Swproxy Url File://00/276/00014276
Swproxy Url File://00/276/00014276
Swproxy Url File://00/276/00014276
Investigation
of UK construction clients' tender evaluation preferences
CHEE H. WONG, GARY D. HOLT & PHI L HARRI S
Built Environment Research Unit, School of Engineering and the Built Environment, University of Wolverhampton, St Peter
Square, Wolverhampton, West Midlands WV1 1RH, UK
Abstract 2 The `lowest-price wins' philosophy has been a
consistent theme of contractor selection over the years.
To comprehensively elucidate this selection preference
and compare it with the use of a multi-criteria selection
(MCS) approach in the tenderer evaluation process, this
paper investigates MCS tender price selection
preferences. That is, project-specic criteria (PSC) and
lowest-price wins selection practices of UK construction
clients, in both building and civil engineering works at
in detail via results of the empirical survey. The
investigation provides further insight into the evaluation
of contractors' attributes (i.e. PSC). Levels of importance
assigned (LIA) for each criterion were analysed (i.e.
quantitative analysis of the differences in opinions and,
variance amongst the respondents) in a multivariate
statistical method. Importance attached by construction
clients to the `lowest-price wins' philosophy is also
presented. Contrast was made between the MCS
approach and the `lowest-price wins' option amongst the
surveyed construction clients. It was found that increased
awareness of the use of PSC prevailed amongst the
survey construction clients. This indicated that cost has
to be tempered with the evaluation of PSC and the
attempt of construction clients searching for a new
evaluation paradigm (i.e. adoption of MCS approach
rather than basing on the lowest-price wins alone).
Keywords contractor evaluation, evaluation criteria,
lowest-price wins, multi-criteria selection, project-
specic criteria
TENDER EVALUATI ON
Tender evaluation is performed once pre-qualied
tenderers have submitted their formal tender. The
scrutiny team may consist of in-house experienced
personnel, or clients' representatives (consultants and
construction specialists). Time and cost incurred in this
contractor assessment mainly rely on the nature of
tenderers' information and for types of assessment
methods used during this particular evaluation process.
Contractor evaluation has received a minimal amount
of attention in the UK industry (Merna & Smith 1990;
Holt et al. 1995). It has been the tendency that award of
contracts is merely on the comparison of tender price,
i.e. the `lowest-price wins' practice (ibid.). They found
that such practice allowed all tenderers who entered
into tender competition, very often taking little account
of other parameters (e.g. a contractor's nancial sound-
ness, management capabilities, technical expertise/
capability, etc.) during tender evaluation.
However, lowest-price does not guarantee the overall
lowest project cost upon project completion (Vorster
1977; Lewis 1981 3 ; Pearson 1985; Grogan 1992;
Dawood 1994; Klein 1994). Further, such a philosophy
poses a high risk to the client because there is an
increased possibility of nancial collapse of contractor,
bad performance, delay in completion, time and cost
over-runs and so on (Russell & Jaselskis 1992; Kwakye
1994; Holt et al. 1995).
Williamson (1975) dened the cost of transaction as
the cost of tendering, negotiating and compilation of
the contract, i.e. ex-ante cost; whilst the cost for
executing of the contract and its policy of resolving
disputes arising from the contracted work as ex-post cost.
Meanwhile Lingard et al. (1998) investigated the
intrinsic link between ex-ante cost for entering a
contractual relationship and the use of a variety of
contractor evaluation methods. They found that meth-
ods used in contractor evaluation have a vital impact on
the cost of a transaction; the ex-ante cost could be higher
than the ex-post cost in multi-criteria contractor selec-
tion models (compared with traditional competitive
tender methods). One reason for this is that quantita-
tive multi-criteria evaluation needs to address a broader
range of contractors' information (e.g. contractors'
likelihood of successful project execution, identication
of uncertainty and assessment of competence). Never-
theless, the measure reduces substantial ex-post costs by
Engineering, Construction and Architectural Management 2001 8 4, 257271
257
2001 Blackwell Science Ltd
avoiding contract(s) being awarded to `risky' contrac-
tor(s) and eliminating the multitude of problems that
inevitably follow the selection of an incompetent and/or
unqualied contractor.
Clearly, the rationale for using an objective tender
evaluation method is that, clients may accomplish most
of the objectives. That is, reduce ex-post cost and
minimize contract failure.
Empirical survey
Perceptions regarding the role of PSC were initially
observed from a thorough literature review. This
literature review includes good practice guidance use
in tender evaluation for building and civil engineering
works (e.g. IOB 1979; ICE 1980; CIC 1993; CIB
1997; CIRIA 1998). Identication of PSC also
observed from both the recommendations of industrial
practitioners and empirical ndings (e.g. Hunt et al.
