds. This paper complements that
increased activity by presenting a review of contractor evaluation and selection modelling meth-
odologies. These methodologies include: Bespoke approaches, Multi-attribute analysis, Multi-
attribute utility theory, Cluster analysis, Multiple regression, Fuzzy set theory, and Multivariate
discriminant analysis. The merits/ demerits and previous/possible future applications of each
methodology are discussed, and indications for future research given. © 1998 Elsevier Science
Ltd and IPMA. All rights reserved.
Keywords: contractor evaluation, contractor selection,construction tenders, decisional modelling theory

Introduction blem. (In this context evaluation is the process of in-


vestigating or measuring contractor attributes.
Construction project outcome may be measured in
Selection is the process of aggregating the results of
terms of time, cost and quality achieved) Arguably,
the construction owner decision most impacting these evaluation to identify optimum choice). In performing
this review, the paper also highlights merits/demerits
superlative 'value' criteria, is that of selecting the
and gives example applications of the methodologies
appropriate, i.e. best, contractor) 3 (In this context the
term 'owner' means the client, or professional advisors considered. These methodologies include: Bespoke
acting on the client's behalf). Hence, there is a growing approaches (BA), Multi-attribute analysis (MAA),
realisation that value achieved is as much a function Multi-attribute utility theory (MAUT), Multiple re-
gression (MR), Cluster analysis (CA), Fuzzy set theory
of the quality of contractor employed, as of project
(FST), and Discriminant analysis (MDA).
cost. 4 However, the fundamental problem is." what con-
stitutes a sound contractor selection methodology?
Earlier investigation of this problem yielded much
proffering and philosophising, with regard to tender Methodologies review
evaluation and contractor selection: -~9 These works Absolute classification of some selection methodologies
have since been expanded upon both by the industry is not possible. For example; BA tend to utilise several
(e.g.4"2°-24), and the academic world. Regarding the lat- approaches, 42 so an element of 'overlap' occurs.
ter, a two pronged attack on this critically important However, distinct classification of techniques is
decision problem, has been witnessed over recent attempted as far as possible. Further, it must be
years. Specifically; within the USA there has been accepted that definitive description is not possible in
much work to improve the prequalification process 25-34 the space available. Therefore, references are compre-
whilst the U.K. has witnessed development of the hensively cited throughout the narrative so that the
H.O.L.T. (Highlight Optimum Legitimate Tender) interested reader may follow up additional detailed, or
technique, designed to embrace the entire contractor specific, aspects elsewhere.
selection process) 5-41
Further investigations on this theme are being
undertaken worldwide, including the UK, Cuba, Bespoke approaches ( B A )
Korea, Australia and Botswana. The objective of this Generally, BA tend to be evidenced in terms of ad hoc
paper therefore, is to complement this increased ac- evaluation/ selection methods having been developed
tivity via critical review of some of the decisional by, and therefore familiar to, a particular construction
methodologies that are being, or show potential for, owner. They tend to incorporate several decisional
application to the contractor evaluation/selection pro- techniques simultaneously, and evidence much var-

153
Which contractor selection methodology." G D Holt

INITIAL a typical question. For each of these decomposed


I SUBMISSION aspects, second stage binary decisions as per those
described above are often made, again, with non-con-
/

Eg.,ISO9000 / . . . . Are all 'musts' Ill forming tenders being rejected. At this juncture and
Accreditation satisfied'? with respect to conforming contractors then either; (i)
I
where prequalification is being performed per project
NO (cf. ~) submissions may have final review to select the
desired number of tenderers; or (ii) contractors may be
~ YES placed onto a select or standing list for ultimate invita-
tion to tender (refer Figure 1). It is these latter stages
Divide tender ~ that have the greatest dependence on subjectivity - -
into components
often a function of client/practitioner past experience
with the contractor, and/or practitioner judgment
based purely on experience of the process. 34 One may
consider the practitioner's role in this respect as being
analogous to that of contractor's estimator; where ex-

I All components
satisfactory? i perience, intuition and hunch play as important roles
as academic ability. 5~
In summary: BA are commonplace in industry; exhi-
NO I b bit substantial variance; and rely heavily on binary de-
cisions and subjective interpretation. As a result of
YES
their implementation the owner may, or may not, con-
tract with the best firm but (in the absence of a formal
Forproject I
prequalgicatien..,
[ preqnalification...
F°rsb°"l"t/ evaluation/output mechanism) may not know this until
the project is well under way! Clearly, a numeric
measure, yardstick, datum or score would improve the
I
Select (best) ~
¢
Place contractor BA approach, if only as a means of comparing con-
required number onto select / standing tractors relative to each other.
of tendecers or rotattonal list
i
i
l
INVITE ~
'
i
Multi-attribute analysis ( M A A )
TENDERS ~ ...........
M A A considers a decision alternative with respect to
several of that alternative's attributes. In this respect
Figure 1 Typical stepwise procedure for bespoke approach an attribute is a characteristic that can be measured.
An objective is a characteristic against which an attri-
iance.~ Therefore, because of their disparate nature BA bute is measured and should be pursued to its fullest.
are difficult to classify in terms of a single parameter. Hence, a contractor attribute represents one aspect of
However, most tend to follow a procedure akin to that a decision option with respect to a client objective. 52
exhibited in Figure 1. Attributes may be measured quantitatively or other-
The initial stage of BA usually concerns adjudication wise. Indeed, a feature of M A A is that some or all of
- - e i t h e r by an individual decision maker (DM) or, an the attributes may not be quantifiable. 53 The most
amalgam of D M ' s in the case of larger construction simple M A A equation may be expressed as;
owners. Typical, is investigation of contractors' sub- n
missions for preliminary 'conforming' criteria, often ACrj = (1)
referred to as cut-off points 43 or musts. 44 Such criteria i=l
are often a function of client predilection. For
where: ACrj=aggregate score for contractor j;
example, has the contractor achieved International
quality assurance accreditation? 45 A binary decision VU= variable (attribute) i score in respect of contractor
j; and n = the number of attributes considered in the
(YES/NO) normally follows, with non-conforming ten-
analysis. These are termed 'simple scoring' M A A
ders (i.e. a N O in one criterion) being instantly
models and because of their simplicity are frequently
rejected. On the face of it, this is a logical and effective
employed by D M ' s in industry. 52 An example of a
way of reducing an original (typically large) set in ac-
simple scoring contractor selection model may be
cordance with owner prerequisites or predilections. observed in Janssens. 43 Their biggest failing is that Vi
Nonetheless, the risk is ever present that a 'good' con- (albeit numeric) is often a very subjective measure.
tractor may be wrongly precluded early in the proceed- That is;
ings.
Conforming tenders (i.e. those not excluded) are
Vi = ~--~f(xi) (2)
then normally decomposed (e.g. into financial/health i=1
and safety aspects), to enable scrutiny of specific areas.
For example, observation of accounting ratios 37'46-4s where: f(xi) are the n functions of V; normally subjec-
and/or, calculation of workload capacity 31"49 is typical tively (often implicitly) considered by the practitioner
for financial examination. Regarding health and safety, during evaluation. An improvement on Equation (1).
then in a U K context; ' H a s the contractor formulated is to attach weighting indices ( W i) to Vi thereby
an internal safety policy in accordance with Section accentuating contractors' aggregated scores who per-
2(3) of The Health and Safety at W o r k Act? '5° is also form better in higher weighted criteria and vice versa.

