Eciency measurement of selected Australian and other

international ports using data envelopment analysis
q
Jose Tongzon
*
Department of Economics, National University of Singapore, 10 Kent Ridge Crescent, Singapore 0511, Singapore
Received 12 October 1998; received in revised form 15 July 1999; accepted 26 July 1999
Abstract
Available studies have not provided a satisfactory answer to the problem of making international
comparisons of port eciency. This study applies data envelopment analysis (DEA) to provide an eciency
measurement for four Australian and twelve other international container ports. While DEA has been
applied to a wide number of di€erent situations where eciency comparisons are required, this technique
has not previously been applied to ports. The DEA technique is useful in resolving the measurement of port
eciency because the calculations are non-parametric, can handle more than one output and do not require
an explicit a priori determination of relationships between output and inputs, as is required for conven-
tional estimation of eciency using production functions. The ports of Melbourne, Rotterdam, Yokohama
and Osaka are found to be the most inecient ports in the sample, based on constant and variable returns
to scale assumptions, mainly due to the enormous slack in their container berths, terminal area and labor
inputs. The study also draws some policy implications for ports and recommends certain areas for future
research. Ó 2000 Elsevier Science Ltd. All rights reserved.
Keywords: DEA; Port eciency; Measurement; Australia; International Ports
1. Introduction
In order to support trade oriented economic development, port authorities have increasingly
been under pressure to improve port eciency by ensuring that port services are provided on an
Transportation Research Part A 35 (2001) 107±122
www.elsevier.com/locate/tra
q
The author would like to express his deep appreciation to the anonymous referees for their valuable comments and
suggestions made on the earlier version of this paper. However, the author alone is responsible for any errors and
omissions made in this paper.
*
Tel.: +65-874-3955; fax: +65-775-2646.
E-mail address: ecsjt@nus.edu.sg (J. Tongzon).
0965-8564/00/$ - see front matter Ó 2000 Elsevier Science Ltd. All rights reserved.
PII: S0965- 8564( 99) 00049- X
internationally competitive basis. Ports form a vital link in the overall trading chain and, con-
sequently, port eciency is an important contributor to a nation's international competitiveness
(Tongzon, 1989; Chin and Tongzon, 1998). Thus, monitoring and comparing oneÕs port with
other ports in terms of overall eciency has become an essential part of many countriesÕ
microeconomic reform programmes.
This study hopes to contribute to this important task by applying an innovative approach to
port eciency ratings covering a selected sample of ports. Relying on mathematical programming
techniques, this approach, called data envelopment analysis (DEA), has been applied to a wide
number of di€erent situations where eciency comparisons are required due to its inherent ad-
vantages compared with conventional approaches.
1
Firstly, the characteristics of DEA, such as
its ability to analyze several outputs and inputs simultaneously and to derive eciency rating
within a set of analyzed units, are particularly suitable for measuring port eciency. Port output
can be multi-dimensional depending on the objective that ports want to achieve. Secondly, DEA
does not require the development of ``standards'' against which eciency is measured, although
such standards can be incorporated in the DEA analysis. Roll and Hayuth (1993) have advocated
the use of this approach to the measurement of port eciency, and demonstrated, based on
hypothetical port data, how the relative eciency ratings of ports could be obtained. This paper
builds on their work by applying the DEA analysis to actual performance data for selected ports.
Given the multiplicity of ports and cargoes handled, it is necessary to restrict the scope of
analysis to a limited number of ports and a speci®c type of cargo. This study examines eciency
with respect to containerized cargoes across ports recognized for their high level performance (in
terms of throughput) in Asia and Europe for which data are available.
2
Data availability is
particularly important since many of the ports surveyed for data via questionnaires refused to
reveal information on some aspects of their port operations due to con®dentiality. Thus, apart
from the data obtained from the survey, the study has to depend on secondary sources. The
following are the secondary sources of data for this study: the Australian Bureau of Transport
and Communications Economics (1996) survey data on four Australian ports and selected Asian
and European ports for data on reliability and speed; Containerisation International Yearbook
(1998) and LloydÕs Ports of the World (1998) for data on port infrastructure. The Australian
Bureau of Transport and Communications Economics data on reliability and speed should be
quite reliable and unbiased since these were obtained from the same shipping lines calling at the
selected ports, rather than from their various port authorities or terminal operators. These data
and the sampled ports are shown in Appendix A.
The rest of the paper is organized as follows. Section 2 discusses the output and input measures
used. Section 3 outlines DEA and gives a brief review of related studies which have used this
technique. Section 4 provides the empirical results for DEA applied to 16 container ports. Finally,
Section 5 summarizes the main results in the paper.
1
Conventional approaches to eciency comparisons include the use of ``standards'' against which eciency is
measured, estimation of production functions using regression analysis and the use of partial indicators representing
di€erent aspects of operation.
2
A comparative study of individual terminals within ports would have given us more speci®c information on port
eciency and allowed us to formulate terminal-speci®c strategies, but this study is more concerned with port eciency
representing all terminals combined to draw more aggregate policy implications.
108 J. Tongzon / Transportation Research Part A 35 (2001) 107±122
2. Port output and port input measures
Various studies have compared ports using selected performance and eciency criteria or
measures, for example, the Australian Bureau of Industry Economics (1993, 1995), Australian
Transport Advisory Council (1992), and Tongzon and Ganesalingam (1994). However, these
measures partially re¯ect di€erent aspects of port operation and fail to provide us with an overall
measure of port eciency. DEA analysis can combine all these partial measures without the
setting of a priori weights for the various parameters to produce an overall eciency measure.
Further, in contrast to conventional econometric techniques such as the regression analysis used
to estimate a production function, in DEA more than one output measure can be speci®ed. A
number of di€erent measures of port output are available, depending on which features of port
operation are being evaluated.
This study uses two output and six input measures of port performance for the year 1996 ± the
year for which the latest data on port throughputs are available.
3
The output measures are cargo
throughput and ship working rate. The ®rst output measure is the total number of containers
loaded and unloaded in 20-foot equivalent units (TEUs). This output relates to the need for
cargo-related facilities and services. Further, since ships are major port users, the second output is
the ship working rate. Ship working rate measures the number of containers moved per working
hour per ship and thus is an indicator of the speed with which ships are worked. This measure can
represent the level and quality of port service. Since the container handling aspect of port op-
eration is the largest component of total ship turnaround time, the speed of moving cargoes o€
and onto ships at berth has considerable implications for the port users. Moreover, improving
eciency in this area is consistent with port authority intentions of maximizing berth utilization, a
factor which will in¯uence both port charges imposed on shipowners and the actual throughput
handled.
To produce the above outputs and to facilitate port operations, a variety of inputs are required.
Based on the production framework, port inputs can be generalized as land, labor and capital.
The major capital inputs in port operations are the number of berths, cranes and tugs. The most
fundamental labor input is the number of stevedoring labor. However, due to a lack of infor-
mation on this particular variable, a proxy variable is used represented by the number of port
authority employees for the respective ports.
4
This proxy variable is less dicult to obtain be-
cause it is usually published in the annual reports of some ports.
5
With respect to the land input,
3
Ideally, more than one single year of observations should have used to provide a more representative basis for this
comparative study of ports. However, due to the unavailability of complete data for other years, the study has to rely on
one single year set of observations.
4
Obtaining accurate data on the number of stevedoring labor for various ports is extremely dicult due to the fact
that stevedoring operations in many ports have been privatized with cargo-handling operations being handled by a
number of private companies. These private companies have also hired casual dockworkers the number of which
change from time to time. Some private companies are also reluctant to reveal this type of information for
con®dentiality reasons.
5
Although the use of this proxy variable has certain limitations, the labor input variable is so important that it should
at least be represented in the analysis.
J. Tongzon / Transportation Research Part A 35 (2001) 107±122 109
the study uses the terminal area of the ports.
6
Another important factor in¯uencing port outputs
is the amount of delay time which is the di€erence between total berth time plus time waiting to
berth and the time between the start and ®nish of ship working, and is an indicator of how well
working time is being used. These delays could be due to labor disputes, work practices such as
meal breaks, equipment breakdown, port congestion, perceived ship problems or bad weather.
These output and input variables are also de®ned in Appendix A.
3. Data envelopment analysis
DEA is an eciency evaluation model based on mathematical programming theory. DEA
o€ers an alternative to classical statistics in extracting information from sample observations. In
contrast to parametric approaches such as regression analysis which ®t the data through a single
regression plane, DEA optimizes each individual observation with the objective of calculating a
discrete piece-wise frontier determined by the set of Pareto ecient decision management units
(DMUs).
7
In other words, the focal point of DEA is on individual observations as opposed to
single optimization statistical approaches which focus on averages of parameters. In the present
application, DEA refers to each port as a DMU, in the sense that each is responsible for con-
verting inputs into outputs. DEA analysis can involve multiple inputs as well as multiple outputs
in its eciency valuation. This makes DEA analysis more suitable for port eciency measurement
because ports produce a number of di€erent outputs. Among these outputs are the quantities and
the variety of cargoes handled, the types of ships serviced, the interchange with land transport
modes, the additional services rendered such as warehousing and so on (Roll and Hayuth, 1993,
p. 153). Furthermore, DEA calculations are non-parametric and do not require an explicit a priori
determination of relationships between inputs and outputs, or the setting of rigid importance
weights for the various factors. Benchmarking has also become a part of normal commercial
culture of the port industry. DEA makes benchmarking easier and more realistic because it en-
ables derivation of an eciency envelope, which contains the most ecient ports of the group
analyzed, against which all other ports are compared, rather than just choosing the most ecient
port. The choice of one port representing the best international practice may be unfair due to
di€erences in contexts. For many applications, these features make DEA a more ¯exible tool as
compared to other conventional eciency measures derived from stochastic production frontier
or economic value added (EVA), which are based on production function estimation involving
many inputs but only one output.
Since its introduction by Charnes et al. (1978), there have been many applications of DEA.
Some applications have involved eciency evaluation of organizations with characteristics similar
to ports, such as hospitals (Banker et al., 1986), schools (Ray, 1991), courts (Lewin et al., 1982),
post oces (Deprins et al., 1984), and air force maintenance units (Charnes et al., 1985). DEA
6
For the purpose of analysis, it is assumed here that the factor inputs identi®ed have equal unit productivity across all
ports in the sample.
7
Two of the many studies comparing DEA with the traditional production function approach to measuring eciency
are Gong and Sickles (1992), and Bowlin et al. (1985).
110 J. Tongzon / Transportation Research Part A 35 (2001) 107±122
provides the ¯exibility to permit ``unconventional'' variables such as the number of students
graduated, number of patients served, even journal ranking (Burton and Phimister, 1995) to be
used for eciency evaluation. DEA has also been applied in the transportation sector to airlines
(Banker and Johnston, 1994; Charnes et al., 1996), and railways (Oum and Yu, 1994). A detailed
bibliography related to DEA (1978±1992) can be found in Charnes et al. (1995, ch. 22). Since the
early work of Charnes, Cooper and Rhodes (CCR), there have been a number of extensions to the
DEA model. For example, Charnes et al. (1985) introduced window analysis to handle panel data
sets involving pooled cross section and time series observations.
The concept of DEA is developed around the basic idea that the eciency of a DMU is de-
termined by its ability to transform inputs into desired outputs. This concept of eciency was
adopted from engineering which de®nes the eciency of a machine/process as Output=Input 61.
In this approach, eciency is always less than or equal to unity as some energy loss will always
occur during the transformation process. DEA generalizes this single output/input technical ef-
®ciency measure to multiple outputs/inputs by constructing a relative eciency measure based on
a single ``virtual'' output and a single virtual input. The ecient frontier is then determined by
selecting DMUs which are most ecient in producing the virtual output from the virtual input.
Because DMUs on the ecient frontier have an eciency score equal to 1, inecient DMUs are
measured relative to the ecient DMUs. The eciency measure is relative to other DMUs. It is
not possible to determine if DMUs judged to be ecient are optimizing the use of inputs to
produce outputs.
More formally, assume that there are n DMUs to be evaluated. Each DMU consumes varying
amounts of m di€erent inputs to produce s di€erent outputs. Speci®cally, DMU
j
consumes
amounts fX
j
ˆ x
ij
g of inputs …i ˆ 1; . . . ; m† and produces amounts Y
j
ˆ fy
rj
g of outputs
…r ˆ 1; . . . ; s†. The s ´ n matrix of output measures is denoted by Y, and the m ´ n matrix of input
measures is denoted by X. Also, assume that x
ij
>0 and y
rj
>0. Consider the problem of evalu-
ating the relative eciency for any one of the n DMUs, which will be identi®ed as DMU
0
.
Relative eciency for DMU
0
is calculated by forming the ratio of a weighted sum of outputs to a
weighted sum of inputs, subject to the constraint that no DMU can have a relative eciency score
greater than unity. Symbolically:
max
u;v
P
r
u
r
y
r0
P
i
v
i
x
i0
ˆ
u
T
Y
0
v
T
X
0
; where u ˆ …u
1
; . . . ; u
s
†
T
; v ˆ …v
1
; . . . ; v
m
†
T
subject to
u
T
Y
j
v
T
X
j
ˆ
P
r
u
r
y
rj
P
i
v
i
x
ij
61
for j ˆ 1; 2; . . . ; n; u
r
; v
i
P0 for r ˆ 1; 2; . . . ; s and i ˆ 1; 2; . . . ; n;
where u
r
and v
i
are weights assigned to output r and input i, respectively.
For this fractional programming problem with a potentially in®nite number of optimal solu-
tions, Charnes et al. (1978) were able to specify an equivalent linear programming problem (LP).
This requires the introduction of a scalar quantity (h) to adjust the input and output weights:
h ˆ
1
v
T
X
0
; l
T
ˆ hu
T
; x ˆ hv
T
:
J. Tongzon / Transportation Research Part A 35 (2001) 107±122 111
Appropriate substitutions produce the CCR LP problem:
max
l;v
K
0
ˆ
X
r
l
r
y
r0
ˆ l
T
Y
0
subject to
x
T
X
0
ˆ
X
i
x
i
x
i0
ˆ 1;
X
r
l
r
y
rj
À
X
i
x
i
x
ij
60; l
r
; x
i
P;
where the value of K
0
is the relative eciency of DMU
0
and is a positive constant, called the
non-Archimedian in®nitesimal, which is introduced to facilitate solving of the LP problem. In
DEA, this LP is known as the CCR model, as it was developed by Charnes, Cooper and Rhodes.
In addition to the CCR DEA model, two other DEA models are also often associated with the
DEA methodology (e.g., Ali et al., 1995): the BCC model and the Additive model. The models
di€er mainly in their envelopment surface orientation and projection path to the ecient frontier
for an inecient DMU. The CCR model results in a constant returns to scale, piece-wise linear
envelopment surface with both input and output orientations for projection paths. The BCC
model provides a variable returns to scale, piece-wise linear envelopment surface, which is similar
to the Additive model. However, its projection path has both input and output orientations,
which di€er from the Additive model. The Additive model was introduced by Charnes et al.
(1985). The envelopment surface derived from the Additive model has a piece-wise linear, variable
returns to scale property. The model is based on the concept of a Pareto ecient (minimum)
function. For any particular one of the n DMUs, again denoted by DMU
0
, the LP for the Ad-
ditive model is:
max
l;v;u
0
X
0
ˆ l
T
Y
0
À x
T
X
0
‡ c
0
subject to
l
T
Y À x
T
X ‡ c
0
s 60;
Àl
T
; Àx
T
6 À s …where s is a column vector of 1†
The presence of an unconstrained vector variable c
0
in the objective function results in variable
returns to scale for the Additive model, as opposed to constant returns to scale in the CCR
model.
8
When c
0
ˆ0, the Additive model reduces to the CCR model.
The term ``relative eciency'' is used in DEA because the eciency of each DMU is calculated
with reference to all the other DMUs that are being selected for assessment. For multiple inputs
and/or outputs, the envelopment surface will be multidimensional. All those DMUs that lie on the
frontier have an eciency score of 1 and are considered DEA ecient, while those below will be
classi®ed as DEA inecient and have eciency scores of less than 1. For an inecient DMU, the
facet is the combination of ecient DMUs which are used to determine the relative ineciency.
8
A variation on the Additive model, referred to as the multiplicative model, yields a piece-wise Cobb±Douglas
(variable returns to scale) or piece-wise log-linear (constant returns to scale) envelopment surface which results from the
application of the Additive model to the logarithms of the data, with (for variable returns to scale) or without (for
constant returns to scale) the unconstrained variable k
0
.
112 J. Tongzon / Transportation Research Part A 35 (2001) 107±122
The relative weightings of the ecient DMUs in the facet, the lambdas, are available for the
Additive model, because the convexity constraint imposes the condition that the sum of the
lambdas equals 1. No such restriction is available for the CCR model. One useful feature of DEA
is the power to identify the sources and amount of wastage in inputs, or shortfalls in outputs, for
each DEA-inecient DMU. The level of ineciency is determined by comparison to the facet
located on the ecient frontier. In practice, the sources of ineciency can be important piece of
information as it can enable managers to identify the problem areas for these inecient DMUs
and provide precise information about the corrective e€orts required to achieve ecient perfor-
mance.
4. Empirical results
As discussed previously, the DEA empirical analysis uses two output measures: TEUs handled
(the number of twenty foot container equivalent units handled), and shiprate (ship working rate
which measures the number of containers moved per working hour). The input measures used are:
nocranes (number of cranes), noberths (number of container berths), notugs (number of tugs),
termiarea (terminal area), delaytime (delay time) and labor (proxied by the number of port au-
thority employees. The DEA software employed was developed at the Centre for Cybernetic
Studies, University of Texas and is described in Assad (1986).
Without precise information on the returns to scale of the port production function, two sets of
results, for the CCR and Additive DEA models, are presented and discussed. The eciency
rankings calculated using these two approaches are given in Table 1(2). Comparison of these
results reveals that the CCR model identi®es slightly more inecient ports (6 vs. 3) than the
Additive model. Following the discussion in Section 3, this is not surprising, as the CCR model
®ts a linear production technology and the Additive model features variable returns to scale,
which are more ¯exible and will, typically, require a larger number of ports to de®ne the eciency
frontier. An analogy to conventional economics would feature a linear production possibilities
frontier with a piecewise convex frontier (e.g., Favero and Papi, 1995, p. 387). The linear frontier
would be tangent to the convex frontier only over a segment, being above the convex frontier
elsewhere.
Interpreting the results of Table 1(2) depends on assumptions made about the production
technology for ports. Ports that are judged to be inecient with variable returns to scale will also
be inecient with linear production relations, but not the converse. But the results presented in
Table 1(2) are quite surprising. There are more ports identi®ed to be ecient than inecient. The
results seem to suggest that the model is over-speci®ed for the sample size, i.e., the sample data are
not enough to develop a meaningful facet or frontier. Due to data constraints, it is impossible to
increase the sample size for this study. Thus, to resolve this problem, only one output is used ± the
number of TEUs handled. The results of the DEA analysis using only one output are presented in
Table 1(3).
On the basis of the results presented in Table 1(3), ports identi®ed as inecient with the
Additive and CCR models are Melbourne, Rotterdam, Yokohama and Osaka. The ports of
Felixstowe, Sydney, Fremantle, Brisbane, Tilbury and La Spezia identi®ed as ecient using the
Additive model, become inecient under CCR. The four ports judged to be inecient with both
J. Tongzon / Transportation Research Part A 35 (2001) 107±122 113
DEA models have opposite characteristics in terms of size and function. The port of Melbourne is
quite small relative to Rotterdam. The port of Rotterdam is a hub port in Western Europe while
the ports of Melbourne, Yokohama and Osaka generate most of their cargo internally.
Six (6) out of sixteen (16) ports found to be ecient using CCR, the group of most ecient
ports, do not have any discernible characteristics. Large and hub ports, such as Singapore and
Hong Kong, and relatively small ports such as Zeebrugge, Tanjung Priok, Keelung and Hamburg,
are included in the group. These ®ndings imply that a portÕs operational eciency level does not
depend solely on its size or its function (i.e., whether it is a hub or feeder port).
In addition to providing eciency measures, DEA also provides other information relevant
for the inecient DMUs. In particular, DEA identi®es the ecient facet being used for com-
parison as well as a combination of the inputs which are being ineciently utilized and the
deviation of speci®c outputs from the ecient level. Because ecient DMUs do not have any
slack, this information is only of interest for inecient DMUs. Table 2 provides the Additive
model results for the inecient DMUs, as well as the facet port numbers and associated
lambdas. As mentioned in Section 3, for the Additive model the convexity constraint requires
the lambdas to sum to one, permitting the lambdas to be interpreted as relative weights. Higher
lambda values indicate which ecient port was more important in determining the ineciency
of the particular inecient DMU. For the port of Melbourne, port #11, Brisbane, has the
highest weight in the facet, while for the port of Rotterdam, port #13, Zeebrugge has the
highest weight. Sensitivity analysis (not reported) was conducted to determine if dropping
Table 1
Relative eciency measures using the CCR and additive DEA models
a
(1) (2) (3)
Port CCR ADDITIVE CCR ADDITIVE
2 Outputs 1 Output
Melbourne 0.5885 0.6633 0.2688 0.6633
Hong Kong 1.0000 1.0000 1.0000 1.0000
Hamburg 1.0000 1.0000 1.0000 1.0000
Rotterdam 0.6644 0.8228 0.5404 0.8228
Felixstowe 1.0000 1.0000 0.6311 1.0000
Yokohama 0.8456 1.0000 0.4898 0.8104
Singapore 1.0000 1.0000 1.0000 1.0000
Keelung 1.0000 1.0000 1.0000 1.0000
Sydney 0.7676 1.0000 0.3057 1.0000
Fremantle 0.8251 1.0000 0.1878 1.0000
Brisbane 1.0000 1.0000 0.1955 1.0000
Tilbury 1.0000 1.0000 0.1734 1.0000
Zeebrugge 1.0000 1.0000 1.0000 1.0000
La Spezia 1.0000 1.0000 0.5261 1.0000
Tanjung Priok 1.0000 1.0000 1.0000 1.0000
Osaka 0.6050 0.6023 0.1954 0.6004
a
Outputs and inputs used in making the eciency evaluations are given in Section 2. CCR is the Charnes et al. (1978)
DEA method and Additive is the additive DEA model of Charnes et al. (1985). A score of 1.0000 indicates the port is
ecient.
114 J. Tongzon / Transportation Research Part A 35 (2001) 107±122
Table 2
Additive DEA results for ports identi®ed as inecient
Value measured Value if ecient Slack
1. Melbourne
Eciency ˆ0.6633 Facet rate ˆ3.4286
Facet: 2 11 7
Lambda ˆ0:015 0:882 0:103
Output
TEUs 904618.0 1748797.8 844179.8
Inputs
nocranes 16.0 16.0 0.0
noberths 12.0 4.7 7.3
notugs 6.0 6.0 0.0
termiare 1184100.0 756840.1 427259.9
delaytime 8.0 5.2 2.8
labor 829.0 288.8 540.2
4. Rotterdam
Eciency ˆ0.8228 Facet rate ˆ2.0622
Facet: 13 7
Lambda ˆ0:646 0:354
Output
TEUs 4935616.0 4935616.0 0.0
Inputs
nocranes 66.0 43.9 22.1
noberths 18.0 11.8 6.2
notugs 15.0 7.5 7.5
termiare 4158000.0 2547402.3 1610597.7
delaytim 1.7 1.5 2.0
labor 981.0 359.5 621.5
6. Yokohama
Eciency ˆ0.8104 Facet rate ˆ1.5755
Facet: 14 2
Lambda ˆ0:758 0:242
Output
TEUs 3911927.0 3911927.0 0.0
Inputs
nocranes 41.0 21.5 19.5
noberths 20.0 9.7 10.3
notugs 34.0 11.9 22.1
termiarea 1823250.0 735764.8 1087485.2
delaytime 6.0 4.0 2.0
labor 472.0 327.5 144.5
16. Osaka
Eciency ˆ0.6004 Facet rate ˆ3.9341
Facet: 2 14 11 7
Lambda ˆ0.128 0:611 0:156 0:105
Output
TEUs 987948.0 3654292.3 2666344.3
J. Tongzon / Transportation Research Part A 35 (2001) 107±122 115
Brisbane from the sample had a signi®cant impact on the eciency measures. The eciency
measures did not change appreciably.
Interpreting the information from the facet is more straight forward than evaluating the sources
of ineciency. Based on the results presented in Table 2, container berths, terminal area and labor
in particular are less eciently utilized resulting in idle capacity in the four inecient ports.
Consequently, port output can be signi®cantly higher. Higher levels of all outputs can be
achieved, with improvements in the usage of these inputs. It is interesting to note that under-
utilization of labor is the main source of ineciency in these ports, with the exception of the port
of Yokohama.
Table 3 provides a list of inecient ports under the CCR DEA model and their sources of
ineciency. A comparison of Tables 2 and 3 results reveals some di€erences. The facets are
sometimes di€erent. For example, the port of Singapore is no longer in the facet for the ports
of Melbourne and Osaka. Further, all Australian ports in the sample are now judged as
relatively inecient. However, the ports of Singapore and Hong Kong still appear in the facet
in nearly all inecient ports, and the CCR results identify similar sources of ineciencies in
inputs. Like the port of Melbourne, the ports of Sydney and Fremantle have an enormous
slack in the terminal area and labor input. But, whereas the port of Sydney underutilizes
considerably its working time (due to considerable delay time), the port of Fremantle is doing
well in this area.
5. Conclusions
This study is an attempt to provide a satisfactory answer to the problem of making eciency
comparisons across ports by applying the DEA analysis to a sample of Australian and other
international ports for which relevant data are available. The eciency results obtained depend on
the type of DEA model employed which, in turn, depends on an assumption made about the
returns to scale properties of the port production function. If a linear technology is assumed, then
ten (10) of the sixteen (16) ports examined are found to be inecient based on 1996 available port
data.
Four ports, ports of Melbourne, Rotterdam, Yokohama and Osaka are found to be the most
inecient ports in the sample using both constant and variable returns to scale assumptions.
Three other Australian ports, Sydney, Brisbane and Fremantle, which are found to be ecient
Table 2 (Continued)
Value measured Value if ecient Slack
Inputs
nocranes 24.0 24.0 0.0
noberths 13.0 8.8 4.2
notugs 10.0 10.0 0.0
termiarea 1154000.0 833389.9 320610.1
delaytime 4.0 4.0 0.0
labor 1070.0 344.5 725.5
116 J. Tongzon / Transportation Research Part A 35 (2001) 107±122
Table 3
CCR DEA results for ports identi®ed as inecient
a
Value measured Value if ecient Slack
1. Melbourne
Facet: 2 15
Lambda ˆ0:067 0:000
Output
TEUs 904618.0 904618.0 0.0
Inputs
nocranes 16.0 4.3 0.0
noberths 12.0 1.2 2.0
notugs 6.0 1.6 0.0
termiarea 1184100.0 147714.5 170576.0
delaytime 8.0 0.3 1.8
labor 829.0 53.7 169.1
4. Rotterdam
Facet: 2 7
Lambda ˆ0:016 0:365
Output
TEUs 4935616.0 4935616.0 0.0
Inputs
nocranes 66.0 35.7 0.0
noberths 18.0 6.5 3.2
notugs 15.0 4.8 3.3
termiarea 4158000.0 1121509.8 1125470.6
delaytime 1.7 0.9 0.0
labor 981.0 369.4 160.7
5. Felixstowe
Facet: 7
Lambda ˆ0.158
Output
TEUs 2042423.0 2042423.0 0.0
Inputs
nocranes 29.0 15.0 3.3
noberths 13.0 2.7 5.5
notugs 3.0 1.9 0.0
termiarea 1432000.0 469934.8 433732.5
delaytime 0.6 0.4 0.0
labor 1824.0 154.1 996.9
6. Yokohama
Facet: 2 13
Lambda ˆ0:286 0:111
Output
TEUs 3911927.0 3911927.0 0.0
Inputs
nocranes 41.0 20.1 0.0
noberths 20.0 6.1 3.6
notugs 34.0 7.4 9.2
termiarea 1823250.0 884962.1 8036.0
J. Tongzon / Transportation Research Part A 35 (2001) 107±122 117
Table 3 (Continued)
Value measured Value if ecient Slack
delaytime 6.0 1.5 1.4
labor 472.0 231.2 0.0
9. Sydney
Facet: 7 2
Lambda ˆ0:029 0:024
Output
TEUs 695312.0 695312.0 0.0
Inputs
nocranes 14.0 4.3 0.0
noberths 11.0 0.9 2.4
notugs 3.0 0.9 0.0
termiarea 1124500.0 138726.7 205035.3
delaytime 9.5 0.2 2.7
labor 635.0 47.4 146.7
10. Fremantle
Facet: 2 15
Lambda ˆ0:012 0:033
Output
TEUs 202680.0 202680.0 0.0
Inputs
nocranes 5.0 0.9 0.0
noberths 7.0 0.4 0.9
notugs 5.0 0.6 0.3
termiarea 273000.0 35639.5 15622.4
delaytime 9.0 1.7 0.0
labor 498.0 58.7 34.8
11. Brisbane
Facet: 15 2
Lambda ˆ0:017 0:017
Output
TEUs 249439.0 249439.0 0.0
Inputs
nocranes 6.0 1.2 0.0
noberths 3.0 0.4 0.2
notugs 5.0 0.6 0.4
termiarea 474000.0 42059.7 50614.6
delaytime 5.5 0.9 0.1
labor 200.0 39.1 0.0
12. Tilbury
Facet: 2 7
Lambda ˆ0:028 0:001
Output
TEUs 394772.0 394772.0 0.0
Inputs
nocranes 11.0 1.9 0.0
noberths 4.0 0.5 0.2
notugs 4.0 0.7 0.0
118 J. Tongzon / Transportation Research Part A 35 (2001) 107±122
based on variable returns to scale assumption, are inecient based on constant returns to scale
assumption. The ports of Hong Kong, Singapore, Hamburg, Keelung, Zeebrugge and Tanjung
Priok are found to be ecient independent of the returns to scale assumption and number of
outputs, indicating that port size or function alone is not the primary determinant of port ef-
®ciency. The enormous slack in the terminal area and labor input found in the ports of Mel-
bourne, Sydney and Fremantle have con®rmed the need to refocus the current government-
initiated waterfront reform in these areas as an essential step towards improving port eciency
levels.
This study has demonstrated that DEA provides a viable method of evaluating relative port
eciency. DEA has recently been successfully applied to a number of di€erent economic eciency
measurement situations. The technique o€ers a signi®cant alternative to classical econometric
approaches to extracting eciency information from sample observations, such as the use of
stochastic frontier production functions. Important features of DEA are that the technique is
non-parametric and that more than one output measure can be speci®ed. In the case of port
eciency, the ability to handle more than one output is particularly appealing because a number
Table 3 (Continued)
Value measured Value if ecient Slack
termiarea 519000.0 65260.9 24743.1
delaytime 4.5 0.1 0.6
labor 750.0 23.6 106.4
14. La Spezia
Facet: 8 2
Lambda ˆ0:006 0:064
Output
TEUs 871100.0 871100.0 0.0
Inputs
nocranes 8.0 4.2 0.0
noberths 7.0 1.2 2.5
notugs 8.0 1.6 2.6
termiarea 270000.0 142039.3 0.0
delaytime 3.7 0.4 1.6
labor 177.0 54.9 38.3
16. Osaka
Facet: 15 2
Lambda ˆ0:008 0:073
Output
TEUs 987948.0 987948.0 0.0
Inputs
nocranes 24.0 4.7 0.0
noberths 13.0 1.4 1.2
notugs 10.0 1.8 0.1
termiarea 1154000.0 161983.5 63559.5
delaytime 4.0 0.8 0.0
labor 1070.0 70.7 138.4
a
Output results are not reported if outputs were ecient. Facet numbers refer to port numbers assigned in Table 1.
J. Tongzon / Transportation Research Part A 35 (2001) 107±122 119
T
a
b
l
e
4
1
9
9
6
P
o
r
t
d
a
t
a
a
;
b
P
o
r
t
s
T
E
U
s
h
a
n
d
l
e
d
S
h
i
p
c
a
l
l
s
S
h
i
p
r
a
t
e
C
r
a
n
e
p
r
o
d
N
o
.
o
f
c
r
a
n
e
s
N
o
.
o
f
c
o
n
t
a
i
n
e
r
b
e
r
t
h
s
N
o
.
o
f
t
u
g
s
T
e
r
m
i
n
a
l
a
r
e
a
(
m
2
)
D
e
l
a
y
t
i
m
e
(
h
)
L
a
b
o
r
(
u
n
i
t
s
)
1
.
M
e
l
b
o
u
r
n
e
9
0
4
6
1
8
8
2
3
2
0
.
8
1
4
.
8
1
6
1
2
6
1
1
8
4
1
0
0
8
.
0
8
2
9
2
.
H
o
n
g
K
o
n
g
1
3
4
6
0
3
4
3
1
2
8
8
0
4
5
.
0
4
5
.
0
6
4
1
8
2
4
2
1
9
8
3
0
0
5
.
0
8
0
0
3
.
H
a
m
b
u
r
g
3
0
5
4
3
2
0
4
1
7
8
3
7
.
2
1
9
.
6
5
2
1
4
2
5
3
0
3
0
0
0
0
0
.
2
1
1
6
8
4
.
R
o
t
t
e
r
d
a
m
4
9
3
5
6
1
6
5
5
4
4
3
2
.
0
1
6
.
0
6
6
1
8
1
5
4
1
5
8
0
0
0
1
.
7
9
8
1
5
.
F
e
l
i
x
s
t
o
w
e
2
0
4
2
4
2
3
2
6
7
7
5
6
.
4
2
3
.
5
2
9
1
3
3
1
4
3
2
0
0
0
0
.
6
1
8
2
4
6
.
Y
o
k
o
h
a
m
a
3
9
1
1
9
2
7
1
1
9
0
8
4
7
.
0
4
7
.
0
4
1
2
0
3
4
1
8
2
3
2
5
0
6
.
0
4
7
2
7
.
S
i
n
g
a
p
o
r
e
1
2
9
4
3
9
0
0
2
4
0
1
5
4
0
.
0
3
9
.
3
9
5
1
7
1
2
2
9
7
9
2
1
1
2
.
3
9
7
8
8
.
K
e
e
l
u
n
g
2
3
2
0
3
9
7
3
1
4
4
2
4
.
0
2
4
.
0
2
3
1
4
9
3
3
9
0
0
0
1
3
.
0
6
9
0
9
.
S
y
d
n
e
y
6
9
5
3
1
2
7
5
9
2
2
.
8
1
3
.
4
1
4
1
1
3
1
1
2
4
5
0
0
9
.
5
6
3
5
1
0
.
F
r
e
m
a
n
t
l
e
2
0
2
6
8
0
6
9
2
1
3
.
3
1
2
.
9
5
7
5
2
7
3
0
0
0
9
.
0
4
9
8
1
1
.
B
r
i
s
b
a
n
e
2
4
9
4
3
9
5
5
6
2
1
.
0
1
2
.
5
6
3
5
4
7
4
0
0
0
5
.
5
2
0
0
1
2
.
T
i
l
b
u
r
y
3
9
4
7
7
2
3
4
7
3
2
.
8
1
8
.
2
1
1
4
2
5
1
9
0
0
0
4
.
5
7
5
0
1
3
.
Z
e
e
b
r
u
g
g
e
5
5
3
1
7
5
1
6
0
8
3
6
.
7
2
6
.
2
1
6
9
5
2
3
1
1
1
0
0
1
.
0
2
1
1
4
.
L
a
S
p
e
z
i
a
8
7
1
1
0
0
1
0
4
5
2
3
.
9
1
7
.
1
8
7
8
2
7
0
0
0
0
3
.
7
1
7
7
1
5
.
T
a
n
j
u
n
g
P
r
i
o
k
1
4
2
1
6
9
3
3
2
3
9
1
8
.
0
1
8
.
0
1
0
6
1
1
3
1
0
0
0
0
5
0
.
0
1
5
1
3
1
6
.
O
s
a
k
a
9
8
7
9
4
8
2
3
7
5
3
2
.
0
3
2
.
0
2
4
1
3
1
0
1
1
5
4
0
0
0
4
.
0
1
0
7
0
a
N
o
t
e
s
:
T
E
U
s
ˆ
2
0
-
f
o
o
t
e
q
u
i
v
a
l
e
n
t
c
o
n
t
a
i
n
e
r
s
;
s
h
i
p
c
a
l
l
s
ˆ
n
u
m
b
e
r
o
f
s
h
i
p
v
i
s
i
t
s
;
s
h
i
p
r
a
t
e
ˆ
s
h
i
p
w
o
r
k
i
n
g
r
a
t
e
w
h
i
c
h
m
e
a
s
u
r
e
s
t
h
e
n
u
m
b
e
r
o
f
c
o
n
-
t
a
i
n
e
r
s
m
o
v
e
d
p
e
r
w
o
r
k
i
n
g
h
o
u
r
p
e
r
s
h
i
p
;
C
r
a
n
e
p
r
o
d
ˆ
c
r
a
n
e
p
r
o
d
u
c
t
i
v
i
t
y
w
h
i
c
h
m
e
a
s
u
r
e
s
t
h
e
n
u
m
b
e
r
o
f
c
o
n
t
a
i
n
e
r
s
m
o
v
e
d
p
e
r
c
r
a
n
e
p
e
r
w
o
r
k
i
n
g
h
o
u
r
;
D
e
l
a
y
t
i
m
e
m
e
a
s
u
r
e
s
t
h
e
a
m
o
u
n
t
o
f
d
e
l
a
y
t
i
m
e
w
h
i
c
h
i
s
t
h
e
d
i
€
e
r
e
n
c
e
b
e
t
w
e
e
n
t
o
t
a
l
b
e
r
t
h
t
i
m
e
p
l
u
s
t
i
m
e
w
a
i
t
i
n
g
t
o
b
e
r
t
h
a
n
d
t
h
e
t
i
m
e
b
e
t
w
e
e
n
s
t
a
r
t
a
n
d
®
n
i
s
h
o
f
s
h
i
p
w
o
r
k
i
n
g
;
l
a
b
o
r
ˆ
n
u
m
b
e
r
o
f
p
o
r
t
a
u
t
h
o
r
i
t
y
e
m
p
l
o
y
e
e
s
.
b
S
o
u
r
c
e
s
:
A
u
s
t
r
a
l
i
a
n
B
u
r
e
a
u
o
f
T
r
a
n
s
p
o
r
t
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
E
c
o
n
o
m
i
c
s
,
W
a
t
e
r
l
i
n
e
,
I
s
s
u
e
s
N
o
.
6
&
7
,
M
a
r
c
h
&
J
u
n
e
1
9
9
6
;
C
o
n
t
a
i
n
e
r
i
z
a
t
i
o
n
I
n
t
e
r
n
a
t
i
o
n
a
l
Y
e
a
r
b
o
o
k
(
1
9
9
8
)
;
L
l
o
y
d
Õ
s
P
o
r
t
s
o
f
t
h
e
W
o
r
l
d
(
1
9
9
8
)
.
120 J. Tongzon / Transportation Research Part A 35 (2001) 107±122
of di€erent measures of port output are available, depending on which features of port operation
are being evaluated. In addition to providing relative eciency rankings, DEA also provides
results on the sources of input and output ineciency, as well as the ports which were used for the
eciency comparison. The ability to identify the sources of ineciency could be useful to port
authority managers in inecient ports, acting as a guide to focusing e€orts at improving port
performance.
Although this study has shown the suitability of DEA for port eciency evaluation and pro-
duced useful ®ndings for certain ports, there is still more scope for future investigation. The lack
of access to stevedoring employment data for most of the sampled ports has constrained the DEA
analysis. It will be interesting to see how port ineciency can be attributed to stevedoring labor
once complete data for this particular variable are available and this variable is incorporated into
the analysis. With availability of more port data and inclusion of more ports, applying the DEA
analysis to similar ports based on a larger sample size is another interesting area for future
research. Ports can be classi®ed into various clusters in terms of size, facilities and function (i.e.,
whether hub or feeder ports), and only ports belonging to the same clusters are included in the
port eciency analysis (Tongzon and Ganesalingam, 1994).
Appendix A
See Table 4.
References
Ali, I., Lerme, C., Seiford, L., 1995. Components of eciency evaluation in data envelopment analysis. European
Journal of Operational Research 462±473.
Assad, M., 1986. Data envelopment analysis software for microcomputers: ccr1, additive, minMax, summary, window.
Research Report CCS 538, Centre for Cybernetic Studies, University of Texas.
Australian Bureau of Industry Economics, 1993. International performance indicators in the waterfront. Research
Report 47. Australian Government Publishing Service, Canberra.
Australian Bureau of Industry Economics, 1995. Waterfront 1995: International Benchmarking Report 95-16.
Australian Government Publishing Service, Canberra.
Australian Bureau of Transport and Communications Economics, 1996. Waterline (various issues).
Australian Transport Advisory Council, 1992. Port performance indicators.
Banker, R.D., Conrad, R.F., Strauss, R.P., 1986. A comparative application of DEA and translog methods: an
illustrative study of hospital production. Management Science 32 (1), 30±44.
Banker, R., Johnston, H., 1994. Evaluating the impacts of operating strategies on eciency in the US Airline Industry.
In: Data Envelopment Analysis: Theory, Methodology, and Application. Kluwer Academic Publishers, Boston.
Bowlin, W.F., Charnes, A., Cooper, W.W., Sherman H.D., 1985. Data envelopment analysis and regression approaches
to eciency estimation and evaluation. Annals of Operations Research 113±138.
Burton, M., Phimister, E., 1995. Core journals: a reappraisal of the diamond list. Economic Journal 105, 361±373.
Charnes, A., Cooper, W.W., Rhodes, E., 1978. Measuring the eciency of decision making units. European Journal of
Operational Research 429±444.
Charnes, A., Cooper, W.W., Clark T., Golany, B., 1985. A developmental study of data envelopment analysis in
measuring the eciency of maintenance units in the US Air Forces. Annals of Operations Research 95±112.
J. Tongzon / Transportation Research Part A 35 (2001) 107±122 121
Charnes, A., Galleous, A., Li, H., 1996. Robustly ecient parameter frontiers: an approximation via the multiplicative
DEA model for domestic and international operations of the Latin American airline industry. European Journal of
Operational Research 88, 525±536.
Charnes, A., Cooper, W.W., Lewin, A.Y., Seiford, L.M., 1995. Data Envelopment Analysis: Theory, Methodology and
Application. Kluwer Academic Publishers, Boston.
Chin, A., Tongzon, J., 1998. Maintaining Singapore as a major shipping and air transport hub. In: Toh, T. (Ed.),
Competitiveness of the Singapore Economy. Singapore University Press, Singapore, pp. 83±114.
Containerization International Yearbook, 1998.
Deprins, D., Simar, L., Tulkens, H., 1984. Measuring labor-eciency in post oces. In: Marchand, M., Pestieau, P.,
Tulkens, H. (Eds.), The Performance of Public Enterprises: Concepts and Measurement. Elsevier, North-Holland,
pp. 243±267.
Favero, C., Papi, L., 1995. Technical eciency and scale eciency in the Italian banking sector: a non-parametric
approach. Applied Economics 27, 385±395.
Gong, B., Sickles, R., 1992. Finite sample evidence on the performance of stochastic frontiers and data envelopment
analysis using panel data. Journal of Econometrics 5, 259±284.
Lewin, A.Y., Morey, R.C., Cook, T.J., 1982. Evaluating the administrative eciency of courts. Omega 401±411.
LloydÕs Ports of the World, 1998. LloydÕs of London Press.
Oum, T., Yu, C., 1994. Economic eciency of railways and implications for public policy. Journal of Transportation
Economics and Policy (May), 121±138.
Ray, S.C., 1991. Resource-use eciency in public schools: a study of connecticut data. Management Science 37 (12),
1620±1628.
Roll, Y., Hayuth, Y., 1993. Port performance comparison applying data envelopment analysis. Maritime Policy and
Management 20 (2), 153±161.
Tongzon, J., 1989. The impact of wharfage costs on Victoria's export-oriented industries. Economic Papers 8, 58±64.
Tongzon, J., Ganesalingam, S., 1994. An evaluation of ASEAN port performance and eciency. Asian Economic
Journal 8 (3), 317±330.
122 J. Tongzon / Transportation Research Part A 35 (2001) 107±122