1966; Helmer & Taylor 1977; Samelson & Levitt 1982 4 ;
Bastidas 1984; Bent 1984; Birrell 1988; Moore
1985a,b; Horgan 1987; Merna & Smith 1990; Holt
et al. 1994a,b,c; Kumaraswamy 1996; Housing Author-
ity List of Building Contractors, 1997; Hatush &
Skitmore 1997; Lo et al. 1998). Nine main PSC
categories were identied from the detailed literature
review. These categories are:
(1) manpower resources;
(2) plant and equipment resources;
(3) project management capabilities;
(4) geographical location knowledge;
(5) location of home ofce;
(6) contractor's capacity;
(7) project execution capabilities to the proposed
project;
(8) technical-economic analysis; and
(9) other relevant PSC (for particular types of work).
Thirty-seven PSC (criteria) attributed to these nine
categories are listed in Tables 1 and 2 for building and
civil engineering works, respectively. Category (8) was
not included in the contractor's questionnaire because
this category concentrated upon investigation of clients'
preferences in comparison with tender prices during
tender evaluation and thus will be discussed separately
from the remaining main PSC categories.
Data collection
The data used for analysis of PSC were obtained from
the nation-wide questionnaire survey. The question-
naire consisted of: the survey of respondents' charac-
teristics; respondents' opinions regarding use of PSC;
and the `lowest-price win' selection preferences. A Likert
scale (15) was used to measure respondents' opinions
regarding LIA for each PSC based on respondents'
experience in procuring building and civil engineering
works over the 2 years prior to the survey, amongst the
UK public/private sector clients and contractors.
Attempts have been made to include different and
widespread construction practitioners and size of
respondents 5 . The respondents were randomly selected
from typical construction professional lists, from public
and private clients and contractor organizations:
Municipal Year Book and Public Service Directory
(2000).
Association for Project Management Yearbook
(1997/1998).
The Property Profession Chartered Surveyor
Regional Directory-Midlands (1997).
Chartered Building Company Directory and
handbook (1997/1998).
Key British Enterprises Dun & Bradstreet (2000).
The survey sample comprised of 250 public clients
200 private clients and 250 contractors. There were
157 (22%) questionnaires returned and completed. Of
these, 68 were from public clients, 50 from private
clients and 39 from contractors. These were 75% from
England, 12% from Wales and Northern Ireland and
13% from Scotland. The surveyed data were analysed
and results compared. Detailed discussion of these
ndings is now presented in the following.
Samples characteristics
The nature of business of respondents varied widely,
with more than 33% from local authorities, which also
represented the biggest sample grouping of the survey.
This was followed by: builders (21%), project manage-
ment consultants (12%), others (10%), quantity survey-
ors (8%), civil engineering and utilities (8%), architects
(6%), and contracting (2%). `Others' were not dened.
The total amount of contracts awarded in the last
2 years was 6069 million for 7683 number of building
contracts and, 1953 million for 1900 number of civil
engineering contracts. These gures equate to an
average of 0.79 million and 1.03 million for building
and civil engineering contracts, respectively. This
shows that building works are responsible for a larger
number of contracts (but smaller average contract
values) as compared with civil engineering works
(which are smaller in number, but larger in value).
The data from respondents also indicate that private
respondents were responsible for larger contract values
compared to public sector respondents. That is, an
Wong, C. H. et al. 258
2001 Blackwell Science Ltd, Engineering, Construction and Architectural Management 8 4, 257271
average of 1.10 million per project (both building and
civil engineering works combined) and an average of
0.27 million per project, respectively.
To summarize, the empirical data consists of a large
number of projects with smaller contract values (i.e.
building works); a smaller number of projects with
larger contract values (i.e. civil engineering works) and
larger contract values in private clients (both in building
and civil engineering works) compared with public
clients. Therefore, the survey encompassed a broad
Table 1 Observed LIA of 37 PSC for building works.
Project-specic criteria Public Rank Private Rank C'tor Rank
Manpower resources
1. Quality and quantity of human resources 3.982 10.0 4.114 9.5 3.913 10.0
2. Quality and quantity of managerial staff 4.000 8.0 4.286 5.0 4.261 5.0
3. Amount of decision-making authority on site 3.471 22.0 3.851 15.0 3.652 17.0
4. Amount of key personnel for the project 3.961 11.0 4.229 6.0 3.913 10.0
Plant and equipment resources
5. Type of plants and equipment available 3.156 29.0 2.971 31.0 2.957 28.5
6. Size of equipment available 2.902 30.0 2.857 33.0 2.739 31.0
7. Condition and availability of equipment 3.431 23.0 2.886 32.0 2.957 28.5
8. Suitability of the equipment 3.567 19.0 3.171 28.0 3.217 22.5
Project management capabilities
9. Number of professional personnel available 3.596 18.0 3.629 20.0 3.696 15.0
10. Type of project control and monitoring procedures 3.692 17.0 4.114 9.5 3.870 12.0
11. Availability of project management software 2.490 35.0 2.829 34.0 2.826 30.0
12. Cost control and reporting systems 3.481 21.0 3.657 19.0 3.565 19.0
13. Ability to deal with unanticipated problems 4.125 6.0 4.543 2.0 4.522 2.5
Geographical familiarities
14. Contractor's familiarity with weather conditions 2.327 36.0 2.543 37.0 2.261 32.0
15. Contractor's familiarity with local labour 2.692 32.0 3.200 27.0 3.261 20.0
16. Contractor's familiarity with local suppliers 2.596 34.0 3.257 25.0 3.217 22.5
17. Contractor's familiarity with geographic area 2.712 31.0 3.057 30.0 3.217 22.5
18. Relationship with Local Authority 3.780 15.0 3.086 29.0 3.217 22.5
Location of home ofce
19. Home ofce location relative to job site location 2.183 37.0 2.771 35.0 3.087 27.0
20. Communication & transportation- ofce to job site 2.683 33.0 2.743 36.0 3.130 26.0
Capacity
21. Current workload 3.750 16.0 3.886 14.0 3.652 17.0
22. Maximum resource/nancial capacity 4.231 3.0 4.143 7.0 3.913 10.0
23. Finance arrangements 4.019 7.0 3.686 18.0 3.739 14.0
Project execution of the proposed project
24. Training or skill level of craftsmen 4.173 4.0 3.833 16.0 3.783 13.0
25. Productivity improvement procedures and awareness 3.327 25.0 3.408 24.0 3.174 25.0
26. Site organization, rules and policies (health and safety, etc.) 4.288 2.0 3.896 13.0 4.043 7.5
27. Engineering co-ordination 3.246 27.0 3.710 17.0 3.652 17.0
Other project-specic criteria
28. Actual quality achieved for similar works 4.157 5.0 4.371 3.0 4.522 2.5
29. Experience with specic type of facility 3.902 13.0 4.114 9.5 4.130 6.0
30. Proposed construction method 3.997 9.0 4.057 12.0 4.043 7.5
31. Ability to complete on time 4.746 1.0 4.686 1.0 4.696 1.0
32. Actual schedule achieved on similar works 3.885 14.0 4.324 4.0 4.391 4.0
Technical-economic analysis
33. Comparison of client's estimate with tender price 3.942 12.0 4.114 9.5
34. Comparison between proposal and average tender prices 3.173 28.0 3.600 23.0
35. Comparison for client's and proposed direct cost 3.500 20.0 3.617 21.0
36. Contractor's errors proposed construction method/procedure 3.383 24.0 3.600 22.0
37. Proposals review unit price/labour cost/resources schedule 3.294 26.0 3.229 26.0
Note: A Likert scale from 1 to 5 is used: 1 = not important, 3 = moderate, 5 = extremely important.
UK construction clients' tender evaluation preferences 1 259
2001 Blackwell Science Ltd, Engineering, Construction and Architectural Management 8 4, 257271
sample of `construction' respondents with a divergence
of projects and opinions, regarding the use of PSC and
the lowest-price wins preferences in both building and
civil engineering works.
Analysis of PSC
The use of statistical treatments presented in this paper
is a multivariate statistical analysis. That is, to investi-
gate the correlation and signicant differences between
Table 2 Observed LIA of 37 PSC for civil engineering works.
Project-specic criteria Public Rank Private Rank C'tor Rank
Manpower resources
1. Quality and quantity of human resources 4.000 9.5 4.200 9.5 4.000 12.0
2. Quality and quantity of managerial staff 4.120 8.0 4.400 4.5 4.286 6.5
3. Amount of decision-making authority on site 3.760 16.5 3.919 17.0 3.643 19.0
4. Amount of key personnel for the project 3.840 15.0 4.400 4.5 4.143 9.0
Plant and equipment resources
5. Type of plants and equipment available 3.400 24.0 3.067 32.5 3.214 25.0
6. Size of equipment available 3.000 31.0 2.933 35.5 2.929 28.5
7. Condition and availability of equipment 3.480 23.0 3.067 32.5 3.214 25.0
8. Suitability of the equipment 3.560 20.0 3.467 22.5 3.643 19.0
Project management capabilities
9. Number of professional personnel available 3.760 16.5 3.600 21.0 3.929 13.0
10. Type of project control and monitoring procedures 3.880 13.5 4.400 4.5 3.857 15.0
11. Availability of project management software 2.660 35.0 3.133 30.5 2.929 28.5
12. Cost control and reporting systems 3.500 22.0 3.667 20.0 3.571 21.0
13. Ability to deal with unanticipated problems 4.380 3.0 4.533 2.0 4.857 1.0
Geographical familiarities
14. Contractor's familiarity with weather conditions 2.640 36.0 2.867 37.0 2.429 30.0
15. Contractor's familiarity with local labour 3.080 28.5 3.267 28.5 3.286 22.5
16. Contractor's familiarity with local suppliers 2.960 32.0 3.000 34.0 3.214 25.0
17. Contractor's familiarity with geographic area 2.680 34.0 3.133 30.5 3.286 22.5
18. Relationship with Local Authority 3.351 26.0 2.933 35.5 3.143 27.0
Location of home ofce
19. Home ofce location relative to job site location 2.220 37.0 3.267 28.5 2.214 32.0
20. Communication and transportation ofce to job site 2.780 33.0 3.333 26.5 2.357 31.0
Capacity
21. Current workload 3.920 12.0 4.133 12.0 3.786 16.5
22. Maximum resource/nancial capacity 4.440 2.0 4.200 9.5 4.286 6.5
23. Finance arrangements 4.200 5.0 3.800 18.0 3.923 14.0
Project execution of the proposed project
24. Training or skill level of craftsmen 4.160 6.5 4.076 13.0 3.786 16.5
25. Productivity improvement procedures and awareness 3.080 28.5 3.419 24.0 3.643 19.0
26. Site organization, rules and policies (health and safety, etc.) 4.160 6.5 4.224 7.0 4.143 9.0
27. Engineering co-ordination 3.592 19.0 3.924 16.0 4.071 11.0
Other project-specic factors/criteria
28. Actual quality achieved for similar works 4.000 9.5 4.400 4.5 4.643 3.0
29. Experience with specic type of facility 3.880 13.5 4.200 9.5 4.429 4.5
30. Proposed construction method 3.956 11.0 4.000 14.5 4.143 9.0
31. Ability to complete on time 4.560 1.0 4.667 1.0 4.786 2.0
32. Actual schedule achieved on similar works 3.712 18.0 4.200 9.5 4.429 4.5
Technical-economic analysis
33. Comparison of client's estimate with tender price 4.240 4.0 4.000 14.5
34. Comparison between proposal and average tender prices 3.040 30.0 3.333 26.5
35. Comparison for client's and proposed direct cost 3.394 25.0 3.467 22.5
36. Contractor's errors proposed construction method/procedure 3.517 21.0 3.733 19.0
37. Proposals review unit price/labour cost/resources schedule 3.217 27.0 3.400 25.0
Note: A Likert scale from 1 to 5 is used: 1 = not important, 3 = moderate, 5 = extremely important.
Wong, C. H. et al. 260
2001 Blackwell Science Ltd, Engineering, Construction and Architectural Management 8 4, 257271
each PSC (based upon the LIA) and amongst the
sample groupings. The analysis was performed in the
most widely used statistics package, i.e. SPSS Version
9.0. The statistic procedures use include:
Spearman rank correlation coefcient (SRCC) test;
Two-way analysis of variance (ANOVA);
Error bar chat; and
Post hoc multiple comparisons.
The main features and mechanisms of the above tests
can be found in the investigations by Norusis (1995)
and Wong et al. (2000a, 2001). The following analyses
will be mainly concentrated on the quantitative analysis
of the survey data and subsequent discussion of the
ndings.
Correlation test
Tables 1 and 2 show the 37 numbers of PSC, arranged
under nine main PSC headings for both building and
civil engineering works, respectively, in public/private
clients and contractor organization. To begin with, the
observed LIA for each PSC were calculated based upon
the overall LIA scores observed from respondents for all
evaluation criteria. The ranking exercise was then
carried out based on the magnitude of LIA mean
response for each particular PSC (i.e. higher mean
response higher rank and vice-versa). This will then
be followed by the SRCC test. Results of the test are
shown in Tables 3 and 4.
Tables 3 and 4 show the correlation coefcients (r)
for building works. That is, 0.83 between public and
private clients, 0.86 between public clients and con-
tractor respondents and 0.96 between private client
respondents and contractor organizations (0.84, 0.85
and 0.90, respectively, for civil engineering works. All
results are signicant at the 0.01 level). Clearly, these
statistics show a high association in any pair of the
sample groupings, regarding the use of PSC measured
on the particular ranks in both building and civil
engineering works during tender evaluation.
Notwithstanding the strong association as shown in
Tables 3 and 4, it is essential to distinguish `associ-
ation' and `causation' relationships of these two sets of
data (Bryman & Cramer, 1999, pp. 181182). There-
fore, the investigation of `causation' is focused upon in
the following. Table 5 shows the coefcients of deter-
mination r
2
for the three pairs of combination sample
Table 3 SRCC test in building works.
Organization
types
Non-parametric
correlation
Public
client
Private
client Contractor
Public client Correlation coefcient 1.000
Sig. (two-tailed)
N (PSC) 32
Private client Correlation coefcient 0.833* 1.000
Sig. (two-tailed) 0.000
N (PSC) 32 32
Contractor Correlation coefcient 0.858* 0.959* 1.000
Sig. (two-tailed) 0.000 0.000
N (PSC) 32 32 32
* Correlation is signicant at the 0.01 level (two-tailed).
Table 4 SRCC test in civil engineering
works.
Organization
types
Non-parametric
correlation
Public
client
Private
client Contractor
Public client Correlation coefcient 1.000
Sig. (two-tailed)
N (PSC) 32
Private client Correlation coefcient 0.838* 1.000
Sig. (two-tailed) 0.000
N (PSC) 32 32
Contractor Correlation coefcient 0.854* 0.895* 1.000
Sig. (two-tailed) 0.000 0.000
N (PSC) 32 32 32
* Correlation is signicant at the 0.01 level (two-tailed).
UK construction clients' tender evaluation preferences 1 261
2001 Blackwell Science Ltd, Engineering, Construction and Architectural Management 8 4, 257271
groupings in building and civil engineering works. The
coefcients of determination for the use of PSC in all
sample groupings are signicantly close. In building
works, it was found that there were 69, 74 and 92%,
respectively, between public and private clients, public
clients and contractors, and private clients and con-
tractors (there were 71, 72 and 81%, respectively, in
civil engineering works). These gures show a high
causation of LIA variation of the PSC (in any pair of the
sample groupings combinations) was accounted by the
variation of LIA of the PSC from each respective
sample groupings. Thus, there is a strong association
regarding the ranking of criteria based upon LIA
amongst the respondent types.
However, the above ndings do not show which PSC
are causing this strong association; or which are
considered signicantly different from others across
the respondent types. The following ANOVA and mul-
tiple comparison analysis were used to conrm these
differences.
Tests for violation assumption
The parametric and non-parametric procedures were
used in this multivariate analysis for analysing the
extent of violation (i.e. multivariate assumptions) of the
surveyed data. It is concluded that if the results from
both tests are found not to differ greatly, then it can be
assumed that the data have been drawn from popula-
tion which did not violate the multivariate assumptions
(Bryman & Cramer, 1999, p. 119; Wong et al. 2000a,
2001). In addition, the Levene test was also used in this
instance to test the equal variance assumption of
multivariate data (Bryman & Cramer, 1999, p. 145).
The Levene tests
Table 6 shows the Levene test results for PSC in
building and civil engineering works. There are three
numbers of PSC, respectively, in building and civil
engineering works having unequal variance. There are
PSC3, PSC17 and PSC20 in Building Works. In civil
engineering works there are PSC3, PSC13 and PSC28.
The ndings reveal that there were approximately 9%
of the overall respondents' LIA opinions to the partic-
ular PSC in both building and civil engineering works
exhibiting the unequal variances. This indicates that
only a small portion of the survey data has unequal
variation in both different sectors of works. However, to
`extensively' evaluate the violation of multivariate
assumptions. Parametric and non-parametric tests for
validating the violation effects of multivariate assump-
tions are discussed 6 .
Parametric and non-parametric tests
Table 7 shows the parametric and non-parametric
(ANOVA) tests for data related to building works. It is
interesting to note that, in building works, PSC with
signicant differences amongst the sample groupings in
the non-parametric test are actually the same as those
for the parametric test.
In civil engineering works, the results are almost
identical. There are six numbers of PSC in parametric
test and ve in non-parametric tests that are signi-
cantly different across the sample groupings. Amongst
these, it was found that the PSC4, PSC19, PSC29 and
PSC32 are the criteria that are signicantly different
from LIA means amongst the 32 PSC. However,
PSC13 is the only criterion found signicantly different
amongst the respondents in parametric test (Table 8).
From the test results above, it is again apparent that
the data used in the analysis have little violation upon
the power of test statistics, regardless of statistic
procedures (parametric or non-parametric tests) used.
Therefore, the data can be assumed normally distri-
buted and of equal variance. Having proof that this data
is valid for use in multivariate analysis, the following
analysis used parametric procedures for nding the
signicant difference in opinions (i.e. LIA) regarding
PSC amongst the respondents.
Interaction plot
The interactions of PSC and organization types are
plotted in Figs 1 and 2. The vertical axis represents LIA
that had reported by different organizations, and the
horizontal axis represents the 32 variables (PSC).
Cursory perusal of the interaction plots, show that the
LIA affected by the effects of the organization types and
different PSC used during tender evaluation.
However, whether these effects are statistically
signicant can only be determined by testing them via
a two-way ANOVA statistical test. The two-way ANOVA
conrms whether the population means of LIA are
Table 5 Coefcient of determination (r
2
).
Building
Civil engineering
works
Organization types r r
2
r r
2
Public client/private client 0.83 0.69 (69) 0.84 0.71 (71)
Public client/contractor 0.86 0.74 (74) 0.85 0.72 (72)
Private client/contractor 0.96 0.92 (92) 0.90 0.81 (81)
Values in bracket are percentages.
Wong, C. H. et al. 262
2001 Blackwell Science Ltd, Engineering, Construction and Architectural Management 8 4, 257271
Table 6 Test of homogeneity of variances.
Levene
Building Civil
PSC1 0.441 2 107 0.644 PSC1 0.896 2 51 0.415
PSC2 0.291 2 107 0.748 PSC2 0.155 2 51 0.857
PSC3 3.440 2 107 0.036* PSC3 3.606 2 51 0.034*
PSC4 0.065 2 107 0.937 PSC4 0.719 2 51 0.492
PSC5 0.187 2 107 0.830 PSC5 0.529 2 51 0.593
PSC6 0.038 2 107 0.963 PSC6 0.285 2 51 0.753
PSC7 0.195 2 107 0.823 PSC7 0.447 2 51 0.642
PSC8 0.355 2 107 0.702 PSC8 1.385 2 51 0.260
PSC9 0.980 2 107 0.379 PSC9 0.914 2 51 0.408
PSC10 1.967 2 107 0.145 PSC10 0.230 2 51 0.796
PSC11 0.672 2 107 0.513 PSC11 0.199 2 51 0.820
PSC12 1.803 2 107 0.170 PSC12 1.973 2 51 0.149
PSC13 0.292 2 107 0.748 PSC13 10.582 2 51 0.000*
PSC14 1.121 2 107 0.330 PSC14 1.356 2 51 0.267
PSC15 2.916 2 107 0.058 PSC15 0.131 2 51 0.878
PSC16 1.565 2 107 0.214 PSC16 0.009 2 51 0.991
PSC17 6.422 2 107 0.002* PSC17 0.334 2 51 0.717
PSC18 0.965 2 107 0.384 PSC18 1.999 2 51 0.146
PSC19 0.326 2 107 0.723 PSC19 0.183 2 51 0.833
PSC20 3.795 2 107 0.026* PSC20 0.472 2 51 0.627
PSC21 0.616 2 107 0.542 PSC21 0.910 2 51 0.409
PSC22 0.030 2 107 0.971 PSC22 0.330 2 51 0.721
PSC23 1.416 2 107 0.247 PSC23 1.692 2 51 0.194
PSC24 0.483 2 107 0.618 PSC24 0.050 2 51 0.951
PSC25 1.771 2 107 0.175 PSC25 0.701 2 51 0.501
PSC26 0.070 2 107 0.932 PSC26 1.569 2 51 0.218
PSC27 1.268 2 107 0.286 PSC27 0.757 2 51 0.474
PSC28 0.508 2 107 0.603 PSC28 4.736 2 51 0.013*
PSC29 1.063 2 107 0.939 PSC29 0.178 2 51 0.837
PSC30 0.386 2 107 0.681 PSC30 0.097 2 51 0.908
PSC31 1.416 2 107 0.247 PSC31 2.037 2 51 0.141
PSC32 1.310 2 107 0.274 PSC32 0.008 2 51 0.992
* Signicantly different at 0.05 level.
Table 7 Parametric and non-parametric tests in building works.
ANOVA (parametric) KruskalWallis (non-parametric)
Sum of square Mean square F Signicance Chi-square Signicance
PSC7* Between groups 7.408 3.704 3.330 0.040 PSC7* 6.612 0.037
Within groups 119.009 1.112
PSC13* Between groups 4.604 2.302 5.256 0.007 PSC13* 9.496 0.009
Within groups 46.862 0.438
PSC15* Between groups 7.806 3.903 3.899 0.023 PSC15* 6.834 0.033
Within groups 107.112 1.001
PSC16* Between groups 11.437 5.718 4.970 0.009 PSC16* 9.438 0.009
Within groups 123.118 1.151
PSC18* Between groups 11.530 5.765 5.546 0.005 PSC18* 11.308 0.004
Within groups 111.236 1.040
PSC19* Between groups 15.354 7.677 6.816 0.002 PSC19* 12.635 0.002
Within groups 120.512 1.126
PSC32* Between groups 6.004 3.002 4.791 0.010 PSC32* 11.258 0.004
Within groups 67.042 0.627
Note: All tests signicant at 0.05 level.
* Signicantly different in the means of LIA across the organization types in both parametric and non-parametric tests.
UK construction clients' tender evaluation preferences 1 263
2001 Blackwell Science Ltd, Engineering, Construction and Architectural Management 8 4, 257271
equal to the corresponding PSC amongst the organiza-
tion types and, whether there is an interaction between
PSC and organization types to give equal or unequal
effects on LIA.
Two-way ANOVA
Two-way ANOVA is designed to test differences between
the means of variables based upon the interaction and
main effects of two or more factors, i.e. independent
variables (which account for the variability of LIA
means) (Norusis, 1995, pp. 304316). Investigation of
the null hypothesis for main effects describes the effects
of PSC and organization types to the LIA are all the
same, whilst the null hypothesis for two-way interaction
is that the effect of PCS type on the LIA means is the
same for all organization types (i.e. no interaction exists
to contribute an effect on LIA).
The output of two-way ANOVA for both main effects
and interaction effects in PSC and organization types for
building and civil engineering works are shown in
Tables 9 and 10, respectively. Test results in both
building and civil engineering works show signicant
main effects of PSC and organization types
(P < 0.0005). This identies that, the population
means of LIA is not equal to all PSC (amongst the
three organization types), that is, the effects of PSC and
organization types to the LIA were different.
However, interaction effects test of PSC and organiza-
tion types upon the LIA mean populations was signi-
cantly different (P < 0.0005) in building works, but
not in civil engineering works (P 0.376). These
statistics show that, in building works, LIA means
depend on the interaction effects of organization types
Table 8 Parametric and non-parametric tests in civil engineering works.
ANOVA (parametric) KrussallWallis (non-parametric)
Sum of quare Mean square F Signicance Chi-square Signicance
PSC4* Between groups 3.029 1.515 3.407 0.041 PSC4* 6.302 0.043
Within groups 22.674 0.445
PSC13 Between groups 2.047 1.023 3.244 0.047 PSC19* 8.286 0.016
Within groups 16.088 0.315
PSC19* Between groups 11.915 5.957 4.723 0.013 PSC28* 11.864 0.003
Within groups 64.330 1.261
PSC28* Between groups 4.019 2.010 6.918 0.002 PSC29* 6.205 0.045
Within groups 18.833 0.290
PSC29* Between groups 2.865 1.432 3.569 0.035 PSC32* 6.950 0.031
Within groups 20.469 0.401
PSC32* Between groups 5.185 2.593 3.593 0.035
Within groups 36.795 0.721
Note: All tests signicant at 0.05 level.
* Signicantly different in the means of LIA across the organization types in both parametric and non-parametric tests.
Figure 1 Interaction plot of LIA for building works.
Figure 2 Interaction plot of LIA for civil engineering works.
Wong, C. H. et al. 264
2001 Blackwell Science Ltd, Engineering, Construction and Architectural Management 8 4, 257271
and PSC. Therefore, there were interaction effects of
PSC and organization types towards the LIA. However,
this is not the case in civil engineering works.
To summarize, in building works, respondents
viewed PSC differently in LIA attributed particular
criteria, the LIA variation relies on which PSC used
and types of organizations. This, however, was differ-
ent in civil engineering works, where the effect of the
type of PSC on the mean of LIA seems to be
statistically similar for all organization types. The
following sections discuss how and where these signi-
cant differences occur. The error bar chart analysis
and post hoc multiple comparison procedure are used
to achieve the above aims.
Error bar chart
An error bar is centred on the mean of a distribution
and extends above and below to show a condence
interval (CI) or a specied number of standard errors
(or standard deviations). The error bar displays both
the central tendency (i.e. mean) and variability of the
data (i.e. lower and upper bounds of the mean
variability). The objective is to nd out how represen-
tative a sample means is, via inspecting the variability of
CIs with regard to the population from which the
sample was drawn (Norusis, 1995, p. 541).
Figs 3 and 4 show the 95% CIs of LIA for client
organizations. Public and private clients are combined
in this instance, to give brevity and reduce the number
of error bars. These LIA mean populations distributed
between the upper and lower bounds of 95% of CI,
mean that it is 95% condence that the observed LIA
mean populations fall into in this region.
Table 9 Two-way ANOVA for PSC in build-
ing works. 21
LIA
Sum of
squares d.f.
Mean
square F Signicance
Main effects
(Combined) 1061.812 33 32.176 36.727 0.000
Organization types 8.358 2 4.179 4.77 0.009
PSC 1053.454 31 33.982 39.788 0.000
Two-way interactions
Organization types/PSC 108.283 62 1.747 1.994 0.000
Model 1326.746 95 13.966 15.941 0.000
Residual 2999.753 3424 0.876
Total 4326.498 3519 1.229
Table 10 Two-way ANOVA for PSC in civil
engineering works.
LIA
Sum of
squares d.f.
Mean
square F Signicance
Main effects
Combined 542.919 33 16.452 18.621 0.000
Organization types 8.48 2 4.24 4.799 0.008
PSC 534.439 31 17.240 19.513 0.000
Two-way interactions
Organization types/PSC 57.432 62 0.926 1.048 0.376
Model 630.199 95 6.634 7.508 0.000
Residual 1441.891 1632 0.884
Total 2072.09 1727 1.200
Figure 3 Error bar chart plot: PSC condence interval for building
works.
UK construction clients' tender evaluation preferences 1 265
2001 Blackwell Science Ltd, Engineering, Construction and Architectural Management 8 4, 257271
In this instance, the error bar is plotted to compare
respondents' opinions regarding LIA means for each
PSC. Thus, by visually inspecting the error bar
patterns, it was found that most of the above and
below bounds of the error bars are similar, between
clients and contractors. The variability of LIA mean
populations in both clients and contractor organiza-
tions can be clearly observed within the upper and
lower bounds of bar errors.
It is apparent that there is a strong correlation in
opinion regarding the use of PSC between clients and
contractor respondents. Also, it shows that the opinion
regarding the use of the PSC in client respondents are
very similar to that reported in contractor organiza-
tions. This strong correlation was also observed in the
earlier SRCC investigation. These ndings show the
close variability of LIA (for each PSC) both in clients
and contractors, but do not dene precisely which LIA
mean is statistically different from others, respectively,
in public and private clients and contractor organiza-
tions. The post hoc multiple comparison procedure was
used in the following to nd the signicant differences
amongst the sample groupings with regard to LIA of
the PSC.
Investigation of true differences
One-way ANOVA is used to test a hypothesis about two
or more population means (refer Tables 7 and 8), to
conrm signicant differences of means in two or more
variables. But it does not reveal exactly where such
differences exist amongst the organization types (i.e.
public clients, private clients and contractor organiza-
tions). In this analysis, for example, it may be that the
LIA mean differs from all of the three different types of
respondents. Or, it may be that only one or two of the
sample groupings differ from others.
Therefore, to pinpoint exactly which sample group-
ings statistically differ from others, in the LIA, in that
particular observed PSC, multiple comparison proce-
dures were used to dene exactly where these differ-
ences are. There are many multiple comparison
procedures available in SPSS. The Bonfferroni proce-
dure is used in this analysis. A 0.017 level of signi-
cance was used as a measure of strength or `cut off'
value to distinguish the signicant differences amongst
three sample groupings (see Bryman & Cramer 1999,
p. 106 and Russell et al. 1992, 7 works for details).
Table 11 shows those PSC with signicant `true'
difference amongst the respondent groupings.
Table 11 conrms three numbers of PSC in building
works and one in civil engineering works statistically
different in LIA means amongst the three sample
groupings. In building works, private respondents
viewed contractor's ability to deal with unanticipated
problems (PSC13) as being of vital importance and
signicantly different from public clients. Perhaps this
reects the fact that private clients might consider a
contractor to be able to resolve problems and conicts
Figure 4 Error bar chart plot: PSC condence interval for civil
engineering works.
Table 11 Post hoc multi-comparisons in building and civil engineering works. 22
Mean
SE Signicance
Building PSC*
PSC13 Private client 4.543 Public client 4.125 0.42 0.145 0.014
PSC18 Public client 3.780 Private client 3.086 0.69 0.223 0.007
PSC19 Contractor 3.087 Public client 2.183 0.90 0.266 0.003
Civil engineering works PSC*
PSC28 Contractor 4.643 Public client 4.000 0.64 0.18 0.002
* All PSC arranged in the sequence as per Tables 1 or 2 according to the number cited.