Journal of Cleaner Production 142 (2017) 2804e2815

Contents lists available at ScienceDirect

Journal of Cleaner Production

journal homepage: www.elsevier.com/locate/jclepro

Total factor carbon emission performance measurement and

development
Xiancun Hu a, Tongguang Si b, Chunlu Liu a, *
a
School of Architecture and Built Environment, Deakin University, Geelong, Australia
b
School of Management Engineering, Shandong Jianzhu University, Jinan, China

a r t i c l e i n f o a b s t r a c t

Article history: Few studies focused on carbon emission performance measurement can be identified in current research.
Received 4 July 2016 This paper presents a measurement that evaluates carbon emission performance in a total factor pro-
Received in revised form duction framework. The Malmquist index based on Data Envelopment Analysis and sequential bench-
28 October 2016
mark technology is used to quantify the performance. Sensitivity analysis on the basis of lower bound
Accepted 30 October 2016
Available online 1 November 2016
DEA models is to test the performance results. Moreover, in order to develop the performance, driving
forces for the performance changes, including technological change, pure technical efficiency change and
scale efficiency change, are discussed. The Australian construction industry during the period 1990e2013
Keywords:
Carbon emissions
is employed as an example to measure the performance which has improved from 1990 to 2013, pri-
Data envelopment analysis marily resulting from technological and scale efficiency improvements. Technology benchmarks are then
Malmquist index identified according to a series of conditions. It would be a feasible practice to further promote perfor-
Performance measurement mance through enhancing pure technical efficiency and learning low-carbon construction techniques
from the technology benchmarks in the Australian construction industry. The approach developed in this
study is generic and could measure, benchmark and evaluate performance especially for undesirable
outputs, and then to identify the driving forces for improving performance.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction
Decreasing greenhouse gas emissions is a vital pathway to
solving the problem of climate change, which is a severe threat to
human development. “Carbon” can be used as a shorthand term to
refer to carbon dioxide equivalent (CO2-e) (Nomenclature is in
Appendix A.) and greenhouse gases in general (Brander and Davis,
2012), as done in this study. Many previous studies have indicated
that the construction sector must play an important part in prop-
erly controlling and reducing carbon emissions through construc-
tion activities (Wong and Zapantis, 2013). The United Nations
Environment Programme e Sustainable Buildings and Climate
Initiative (2008) indicated that 30% of all greenhouse gas emis-
sions result from the global building sector and this will more than
double in the next 20 years if conditions remain unchanged.
Consequently, the reduction of carbon emissions is a serious chal-
lenge in the construction industry.
In the construction domain, investigations of carbon emissions
* Corresponding author.
E-mail address: chunlu@deakin.edu.au (C. Liu).
can be divided into two clusters. The first cluster of research focuses
on carbon performance related to the built environment, which is a
popular topic in construction research. In order to build a low-
carbon or zero-carbon home, numerous studies provide many
methods and measures, such as implementing carbon policies (e.g.
carbon tax, emissions trading, emission standards) (Lu et al., 2012),
innovating products, practices and processes in construction en-
terprises (Killip, 2013), building on the moderating role of organi-
zational culture (Wong and Zapantis, 2013), and creating a policy
framework and support mechanisms (Heffernan et al., 2015). The
second research cluster studies carbon performance in construction
activities; this is rare in the construction domain. The carbon per-
formance of a hotel construction project in South Wales was eval-
uated with the intention of mapping carbon emissions from
construction activities and measuring the carbon footprint of those
activities (Ren et al., 2012). Mukherjee (2012) proposed a frame-
work to estimate the carbon footprints of typical construction
work-items by applying life-cycle assessment methods and in-
ventories. Nevertheless, studies that quantify the performance
measurement of carbon emissions in the construction industry
cannot be identified.

http://dx.doi.org/10.1016/j.jclepro.2016.10.188
0959-6526/© 2016 Elsevier Ltd. All rights reserved.
 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815
In modelling energy and environmental performance, Data
Envelopment Analysis (DEA) has gained great popularity (Zhou
et al., 2008). A full list of abbreviations is provided in Appendix B.
DEA, which is a non-parametric frontier method, was first intro-
duced on the basis of the economic theory of Pareto optimality by
Charnes et al. (1978). DEA is data based and is generally employed
to evaluate production performance using linear programming
techniques (Fa €re et al., 2011). DEA, which is utilised to investigate
the efficiency of a DMU, can be applied to various entities: pro-
ductive and non-productive, public and private, profit and non-
profit organisations (Sun, 2011). Especially, since the Malmquist
productivity index was defined by Caves et al. (1982), DEA has also
been implemented in diverse contexts to estimate productivity
performance. The original Malmquist index is a quantity index on
the basis of a consumer theory, proposed by Malmquist (1953). The
index does not depend upon price information and equilibrium
assumptions and can be further decomposed into factors in
measuring productivity changes. Moreover, building a reasonable
production technology set is fundamental in applying the DEA
technique. Benchmark technologies are applied in building the set,
mainly consisting of contemporaneous benchmark technology,
intertemporal benchmark technology, sequential benchmark
technology, window benchmark technology and global benchmark
technology (Hu and Liu, 2015). This research has selected sequen-
tial benchmark technology to constitute the production technology
set. The sequential benchmark technology seeks benchmark(s) and
envelopes the production technology set from all the previous and
current DMUs. Therefore, the identified benchmark(s) can be
feasibly transformed in practice. This also solves the problem of the
shortage of sample data in the DEA calculation for some situations.
Controlling carbon emissions in the construction industry has
positive environmental and economic implications for sustainable
development in not only reducing pollution but also decreasing
costs of input resources. Furthermore, the carbon tax would be
reduced by lessening carbon emissions. The purpose of this
research project is to measure the trend of carbon emission per-
formance with the objective of reducing carbon emissions, in
consideration of input constraints and without discouraging
desirable output increases. The research framework is displayed in
Fig. 1. Carbon emission efficiency is first measured from a total
factor perspective through applying the DEA technique in a
sequential benchmark technology. Carbon emission performance
change is then defined by using the Malmquist index and further
decomposed into technological change, pure technical efficiency
change and scale efficiency change in order to identify driving
forces for performance development. Moreover, sensitivity analysis
is provided to test the consequences of carbon emission perfor-
mance depended on the efficiency results from lower bound DEA
models. Finally, benchmark analysis applied aims to enhance car-
bon emission performance. In summary, this study primarily re-
defines the performance of CO2-e emissions in a total factor
Fig. 1. The research framework.
2805
production framework and models the performance in a new
measurement and decomposition technique.
2. Literature review
2.1. Carbon emission performance measurement
An important foundation of reducing carbon emissions is un-
derstanding the patterns of carbon emissions and monitoring the
performance of carbon emissions (Zhang et al., 2015). In order to
evaluate carbon emission performance trends, various indicators
have been developed and applied in previous studies (Zhou et al.,
2010). Busch (2010) reviewed carbon performance indicators at
the corporate level and believed that carbon performance in-
dicators could provide company-specific information to develop
performance on the micro level. Riccardi et al. (2012) studied the
efficiency of carbon emissions and pointed out that efficiency can
be improved by investing in new technologies and using alternative
fuels and raw materials. Busch et al. (2011) developed a generic
framework for the assessment of corporate carbon performance
with the purpose of discussing carbon reduction strategies and
policies. In order to build supply chains that are more eco-friendly.
Hsu et al. (2014) employed the indicator of carbon performance to
evaluate suppliers. Ozawa-Meida et al. (2013) measured carbon
performance through a consumption-based carbon footprint using
a case study of a UK university. Zhang et al. (2015) measured the
total-factor carbon emission performance for the Chinese trans-
portation industry and defined the total-factor carbon emission
performance as the ratio between the potential target for carbon
intensity and actual carbon intensity; they concluded that the
overall performance decreased 32.8% over the research period
because of technological decline.
However, through reviewing those studies, two aspects can be
further developed. Firstly, many indicators for representing carbon
emissions measure performance from a partial perspective and so
merely reflect the partial performance of carbon emissions (Zhou
et al., 2010). Secondly, numerous previous studies evaluated car-
bon performance through measurement of carbon emissions and
desirable outputs simultaneously, such as a popularly used indi-
cator of carbon intensity that is defined as carbon emissions per
unit of GDP. This type of evaluation reflects production perfor-
mance from the two dimensions of carbon reduction and economic
growth, and therefore cannot focus on evaluating the real perfor-
mance of carbon emissions. The performance measurement of
carbon emissions should only aim to investigate the real situation
of carbon emissions. Recently, a few studies have noted the
importance of concentrating on carbon emission evaluation and
begun to measure carbon performance from a total factor view-
point. Zhou et al. (2010) studied the total-factor carbon emission
performance of the top 20 emitters worldwide through focusing on
the efficiency of carbon emissions, and subsequently showed that
2806 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815

global performance has been driven by technological progress and
improved by an average value of 24%. Therefore, the performance of
carbon emissions should be measured by focusing on CO2-e emis-
sions themselves in a total factor production framework specifically
without decreasing desirable outputs or wasting input resources.
2.2. The DEA-based Malmquist index in the construction industry
index method to investigate the factors affecting the energy pro-
ductivity of the Australian construction industry and compared
these factors with those decomposed from total factor productivity.
This study employs a DEA-based Malmquist index to evaluate the
performance changes and influencing factors in relation to carbon
emissions in a total factor framework for the Australian construc-
tion industry.

The DEA technique has also been widely applied in the con-
struction domain, such as investigations of energy consumption 3. Methodology
efficiency (Xue et al., 2015), total factor productivity (Li and Liu,
2010), eco-efficiency (Li et al., 2010) and CO2 reduction potentials 3.1. Measuring the efficiency of undesirable outputs
(Hu and Liu, 2015). The main advantages of DEA are that DEA can
avoid the assumptions of measurement functional forms and prior Consider DMUj j ¼ 1, 2, …, J observed in G time periods, G ¼ 1, 2,
conditions (Molinos-Senante et al., 2016). DEA does not require the …, T. Assume the inputs X ¼ (x1, x2 …, xN) 2 RN N
þ ¼ (x: x2 Rþ , x 0)
same measurement units for all variables and can clearly identify of each DMU can produce desirable outputs Y ¼ (y1, y2 …, yM) 2
the benchmark technologies. As a deterministic method, DEA RM M
þ ¼ (y: y2 Rþ , y 0) and undesirable outputs B ¼ (b1, b2, …, bW)
compares every individual production technology so as to identify
2 RW W
þ ¼ (b: b2 R þ , b  0) in a production technology set F, i.e.
the best-practice technologies and inefficient DMUs. Therefore, it
FG ¼ {(XG; YG, BG), XG can produce YG and BG at time G}. N, M, W and
provides a potentially useful tool for decision-makers to make and
T are the numbers of the variables of inputs, desirable outputs,
implement plans/programs for improving their production. More-
undesirable outputs and time periods, respectively.
over, the DEA-based Malmquist index has also been extensively
Suppose there is an observed DMU in period t; the sequential
used in measuring production performance changes. This index
benchmark technology set can be structured as Ft ¼ {F1C U/… UFG C
provides the opportunity to compare DEA results, and de-
U … UFtC }, where FG G G G G G G
C ¼ {(X1 ; Y 1 , B1 ) U/… U (Xj ; Y j ; , Bj ) U… U
compositions of the index can identify the influencing factors for
(XG G G
J ; Y J ; BJ )}, G ¼ 1, 2, …t. Let the i-th observation or DMUi be
performance change. In addition, the Malmquist index relates to
€rnqvist index and the Fisher index, which has also contrib- evaluated at period t in the set of Ft. The calculating model (1),
the To
which is used to measure the efficiency of undesirable outputs for
uted to the popularity of the Malmquist index (Lovell, 2003).
DMUi under constant returns to scale (CRS), can be constructed by
As a quantitative index, the DEA-based Malmquist index has
the objective function Dt(Xt; Yt,BtjCRS) ¼ (inf{l:(Xt; Yt, lbt)2Ft})
been applied in diverse contexts to estimate production perfor-
subject to constraint conditions:
mance changes, especially in measuring changes in productivity
and efficiency. At the macroscopic level, Coelli and Rao (2005)
 
Dt Xt ; Yt ; bt CRS ¼ min l (1)
examined total factor productivity growth for 93 countries using
Malmquist index analyses; Zhou et al. (2010) calculated the
Malmquist index of total carbon emissions; and Fa €re et al. (2012) s:t:
employed the Malmquist index to analyse the timing of carbon
Xt XJ
reduction and proposed that reduction costs would be lower if G G  xt
x m (1a)
countries allowed production to be redistributed. At the meso- G¼1 j ¼ 1 jn j in

scopic level, the Malmquist index has been widely used in different
Xt XJ
types of businesses, such as the efficiency of coal consumption in G
y m G  yt (1b)
G¼1 j ¼ 1 jm j im
Chinese provinces (Long et al., 2016) and the road safety of US states
(Egilmez and McAvoy, 2013). At the microscopic level, Chang et al. Xt XJ G G  lbt
(2009) measured changes in the productivity and efficiency of US
G¼1
b m
j ¼ 1 jw j iw (1c)
accounting firms through utilising the index; and del Mar Salinas-
Jimenez (2004) analysed the effects of public infrastructure on
private factor productivity and efficiency in Spanish regions using n2N; m2M; w2W; t2T; i2J and mGj  0
the Malmquist decomposition method.
Here, l represents the efficiency of undesirable outputs and
In the construction domain, the Malmquist index is primarily
moreover indicates the target emission ratio of undesirable outputs
employed to measure total factor productivity changes. The index
that can be achieved for the given inputs and desirable outputs
has been used in measuring the total factor productivity changes in
using benchmark technology. mG j
, that is, the intensity factor of each
the Chinese construction industry (Wang et al. (2013)), and of the
DMU, is used to structure the frontier of the sequential production
construction companies in Libya (Eltaief and Ahmad, 2011). More-
technology set Ft, which is from the beginning period to period t.
over, according to different decomposition methods, the Malmquist
The inequality constraints of (1a) and (1b) respectively require that
index can be decomposed into various sections. Li and Liu (2010)
the target values of DMUi should use no more inputs and produce
measured the total factor productivity changes in the Australian
no fewer desirable outputs than their actual values. The inequiva-
construction industry using a DEA-based Malmquist index, where
lent constraint in (1c) demonstrates the requirement for the un-
the index was decomposed into the output bias index, input bias
desirable outputs to be reduced to the maximum level.
index, neutral technical change index, pure technical efficiency
When the objective function of Dt(Xt; Yt, btjVRS) is under vari-
index and activity effect index. Horta et al. (2013) applied DEA to
able returns to scale (VRS), an extra constraint condition
the evaluation of efficiency and employed the Malmquist index to Pt PJ G
G ¼ 1 j ¼ 1 mj ¼ 1 should be added. Similarly, the efficiency of (X ;
t
estimate productivity changes in the construction industry world-
wide, where the index was decomposed into efficiency change and Yt,Bt) in the set Ftþ1 is defined as Dtþ1(Xt; Yt,Bt) ¼ (inf{l:(Xt; Yt,lbt)
technological change. However, so far no research has employed 2Ftþ1}); and the efficiency of (Xtþ1; Ytþ1,btþ1) in the set Ft is
the Malmquist index to evaluate production performance for one defined as Dt (Xtþ1; Ytþ1,btþ1) ¼ (inf{l:(Xtþ1; Ytþ1,lbtþ1)2Ft}).
aspect in construction. Hu and Liu (2016a) adopted the Malmquist Accordingly, the efficiency l of undesirable outputs can be
 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815
respectively calculated for Dt(Xtþ1; Ytþ1,btþ1), Dtþ1(Xt; Yt,bt) and
Dtþ1(Xtþ1; Ytþ1,btþ1) under CRS and VRS by following the previous
procedures.
3.2. The DEA-based Malmquist index and its components
According to the definition of a DEA-based Malmquist index
(Caves et al., 1982), a Malmquist performance change index (MPCI)
is defined to evaluate the performance changes of undesirable
outputs over time based on the efficiency results. When measuring
a performance change between t and tþ1, MPCI is presented as the
geometric mean of performance changes at period t and tþ1, as
follows in Equation (2). This definition avoids the assumption of
Hicks-neutral technical change that requires that a production
change does not affect the assumed production function. Following
the Malmquist decomposition method that was produced by Fa €re
et al. (1994), MPCI is decomposed into the pure technical effi-
ciency change index (PTECI), scale efficiency change index (SECI)
and technological change index (TCI), as shown in Appendix C. The
calculations of PTECI, SECI and TCI are respectively provided in
Equations (3) (4) and (5).

 
  1=2
D Xtþ1 ; Ytþ1 ;btþ1 CRS Dtþ1 Xtþ1 ; Ytþ1 ;btþ1 CRS
t
MPCI t;tþ1 ¼
Dt ðXt ; Yt ;bt jCRSÞ Dtþ1 ðXt ; Yt ;bt jCRSÞ

 
Dtþ1 Xtþ1 ; Ytþ1 ; btþ1 VRS
PTECI t;tþ1
¼ t t t t
D ðX ; Y ; b jVRSÞ

 

t;tþ1 Dtþ1 Xtþ1 ; Ytþ1 ;btþ1 CRS Dtþ1 Xtþ1 ; Ytþ1 ;btþ1 VRS
SECI ¼
Dt ðXt ; Yt ;bt jCRSÞ=Dt ðXt ; Yt ;bt jVRSÞ

 
  1=2
t;tþ1 Dt Xtþ1 ; Ytþ1 ; btþ1 CRS Dt Xt ; Yt ; bt CRS
TCI ¼ tþ1
tþ1 tþ1 tþ1   bound for DEA models to measure the production efficiency of
X ; Y ; b CRS Dtþ1 ðXt ; Yt ; b jCRSÞ
t
D carbon emissions, which is similar to Wu et al. (2015). The larger
t,tþ1
The MPCI of Equation (2) measures the performance changes
in undesirable outputs between two adjacent periods t and tþ1. If
MPCIt,tþ1 > 1, the performance is incremental from period t to
period tþ1, which indicates that the DMUi production becomes
more environmental and efficient. In other words, fewer undesir-
able outputs are produced from t to tþ1. Similarly, if
MCEPIt,tþ1 < or ¼ 1, MPCI is in decline or stagnation, respectively.
The changes in PTECI, SECI and TCI in Equations (3) (4) and (5)
indicate the same interpretations. As MPCI is measured in CRS
production technology, where all the efficiency results are calcu-
lated by the DEA models of total factor production, MPCI is a total
factor performance index for undesirable outputs (Zhou et al.,
2010). In particular, the CRS condition which is used in construct-
ing DEA models is a necessary condition for measuring true indices
(Chung et al., 1997). PTECI mainly embodies organizational capacity
in using internal resources such as operational strategies, schedules
and policies, and management ability, SECI can be influenced
by internal scale factors, for example, business capital and pro-
duction capacity; and external scale factors such as the policy
environment and production demands. TCI can be enhanced by
introducing advanced equipment and techniques, and developing
2807
new production technologies, methods, theories, and process
flows.
3.3. Sensitivity analysis for carbon emission performance based on
lower bound DEA models
In order to test the reliability of the results in Model (1), sensi-
tivity analysis is employed to evaluate the effect in calculating the
carbon emission performance without consideration of desirable
outputs in the DEA models. Sensitivity analysis that is used to
measure the effect of change in one factor on another factor can
potentially be applied to all stages of the model process, model
parameters, model formulation, model calibration and model
verification (McCuen, 1973). This approach explores the perfor-
mance of “what-if” situation of removing one or more variables
from DEA models to determine the efficiency changes (Pahwa et al.,
2003), which is similar to another sensitivity analysis approach of
removing one DMU (e.g. Egilmez et al., 2016). Accordingly, the ef-
ficiency of carbon emissions can be measured by the following DEA
model.

 
Dt Xt ; bt CRS ¼ max q (6)
s:t:
(2) Xt XJ
G G  xt
x m (6a)
G¼1 j¼1 jn j in
Xt XJ G G  qbt
(3)
G¼1
b m
j¼1 jw j iw (6b)
 n2N; w2W; t2T; i2J and mGj  0
Compared to Model (1) which considers the most efficient car-
bon emissions as the production frontier, Model (6) allows the least
(4)
efficient carbon emissions to erect the production frontier in the
sequential benchmark technology set under all the constraints. The
two pathways respectively construct the upper bound and lower
(5) the value of q, the higher efficiency of carbon emissions is. More-
over, Model (6) does not comprise the constraint of desirable
output. The other efficiencies of Dt(Xtþ1; btþ1jCRS), Dtþ1(Xt; btjCRS)
and Dtþ1(Xtþ1; btþ1jCRS) without consideration of desirable out-
puts can accordingly be calculated from the respective lower bound
DEAs. Therefore, MPCI of carbon emissions can be obtained from
Equation (2) using the efficiency results of all the lower bound DEA
models so as to test sensitivity analysis.
4. Empirical study
4.1. Variable selection and data sources
The construction industries of the Australian states and terri-
tories are selected for observation, including New South Wales
(NSW), Victoria (Vic.), Queensland (Qld), Western Australia (WA),
South Australia (SA), Tasmania (Tas.), the Northern Territory (NT)
and the Australian Capital Territory (ACT). The study period is from
1990 to 2013, due to data availability. The measurement variables
comprise two input factors, employed persons and construction
work done, a desirable output factor named industry gross value
added and an undesirable output factor called CO2-e emissions.
These DEA variables have been applied in previous studies for the
2808 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815
Australian construction industry (e.g. Li and Liu, 2010). The factor of
construction work done is the sum of all building and engineering
work done, comprising the costs of materials, labour and architects,
plus work done by subcontractors and firms’ own workforces. The
factor of employed persons includes full-time, part-time and self-
employed workers. Industry gross value added measures the
value of industry production and is used to measure the contri-
bution of industry. The data for CO2-e emissions consist of carbon
dioxide, methane, nitrous oxide and others.
The data for employed persons, construction work done and
industry gross value added have been collected from the Australian
Bureau of Statistics (ABS). The data for CO2-e have been directly
gathered from the Australian Government Department of the
Environment. CO2-e emissions have increased slightly from 1990 to
2013 in the Australian construction industry. The summary statis-
tics for all variables are given in Table 1. Significant regional im-
balances and differences among the construction industries of the
Australian states and territories can be observed from Table 1 in
terms of all the variables, which show that there are dramatic
construction scale diversities in Australia.
In order to ensure the validity of the selected input and output
variables, the functions relating to the variables must comply with
the property of isotonicity in the DEA method (Li and Shi, 2014).
The Pearson's correlation coefficients among the selected input and
output variables are summarised in Table 2. It can be seen that the
values of all the coefficients are from 0.7796 upwards and this
denotes statistical significance at the 0.01 level. As a result, there
are significant positive relationships between the input and output
variables. In other words, an increase in any input will lead to some
increase in any output. Therefore, the property of isotonicity is
tested and complied with, and the selected input and output vari-
ables are justified in this study.
All mathematical operations have been implemented in
Microsoft Excel worksheets and individually solved using MS-Excel
Solver. As a consequence, the annual results of the total factor CO2-e
emissions performance indices, including MPCI, PTECI, TCI and
SECI, have been obtained for all construction industries in the
Australian states and territories.
4.2. Analyses of the Malmquist carbon emission performance
change
The MPCI values for the total-factor carbon emission perfor-
mance in all construction industries are shown in Table 3, which is
calculated from Equation (2). Because some TCI results could not be
measured in some regions and periods, the MPCI could not be
identified. As a result, the ACT, the NT and WA had better perfor-
mance than other regions. For instance, the construction industry
of the NT reduced net energy usage from 0.9 PJ in 1990 to 0.3 PJ in
2013 (Australian Government Department of Industry and Science,
2015). This decrease in energy usage led to a reduction and per-
formance improvement of CO2-e emissions. Furthermore, the
construction industries of Tas., SA, Qld, Vic. and NSW annually
improved by 10.74%, 8.11%, 7.1%, 5.93% and 3.85% from 1990 to 2013
according to their average values, respectively. More importantly,
Table 1
Descriptive statistics of the Australian construction industry: 1990e2013.
Variable Unit Mean
Construction work done $ Millions 10961.50
Employed persons '000 92.87
Industry gross value added $ Millions 6982.91
CO2-e emissions Gg 990.07
the MPCI for the whole Australian construction industry was
collectively boosted in 2002e05 and 2011e12. Overall, the MPCI for
the whole construction industry demonstrated a trend of growth
over the study period. This means that the CO2-e emission perfor-
mance improved in the Australian construction industry.
The cumulative growth of the MPCI in the Australian construc-
tion industry, except for the ACT and the NT due to not having
calculated results for most of research period, is shown in Fig. 2.
Supposing that all the MPCI values in 1990e91 are transformed into
one unit from the initial values, the transformed MPCI value in
another period for each construction industry is the ratio between
its initial MPCI value and the MPCI value in 1990e91. The cumu-
lative results are calculated from the products of all the trans-
formed MPCI values. It can be seen that these regions experienced
continuous and gradually increasing periods over 1990e2013,
particularly from 2000 to 01. The construction industry of WA ob-
tained the fastest growth among all areas, especially from 2005.
The construction industry of Tas. clearly showed improved perfor-
mance, although it experienced various fluctuations during the
activity periods. For example, the Tas. construction industry
released CO2-e (Gg) of 256.9, 154.59 and 274.69 in 2005, 2006 and
2007 (Australian Government Department of the Environment,
2015), respectively. As a result, the MPCI in the Tas. construction
industry attained a peak in 2005e06. The reason may be that Tas.
implemented a series of actions for reducing carbon emissions such
as the Draft Climate Change Strategy, effective volunteering activ-
ities and changes to management (Tasmanian Climate Change
Office, Department of Premier and Cabinet, 2008). Another reason
could be data errors. Moreover, the cumulative MPCI in the other
construction sectors including Qld, SA, Vic. and NSW display a
gradual upward trend during 1990e2013.
For the purpose of the sensitivity analysis, the Malmquist
change values without consideration of desirable outputs are
measured on the basis of the efficiencies of Dt(Xt; btjCRS), Dt(Xtþ1;
btþ1jCRS), Dtþ1(Xt; btjCRS) and Dtþ1(Xtþ1; btþ1jCRS) calculated by
the respective lower bound DEA models. Table 4 displays the
Malmquist change values of the total factor CO2-e emission per-
formance in the Australian construction industry. Compared results
from both Tables 3 and 4, three primary conclusions can be sum-
marised. Firstly, both of tables indicate the increased performance
of CO2-e emissions. It demonstrates the Australian construction
industry has enhanced the production efficiency of CO2-e emissions
without influencing production for desirable outputs. Secondly,
there are no obvious changes to the ranking of all sectional con-
struction industries according to the average values in the two ta-
bles. However, ACT shows lower values in Table 4 than in Table 3,
which demonstrates ACT has considerably developed industry
value added even though with limited improvement in carbon
emission efficiency. Finally, using a paired t-test without consid-
eration na values in Table 3, no significant differences are found
between the results of calculating from the lower and upper
bounds in DEA models. In summary, the results of the sensitivity
test in Table 4 provide supports that are consistent with the main
findings in Table 3. CO2-e emission performance in the Australian
construction industry has been clearly improved within or without
Std dev. Median Maximum Minimum
12879.03 5435.94 55901.89 383.33
87.53 63.17 296.54 5.00
7404.75 4146.00 32484.00 386.00
820.76 731.41 2485.13 65.16
 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815 2809

Table 2
Pearson correlation coefficients among the selected input and output variables.

Construction work done Employed persons Industry gross value added CO2-e emissions

Construction work done 1.0000

Employed persons 0.8387* 1.0000
Industry gross value added 0.9911* 0.8706* 1.0000
CO2-e emissions 0.7417* 0.9583* 0.7796* 1.0000

Note: * denotes significance of the coefficients at the 0.01 level.

Table 3
Malmquist performance change of total-factor carbon emissions.

Year NSW Vic. Qld SA WA Tas. NT ACT

1990e91 1.1049 1.4376 1.0688 1.2487 1.2583 1.2304 na na

1991e92 0.8833 0.8072 0.9612 0.8366 1.0326 1.1353 0.8396 na
1992e93 0.9625 0.9754 1.0556 0.7223 0.9852 0.9203 0.8103 na
1993e94 1.0857 0.9518 1.0369 1.3099 0.9012 1.0017 na 1.2104
1994e95 1.1035 1.0794 1.0589 1.5611 1.2631 2.1078 0.7074 na
1995e96 1.0426 1.0377 1.0305 1.5557 0.9455 0.9754 0.8244 na
1996e97 1.0479 1.0369 1.0847 0.6843 1.0674 1.5544 1.6579 1.0838
1997e98 1.0230 1.0760 1.0073 0.9133 1.0210 0.5591 0.5796 0.8558
1998e99 1.0135 1.0780 1.0355 0.9281 0.9774 0.9342 na na
1999e00 1.1621 1.0751 1.1320 1.1252 1.1161 0.8955 1.8163 na
2000e01 0.7952 0.8811 0.8859 0.9267 1.0892 0.9526 na na
2001e02 1.1973 1.2392 1.1777 1.2116 0.9870 1.1252 na na
2002e03 1.1275 1.1570 1.1401 1.0275 1.0645 1.0012 na na
2003e04 1.1090 1.0723 1.1225 1.2740 1.2370 1.2096 na na
2004e05 1.0857 1.1045 1.1225 1.1175 1.1171 1.2125 na 1.0735
2005e06 0.9767 1.1064 1.1520 0.9947 1.2180 1.8730 0.9025 na
2006e07 0.9659 1.0594 1.1388 1.1138 1.3062 0.6767 1.0871 0.9877
2007e08 1.0758 1.0351 1.0885 1.0368 1.0992 0.9747 na na
2008e09 1.0985 1.0445 1.1234 1.0534 1.2446 1.1005 0.9724 1.2032
2009e10 1.0411 1.1059 1.0214 1.1572 1.0993 1.1915 0.6099 1.3753
2010e11 0.9435 0.9220 1.1075 1.0687 1.1939 0.9719 1.0160 na
2011e12 1.0179 1.0093 1.0583 1.0200 na 1.0138 1.6839 na
2012e13 1.0234 1.0722 1.0227 0.9792 na 0.8529 na na
Mean 1.0385 1.0593 1.0710 1.0811 na 1.1074 na na

Fig. 2. The cumulative results of Malmquist performance change index (MPCI) for carbon emissions.

desirable outputs in the measurement.
4.3. Driving forces analysis for the total-factor carbon emission
performance
4.3.1. Analysis of pure technical efficiency change
The PTECI values for the total factor CO2-e emissions perfor-
mance for all the construction sectors are given in Table 5. The NT's
construction industry achieved the fastest improvement among all
construction industries in the research period, mainly promoted by
sharp improvements in 1996e97, 1999e2001 and 2011e13. The
construction industry of Tas. was the second fastest one, primarily
caused by rapid improvements in 1994e95 and 2004e06. The
construction industry of WA also attained some obvious improve-
ments. Some regions in some phases remained unchanged, such as
the ACT over the total research period except in 1997e99 and the
NT and WA in many study phases. Although NSW indicated un-
changed trends in most of the study periods, it declined slightly
over the whole research period due to a sharp downgrade in
2010e11. Before 2011, the gross value added of NSW was better
2810 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815

Table 4
Malmquist performance change index based on lower bound DEA models.

Year NSW Vic. Qld SA WA Tas. NT ACT

1990e91 1.0762 0.8943 0.9635 1.0733 0.8685 1.0472 1.1922 0.9957

1991e92 0.8734 0.7979 1.0037 0.8087 0.9443 1.1077 0.9392 1.2035
1992e93 0.9750 1.0031 1.1227 1.0336 1.1307 0.9761 0.9935 0.8868
1993e94 1.0040 1.0702 1.0913 0.9966 1.1823 0.9830 0.9924 0.9046
1994e95 1.0721 1.0644 1.0972 0.9443 0.9441 0.8970 1.2545 1.0787
1995e96 1.0282 1.0109 0.8859 0.9117 1.1212 1.0700 1.1497 0.8368
1996e97 0.9719 1.0614 1.0733 1.0862 1.0337 0.9532 0.7899 1.1147
1997e98 1.0104 1.0906 0.9745 1.1975 1.0230 0.8720 1.2695 0.9091
1998e99 1.0116 1.0192 1.0596 0.9261 0.9107 1.0237 1.0196 0.7917
1999e00 1.1295 1.0650 1.0457 1.3132 1.1114 1.0660 0.9629 1.0164
2000e01 0.9489 1.0378 0.9717 0.8760 0.9790 0.9535 0.8770 1.4702
2001e02 1.0132 1.0937 0.9799 1.2418 1.0793 1.3838 0.8210 0.7570
2002e03 1.0192 1.0020 1.0808 1.0161 0.9460 1.0245 1.4399 1.2783
2003e04 1.1166 1.0193 1.0756 1.0382 1.0516 0.9601 1.1150 0.8986
2004e05 1.0794 1.0485 1.1227 1.0484 1.0893 1.1325 1.0673 1.3671
2005e06 0.9291 1.1534 1.1407 1.0280 1.0443 1.6576 1.1043 1.0436
2006e07 1.0249 1.0442 1.0651 1.0372 1.0084 0.5921 0.8924 0.9701
2007e08 1.0053 0.9818 1.0311 0.9857 1.0912 1.0270 1.2083 1.1779
2008e09 1.0400 1.0164 1.0697 1.1303 1.0202 1.2002 1.0824 0.8485
2009e10 0.9502 1.0455 0.9543 1.0565 1.0207 1.0177 1.2186 0.9785
2010e11 0.9383 1.0151 0.9826 1.0656 1.0263 0.9847 1.0273 1.0906
2011e12 1.0291 0.9461 0.9391 0.9196 0.9619 1.0311 0.9729 0.9392
2012e13 0.9516 0.9799 1.0012 1.0064 1.0029 0.8696 1.1750 0.9564
Mean 1.0086 1.0200 1.0318 1.0322 1.0257 1.0361 1.0680 1.0223

Table 5
Pure technical efficiency change in total-factor carbon emission performance.

Year NSW Vic. Qld SA WA Tas. NT ACT

1990e91 1.0000 1.0000 0.8585 1.0000 1.0000 1.0425 1.0000 1.0000

1991e92 1.0000 0.7912 0.9876 0.8181 1.0000 1.0816 1.0000 1.0000
1992e93 0.8748 0.9890 0.9866 0.7950 0.9978 0.8633 1.0000 1.0000
1993e94 1.1431 0.9694 0.9915 1.1411 1.0023 0.9885 1.0000 1.0000
1994e95 1.0000 0.9721 1.0410 1.3474 1.0000 1.7401 0.8410 1.0000
1995e96 1.0000 1.0553 0.9874 1.0000 0.9963 0.5397 0.8506 1.0000
1996e97 1.0000 0.8540 1.0362 1.0000 0.8280 1.2669 1.2420 1.0000
1997e98 1.0000 1.0589 0.9850 0.7708 0.9548 0.9855 0.7950 0.9640
1998e99 1.0000 1.0660 1.0304 0.8920 1.2697 0.9347 0.7611 1.0373
1999e00 1.0000 0.9247 0.9699 0.7801 1.0000 0.9613 1.5535 1.0000
2000e01 0.7901 0.8758 0.8802 0.9228 1.0000 1.1137 1.1974 1.0000
2001e02 1.2657 1.2507 1.7737 1.2369 0.8683 0.6730 1.0000 1.0000
2002e03 1.0000 1.0758 0.7179 0.8769 1.1516 0.9874 1.0000 1.0000
2003e04 1.0000 0.9843 1.0390 1.1964 1.0000 0.5809 1.0000 1.0000
2004e05 1.0000 1.2430 1.0207 1.0614 1.0000 1.3153 1.0000 1.0000
2005e06 1.0000 0.8583 1.0863 0.8033 1.0000 1.6618 0.9515 1.0000
2006e07 1.0000 1.0063 1.0586 1.0456 1.0000 0.5689 1.0510 1.0000
2007e08 1.0000 0.8586 1.0602 0.8798 1.0000 0.9778 1.0000 1.0000
2008e09 1.0000 0.7451 0.9115 0.7873 1.0000 1.0445 0.9634 1.0000
2009e10 1.0000 0.9682 0.6912 1.0485 1.0000 1.0352 0.6514 1.0000
2010e11 0.4297 0.7595 0.9231 0.8949 1.0000 0.9918 0.9913 1.0000
2011e12 0.8666 0.8341 0.8762 0.8438 1.0000 1.0049 1.1114 1.0000
2012e13 0.9794 1.0627 1.0117 0.9623 1.0000 0.9806 1.4464 1.0000
Mean 0.9717 0.9653 0.9967 0.9611 1.0030 1.0148 1.0177 1.0001

than that of WA; then the benchmark for NSW would be itself, with
the largest construction market in Australia, and the CO2-e reduc-
tion amount would be limited. In and after 2011, the gross value
added of NSW was less than that of WA; then the benchmark for
NSW would be WA and the CO2-e reduction amount would be
greater. As a result, there was an abrupt drop in PTECI for NSW from
2010 to 2011. Moreover, Qld with a limited decrease and Vic. and SA
with obvious decreases can be observed from their average values
of PTECI over the whole research phase.
4.3.2. Analysis of scale efficiency change
Table 6 indicates the SECI values for the total factor CO2-e
emission performance in the Australian construction industry. The
SECI in all areas demonstrates a trend towards an increase over the
research period from the mean values. This implies that the SECI
plays a positive role in the total factor CO2-e emission performance
changes in the whole construction sector. The highest average in-
cremental value was for the NT with 2.51%, followed by NSW with
2.41. The SECI of the construction industry in Vic. was also clearly
enhanced because of continual and gradual upward forces from
1996 to 2000 and from 2006 to 2013, followed by that of Qld. The
construction industries of SA and WA also demonstrate a trend of
improvement. Although the Tasmanian construction industry
experienced a series of reductions in most research years, it ach-
ieved a slight improvement over the whole surveyed period from
the mean value due to several abrupt increases. The SECI in the ACT
remained steady over the whole surveyed period except for
1997e99 and 2006e08. Furthermore, the SECI in the overall
 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815 2811

Table 6
Scale efficiency change in total-factor carbon emission performance.

Year NSW Vic. Qld SA WA Tas. NT ACT

1990e91 1.0515 1.2629 1.0728 0.9993 1.0162 0.9562 1.0000 1.0000

1991e92 0.8389 0.9966 0.9362 1.0091 1.0186 1.0290 0.8391 1.0000
1992e93 1.0651 0.8354 1.0266 0.8039 0.8971 0.9264 0.7946 1.0000
1993e94 0.9499 0.9152 1.0458 0.9964 0.7832 0.8693 1.4999 1.0000
1994e95 1.1035 1.1102 1.0172 1.0938 1.2015 1.1515 0.8412 1.0000
1995e96 1.0146 0.9529 1.0156 1.0107 0.8288 0.9383 0.6627 1.0000
1996e97 0.9928 1.1502 0.9918 0.6244 1.2212 1.1449 1.2992 1.0000
1997e98 1.0230 1.0162 1.0226 1.1849 1.0693 0.5673 0.7291 0.8878
1998e99 1.0057 1.0035 0.9972 1.0324 0.7638 0.9931 1.7398 1.1264
1999e00 1.1341 1.1346 1.1390 1.4077 1.0892 0.9091 1.0790 1.0000
2000e01 0.9591 0.9587 0.9591 0.9570 1.0379 0.8151 1.0088 1.0000
2001e02 0.9237 0.9675 0.6484 0.9565 1.1100 1.6327 1.0000 1.0000
2002e03 0.8516 0.8123 1.1995 0.8850 0.6981 0.7658 0.9559 1.0000
2003e04 1.0469 1.0284 1.0200 1.0053 1.1678 1.9657 1.0461 1.0000
2004e05 0.9421 0.7710 0.9544 0.9137 0.9694 0.8000 1.0000 1.0000
2005e06 0.7056 0.9313 0.7661 0.8945 0.8800 0.8142 0.9079 1.0000
2006e07 0.9659 1.0528 1.0757 1.0653 1.3062 1.1894 1.0344 0.9877
2007e08 0.9064 1.0157 0.8651 0.9929 0.9261 0.8399 1.0648 1.0124
2008e09 1.0443 1.3327 1.1717 1.2719 1.1129 1.0016 0.9822 1.0000
2009e10 0.9306 1.0210 1.3208 0.9865 0.9826 1.0288 0.7713 1.0000
2010e11 1.9998 1.1057 1.0928 1.0877 1.0874 0.8926 0.9335 1.0000
2011e12 1.0552 1.0871 1.0850 1.0860 1.0796 0.9063 1.0885 1.0000
2012e13 1.0449 1.0089 1.0109 1.0176 1.0210 0.8697 1.2991 1.0000
Mean 1.0241 1.0205 1.0189 1.0123 1.0116 1.0003 1.0251 1.0006

industry showed no growth in 2004e06 and no decline in
2003e04.
4.3.3. Analysis of technological change
Table 7 shows the TCI values for the total factor CO2-e emission
performance for all construction industries. Owing to sequential
benchmark technology, which demands that benchmark technol-
ogies do not go backwards in the production set, the TCI in all
construction sectors illustrated continuous growth over the
research period. Importantly, some TCI results could not be
measured, such as in the construction industries of the ACT and the
NT in most observed periods and WA in 2011e13. This is because
their technological progress occurred so rapidly in period tþ1 that
their performances could not be defined from the production set in
period t. In other words, the production technology related to CO2-e
emissions moved forward in these regions and periods. This situ-
ation has been discussed in previous studies (e.g. Fa €re et al., 2006;
Chung et al., 1997). Except for the ACT, the NT and WA, the growth
in Tas. was the fastest among all the regions, followed by SA, Vic.,
Qld and NSW. Moreover, if the TCI result is equal to 1.0000, the
technology did not increase between the two periods. No advances
occurred in 1997e98 or 2006e07 for the whole construction in-
dustry, which indicates that benchmark technology did not
advance in those phases.
In order to investigate the driving forces for the total-factor CO2-
e emission performance, the relationships and comparisons be-
tween the MPCI and three components of PTECI, SECI and TCI
needed be investigated in the Australian construction industry,

Table 7
Technological change in total-factor carbon emission performance.

Year NSW Vic. Qld SA WA Tas. NT ACT

1990e91 1.0508 1.1384 1.1604 1.2497 1.2382 1.2342 na na

1991e92 1.0529 1.0237 1.0396 1.0133 1.0138 1.0201 1.0007 na
1992e93 1.0330 1.1807 1.0422 1.1301 1.1007 1.1507 1.0197 na
1993e94 1.0000 1.0728 1.0000 1.1521 1.1481 1.1658 na 1.2104
1994e95 1.0000 1.0001 1.0000 1.0592 1.0513 1.0519 1.0000 na
1995e96 1.0277 1.0319 1.0277 1.5393 1.1451 1.9261 1.4624 na
1996e97 1.0556 1.0556 1.0556 1.0961 1.0556 1.0717 1.0274 1.0838
1997e98 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
1998e99 1.0078 1.0078 1.0078 1.0078 1.0078 1.0065 na na
1999e00 1.0247 1.0247 1.0247 1.0247 1.0247 1.0247 1.0836 na
2000e01 1.0494 1.0494 1.0494 1.0494 1.0494 1.0494 na na
2001e02 1.0241 1.0241 1.0241 1.0241 1.0241 1.0241 na na
2002e03 1.3240 1.3240 1.3240 1.3240 1.3240 1.3240 na na
2003e04 1.0593 1.0593 1.0593 1.0593 1.0593 1.0593 na na
2004e05 1.1524 1.1524 1.1524 1.1524 1.1524 1.1524 na 1.0735
2005e06 1.3842 1.3842 1.3842 1.3842 1.3842 1.3842 1.0447 na
2006e07 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000
2007e08 1.1868 1.1868 1.1868 1.1868 1.1868 1.1868 na na
2008e09 1.0519 1.0519 1.0519 1.0519 1.1184 1.0519 1.0276 1.2032
2009e10 1.1188 1.1188 1.1188 1.1188 1.1188 1.1188 1.2138 1.3753
2010e11 1.0979 1.0979 1.0979 1.0979 1.0979 1.0979 1.0979 na
2011e12 1.1132 1.1132 1.1132 1.1132 na 1.1132 1.3919 na
2012e13 1.0000 1.0000 1.0000 1.0000 na 1.0000 na na
Mean 1.0789 1.0912 1.0835 1.1232 na 1.1397 na na
2812 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815

according to their mean values shown in Table 8. It can be seen that
the MPCI in all the construction industries improved over the
observed period, mainly because of a noticeable improvement in
technological change. Previous studies have shown that technology
innovation plays an important role in carbon emission performance
development (e.g. Riccardi et al., 2012; Zhang et al., 2015). Tech-
nological innovation especially in developing energy conservation
technologies would enhance the energy efficiency of the industrial
sectors and then decline total industrial carbon emissions (Chen
et al., 2016). For instance, outdated technologies and equipment
powered by petroleum products could be replaced by advanced
ones powered by renewable energy sources, which embodies the
process of developing energy consumption structure and further
promotes carbon emission performance.
According to Table 8, the development of the SECI furthermore
promoted MPCI progress in all the construction industries. Con-
struction scale factors such as residents’ housing demands and
public project investment could also have played a positive role in
MPCI development. For instance, the total value of all types of
buildings approved in 2013 more than tripled compared with 1990
(source: ABS, 2015). Moreover, in the NT and Tas., the PTECI also
provided a positive force for MPCI growth. This suggests that the
construction sectors in the NT and Tas. reinforced internal con-
struction activities to promote CO2-e emissions efficiency, for
example, improvements in construction enterprise management,
engineering project control and workforce skills. However, the
function of the PTECI had a clear negative effect on MPCI
advancement in NSW, Vic., Qld and SA, and a slight positive effect in
WA and the ACT. This shows that these construction industries
introduced advanced equipment and machinery (the TCI
improved), but they did not utilise this equipment and machinery
efficiently or sufficiently (the PTECI declined). Specific technical
factors, such as project operational strategies, construction de-
cisions, construction scheduling and management programs, typi-
cally affect project scope, construction process and input resources,
therefore influencing performance in relation to carbon emissions.
4.4. Benchmark analysis for the total-factor carbon emission
performance
to tþ1. Equation (7b) shows that the DMU (Xtþ1; Ytþ1,btþ1) at period
tþ1 is located outside the production frontier of the production set
at period t. In this case, this DMU at period tþ1 is better than all
DMUs of the production set at period t. The DMUs, which could not
be measured in Tables 3 and 7, is in this situation of Equation (7b).
Equation (7c) requires that this DMU at period tþ1 should be
located on the production frontier of the production set at period
tþ1. Table 9 provides the results of the technology benchmarks
identified in the Australian construction industry from 1990 to
2013.
From Table 9, it can be seen that the ACT was a technology
benchmark in every period and the NT was a technology bench-
mark in some periods. In each period, the construction industry of
the ACT released the least CO2-e but was ranked sixth in terms of
gross value added. For instance, in 1990 and 2013, the percentage of
CO2-e emissions of the ACT's construction industry was 0.92%
compared with 0.94% in the whole industry, but the percentages of
gross value added were 2.02% and 2.89%, respectively. Therefore, it
can be concluded that CO2-e emissions in the ACT's construction
industry were efficient and effective. The situation of the NT was
similar to that of the ACT. Therefore, the learning of techniques
from the ACT and the NT is a significant and effective pathway for
improving CO2-e emission performance in the Australian con-
struction industry.
Combining with the reduction potentials of carbon emissions
(Hu and Liu, 2015) and the performance of carbon productivity (Hu
and Liu, 2016b) in the Australian construction industry, it can be
further concluded that the ACT, the NT and WA displayed the best
carbon emission performance from 1990 to 2013. For example, the
ACT government has been developing building and planning reg-
ulations which play a central part to achieve Territory medium to
long term climate change and energy reduction objectives, and has
also been working with construction industry to encourage de-
signers and practitioners to exceed minimum sustainability stan-
dards (ACT Government Environment and Sustainable
Development, 2011). Moreover, the technological factor played a
significant positive role in the carbon emission performance
changes. The Australian construction industry could have actively
and appropriately applied advanced construction technologies,
such as interoperable technology (e.g. Building Information

Identifying construction benchmarks is the foundation of

continuous performance improvement (Park et al., 2005). The Table 8
recognition of benchmarks and the learning of techniques from Driving forces for the total factor carbon emission performance growth.
benchmarks are conducive to promoting construction growth and
NSW Vic. Qld SA WA Tas. NT ACT
reducing CO2-e emissions. DEA is one of the best methods for
realising benchmarking techniques that can be achieved for other MPCI 1.0385 1.0593 1.0710 1.0811 na 1.1074 na na
PTECI 0.9717 0.9653 0.9967 0.9611 1.0030 1.0148 1.0177 1.0001
DMUs (Lin and Huang, 2009). Sequential benchmark technology in
SECI 1.0241 1.0205 1.0189 1.0123 1.0116 1.0003 1.0251 1.0006
DEA can refrain from searching for benchmarks in future DMUs, so TCI 1.0789 1.0912 1.0835 1.1232 na 1.1397 na na
that the identified benchmarks are available and achievable for
inefficient DMUs to learn techniques from the benchmark DMUs. A
DMU is considered a technology benchmark if the three following Table 9
conditions are achieved (Zhang et al., 2015): Technology benchmarks in the Australian construction industry.

Year Technology benchmark Year Technology benchmark

TCIt; tþ1 > 1 (7a)
1990e91 NT, ACT 2002e03 NT, ACT
  1991e92 ACT 2003e04 ACT
Dt Xtþ1 ; Ytþ1 ; btþ1 jCRS > 1 (7b) 1992e93 ACT 2004e05 NT, ACT
1993e94 NT, ACT 2005e06 NT, ACT
1994e95 ACT 2006e07 ACT
 
1995e96 ACT 2007e08 ACT
Dtþ1 Xtþ1 ; Ytþ1 ; btþ1 jCRS ¼ 1 (7c) 1996e97 ACT 2008e09 NT, ACT
1997e98 ACT 2009e10 ACT
Equation (7a) indicates that the technological factor should in- 1998e99 ACT 2010e11 ACT
crease from period t to period tþ1, which suggests that the pro- 1999e00 ACT 2011e12 ACT
duction frontier-shift effect has positively improved. In other 2000e01 ACT 2012e13 NT, ACT
2001e02 NT, ACT
words, the production frontier has moved upwards from periods t
 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815 2813

Modelling), prefabrication, preassembly, modularisation, and off- worldwide to investigate performance changes in undesirable
site fabrication techniques and processes. However, labour- outputs and to identify the influencing factors for improving sus-
intensive practices and lacking standard production methods hin- tainable development. Introducing advanced technologies can
dered productivity improvement in the Australian construction promote total-factor carbon emission performance, but the tech-
industry (Li and Liu, 2010). The pure technical efficiency should be nologies should be used efficiently and efficiently in construction
improved in developing carbon emission performance, for instance, project management so as to further enhance the performance.
enhancing job-site efficiency through more effective interfacing of Finally, the study also contributes to the performance measurement
people, processes, materials, equipment and information. method in systematically applying DEA, the Malmquist index, the
sequential benchmark technology, sensitivity analysis and bench-
5. Conclusions mark analysis. DEA provides pathways for measuring and
comparing efficiency performance and identifying production
Managing carbon emissions without discouraging production benchmarks. The sequential benchmark technology would extend
improvement has positive environmental and economic implica- the DEA application. The Malmquist index and decomposition can
tions in sustainable development. Effective performance mea- investigate the influencing factors for production performance
surement can support the improvement of production performance changes. Therefore, the research approach presented in this study is
and the reduction of carbon emissions. This study has examined the generic and can be applied to other organisations, sectors or re-
carbon emission performance of the Australian construction in- gions in evaluating the performance of undesirable outputs in
dustry in a total factor production framework. In the Australian multi-input and multi-output production. However, the specific
construction industry, the total-factor CO2-e emission performance techniques from the benchmarks have not been summarised and
has improved from 1990 to 2013, primarily resulting from tech- surveyed in this research, which would be achieved by using
nological advancement. It can be asserted that the Australian con- qualitative research methods such as participant observation and
struction industries have introduced a number of advanced in-depth interviews. Future work could use additional measure-
technologies during the research period. The scale efficiency factor ment techniques such as panel data analysis and factor analysis in
has also played a positive effort in promoting the performance measuring carbon emission performance and investigating specific
progress of CO2-e emissions. However, the pure technical efficiency factors and measures for further improving the performance.
factor did not evidently improve in most of the Australian con-
struction industries.
Appendix A. Nomenclature
The Australian construction industry overall could promote
technical efficiency through developing management ability in
bG
jw The w-th undesirable output of DMUj at period G
internal construction activities, mainly including technological
Dt(Xt; Yt,btjCRS) The efficiency of (Xt; Yt,bt) compared to the
capability, human resource management capability and resource
frontier of Ft under CRS
allocation capability. In construction project management,
Dt(Xt; Yt,btjVRS) The efficiency of (Xt; Yt,bt) compared to the
improving carbon emission performance could be achieved by
frontier of Ft under VRS
effectively and efficiently managing equipment, machinery, fleet,
D (X ; Y ,b jCRS) The efficiency of (Xtþ1; Ytþ1,btþ1)compared
t tþ1 tþ1 tþ1
site layout, energy utilisation and material supply. Furthermore, the
to the frontier of Ftunder CRS
measures and policies from the technology benchmarks of the ACT
D (X ; Y ,b jCRS) The efficiency of (Xt; Yt,bt)compared to the
tþ1 t t t
and the NT can promote CO2-e emission performance. For instance,
frontier of Ftþ1under CRS
according to ACT Sustainable Energy Policy, the ACT Government
D (X ; Y ,b jCRS) The efficiency of (Xtþ1; Ytþ1,btþ1)
tþ1 tþ1 tþ1 tþ1
has been positively encouraging construction designers and prac-
compared to the frontier of Ftþ1under
titioners to exceed minimum sustainability standards by devel-
CRS
oping, building and planning regulations such as new energy
Dtþ1(Xtþ1; Ytþ1,btþ1jVRS) The efficiency of (Xtþ1; Ytþ1,btþ1)
efficiency standards for all buildings and 6-star energy rating for
compared to the frontier of Ftþ1under
new detached dwellings. In the Carbon Neutral ACT Government
VRS
Framework, tender evaluation criteria of selecting major con-
Ft The sequential production technology set at period t
struction works consist of evaluating tender commitment and ca-
J Number of DMUs
pacity of providing effective climate change outcomes, appropriate
M Number of desirable outputs
embodied energy of construction projects and sustained reductions
N Number of inputs
of carbon emissions. Therefore, although CO2-e emission perfor-
T The t-th observed time
mance improved steadily over the research period, it would be a
T Number of the observed time periods
feasible practice to further promote CO2-e emission performance
W Number of undesirable outputs
for the whole construction industry through enhancing pure
xG The n-th input of DMUj at period G
technical efficiency and learning low-carbon construction tech- jn
yG The m-th desirable output of DMUj at period G
niques from the benchmarks. Although this research focuses on the jm
Australian construction industry, experiences and knowledge
mGj The intensity factor of DMUj in structuring Ft
learned could help other construction industries to formulate
l The efficiency of by-producing undesirable outputs when
consideration of desirable outputs
appropriate strategies and pathways for improving their carbon
performance.
q The efficiency of by-producing undesirable outputs when
no consideration of desirable outputs
This research is a leading study of quantifying the performance
measurement of CO2-e emissions in the construction domain. First
of all, the research clearly redefines that the performance mea- Appendix B. Abbreviations
surement of CO2-e emissions should focus on CO2-e emissions
themselves. The measurement results can prevent straight forward ABS Australian Bureau of Statistics
decisions about ways to reduce CO2-e emissions being made at the ACT Australian Capital Territory
expense of decreasing desirable outputs or wasting input resources. CO2-e Carbon dioxide equivalent
Furthermore, the research can assist construction industries CRS Constant returns to scale
2814 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815

DEA Data Envelopment Analysis €re, R., Grosskopf, S., Margaritis, D., 2011. Malmquist productivity indexes. In:
Fa
Cooper, W.W., Seiford, L.M., Zhu, J. (Eds.), Handbook on Data Envelopment
DMU Decision-making unit
Analysis. Springer, US, pp. 127e149.
GDP Gross domestic product €re, R., Grosskopf, S., Margaritis, D., Weber, W.L., 2012. Technological change and
Fa
MPCI Malmquist performance change index timing reductions in greenhouse gas emissions. J. Prod. Anal. 37, 205e216.
NSW New South Wales Heffernan, E., Pan, W., Liang, X., de Wilde, P., 2015. Zero carbon homes: perceptions
from the UK construction industry. Energy Policy 79, 23e36.
NT Northern Territory Horta, I.M., Camanho, A., Johnes, J., Johnes, G., 2013. Performance trends in the
PTECI Pure technical efficiency change index construction industry worldwide: an overview of the turn of the century.
Qld Queensland J. Prod. Anal. 39, 89e99.
Hsu, C., Kuo, R., Chiou, C., 2014. A multi-criteria decision-making approach for
SA South Australia evaluating carbon performance of suppliers in the electronics industry. Int.
SECI Scale efficiency change index J. Environ. Sci. Technol. 11, 775e784.
Tas Tasmania Hu, X., Liu, C., 2015. Managing undesirable outputs in the Australian construction
industry using data envelopment analysis models. J. Clean. Prod. 101,
TCI Technological change index 148e157.
Vic Victoria Hu, X., Liu, C., 2016a. Energy productivity and total-factor productivity in the
VRS Variable returns to scale Australian construction industry. Archit. Sci. Rev. 59, 432e444.
Hu, X., Liu, C., 2016b. Carbon productivity: a case study in the Australian con-
WA Western Australia struction industry. J. Clean. Prod. 112, 2354e2362.
Killip, G., 2013. Products, practices and processes: exploring the innovation po-
Appendix C. The decomposition process of the Malmquist tential for low-carbon housing refurbishment among small and medium-sized
enterprises (SMEs) in the UK construction industry. Energy Policy 62,
performance change index 522e530.
Li, Y., Liu, C., 2010. Malmquist indices of total factor productivity changes in the
Consider Dt(Xt; Yt,btjCRS), Dt(Xtþ1; Ytþ1,btþ1jCRS), Dtþ1(Xt; Australian construction industry. Constr. Manag. Econ. 28, 933e945.
Li, H., Shi, J.-F., 2014. Energy efficiency analysis on Chinese industrial sectors: an
Yt,btjCRS), Dtþ1(Xtþ1; Ytþ1,btþ1jCRS), Dt(Xt; Yt,btjVRS) and Dt(Xtþ1; improved Super-SBM model with undesirable outputs. J. Clean. Prod. 65,
Ytþ1,btþ1jVRS) represented by “a”, “b”, “c”, “d”, “e” and “f”, respec- 97e107.
tively. Therefore. Li, D., Hui, E.C., Leung, B.Y., Li, Q., Xu, X., 2010. A methodology for eco-efficiency
evaluation of residential development at city level. Build. Environ. 45, 566e573.
MPCIt,tþ1 ¼ [b/a  d/c]1/2 ¼ d/a  [b/d  a/c]1/2 ¼ f/e  d/f/a/e Lin, C.-L., Huang, H.-M., 2009. Improved baseline productivity analysis technique.
 [b/d  a/c]1/2 ¼ PTECIt,tþ1  SECIt,tþ1  TCIt,tþ1 J. Constr. Eng. M. ASCE 136, 367e376.
Long, R., Wang, H., Chen, H., 2016. Regional differences and pattern classifications in
the efficiency of coal consumption in China. J. Clean. Prod. 112, 3684e3691.
References Lovell, C.K., 2003. The decomposition of Malmquist productivity indexes. J. Prod.
Anal. 20, 437e458.
Australian Government Department of the Environment, 2015. National Inventory Lu, Y., Zhu, X., Cui, Q., 2012. Effectiveness and equity implications of carbon policies
by Economic Sector. Canberra. Retrieved from. http://bit.ly/2cbtp7P (accessed in the United States construction industry. Build. Environ. 49, 259e269.
13 October 2016). Malmquist, S., 1953. Index numbers and indifference surfaces. Trab. Estad. Investig.
ACT Government Environment and Sustainable Development, 2011. ACT Sustain- Oper. 4, 209e242.
able Energy Policy e Energy for a Sustainable City 2011-2020. Canberra. McCuen, R.H., 1973. The role of sensitivity analysis in hydrologic modeling. J. Hydrol.
Retrieved from. http://bit.ly/2cnORGh (accessed 13 October 2016). 18, 37e53.
Australian Bureau of Statistics, ABS, 2015. 8731.0-Building Approvals, Australia, Jul Molinos-Senante, M., Sala-Garrido, R., Hern andez-Sancho, F., 2016. Development
2015. ABS, Canberra. and application of the Hicks-Moorsteen productivity index for the total factor
Australian Government Department of Industry and Science, AGDIS, 2015. Austra- productivity assessment of wastewater treatment plants. J. Clean. Prod. 112,
lian Total Net Energy Consumption, by State, by Industry, Energy Units. AGDIS, 3116e3123.
Canberra. Mukherjee, A., 2012. The project emissions estimator (PE-2): a tool to aid con-
Brander, M., Davis, G., 2012. Greenhouse gases, CO2, CO2-e, and carbon: What do all tractors and agencies benchmark carbon emissions for highway construction
these terms mean. Ecometrica Press, London. projects. In: ICSDEC 2012: developing the Frontier of Sustainable Design, En-
Busch, T., 2010. Corporate carbon performance indicators revisited. J. Ind. Ecol. 14, gineering, and Construction. ASCE, pp. 144e154.
374e377. Ozawa-Meida, L., Brockway, P., Letten, K., Davies, J., Fleming, P., 2013. Measuring
Busch, T., Weinhofer, G., Hoffmann, V.H., 2011. The carbon performance of the 100 carbon performance in a UK University through a consumption-based carbon
largest US electricity producers. Util. Policy 19, 95e103. footprint: de Montfort University case study. J. Clean. Prod. 56, 185e198.
Caves, D.W., Christensen, L.R., Diewert, W.E., 1982. The economic theory of index Pahwa, A., Feng, X., Lubkeman, D., 2003. Performance evaluation of electric distri-
numbers and the measurement of input, output, and productivity. Econo- bution utilities based on data envelopment analysis. IEEE T. Power Syst. 18,
metrica 50 (6), 1393e1414. 400e405.
Chang, H., Choy, H.L., Cooper, W.W., Ruefli, T.W., 2009. Using Malmquist Indexes to Park, H.-S., Thomas, S.R., Tucker, R.L., 2005. Benchmarking of construction pro-
measure changes in the productivity and efficiency of US accounting firms ductivity. J. Constr. Eng. M. ASCE 131, 772e778.
before and after the SarbaneseOxley Act. Omega 37, 951e960. Ren, Z., Chrysostomou, V., Price, T., 2012. The measurement of carbon performance
Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the efficiency of decision of construction activities: a case study of a hotel construction project in South
making units. Eur. J. Oper. Res. 2 (6), 429e444. Wales. Smart Sustain. Built Environ. 1, 153e171.
Chen, L., Xu, L., Xu, Q., Yang, Z., 2016. Optimization of urban industrial structure Riccardi, R., Oggioni, G., Toninelli, R., 2012. Efficiency analysis of world cement in-
under the low-carbon goal and the water constraints: a case in Dalian, China. dustry in presence of undesirable output: application of data envelopment
J. Clean. Prod. 114, 323e333. analysis and directional distance function. Energy Policy 44, 140e152.
Chung, Y.H., Fa €re, R., Grosskopf, S., 1997. Productivity and undesirable outputs: a Sun, C.C., 2011. Evaluating and benchmarking productive performances of six in-
directional distance function approach. J. Environ. Manag. 51, 229e240. dustries in taiwan hsin chu industrial science park. Expert Syst. Appl. 38 (3),
Coelli, T.J., Rao, D., 2005. Total factor productivity growth in agriculture: a Malm- 2195e2205.
quist index analysis of 93 countries, 1980e2000. Agric. Econ. 32, 115e134. Tasmanian Climate Change Office e Department of Premier and Cabinet, TCCO-
del Mar Salinas-Jime nez, M., 2004. Public infrastructure and private productivity in DPAC, 2008. Tasmanian Framework for Action on Climate Change. TCCO-DPAC,
the Spanish regions. J. Pol. Model. 26, 47e64. Hobart, AU.
Egilmez, G., McAvoy, D., 2013. Benchmarking road safety of US states: a DEA-based United Nations Environment Programme e Sustainable Buildings and Climate
Malmquist productivity index approach. Accid. Anal. Prev. 53, 55e64. Initiative, UNEP-SBCI, 2008. Buildings and Climate Change: Summary for De-
Egilmez, G., Gumus, S., Kucukvar, M., Tatari, O., 2016. A fuzzy data envelopment cision-makers. UNEP-SBCI, Paris.
analysis framework for dealing with uncertainty impacts of inputeoutput life Wang, X., Chen, Y., Liu, B., Shen, Y., Sun, H., 2013. A total factor productivity measure
cycle assessment models on eco-efficiency assessment. J. Clean. Prod. 129, for the construction industry and analysis of its spatial difference: a case study
622e636. in China. Constr. Manag. Econ. 31, 1059e1071.
Eltaief, A., Ahmad, A., 2011. Efficiency and Productivity Analysis of Construction Wong, P.S., Zapantis, J., 2013. Driving carbon reduction strategies adoption in the
Companies in Libya/Abdullah Ahmad a. Eltaief. Universiti Teknologi MARA. Australian construction sectoreThe moderating role of organizational culture.
€re, R., Grosskopf, S., Norris, M., Zhang, Z., 1994. Productivity growth, technical
Fa Build. Environ. 66, 120e130.
progress, and efficiency change in industrialized countries. Am. Econ. Rev. 84, Wu, J., Xiong, B., An, Q., Zhu, Q., Liang, L., 2015. Measuring the performance of
66e83. thermal power firms in China via fuzzy Enhanced Russell measure model with
€re, R., Grosskopf, S., Margaritis, D., 2006. Productivity growth and convergence in
Fa undesirable outputs. J. Clean. Prod. 102, 237e245.
the European union. J. Prod. Anal. 25, 111e141. Xue, X., Wu, H., Zhang, X., Dai, J., Su, C., 2015. Measuring energy consumption
 X. Hu et al. / Journal of Cleaner Production 142 (2017) 2804e2815 2815

efficiency of the construction industry: the case of China. J. Clean. Prod. 107,
509e515.
Zhang, N., Zhou, P., Kung, C.-C., 2015. Total-factor carbon emission performance of
the Chinese transportation industry: a bootstrapped non-radial Malmquist in-
dex analysis. Renew. Sustain. Energy Rev. 41, 584e593.
Zhou, P., Ang, B.W., Poh, K.-L., 2008. A survey of data envelopment analysis in en-
ergy and environmental studies. Eur. J. Oper. Res. 189, 1e18.
Zhou, P., Ang, B., Han, J., 2010. Total factor carbon emission performance: a
Malmquist index analysis. Energy Econ. 32, 194e201.