154
Which contractor selection methodology: G D Holt

Table 1 Hypothetical MAA example


Input Output
V1 W1 V2 W2 V3 W3 V4 W4 V5 W5 ACri ACrimax UACrj Rank
Crl 8 8 2 10 10 7 8 10 6 9 288 440 0.652 2
Cr2 7 8 3 10 10 7 3 10 3 9 213 440 0.484 4
Cr3 4 8 10 10 4 7 5 10 5 9 255 440 0.583 3
Cr4 7 8 8 10 6 7 7 10 8 9 320 440 0.731 1
Vi & Wi scored/weighted 1 to 10

Hence; Multi-attribute utility theory ( MA UT)


M A U T is an extension of M A A and as will be shown,
ACrj = ~VijW i (3) is particularly suited to the problem under discussion.
i=1 M A U T utilises 'utility' to quantify the subjective com-
ponents of MAA.
Wi may be a function of;
First, it is necessary to appreciate the concept of uti-
1. sole practitioner experience/predilection lity. Utility (normally expressed as Ui) is a measure of
2. group consensus opinion desirability or satisfaction; of a characteristic (attri-
3. survey and analysis of data, from a sample perti- bute), of an alternative (contractor)J 7 A utility value is
nent to the selection setting in which the model will an abstract equivalent of the attribute being considered
be applied. 39 from natural units such as years, or £, into a series of
commensurable units (utiles) on an interval scale of
Where the components (Vi and kV~ in this instance)
zero to 1.0. Such transformation of values may, or
are represented by an infinite range of integers then a
may not be, a linear function - - t h i s is primarily
unified aggregate contractor score (designated UACrj,
dependent upon the DM(s) or expert(s) from which
i.e. 0 < UACrj<I.O) m a y be achieved via;
such functions were derived (e.g. risk seeking /risk
UACrj -- ACrj (4) averse) - - s e e Figure 2.
A Crjmax The overall concept of M A U T can be very complex
with respect to both models and utility weight deri-
where: the components are as previously described; vation. Hence, the reader is directed in the first
and ACrjm~x is the m a x i m u m attainable aggregate instance to; 5~59. M A U T may be applied to the con-
score utilising HI,.. tractor selection problem in one of two fundamental
Table 1 shows the hypothetical evaluation of four ways. Consider;
contractors in respect of five attributes utilising all the n

concepts described above. Notwithstanding that M A A ACrj = y ~ U i (5)


has previously been researched with respect to this i=l
problem setting, the derivation of essential Vi and, the
strength of W~ should be further investigated (e.g. with where: U; represents the n attributes considered by the
respect to geographical location of project, nature of DM. These are a function of several dimensions i.e.
work, form of procurement option employed etc.). A Ui = Y~J(uxi). (ref. Equation (2,)). Hence, the sum of
general discussion on M A A decision making may be integers Ui yield an aggregate score which may if
observed in M o o r e and T h o m a s 54 whilst detailed treat- necessary, also be unified as described in Equation (4).
ment is given elsewhere. 52-56 Alternatively, Ui may be used in conjunction with scal-

risk seeking functions

Max Ui 1.00

0.75 risk neutral


function
Abstract
expressed
equivalent 0.50 i
in utiles
(Ui)
0.25 averse functions

Min Ui 0.00 I

Min natural unit Xi Max natural unit


Figure 2 Characteristic utility curves. For the utility curves shown then natural unit (Xi) has utility values in the ranges; risk
averse: zero to 0.2, risk neutral: 0.6, risk seeking 0.9 to 1.0

155
Which contractor selection methodology: G D Holt

ing constants or weighting coefficients as follows;


