You are on page 1of 8

Energy Economics 34 (2012) 1392–1399

Contents lists available at SciVerse ScienceDirect

Energy Economics
journal homepage: www.elsevier.com/locate/eneco

Decomposition of the environmental inefficiency of the meta-frontier with


undesirable output
Ching-Ren Chiu a, Je-Liang Liou b, Pei-Ing Wu c, Chen-Ling Fang d,⁎
a
Department of Business Administration, National Taiwan University of Science and Technology, No.43, Sec.4, Keelung Rd., Taipei, 106, Taiwan
b
International Division, Chung-Hua Institution for Economic Research, No.75 Chang-Hsing St., Taipei, 106, Taiwan
c
Department of Agricultural Economics, National Taiwan University, No.1 Sec. 4 Roosevelt Rd., Taipei, 106, Taiwan
d
Department of Finance and Cooperatives Management, National Taipei University, No. 151, University Rd., New Taipei City, 237, Taiwan

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

Article history: In this paper, we present an alternative analysis framework to evaluate the effects of technology heterogene-
Received 8 December 2011 ities and undesirable output on environmental efficiency measurement. The proposed framework combines
Received in revised form 18 May 2012 the directional distance function and a meta-frontier analysis. It can be used to measure efficiency improve-
Accepted 1 June 2012
ments brought about by enhanced technical management and technological advances. For demonstration
Available online 20 June 2012
purposes, we used the framework to measure the environmental efficiency in 90 countries worldwide for
JEL classification:
the 2003–2007 period. The results showed that when the meta-technology set is used as the evaluation
C6 basis, the average environmental efficiency of high competitiveness countries is greater than that of lower-
O57 middle, low, and upper-middle competitiveness countries. The upper-middle competitiveness countries per-
Q43 form worse than the lower-middle and low competitiveness countries because of the excessive labor force
Q54 usage and carbon dioxide emissions in these countries. We also found that the environmental inefficiency
of the meta-frontier for high competitiveness countries can be attributed to managerial failure in the produc-
Keywords: tion process, whereas that for upper-middle, lower-middle, and low competitiveness countries can be attributed
Environmental efficiency
to technological differences.
Meta-frontier inefficiency
Carbon dioxide emissions
© 2012 Elsevier B.V. All rights reserved.
Directional distance function

1. Introduction mostly generated from the energy conversion of fossil fuel, which is
the driving force behind economic growth. A reduction in energy use
In recent years, considerable attention has been devoted to the could reduce CO2 emissions, with no change in production technolo-
importance of controlling global warming while maintaining sustain- gies and environmental efficiency. Such a reduction, however, comes
able economic growth. Since global warming caused by greenhouse at the price of an economic slowdown (Kraft and Kraft, 1978; Yu and
gas emissions, the by-products of energy use, affect economies at Choi, 1985), which is why most countries remain noncommittal
both national and regional levels, related regulations and environ- about reducing their CO2 emissions. From a methodological perspec-
mental policies must be integrated into national as well as regional tive, some alternative approaches have been developed to measure
economic policies. To formulate effective environmental economic the relationship between environmentally harmful by-products and
policies, a significant amount of research has been done to investigate economic growth in various nations and regions. Among the ap-
the economic effects brought about by reducing the level of harmful proaches adopted in this field, environmental efficiency measurement
by-products in different nations as well as regions. could be used to investigate the trade-off between the by-products
Excessive carbon dioxide (CO2) emissions are responsible for the and economic growth based on an input–output production process.
greenhouse effect, and it is necessary to reduce these emissions to pre- It could also provide more policy insights to improve the efficiency of
vent catastrophic climate change. In 1992, many countries signed the resource utilization. As a result, environmental efficiency measurement
United Nations Framework Convention on Climate Change (UNFCCC) has been regarded as an effective tool.
in Rio de Janeiro. Of these, 38 went on to sign the Kyoto Protocol in Conceptually, environmental efficiency measurement is based on
1997, which required all signatories to reduce greenhouse gas emis- the fundamental concept of the environmental efficiency index con-
sions between 2008 and 2012 (i.e., the first stage of emission reduc- structed from the distance function. The environmental efficiency
tion) to an average of 5.2% below their 1990 levels. CO2 emissions are index defines the efficiency of a specific decision-making unit (DMU)
as being better than that of others if it produces more desirable outputs
⁎ Corresponding author. Tel.: + 886 2 26748189; fax: + 886 2 86715905. (e.g., economic growth) and fewer undesirable outputs (e.g., pollution
E-mail address: faling@totalbb.net.tw (C-L. Fang). or other environmentally harmful by-products) given the same amount

0140-9883/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.eneco.2012.06.003
C-R. Chiu et al. / Energy Economics 34 (2012) 1392–1399 1393

of inputs. In the pioneering work of Chung et al. (1997), the effect of DMUs. In order to take technology heterogeneity into account in an
undesirable output on environmental efficiency was evaluated using analysis framework, Battese and Rao (2002) proposed the meta-
the directional distance function. Because the directional distance func- frontier production function framework, which utilizes the stochastic
tion is applied on the basis of quantitative data without requiring any frontier approach (SFA) to investigate the environmental efficiency of
input/output price data and because it does not need to impose any as- firms in groups that have different technologies. The meta-frontier
sumptions of functional form on production function, it has been widely approach applied by the SFA provides an effective framework for the
used in subsequent energy–environment-related studies (Arcelus and analysis of technology heterogeneity in different groups. However, due
Arocena, 2005; Boyd et al., 2002; Färe et al., 1996, 2006; Halicioglu, to the nature of the SFA, it can only deal with single-output cases, and
2009; Kumar, 2006; Lee et al., 2002; Lozano and Gutiérrez, 2008; thus, it fails in the cases in which multiple outputs existed. In another
Picazo-Tadeo et al., 2005; Sueyoshi and Goto, 2011; Zofio and Prieto, work, O'Donnell et al. (2008) constructed the meta-frontier based on a
2001). Yörük and Zaim (2005) utilized both the Malmquist (M) and data envelopment approach basis (DEA) instead of the SFA. Since the
Malmquist–Luenberger (ML) productivity indices to analyze the pro- DEA can evaluate the cases with multiple outputs, it is preferable to
ductivity growth of 28 Organization for Economic Cooperation and the SFA for constructing the meta-frontier. However, these studies still
Development (OECD) countries from 1985 to 1998. They found that only focus on dealing with the problem of technology heterogeneity,
the M index does not suitably measure productivity in the presence of and they never apply the meta-frontier approach to measure environ-
negative externalities, and the ML index may have a linear program- mental efficiency.
ming infeasibility problem in measuring cross-period directional distance Thus far, only one study takes the impact of undesirable outputs
functions. Oh (2010a) proposed the global Malmquist–Luenberger index into consideration to measure environmentally sensitive productivity
to overcome this weakness of the ML index. This index was employed to growth in a meta-frontier analysis framework. Oh (2010b) indicated
measure the environmentally sensitive productivity growth of 26 OECD that although the ML index can deal with the effect of environmentally
countries for the period of 1990–2003. It was found that productivity harmful by-products in productivity growth analysis, it does not incor-
growth mainly stems from technical change. porate ex ante group heterogeneities. Hence, he proposed the meta-
In addition to the directional distance function, Zaim and Taskin frontier Malmquist–Luenberger productivity growth index to consider
(2000) developed a hyperbolic efficiency measure to calculate the group heterogeneity. He found that while European countries are good
CO2 emissions efficiency of OECD countries. Zhou et al. (2006) pro- at innovating and take the lead in the world frontier technology, Asian
posed a slacks-based efficiency measure to model CO2 emissions perfor- countries are good at catching up with the world frontier technology.
mance that accounts for economic inefficiency. Most of the literature on Given the aforementioned research background, this article com-
energy and environmental studies has focused on the energy efficiency bines the directional distance function and the meta-frontier frame-
and productivity of the electricity industry (Abbott, 2006; Barros, 2008; work to construct a comparative environmental efficiency analysis
Barros and Peypoch, 2008; Nag, 2006; Pombo and Taborda, 2006; framework. The proposed framework not only deals with environ-
Ramos-Real et al., 2009; Sueyoshi et al., 2010; Vaninsky, 2006). After mentally harmful outputs, but it also incorporates the technology
the issue of CO2 emissions arose, extensive research was conducted on heterogeneity effect into the evaluation of environmental efficiency.
the environmental effects of these emissions (Lozano and Gutiérrez, As a result, the trade-off between desirable and undesirable outputs
2008; Lozano et al., 2009; Sueyoshi and Goto, 2010; Zhou et al., 2010). in various nations or regions can be integrated into the process of
It also became crucial for environmental performance measurement measuring environmental efficiency. In addition, the post studies
studies to identify the factors that affect changes in environmental that applied the traditional meta-frontier framework never indentify
efficiency. Some studies have recently explored decomposition of the the sources of inefficiency. So as to provide more policy implications,
changes in undesirable outputs (Kim and Kim, in press; Pasurka, an approach is developed herein to decompose the environmental
2006; Zhou and Ang, 2008). Pasurka (2006) applied the output dis- inefficiency of the DMUs within the meta-frontier in this article. For
tance functions to decompose the changes in nitrogen oxide (NOx) illustrative purposes, the proposed framework is applied to measure
and sulfur dioxide (SO2) emissions from coal-fired power plants. the environmental efficiency of 90 countries in worldwide during
Zhou and Ang (2008) proposed production-theoretical decomposition the years 2003–2007. With the desirable output of GDP taken into
analysis (PDA) to assess the effects of production technologies on CO2 account along with the undesirable output of CO2 emissions, the envi-
emissions and adopted two input distance functions to decompose ronmental efficiency measurement results can be used to present an
the change in aggregate CO2 emissions over time. Kim and Kim (in integrated view of the conflict between environmental protection
press) also used PDA to estimate the energy mix effect and industrial and economic development. This framework can be used to identify
structure effect on the production technology of CO2 emissions. What how advances in technology as well as enhanced technical manage-
these three studies have in common is that they combine the multipli- ment improve environmental efficiency.
cative identity, radial index, and Malmquist productivity indexes. Thus, The remainder of this paper is organized as follows. Section 2
the decomposition results are straightforward and can provide useful introduces the methodological models of the paper, namely, the
policy implications by measuring the effects of all the factors. Dissimilar meta-frontier and group frontiers of the directional distance function,
to the above studies, the decomposition of environmental efficiency in and the source of the meta-frontier inefficiency. The analysis results
this study inherits the structure of meta-frontier analysis framework; are presented in Section 3. Section 4 summarizes the conclusions.
the driving forces of environmental efficiency can be further decomposed
into management improvement and advance in technology. 2. Methodology
Although the distance function and directional distance function
are widely used, evaluations of environmental efficiency using these 2.1. The meta-frontier and group frontiers of the directional
functions do not consider the possible technology heterogeneity among distance function
different groups. Prior evaluations usually assume that the DMUs of
different groups are equipped with similar levels of technology. How- When using the conventional directional distance function to as-
ever, given the different institutional environments and production sess efficiency, it is usually assumed that all producers possess the
frontiers of different countries and regions, this assumption is not a same level of production technology. However, the assessed DMUs
realistic one and could lead to biased evaluation results under an inap- usually have different production technologies due to differences in
propriate efficiency frontier. In other words, without taking technology geographical location, national policy, and socioeconomic conditions.
heterogeneity into account, it would be impossible to determine Therefore, O'Donnell et al. (2008) applied the meta-frontier concept
whether this heterogeneity among groups is what causes inefficient in estimating DEA efficiency, gauged a meta-frontier through the use
1394 C-R. Chiu et al. / Energy Economics 34 (2012) 1392–1399

of overall samples, divided the DMUs, and estimated the group frontiers set, respectively and N1 + N2 + ⋯ +NK = N. Consequently, the environ-
of group samples. If we assume that there are N peer countries, each mental efficiency of DMUo measured on the basis of the meta-frontier
with M inputs, S desirable outputs, and F undesirable outputs, then let and the group frontiers is defined as MEE =1 −βm and GEE= 1− βk,
M
xij, yrj, and bfjdenote the values of the ith inputs (i ∈ R+ ), the rth desir- and the environmental efficiency could be between zero and unity.
S F
able outputs (r ∈ R+), and the fth undesirable outputs (f ∈ R+ ) of
N
DMUj (j ∈ R+ ). We assume that the production possibility set (PPS) of 2.2. Meta-technology ratios and the source of the
inputs and outputs is weak disposal. Färe et al. (2005) and Picazo- meta-frontier inefficiency
Tadeo et al. (2005) proposed the directional technology function,
which allows for desirable outputs to be proportionately increased, Because the meta-frontier is enveloped in the K group-specific
while undesirable outputs are proportionately decreased at the same frontiers, the environmental efficiency measured on the basis of the
time. The directional meta-distance function is defined as follows: meta-frontier is less than that of the group frontiers. It is given as
follows:
→   n   o
Dm x; y; b; gy ; gb ¼ max β m : x; y þ β m gy ; b−β m g b ∈T m ðx; y; bÞ
MEE≤GEE:
where the nonzero direction vector g = (gy, gb) determines
The ratio between the environmental efficiency of the meta-
which desirable outputs, and undesirable outputs are scaled for
S F frontier and that of the group frontiers is referred to as the meta-
DMU and g ∈ R+ × R+ . The meta-technology set is Tm = {(x, y, b) :
technology ratio (MTR), which can be described as follows:
x can produce y and b}. We assume that there are K technology set (sub-
group) differences among technological competitiveness indicators, MEE
0bMTR ¼ ≤1: ð3Þ
whereas the K group directional distance function is defined as follows: GEE
→k   n
k

k k

k
o
D x; y; b; g y ; g b ¼ max β : x; y þ β g y ; b−β gb ∈T ðx; y; bÞ ; The closer the MTR is to 1, the smaller is the technology heteroge-
neity, which means that the environmental efficiency of that group's
k ¼ 1; 2; …; K frontiers is closer to the environmental efficiency of the meta-
frontier. Conversely, the smaller the MTR, the greater the technology
heterogeneity, and therefore, the environmental efficiency of the
The group technology set is Tk ={(x,y,b):x can be used by DMUs
meta-frontier will be considerably lower than that of the group
in group k to produce y and b} and the meta-technology set consists of
frontiers.
the K group technology set Tm ={T1 ∪T2 ∪⋯∪Tk}.
The MTR evaluated based on the meta-frontier and group-specific
The direction vector g = (gy, gb) should be chosen by the researcher
frontiers indicates that DMUo in group-specific frontiers are only dis-
(Färe et al., 2008) before evaluating the directional distance function.
tinguished by technology heterogeneity between the two frontiers, so
In the paper, we let the direction vector g = (gy = y, gb = b) (Oh,
the source of the meta-frontier inefficiency cannot be identified.
2010b). Therefore, the inefficiency measure of the DMUo of the meta-
Therefore, the inefficiency of the DMUo under the meta-frontier is
technology set and the group technology set under the assumption of
decomposed further into technology gap inefficiency (TGI) and mana-
variable returns to scale (Zhou et al., 2008) can be represented by the
gerial inefficiency (GMI) of group-specific frontiers. The TGI represents
following two linear programs:
the inefficiency of the DMUo in group-specific frontiers originating from
→   the technical gap between the meta-frontier and the group-specific
Max D m xkio ; ykro ; bkfo ¼ βm frontiers. The cause of inefficiency is attributed to technological differ-
K X
X Nk ences, which are formulated as follows:
k k k
s:t: λn xin ