Predicting Performance of Design Build A
Predicting Performance of Design Build A
Predicting Performance of Design Build A
Abstract: Design-build ~DB! and design-bid-build ~DBB! are two principal project delivery systems used in many countries. This paper
reports on models constructed to predict performance of DB and DBB projects on 11 areas, using project-specific data collected from 87
Downloaded from ascelibrary.org by CASA Institution Identity on 04/20/19. Copyright ASCE. For personal use only; all rights reserved.
building projects. The study included collecting, checking, and validating industry data, and the statistical development of multivariate
linear regression models for predicting project performance. Robust models are developed to predict construction and delivery speeds of
DB and DBB projects. Gross floor area of the project is the most significant factor affecting speed. Besides this, for DBB projects,
contractors’ design ability, and adequacy of plant and equipment would ensure speedy completion of the projects. For DB projects, if the
contract period is allowed to vary during tender evaluation, this would slow down the project. Robust models to predict turnover and
system quality of DB projects are also constructed. A DB contractor’s track record is an important variable. They must have completed
past projects to acceptable quality and have ability in financial, health and safety management.
DOI: 10.1061/~ASCE!0733-9364~2004!130:1~75!
CE Database subject headings: Performance evaluation; Design/build; Project management; Models.
Introduction predict what the likely project performance level will be. This is
useful because based on the predicted project performance, own-
The design-bid-build ~DBB! procurement method is the prevalent
ers and A/Es will be able to decide if they should use the DBB or
procurement method in many countries such as Singapore, the
DB procurement method in order to obtain the desired results. If
U.K., and the United States. DBB is the traditional project deliv-
they have already decided on a certain procurement method, the
ery system where the owner contracts separately with a designer
models will help them decide what the key variables which need
and a constructor to design and construct the facility, respectively
to be controlled in order to obtain good project performance. Per-
~Mohsini and Davidson 1992!. One of the alternative procurement
formance of a project is multifaceted. 11 possible performance
systems is the design-build ~DB!, whereby the owner contracts
measures are shown in Table 1, and grouped into four categories:
with a single entity to perform both design and construction under
cost, time, quality, and others.
a single DB contract ~Janssens 1991!.
All the projects investigated in this study were based in Sin-
The objectives of this paper are ~1! to find explanatory vari-
gapore. They were all grass-root building construction projects
ables that significantly affect project performance and ~2! to con-
~i.e., not renovation works! exceeding $5 million, and were com-
struct models to predict the performance of DB and DBB
pleted between 1993 and 2001. Both private and public sector
projects. The first objective is important because contractors will
projects were investigated.
know the important variables that they must pay very close atten-
tion to in order that their projects can be completed within budget
and schedule, to acceptable level of quality, and to the satisfaction
of the owner and consultants. The second objective is important Literature Review
because the project performance models developed in this study
can help owners, contractors, and architects and engineers ~A/Es! In the U.K., Bennett et al. ~1996! studied DB and DBB project
selection and performance from the owners’ perspective. They
1 constructed three models to predict unit cost, construction speed,
Assistant Professor, Dept. of Building, National Univ. of Singapore,
4 Architecture Dr., Singapore 117566 ~corresponding author!. E-mail: and delivery speed, and obtained R 2 of 0.51, 0.90, and 0.80, re-
[email protected] spectively. The models were developed based on more than 170
2
Assistant Professor, Dept. of Building, National Univ. of Singapore, projects. When trying to predict one performance metric ~example
4 Architecture Dr., Singapore 117566. E-mail: [email protected] construction speed!, the study included other performance metrics
3
Research Assistant, Dept. of Building, National Univ. of Singapore, 4 as predictor variables ~example quality and unit cost!. This made
Architecture Dr., Singapore 117566. the constructed model difficult to use, as the evaluator would not
4
Research Assistant, Dept. of Building, National Univ. of Singapore, 4 have the information of the other independent variables before the
Architecture Dr., Singapore 117566. project starts. As can be seen from Table 1, there are many other
Note. Discussion open until July 1, 2004. Separate discussions must performance metrics that were not reported in the Bennett et al.
be submitted for individual papers. To extend the closing date by one
~1996! study.
month, a written request must be filed with the ASCE Managing Editor.
The manuscript for this paper was submitted for review and possible In the United States, Konchar and Sanvido ~1998! conducted
publication on May 15, 2002; approved on November 7, 2002. This paper an empirical study that examined explanatory and interacting
is part of the Journal of Construction Engineering and Management, variables to predict project performance based on DB, DBB, and
Vol. 130, No. 1, February 1, 2004. ©ASCE, ISSN 0733-9364/2004/1- construction management at risk procurement systems. Using
75– 83/$18.00. multivariate regression analysis, they developed models to predict
Time
Y4 Construction speed ~m2/month! Area/~as-built construction end date2as-built construction start date!
Y5 Delivery speed ~m2/month! Area/total time
Y6 Schedule growth ~%! @Total time2total as-planned time!/total as-planned time#*100
Downloaded from ascelibrary.org by CASA Institution Identity on 04/20/19. Copyright ASCE. For personal use only; all rights reserved.
Quality
Y7 Turnover quality Ease of starting up and extent of call backs.
~55exceed owner’s expectation; 15not satisfactory!
Y8 System quality Performance of building elements, interior space and environment.
~55exceed owner’s expectation; 15not satisfactory!
Y9 Equipment quality Performance of equipment. ~55exceed owner’s expectation; 15not satisfactory!
Others
Y10 Owner’s satisfaction 55exceed owner’s expectation; 15not satisfactory
Y11 Owner’s administrative burden 55minimum burden; 15very heavy burden
Note: Adapted from Konchar and Sanvido ~1998! and Molenaar and Songer ~1998!.
unit cost, construction, and delivery speeds based on 316 projects. mance metrics ~see Table 1! and instead concentrated on several
However, the regression equations, coefficients of variables, and areas of project performance. Many of the studies, even after
R 2 were not reported in detail, thus limiting the usefulness of the developing the regression models, did not provide detailed infor-
models. mation such as regression coefficients. This limits the use of the
Molenaar and Songer ~1998! developed prediction models for models. Based on the above, there appears to be a need to develop
the United States public-sector DB project selection, which re- more tools to be used for predicting how DB and DBB projects
lated specific project characteristics to success. A multiattribute would perform. This study therefore aims to fill this gap.
regression technique was used to develop the prediction models
based on 122 projects. Five models were developed for cost and
schedule growth, conformance to expectations, administrative Research Method
burden, and overall user satisfaction. The models are generally
not robust, with R 2 ranging from 0.28 to 0.47. However, the work From the review of past works, 59 potential factors affecting
is useful because the models and equations are provided for use to project performance were identified ~see Table 2!. The factors
predict performance of future projects. The limitation of this work affecting project success are categorized into attributes relating to
is that the models were constructed using only public sector DB the project, owners, consultants, and contractors. The research
projects. So, private sector DB projects, and DBB projects could method had the aim of identifying the important factors ~listed in
not rely on this study for performance prediction. Moreover, Table 2! that affect project performance ~defined in Table 1!. To
many other performance metrics ~see Table 1! were not reported. achieve this, a retrospective case study questionnaire ~data collec-
In Hong Kong, Chan et al. ~2001! constructed models to pre- tion instrument! was designed. A pilot study was first carried out
dict time and cost performance and overall DB project perfor- to identify possible inadequacies in the data collection instrument.
mance using 19 DB projects, with information supplied by 53 The finalized form provided the basis for the implementation of a
project participants. Thirty-one attributes that may affect project questionnaire survey, which constituted the primary data collec-
performance were reduced, using factor analysis, to six main tion method for this study.
project success factors. Like Molenaar and Songer ~1998!, the Projects were identified from the Singapore Building and Con-
models developed in this study can only be used to predict public struction Authority’s ~BCA! website ~BCA 2002!. All grass-root
sector DB project performance. The study may have used all the building projects exceeding $5 million in value, and com-
53 project participants’ responses to develop the models, and this pletedafter 1992 were identified. These were numbered chrono-
could give rise to double-counting, as there were only 19 projects. logically based on project end-date. Using a table of random num-
When a few respondents provided data for one project, they did bers, 400 projects were randomly identified. In the list of the 400
not average the ratings. The relatively small number of projects projects, many consultants and contractors have undertaken mul-
may cause the findings to be nongeneralizable and violate the tiple projects. A decision was made to request each firm to pro-
assumptions of normality in their statistical analysis. vide information on a maximum of three projects, to avoid fa-
From the brief literature review, a few gaps in knowledge tigue. Sixty contractors were therefore selected to provide
emerge. Many of the studies concentrated only on DB in the information of 180 projects, 57 consultants to provide information
public sector. While Konchar and Sanvido ~1998! and Bennett of 171 projects, and the balance 49 projects were requested from
et al. ~1996! looked into both public and private sectors, like all 40 owners ~35 owners were asked to provide information of one
other previous studies, they did not investigate all the 11 perfor- project each, one owner to provide information of two projects,
Contractor characteristics
X39 Contractor’s experience with similar types of projects 15no similar projects; 55nearly all those types
X40 Contractor’s experience with similar size of projects 15no similar projects; 55nearly all those types
X41 Contractor’s experience with projects in Singapore 15no similar projects; 55nearly all those types
X42 Subcontractors’ experience and capability 15poor; 55excellent
Downloaded from ascelibrary.org by CASA Institution Identity on 04/20/19. Copyright ASCE. For personal use only; all rights reserved.
and four owners to provide information of three projects each!. tor projects. The DB models would apply to projects with a large
Fewer questionnaires were sent to owners because past research range of floor area. However, there is a skew toward public sector
experience showed that very few of them respond to surveys in DB projects. The entire data set was used for analysis in lieu of a
Singapore. To provide the data, respondents had a choice of being smaller focused set because the original data set was not large.
interviewed or self-administering the questionnaires, and sending
them back to the researchers. After the completed questionnaires
were received, multiple linear regression ~MLR! modeling was Multivariate Analysis
undertaken to construct models to predict each of the 11 project
performance measures identified in Table 1. More project data In this study, multivariate regression analysis was used to develop
sets were subsequently collected, and used to validate the con- 11 models to determine the statistical relationship between some
structed models. response variables of DBB projects ~for example cost growth and
construction speed, see Table 1! and the explanatory variables ~for
example gross floor area, type of client, and adequacy of contrac-
Data Sample Characteristics tor’s plant and equipment, see Table 2!. Another 11 models were
developed to predict performance of DB projects. The models
Data sets of 87 projects were received. 27 firms provided data of were developed using traditional regression techniques with the
54 DBB projects, while another 15 firms provided data of 33 DB help of the Statistical Package for Social Sciences software
projects. For DBB projects, 17, six and four contractors, A/Es, ~SPSS!.
and owners, respectively, provided the project data. Seven con- In this research, the independent/predictor variables are the
tractors, six A/Es, and two owners projected data for DB projects. attributes relating to the project, consultants, owners, and contrac-
The details of the projects are shown in Table 3. It can be seen tor listed in Table 2. For each model, the dependent variable is
that the models constructed in this study would be more appli- one of the performance metrics identified in Table 1. Each model
cable to residential projects which are undertaken by both the is expressed by the following equation:
public and private sectors in Singapore. The data of schools and
Y i 5a1b 1 x 1i 1b 2 x 2i 1b 3 x 3i 1¯1b K x Ki 1« i (1)
offices formed a minority, but were not removed from the data set
because the Engineering New Record classifies these together where Y5value of the dependent variable ~one of the performance
with residential construction under General Building. The DBB metrics in Table 1!; a5constant, and the intercept at the Y axis;
models would apply more to medium sized projects up to 100,000 b 1 to b K 5estimated regression coefficients; X 1 to X K 5values of
m2. These models would apply equally to public and private sec- the independent or explanatory variables; « i 5error term, which is
Y3 DBB . Intensity, R 2 50.276, Adj R 2 50.242 X48 12.29 3.15 0.63 3.90 0.002
Constant 81.29 26.9 NA 3.016 0.004 X18 3.377 1.33 0.41 2.531 0.024
X10 216.07 5.14 20.41 23.127 0.003 X13 6.536 2.70 0.39 2.424 0.029
X18 8.898 3.78 0.31 2.351 0.023
Y4 DB . Construction speed, R 2 50.782, Adj R 2 50.765
2 2 Constant 1,628.8 557.7 NA 2.921 0.007
Y4 DBB . Construction speed, R 50.829, Adj R 50.820
Constant 2357.0 498.8 NA 20.716 0.479 X1 0.034 0.004 0.99 9.34 0.000
X1 0.021 0.002 0.91 13.14 0.000 X24 2824.28 322.9 20.27 22.553 0.017
X50 277.3 123.0 0.16 2.254 0.030
Y5 DB . Delivery speed, R 2 50.896, Adj R 2 50.879
2 2 Constant 3,462.2 785.3 NA 4.409 0.000
Y5 DBB . Delivery speed, R 50.934, Adj R 50.930
Constant 2144.9 170.6 NA 20.850 0.402 X1 0.024 0.002 0.895 11.08 0.000
X1 0.017 0.001 0.93 19.62 0.000 X15 2463.57 151.3 20.21 23.065 0.005
X58 132.8 51.8 0.12 2.563 0.015 X24 2443.01 180.8 20.19 22.451 0.022
X18 2180.16 77.55 20.16 22.323 0.029
Y6 DBB . Schedule growth, R 2 50.406, Adj R 2 50.391
Constant 58.37 10.11 NA 5.776 0.000 Y6 DB . Schedule growth, R 2 50.129, Adj R 2 50.100
X49 213.38 2.53 20.64 25.291 0.000 Constant 26.61 10.92 NA 2.645 0.013
X21 25.034 2.58 20.39 22.112 0.043
Y7 DBB . Turnover quality, R 2 50.195, Adj R 2 50.175
Constant 2.71 0.28 NA 9.680 0.000 Y7 DB . Turnover quality, R 2 50.821, Adj R 2 50.783
X2 0.702 0.22 0.44 3.185 0.003 Constant 1.44 0.514 NA 2.804 0.014
X48 0.424 0.135 0.42 3.147 0.007
Y8 DBB . System quality, R 2 50.375, Adj R 2 50.344 X1 253106 0.000 20.52 24.537 0.000
Constant 2.94 0.42 NA 7.022 0.000 X56 0.29 0.114 0.34 2.547 0.023
X55 0.331 0.09 0.46 3.623 0.001
X14 20.184 0.06 20.37 22.973 0.005 Y8 DB . System quality, R 2 50.888, Adj R 2 50.862
Constant 0.114 0.444 NA 0.257 0.801
Y9 DBB . Equipment quality, R 2 50.550, Adj R 2 50.512 X54 1.042 0.159 0.94 6.566 0.000
Constant 2.122 0.62 NA 3.419 0.002 X33 0.265 0.046 0.58 5.81 0.000
X36 0.284 0.08 0.41 3.394 0.002 X50 20.391 0.157 20.37 22.496 0.027
X48 0.379 0.12 0.37 3.056 0.004
X41 20.233 0.08 20.34 22.945 0.006 Y9 DB . Equipment quality, R 2 50.746, Adj R 2 50.706
Constant 20.238 0.661 NA 20.36 0.725
Y10DBB . Owner’s satisfaction, R 2 50.780, Adj R 2 50.746 X54 1.042 0.141 0.75 5.294 0.000
Constant 2.07 0.47 NA 4.417 0.000 X33 20.391 0.102 0.35 2.484 0.027
X57 0.206 0.10 0.23 2.087 0.045
X51 20.232 0.06 20.33 23.945 0.000 Y10DB . Owner satisfaction, R 2 50.720, Adj R 2 50.677
X48 0.483 0.11 0.49 4.435 0.000 Constant 20.348 0.767 NA 20.454 0.657
X24 0.180 0.05 0.31 3.565 0.001 X57 0.701 0.174 0.631 4.038 0.001
X14 20.163 0.06 20.26 22.989 0.005 X56 0.381 0.152 0.392 2.509 0.026
Y11DBB . Owner’s admin burden R 2 50.095, Adj R 2 50.076 Y11DB . Owner’s admin burden, R 2 50.511, Adj R 2 50.509
Constant 2.102 0.51 NA 4.107 0.000 Constant 0.287 0.829 NA 0.347 0.734
X49 0.298 0.14 0.31 2.219 0.031 X48 0.839 0.213 0.713 3.944 0.001
Note: Regression coefficient ~b!, calculated using ordinary least-squares Note: Same notes as Table 4.
method. Standard error ~s! of variable regression coefficient, measures
the dispersion of regression coefficient over sampling distribution. Stan-
dardized regression coefficient ~b!, allows for equal comparison of coef-
ficient weights, when the constant is removed. Value of t-statistic, to be Time Performance
compared to the theoretical t-distribution for accuracy. Significance of
t-statistic. For significance ,0.05, the null hypothesis that b50 is re- In this study, robust models for predicting construction speeds
2 2
jected. There is less than 5% chance that t-statistic is wrong due to ran- were developed (R DBB 50.83; R DB 50.78). The models are ex-
dom occurrence. tracted from Tables 4 and 5, and presented below.
for completion toacceptable quality Model validation was carried out on the two robust DBB models
and four robust DB models presented earlier. A diagnostic test
253106 •Gross Floor Area was conducted, and the residual plots of R ~being actual Y less
predicted Y! versus ‘‘predicted Y’’ show a random distribution.
10.29•Contractor’s health and safety management This confirms that the assumption about normality is valid. As
capability (6) described earlier, when two predictor variables were found to be
significantly correlated, one of them was excluded from the
Y8 DB50.11411.04•Contractor’s ability in financial model. Therefore, multicollinearity is absent from the models. To
validate these prediction models further, six new DBB and five
management10.27•Consultant’s experience new DB project data were collected ~see Table 6!. The predicted
project performance measures ~‘‘predicted Y’’! were derived
with similar projects
mathematically from the models. These predicted results were
20.39•Adequacy of contractor’s plant compared to the actual performance ~‘‘actual Y’’! of the projects.
The relative measures of accuracy used to validate the predic-
and equipment (7) tion results of the models are given below.
Actual value2Model’s predicted value The first limitation is that only 87 project data sets were used to
5 3100% construct the models, which appeared to be small. The value of
Actual value
construction put in place in 2000 in the United States was $815
(8) billion ~U.S. Census Bureau 2002! while that of Singapore was $8
n
( i5n PE billion ~BCA 2002!. Therefore, proportionally, if U.S. studies ana-
Mean percentage error ~MPE)5 (9) lyzed 300 sets of data ~example, Konchar and Sanvido 1998!, the
N
87 data sets collected for this study is not small, when compared
where n5number of predictions. to the overall size of the Singapore construction industry.
n The second limitation is the possible biases that might be cre-
( i5nu PEu
Mean absolute percentage error ~MAPE)5 ated when respondents answer some of the questions. For ex-
n ample, owner’s satisfaction ~Y10! and owner’s administrative
(10) burden ~Y11! may not be accurately answered by contractors or
where uPEu5absolute value of the percentage error. A/Es. Finally, the use of only six and five DBB and DB projects
Table 7 shows that all models achieved relatively accurate rep- to test the models. However, this is not expected to nullify the
resentation of the actual results with MPE ranging from 28.2 to results for two reasons. First, the triangulation rule ~Hammersley
16.22%. The models are robust and appear to be able to predict a and Atkinson 1983! was met. Second, some 54 and 33 DBB and
project’s performance that would closely reflect its actual perfor- DB project data sets were fed into the SPSS to construct the
mance. The robustness of the models does not appear to have multiple linear regression models.
been affected even though the data set contained both residential
and industrial constructions which vary in many aspects.
Conclusion
Practical Application The essence of this study is that significant variables that affect
different aspects of project performance were identified in Tables
The two robust DBB and four robust DB models may be used to 4 and 5, thus achieving the first objective. Certain performance
predict certain aspects of project performance. An example of aspects of DBB and DB projects can be predicted, thus fulfilling
how the model for predicting construction speed of projects is the second objective of this study. The predictions can be made
now given. The construction speed of a project can be predicted through the models shown in Tables 4 and 5. Six robust models
by using the Eqs. ~2! and ~3!. Suppose there is a project with were developed. They have high R 2 , the regression coefficients
gross floor area of 10,000 m2. The contract period is not allowed
to vary. The potential contractor has many plants and equipment.
These information are input into Eqs. ~2! and ~3!, to obtain Table 7. Comparison of Predicted and Actual Performance
2
Y4 DBB5235710.021310,00012773551,238 m /month Mean percentage Mean absolute
(11) error percentage error
Performance measure ~%! ~%!
Y4 DB51,62910.034310,00028243151,145 m2 /month
(12) Design-bid-build construction speed 6.22 25.34
Design-bid-build delivery speed 26.22 28.85
Eqs. ~11! and ~12! above show what the likely construction
Design-build construction speed 23.31 14.11
speeds of the project would be, if DBB and DB procurement
Design-build delivery speed 28.20 9.21
systems are used. If the owner would like the construction speed
Design-build turnover quality 20.76 14.16
to be high, DBB procurement method should be chosen in this
Design-build system quality 0.97 6.24
example.
tract period is allowed to vary during bid evaluation. Turnover project success factors: Multivariate analysis.’’ J. Constr. Eng. Man-
and system quality of DB projects can also be predicted, as shown age., 127~2!, 93–100.
in Table 5. Hammersley, M., and Atkinson, P. ~1983!. Ethnography principles in
Many other attributes which would ordinarily be thought to be practice, Routledge, London.
Janssens, D. E. L. ~1991!. Design-build explained, Macmillan, London.
significant proved instead to be insignificant in predicting project
Kaming, P. F., Olomolaiye, P. O., Holt, G. D., and Harris, F. C. ~1997!.
performance. However, it does not mean that these attributes are
‘‘Factors influencing construction time and cost overruns on high-rise
not important, as correlation analyses show that these are corre- projects in Indonesia.’’ Constr. Manage. Econom., 15~1!, 82–94.
lated to project performance. What is obvious however, is that Konchar, M., and Sanvido, V. ~1998!. ‘‘Comparison of U.S. project de-
they need not be used to predict project performance. livery systems.’’ J. Constr. Eng. Manage., 124~6!, 435– 444.
It is recommended that those who need to decide on whether Kumaraswamy, M. M., and Chan, W. M. ~1995!. ‘‘Determinants of con-
to use DB or DBB rely on the models developed in this study to struction duration.’’ Constr. Manage. Econom., 13~2!, 209–217.
predict the project’s likely performance. The practical application Mohsini, R. A., and Davidson, C. H. ~1992!. ‘‘Determinants of perfor-
of this research finding for contractors is that in order to ensure mance in the traditional building process.’’ Constr. Manage. Econom.,
their projects have high performance, they should concentrate on 10~4!, 343–359.
the important attributes which can be used to predict perfor- Molenaar, K. R., and Songer, A. D. ~1998!. ‘‘Model for public sector
design-build project selection.’’ J. Constr. Eng. Manage., 124~6!,
mance.
467– 479.
This study has extended previous works particularly those of Newbold, P. ~1991!. Statistics for business and economics, 3rd Ed.,
Konchar and Sanvido ~1998! and Chan et al. ~2001!. More mod- Prentice-Hall, Englewood Cliffs, N.J.
els were constructed in this study which were not found in the U.S. Census Bureau. ~2002!. ‘‘Annual value of construction put in place
work of Konchar and Sanvido ~1998!. These include intensity, in the United States: 1996 –2000.’’ USA: Census Bureau, ^http://
turnover quality, system quality, equipment quality, owner’s sat- www.census.pub/const/C30/tab100.txt& ~Apr. 19, 2002!.