tl
P1 = 0.311 + 0.151 I"1 + 0.035 Vs + 0.154V9
ACrj = ZUiWi (6) - 0.159 Vt9 - 0.031 I"2o + 0.232 V2t (8)
i=1
where: P1 = prequalification score; Vl =size of con-
where components are as described previously and n tractor organisation; Vs=quality of bank reference;
are the number of attributes. V9 = quality of creditor references; V19 = past perform-
In short, M A U T is similar to MAA in modelling ance (time overruns); V20=past performance (cost
terms but utilises a quantitative method of determining overruns); and V21=past performance (quality
values for inclusion therein. Hence, M A U T can quan- achieved). For this equation R 2= 0.96.
titatively consider both tangible (e.g. safety record) Obviously, there are essential 'musts' in respect of
and intangible (e.g. contractor image) attributes, contractors desirous to tender (refer BA), such as ade-
during evaluation. The existence of such subjective quate insurance; adequate bonding capacity and finan-
measures characterises the contractor selection pro- cial stability. Hence, using the above M R approach
blem. Regarding future research, there is scope for then prequalification might follow the route as exhib-
developing utility curves in respect of alternative scen- ited in Figure 3. Again, here is an area that yearns for
arios (as intimated in the previous section) and with further research.
respect to differing owner groupings. Equation (8) was derived from a limited sample size
and in respect to 'smaller' projects, hence the potential
for development of M R models based on inter alia; a
larger sample; projects of particular character; and
contractors of a specialist nature. Further, were con-
Multiple regression ( M R ) tractors' past performance measures (time, cost, qual-
M R is a statistical technique whereby an equation is ity) each quantified and used as dependent variables
constructed to observe and ultimately predict the effect (i.e. regressed upon separately) then three M R
of several independent variables upon a dependent equations would result; with potential to predict future
variable. That is, an M R equation will predict a nu- contractor performance in respect of each of these
meric outcome (designated Y*) this being a function of superlative owner objectives.
several independent variables Vl, Vz... Vi. For a given To simplify and rationalise this task, which would
scenario j; represented by several dimensions; Vjj there involve an inordinate number of contractor attri-
will be an actual outcome; Y. It is from the statistical butes, 2'39"6° then the particular method of stepwise re-
analysis of several of these scenarios from which an gression (vis-a-vis simultaneous regression which has
M R equation may be constructed. Clearly, any differ- been described thus far) may be considered. An excel-
ence between predicted (Y*) and actual, may be for- lent overview of the stepwise technique may be found
malised as Y * - Y; such difference(s) being termed in Kinnear and Gray. 61 Mohsini and Davidson 62 may
residuals. The lesser number/smaller magnitude of also be observed for example application of M R in a
these residuals, then the more accurate is the M R construction context.
equation. An M R equation may be formalised as;
n

Y* = Co + y ~ Vici (7) Cluster analysis (CA)


i=1
The nature of the problem under consideration in this
where: Y* represents the dependent variable; V,- are the paper involves a theoretically infinite range (set) of
independent variables; ci are the partial regression contractors, albeit this will be a function of tendering
arrangement employed. 63,64 The principal task there-
coefficients; Co is a constant representing the intercept
upon the y axis of the resulting regression line; and n fore, is one of reducing this original set into a series of
is the number of independent variables within the smaller, manageable sub-sets of like character. By ana-
equation. lysing these sub-sets, the quality (i.e. attributes) of con-
Thought of in the space of an x:y graph, then tractors therein may be observed and the best sub-
changes in V; will result in a change for Y*. Each indi- set(s) identified for subsequent tender invitation if pre-
vidual coefficient c; accounts for partial change in Y* qualification is being performed. Alternatively, the
thus the term 'partial' regression coefficient. When characteristics of sub-set membership would help in
V i = 0 then Y* = Co. Reverting to Y * - Y, then 'good-
assigning contractors to standing lists (e.g. specific
ness of fit' of the regression line is represented by R2: work types or, project sizes). Fundamental benefits of
the coefficient of determination. For example, if a CA approach are threefold;
RZ=0.8 for a given equation then 80% of movement 1. application of a limited number of previously ident-
in Y* can be attributed to movement in Vi. The larger ified controlling criteria (see below) to the entire
R 2 is, then the more accurate a predictor is the original set, rationalises the evaluation process but
equation based on the input data upon which it was facilitates effective investigation of a// members;
built. Having briefly discussed the technique, how can 2. this negates the possibility of rejecting 'good' con-
it be applied to the problem under review? Firstly, tractors at an early stage in the procedure (a risk
there is scope for building M R equations to yield con- synonymous with the binary decision approach - -
tractor prequalification scores, based on contractor refer BA);
(interval) attribute data. Such an approach has been 3. achieving this minimises owner resource commit-
tentatively investigated 6° the resulting equation being ment i.e. maximum yield on the cost of information
of the form; collection and processing; (cf. 52"54) and hence, maxi-

156
Which contractor selection methodology: G D Holt

Contractors desirous
to tender

1" I
Bonding capacity
Ale 'musts' ~ NO [ Reject ]l
satisfied?

4
iI
i I

I YES
Evaluate MR attributes
achieve interval data

Apply MR equation
P1 = Co + ~Vi ci

¼
Rank contractorsbased on •
Prequalificationscores
I
I Select desired number of
tenderers ie., highest
P1 scores

Figure3 Possible stepwise procedure for an MR prequalification model

m u m potential for achieving client satisfaction from are similar to each other is reflected in the hierarchical
selecting the best alternative(s). tree as distinct branches. Distance between branches of
The method takes a given number of contractors, the tree are proportional to the dissimilarity between
each being described by a set of numerical attribute contractors.
scores and, uses a classification algorithm to group the There are various CA algorithms. However, the
contractors into a number of clusters such that con- most straightforward way of establishing degree of
tractors within classes are similar and unlike those difference between contractors, is to compute
from other clusters. 65 Two particular types of CA are Euclidean distances. This measure is computed as;
suitable: jointing-tree clustering and k-means cluster-
ing. I f jointing-tree clustering is firstly applied to the
original set, it establishes the most significant number
of clusters inherent within it. That is, amongst the
pooled contractor data, we assume no priori hypoth- where: D o is the distance between two points i: j; and
esis with regard to number of sub-groups. The output xik is the value of the kth variable for the ith entity. 66
of this initial analysis is a tree diagram known as a In summary, jointing-tree clustering graphically shows
dendrogram. Figure 4. shows a dendrogram for 19 via a dendrogram, the number of 'natural' clusters
contractors (Crl, Cr2,...Crl9), segregated into two prin- within a given original set of contractors.
cipal clusters. The x axis exhibits each contractor in a k-means CA establishes k clusters from the data
class by itself. As we progress upwards, the threshold where k is defined by the practitioner. Therefore, the
regarding when to declare two or more contractors as former method provides a posterior hypothesis for k-
being similar are relaxed, so more and more contrac- means analysis, k-means CA begins with k random
tors are clustered until finally in the uppermost part of clusters, moving contractors between them to minimise
the dendrogram, all contractors are linked together. variability within clusters and, maximise variability
Thus a clear 'structure' in terms of contractors that between clusters. As a result of the analysis, the mean

157
Which contractor selection methodology: G D Holt

Cluster 1 Cluster 2
1.4 .....................

,i!i
1.2 I ,............. i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

0
0.8

0.6
i
0.4

O. o.o, , , , ...........
, ,

Crl7 Crl9 Cr8 Crl2 Cr4 Crl6 Cr2 Crib Cr9 Crl
Crl4 Crl5 Cr7 Crl0 Cr3 Crl I Crl3 Cr6 Cr5

Contractors (Cr)
Figure 4 Jointing-tree CA output: dendrogram. Vertical axis: respective distance from cluster centre

values of each attribute, within each cluster, may be Fuzzy set theory ( F S T )
examined. This; (i) confirms how distinct the k clusters
Another characteristic of the problem under discussion
are; and ii) identifies the most discriminating criteria. is that it encompasses uncertainty, functions of which
That is, the process identifies the controlling variables are: imprecision, randomness and ambiguity. 67
referred to at the outset of this section. Clearly, this Probability theory is often associated with random-
provides m a x i m u m potential for rationalisation of the ness, but such theory attempts prediction of future
purchasing process, because only significant discrimi- events based on previous events. Unfortunately, con-
nators need be considered. tractor selection does not fit neatly into this framework
Table 2. details mean criteria scores and A N O V A because each selection exercise is a one-off and hence
output for the hypothetical selection scenario shown in different. However, FST can model human judgment
the earlier dendrogram. Eight selection criteria (V~, and cope with uncertainty. 67 Further, FST can deal
I1"2..... II8) have been utilised, although in reality a lar- with multi-criteria information which, may also be
ger number of criteria will apply. 39 The Table estab- imprecise and subjective. 68 These features characterise
lishes that there are four controlling criteria the contractor selection problem. ~,~2 FST principles
were introduced by Zadeh 69 and may be described in
(significant at the 95% level), these being: V1, V6, V7
terms of full, or partial membership of a set, for a
and V8. Controlling variables are easily identified on
given case (Crj) in question.
Figure 5., this being a graphical plot of these final clus-
To help explain the concept we shall revert back to
ter centres, in respect of each selection criterion. These
BA for a moment. It is recalled from that example
'cluster profiles' also confirm that cluster 2 is the 'bet-
that binary decisions were made e.g. the contractor
ter' set of contractors; with higher mean scores does, or does not, have a formal safety policy. This is
achieved in all variables except: II5 and Vs. For the a perfect example of classic set theory i.e. the contrac-
purposes of this example it was assumed that contrac- tor is, or is not, a member of the respective (does, does
tor attributes ( V i) were evaluated and scored on an not) sets. FST takes this a stage further i.e. the degree
interval scale where: 0 < Vi<l.O (see37'4°). of membership may be measured and a membership

Table 2 Mean criteria Scores and ANOVA: Clusters l and 2

Variable Mean scores Analysis of variance

Cluster 1 Cluster 2 Between clusters* Within clusterst Significance (P)


Vj 0.42 0.79 0.64 2.50 0.052
Vz 0.90 0.91 < 0.01 0.25 0.911
113 0.80 0.85 < 0.01 0.75 0.650
V4 0.80 0.92 0.06 0.38 0.102
V5 0.33 0.27 0.01 0.49 0.440
V6 -- 0.65 2.00 2.02 0.000
V7 0.42 0.79 0.64 2.50 0.052
1/8 0.73 0.18 1.45 2.76 0.008
*Sum of squares (! dj')
t S u m of squares (17 dJ)

158
Which contractor selection methodology: G D Holt

0.9
0.8
'a
0.7
0.6
• ..... ! .... .... r ....... /-.,:- . . . . ....
• -I . . . . . . . . . ', . . . . . . . . . . . " . . . . . . . . . . . ', . . . . . . . . . . L . . . . . . . . . . . ', ---I--- Cluster 1
0.5
. . . . . . . . . . . :. . . . . . . . . . . ~. . . . . . . . . . . ', . . . . . . . . ix.L . . . . . . i Q Cluster 2
0.4
0.3
0.2
0.1 _/ . . . . . . . . • . . . . . . . . . . . :
x /
0 I !
VI V2 V3 V4 V5 V6 V7 V8
Variables (Vi)
Figure 5 Mean Vi profiles for clusters 1 and 2

value (Mv) assigned, where; 0 <Mv_<l.0 and


1.0 = m a x i m u m strength o f membership.
This is fully elucidated in Figure 6, the fundamental
message being, that F S T facilitates ordinal sorting o f
Crj albeit, overlap m a y occur in terms o f description.
I I I I I 1 T h a t is, there is an element o f 'fuzziness' between sets
Cr 1 Cr2 Cr3 Cr4 Cr5 Cr6 (cf. sets A, B, C). Mathematically, and with reference
I I 1 I i I to the Figure, if U represents the universe o f objects
Does the contractor have a formal health & safety policy? under consideration then set A is a fuzzy set in U

I
YES NO
i i
NO
i
YES YES
i YES
i where;
h = {Crj,l.tA(Crj)} (10)
i I I I I I
when: U=A; Crj e A (in this instance C r 4 only) and;
I #A(Crj) defines the strength o f membership (Crj in A)
being a real n u m b e r on the interval 0.0 .... 1.0 which in
turn is a function o f the membership value (My) o f Crj
Fuzzy set theory in U.
F o r Figure 6 then; llA(Cr2 a n d 3) = 0.0 i.e. non mem-
i!
bership o f D in U and; #xA(Cr4) = 1.0 i.e. full member-
: ( SET I:YES "~ ship o f Cr 4 in A and hence, o f A in U. The Figure also
: ~ Crl, Cr4, Cr5, Cr6 I demonstrates that non-numerical measures (linguistic
r-~ i Mv=l.0forall [ Ii
\ members ,) Extent of policy 1 in this instance) m a y be a c c o m m o d a t e d by FST.
determined and expressed However, the boundaries between sets can cause con-

I
as Mv where; 0.0 = NO fusion, for example, where is the delineation between
& 1.0. = maximum
SET 2: NO 1 formal policy M I N I M U M and S T R O N G ? T h a t is, into which set
_ _ll,.t Cr2 & Cr3 would go a c o n t r a c t o r with an My o f 0.5? The firm
l Mv = 0.0 for 'all might have tentative membership o f the S T R O N G set
k, members
or, m a x i m u m membership o f the M I N I M U M set.
• ~ s,,.¢ "e•
To satisfy this problem, there is scope for future

{Cr4.Mv=lOfJ
--" •" : i
I I development o f F S T 'membership functions' in respect
Se, A ~"" ~. :( SetD "~ o f n c o n t r a c t o r attributes. Such functions could be
MAXIMUM _]
based on the a p p r o a c h o f Dexter and Hales 7° that is,
STRONG
Cr5. Mv = 0.9
t :
'" i ° "ICr2, Mv : 0.0 I
I~r3, Mv = 0.0J each c o n t r a c t o r attribute being d e c o m p o s e d into a
g r o u p o f membership profiles based on either;
Cr6, Mv = 0.7 1. practitioner predilection/expertise;
SetC
MINIMUM 2. g r o u p consensus opinion; or
Crl, Mv = 0.3 3. quantitative derivation via stratified survey o f rel-
evant expert opinion.
Figure 6 The fundamental concepts of classic/fuzzy set the- Figure 7 shows a set o f profiles for the health and
ory safety attribute discussed above. Albeit hypothetical,

159
Which contractor selection methodology." G D Holt

1.0 Contractorwith perceivedlevel of formal safety policy at 48% !


I
achieves Mv of 0.5 which may be described as max I
membership strength of MIN set or, tentativemembership of /
STRONG set

Membership /
value 0.5 ~

; . . . . . . . / MAX
NO = zero / Vl
score
/ ?,i

zero

o lO 20 30 40 50 60 70 80 90 100

Perceivedlevel of formal safety policyexpressed as percentage y

Figure 7 Hypothetical memebership functions for the variable: health and safety policy

the profiles underline the future potential of this This variable is then combined with each of the other
approach. However, FST may only find lukewarm variables in turn until the next variable is found which
reception amongst practitioners because of its complex contributes most to any further discrimination between
nature. For further narrative on the technique the the groups. The process continues in a similar manner
reader may o b s e r v e : 69'7~ 76 until such time as very little discrimination is gained
by inclusion of any further variable. The large amount
of calculation required calls for a computer software
Multivariate discriminant analysis ( M D A ) package such as the SPSS-PC+ (or Windows) suite of
In essence, M D A studies the differences between two programmes, 58 an overview of which may be found in
or more objects with respect to several variables, sim- Kinnear and Gray. 61'8°
ultaneously. 77 In the context of this paper, then by Based on the M D A discussion thus far one might
examining contractors' previous performance on job- question its uniqueness, particularly, with retrospective
site and, the characteristics of those contractors, one comparison to CA and MR. This may be explained by
can determine: observation of Figure 8. That is, above the dashed line
I. which if any of the contractor attributes are discri- on Figure 8 is shown MDA: where no independent
minators i.e. have the ability to predict performance variable (either Vi or Gj) is designated prior to analysis
on future jobsite; (otherwise this would be analogous to MR). Below the
2. how these attributes might be combined into an al- dashed line, the value of V, is a function of Gj this
gorithm in order to achieve (/) and; being an extension of multivariate analysis of variance
3. the accuracy o f any such derived equation, v7'7~ and therefore, analogous to k-means CA earlier.
Hence, the uniqueness of MDA is identified.
In short, we may consider different types, and sizes, M D A has to some extent previously been applied to
of contractors (Crj) in tandem with a range of multi- the problem under discussion. Specifically, to predict
variate selection variables. In such instances of varia- future jobsite performance of Hong Kong contrac-
bility amongst cases and measures, M D A is the most tors. 79 The resulting canonical discriminating function
appropriate techniquefl 9 was o f the form;
The basic prerequisite of M D A is that two or more
groups exist which are presumed to differ on several Zi = Co -~- CI VI --1-C2 V2 --~ ..... -~- C6 V6 (ll)
variables ( V i) and, that these can be at the interval or
ratio level. 77 However, for a given analysis: Vg must be where: Zt is the predictive contractor performance
<_(Crj+ 2). M D A then analyses the differences between index; Co is a constant; CL.,6 are the discriminating
groups and subsequently interprets and/or classifies. coefficients; and V~,..,6 are the discriminating variables
Interpretation studies the differences between group governing contractor performance these being: com-
attributes and therefore, yields the knowledge of how plexity of project; percentage of professionally quali-
discrimination between (say; good, not-so-good and fied staff; project leaders experience; contractors past
bad) groups is achieved. Importantly, M D A identifies performance or image; origin of the company and con-
the controlling criteria i.e. most powerful discrimina- struction owner's control, vs
tors, this bringing with it the benefits of a rationalised The research cited used a retrospective approach for
technique elucidated under CA earlier. validation of the model, based on 'good' and 'bad'
Classification provides a means o f assigning any pre- project outcomes in Hong Kong. However, there is
viously unanalysed contractor into a group it most clo- future scope for inter alia; concurrent validation of a
sely resembles. Hence, we may regard classification a derived discriminant function by using live projects.
subsequent function o f M D A interpretation. The tech- Further, such an approach would facilitate comparison
nique begins by finding the most discriminating Vi. of project outturn (refer superlative value criteria in

160
Which contractor selection methodology: G D Holt

Above dashed line = MDA


relating one nominal
level variable (Gj) to
several interval
level variables (Vi)

Below dashed line is a function


of multivariate analysis of variance
G1
ie., value of Vi dependent upon

designation of Gj

Figure 8 The interface of Gj and Vi in an MDA context

introduction) with previous levels of client satisfaction trolling (prequalification) criteria involved. Finally, the
for the firms analysed (cf.8~), using covariance analysis. technique is not over-complex; it is familiar to most
The reader will find comprehensive, introductory text statistical software suites; and yields a numeric/graphi-
on the M D A concept in Klecka, 77 whilst the works of cal output that may be quickly interpreted by prac-
Skitmore and Marsden 58 and Abidali 47 provide excel- titioners. This is particularly so if one considers the
lent examples of M D A in a construction setting. cluster profiles introduced in Figure 5. A more compre-
hensive description of this rationale may be observed
in Holt. 65
Conclusions
Second stage evaluation has been previously investi-
This paper has reviewed several decisional method-
gated using M A U T . 52 However, as indicated in this
ologies which are being applied, or show potential for
application, to the contractor evaluation and selection paper other methodologies offer equal scope in this
task. The paper does not purport to have discussed all respect. An 'at-a-glance' conclusion to this review is
possible solutions. For example, the application of given in Table 3, which also highlights salient merits/
case-based-reasoning has been considered elsewhere 82 demerits of each methodology with regard practical
whilst the concept of neural networks has not to the application to the contractor selection problem. The
author's knowledge, yet been considered. However, methodologies may be summarised as follows:
based on this paper, it can be established that certain
BA: prolific industrial use but very subjective
approaches are particularly suited to specific aspects of
this decision problem. MAA: academic and industrial use but can be
First, regarding the problem strategy, a two stage subjective
procedure (stage 1: prequalify, stage 2: evaluate ten- M A U T : academic use, scope for derivation of
derers) is desirableJ 's3 The first stage should consider alternative utility functions
all contractors desirous to tender with respect to MR: evidence of academic use, scope for further
'essential' contractor organisational criteria. Such cri- research and industrial application
teria include: past performance, past experience, and CA: limited academic use, scope for further
financial stability. 37 The second stage should consider
research and application to prequalification
specific contractor suitability for the proposed project.
Project specific criteria include: office location with FST: evidence of academic use but may be too
respect to project, experience in the geographical complex for acceptance by industry
region, and experience of the proposed construction M D A : academic use, broader scope for research
methods.2.37, 40 and industrial application.
Based on this review it seems that the C A technique
Future research should be addressed towards those
offers greatest potential for the first stage. Primarily,
methodologies identified as requiring such. That is:
this is because it can deal relatively easily with the po-
tentially large numbers of contractors involved. That development of the selection model; retrospective/con-
is, CA reduces the probability of rejecting a 'good' current/predictive application to the problem; and ulti-
contractor too early in the process via subjective re- mate validation by observation of output. 84 In the final
duction of the often large original set (as is the risk analysis, whichever selection methodology is employed
with BA). Further, CA would offer scope for rational- the best protection against a bad selection decision is a
isation of the selection process by identifying the con- good selection procedure! ~
JPMA 16/3 B 161
Which contractor selection methodology. G D Holt

References
3
1. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., A review o f
¢- contractor selection practice in the UK constructin industry•
o Building and Environment, 1995, 30(4), 553-561.
,.~ .ID
2. Russell, J. S. and Skibniewski, M. J., Decision criteria in con-
tractor prequalification. Journal of Management in Engineering
o9
A.S.C.E., 1988, 4, 148 164.
o9 o9 eo9 3. Odusote, O., An examination of the importance of resource
~ o9 ~ o9

o9
considerations when contractors make project selection de-
~ ~ ~=~ -=~.. cisions. M.Sc. dissertation, University of Bath, UK, 1990.
o9.- o9 ",~,
4. C.I.C., The Procurement of Professional Services. Guidelines for
the Value Assessment of Competitive Tenders. Construction
Industry Council, UK, 1994.
5. Harper, D. R., Evaluation of alternative methods of contractor
s ,~ o
selection. In Proceedings of the International Building
Exhibition, 23 November, London, 1971.
6. Vorster, M., When the lowest tender isn't the lowest Journal of
Construction in South Africa, 1977, February, 37-43.
7. Helmer, T. F. and Taylor, R. L., The evaluation of contractor
._~
management during source selection. Dept. of Economics,
Geography and Management, U.S.A.F. Academy, Colorado
"cI
80840. Proceedings of the A.I.I.E. Spring Annual Conference,
1977.
8. Spellman, K., Predicting the failure of a construction company.
Z
Journal of AccountamT, 1978, August.
9. I.O.B.Contractor selection a guide to good practice
Estimating information service 1979 34.
10. Mason, R. J. and Harris, F. C., Discussion: predicting company
e-
-~ ~ ~-_ failure in the construction industry. Proceedings of the
Institution of Civil Engineers, Part 1, 1980, 68(February), 149-
154.
r, ~ ~o s .~ ~.- o9
11. Diekmann, J. E., Cost-plus contractor selection: A case study,
Journal of the Technical Councils of A.S.C.E., 1981, 107(TC1).
o9
12. Diekmann, J. E., Cost-plus contractor selection: an analytical

~-, ;>
• .,.-, ~
o o9
method. Engineering costs and production economics, 1983, 7,
147 158.
Z 13. Bovis, Selection of a Management Contractor for the Public
e. Sector. Bovis Construction, Harrow, UK, 1981.
o
14. Bent, J. A., Contractor proposal evaluation program. Annual
m Association of Cost Engineers Transactions, 0-4, 0.4.1.-0.4.9.,
1984.
g
.y. 15. Mascoll, S. G., Organisational structures and contracts for con-
.,~, .0 ,.~ struction projects and their selection. Ph.D. thesis, Dept. of
Civil Engineering, University of Leeds, UK, 1984.
... 16. Baker, M. and Orsaah S., How do the customers choose a con-
tractor? Building Magazine, 1985 May 30- 31.
o
17. Flanagan, R. and Norman, G., Sealed bid auctions: an appli-
cation to the building industry. Construction Management and
~, ~ .=-~ ~ -o9~
~. Economics, 1985, 3, 145-161.
18. Birrell, G. S., Bid appraisal incorporating quantified past per-
formances by contractors. Annual Association of Cost Engineers
Transactions, D-I, D.I.1. D.1.6., 1988.
~ ~° ~ ° z.~ CY 19. Merna, A. and Smith, N. J., Bid evaluation for U K public sec-
tor construction contracts. Proceedings of the Institution of Civil
Engineers, Part 1, 1988, 88(February), 91-105.
20. N.J.C.C., Code of Procedure for Single Stage Selective
Tendering. National Joint Consultative Committee for Building.
~ .z ._~ ~ .~_ [- R.1.B.A., London, 1989a.
- 21. N.J.C.C., Code of Procedure for the Letting and Management of
Domestic Sub-contract Works. National Joint Consultative
~0 Committee for Building. R.I.B.A., London, 1989b.
..- 22. N.J.C.C., Code of Procedure for the Selection of a Management
Contractor and Works Contractors. National Joint Consultative
O
Committee for Building. R.I.B.A., London, 1991.
._. ~ ~ -r. r. i_~ r. e~
23. Latham, Sir M., Trust and Money. Interim report of the
Government/Industry review of procurement and contractual
arrangements in the U K construction industry. HMSO,
London, 1993.
24. Latham, Sir M., Constructing the Team. Final report of the
Government/Industry review of procurement and contractual
arrangements in the U K construction industry. HMSO,
London, 1994.
25. Russell, J. S. and Skibniewski, M. J., A structured approach to
the contractor prequalification process in the U.S.A. In
e~
Proceedings of the CIB-SB1 4th International Symposium, 1987.
26. Russell, J. S. and Skibniewski, M. J., QUALIFIER-l: contrac-
tor prequalification model. Journal of Computing in Civil
Engineering, A.S.C.E., 1990, 4(1), 77-90.

162

. . . . I 411
Which contractor selection methodology: G D Holt

27. Russell, J. S. and Ahmad, I., A PERT approach to contractor Dept. of Civil Engineering, Loughborough University of
prequalification analysis. Thirty ,[ourth Annual Association of Technology, UK, 1994.
Cost Engineers, Boston, MA, D 1. I-D 1.6., 1989. 52. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., Applying
28. Russell, J. S. and Jaleskis, M., Quantitative study of contractor multi-attribute analysis to contractor selection decisions.
evaluation programs and their impact. Journal of Construetion European Journal of Purchasing And Supply Management,
Engineering and Management, 1992, 118(3), 612-624. 1995b, 1(3), 139-148.
29. Russell, J. S., Skibniewski, M. J., Killen, T. S. and Robinson, 53. Hwang, C. and Yoon, K., Multiple Attribute Decision Making.
J. H., Evaluating alternatives in construction management. A State of the Art Survey. Springer Verlag, Berlin, 1981.
Presented at the construction congress 1, A.S.C.E., San 54. Moore, P. G. and Thomas, H., The Anatomy of Decisions.
Francisco, CA, 1989, pp. 340 348. Penguin Books, Harmondsworth, 1976.
30. Russell, J. S., Skibniewski, M. J. and Cozier, D. 55. Hutton Barron, F., Selecting a best multi-attribute alternative
R., QUALIFIER-2: knowledge based system for contractor with partial information about attribute weights. In Current
prequalification. Journal of Construction Engineering and Themes in Psychological Decision Research. Selected Proceedings
Management, A.S.C.E., 1990, 116(1), 155 169. of the 13th Research Conference on Subjective Probability.
31. Russell, J. S., Hancher, D. E. and Skibniewski, M. Utility and Decision Making, Fribourg, Switzerland, August,
J., Contractor prequalification data for construction owners. 1991.
Construction Management and Economics, 1992, 10, 117 135. 56. Timmerrnans, D. and Vlek, Ch., Multi-attribute decision sup-
32. Russell, J. S., A knowledge based system approach to the con- port and complexity: An evaluation and process analysis of
tractor prequalification process. Ph.D. thesis, Purdue aided versus unaided decision making. In Current Themes in
University, West Lafayette, IN, 1988. Psychological Decision Research. Selected Proceedings of the
33. Russell, J. S., Contractor failure: analysis. Journal of 13th Research Canference on Subjective Probability, Utility and
Performance in Construction, A.S.C.E., 5(3). Decision Making, Fribourg, Switzerland, August, 1991.
34. Russell, J. S., Decision models for analysis and evaluation of 57. Moselhi, O. and Martinnelli, A., Analysis of bids using multi-
construction contractors. Construction Management and attribute utility theory. In Proceedings of the International
Economics, 1992, 10, 185-202. Symposium on Building Economics and Construction
35. Holt G. D., Olomolaiye, P. O. and Harris, F. C., Tendering Management, Sydney, Australia, 1990.
practice - - exploring alternatives, Faculty ~?f Building Journal, 58. Skitmore, R. M. and Marsden, D. E., Which procurement sys-
1993, Autumn, 28-30. tem? Towards a universal procurement technique Construction
36. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., A conceptual Management and Economics, 1988, 6, 71-89.
alternative to current tendering practice. Building Research and 59. Finch, E., Environmental assessment of construction projects.
Information, 1993, 21(3), 167 172. Construction Management and Economics, 1992, 10, 5 18.
37. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., Evaluating 60. Holt, G, D., A methodology for predicting the performance of
prequalification criteria in contractor selection. Building and construction contractors. Unpub. Ph.D. thesis. University of
Environment, 1994, 29(4), 437 448. Wolverhampton, UK, 1995.
38. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., Evaluating 61. Kinnear, P. and Gray, C., SPSS for Windows Made Simple.
performance potential in the selection of construction contrac- Lawrence Erlbaum Associates Ltd, Sussex, 1995.
tors. Engineering Construction and Architectural Management, 62. Mohsini, R. A. and Davidson, C. H., Determinants of perform-
1994, 1(I), 29 50. ance in the traditional building process. Construction
39. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., Factors influ- Management and Economics, 1992, 10, 343 359.
encing UK construction clients choice of contractor. Building 63. Franks, J., Building Procurement Systems. Chartered Institute
and Environment, 1994, 29(2), 241 248. of Building, Englemere, UK, 1990.
40. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., Incorporating 64. Buchan, R., Fleming, F, and Kelly, J., Estimating Jbr Builders
project specific criteria and client utility, into the evaluation of and QuantiO' Surveyors. Butterworth Heinemann Ltd, Oxford,
construction tenderers. Building Research and Information, 1994, 1991.
22(4), 214 221. 65. Holt, G. D., Applying cluster analysis to contractor classifi-
41. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., A generic cation. Building and Environment. Pergamon Press, Oxford,
approach to the selection of construction contractors. Final 1996, 31(6), 557-568.
report of a research sponsored by the Leverhuhne Trust. 66. Everitt, B., Cluster Analysis, 2nd edn. Heinemann Educational
School of Construction, Engineering and Technology, Ltd, 1980.
University of Wolverhampton, UK, 1994e. 67. Wang, W. and McCarthy, C., Fuzzy logic, clear thinking: con-
42. Bohanec, M., Urh, B. and Rajkovic, V., Evaluating options by tract strategy and the new engineering contract. ~Project'.
combined qualitative and quantitative methods. In Current Journal of the Association of Project Managers, UK, 1994, 10-
themes in Psychological Decision Research. Selected Proceedings 12.
of the 13th Research Conference on Subjective ProbabiliO', 68. Nguyen, V. U., Tender evaluation by fuzzy sets. Journal of
UtiliO' and Decision Making, Fribourg, Switzerland, August, Construction Engineering and Management. A.S.C.E., 1985, 111,
1991. 231 243.
43. Janssens, D., Design Build Explained. Macmillan, London, 69. Zadeh, L. A., Fuzzy sets. Information and control, 1965, 8, 338-
1991. 353.
44. Harris, F. C. and McCaffer, R., In Modern Construction 70. Dexter, A. L. and Haves, P., Building control systems: an
Management, Chap. 7. BSP Professional Books, Oxford, 1989. evaluation of performance using an emulator. Building Services
45. I.S.O. (9000). Standard 9000 Quality Systems. International Engineering Research and Technology, 1994, 15(3), 131-140.
Standards Organisation. 1, Rue Varembe, Case Postal 56, Ch 71. Unwin, S. D. (1984). An introduction to fuzzy set theory with a
1211, Geneva 20, Switzerland. view to the quantification and propagation of vagueness in
46. Holmes, G. and Sugden, A., Interpreting Company Reports and probabilistic risk and reliability. Report: SRD R301. United
Accounts, 4th edn. Cambridge: Woodhead/Faulkner, 1990. Kingdom Atomic Energy Authority Safety and Reliability
47. Abidali, A. F., A model for predicting company failure in the Directorate.
construction industry. Unpub. Ph.D. thesis, Dept. of Civil 72. Schmucker, K. J., Fuzzy Sets, Natural Language Computations
Engineering. Loughborough University of Technology, UK, and Risk Analysis. Computer Science Press, George Washington
1990. University, 1984.
48. Langford, D., lyagba, R. and Komba, D., Prediction of sol- 73. Paek, J. H. and Yong, H. O., Pricing construction risk: fuzzy
vency in construction companies. Construction Management and set application Journal of Construction Engineering and
Economics, 1993, 11, 317-325. Management, 1993, 119(4).
49. Harris, P. T., An investigation of contractors' workload ca- 74. Wang, B., Study of fuzzy design of structure. In Proceedings of
pacity. M.Sc. dissertation, School of Construction, Engineering the EASEC Conference on Building for the 21st Century,
and technology, University of Wolverhampton, UK, 1995. Griffith University, Australia, 1995, pp. 1067-1072.
50. H.A.S.W.A., Writing a Safety Policy Statement. Ref: C1000 8/ 75. Fayek, A., Duffield, C. F. and Yang, D. M., The design and
91. Health and Safety Commission, Health and Safety implementation of an integrated computer-aided estimating and
Executive, UK, 1974. tendering system that utilises fuzzy logic. In Proceedings of the
51. Humphries, J., Contractors' understanding of the factors which EASEC Conference on Building for the 21st Century, Griffith
affect tendering levels of construction works. M.Sc. dissertation, University, Australia, 1995, pp. 859-864.

163
W h i c h contractor selection m e t h o d o l o g y : G D H o l t

76. Sugiyama, T., Mizutani, J. and Kumagai, S., Judgmen! of 83. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., Tendering
agreement between measured and predicted values of defor- procedures, contractual arrangements and Latham: the contrac-
mation of retaining walls based on fuzzy set theory. In tors" view. Engineering Construction and Architectural
Proeeedings o[" the E A S E C Con[~renee on Building Jbr the 21st Management, 1995, 3(1 and 2), 97 115.
Centuo', Griffith University, Australia, 1995, pp. 193 198. 84. Holt, G. D., Construction research --what is the point? Faculty
77. Klecka, W. R., Discriminant analysis. In Quantitative o f Building Journal, 1994, Winter, 28 31.
Applications in the Social Seienees. Sage University Press,
London, 1980.
78. Tam, C. M. and Harris, F. C., Discriminant analysis model for Gao' D Holt is Senior Research
predicting contractor performance in Hong Kong. Conseil Fellow within the School of
International du Batiment, 1992 World Building Congress. Engineering and the Built
79. Tam, C. M., Discriminant analysis model for predicting con- Environment at The UniversiO' o['
tractor performance in Hong Kong. Unpub. Ph.D. thesis, Dept. Wolverhampton, UK. He has exten-
of Building and Construction, City Polytechnic of Hong Kong, sive management experience, in par-
1993. titular, within the UK speeulatiw"
80. Kinnear, P. and Gray, C., S P S S P C + Made Sinq~le. Lawrence housing sector. He has published
Erlbaum Associates Ltd, Sussex, 1992. extensively as a result q[ his
81. Holt, G. D., Olomolaiye, P. O. and Harris, F. C., Applicalion research, which eentres around the
of an alternative contractor selection model. BuiMin~, Research subject q[' contractor evaluation and
and h~/brmation, 1995, 23(5), 255---264. subsequent modelling of the selec-
82. Ng, S. T., Smith, N. J. and Skitmore, R. M., Case based tion process. A Chartered Buihh, r,
reasoning for contractor prequalification a feasibility study. he holds a first in Building
In Developments in Artificial httelligence /br Civil aml Structm'al Management and a Ph. D. in
Engineering. Civil-Comp Press, Edinburgh, 1995. ('onstruelion Management.

164

S-ar putea să vă placă și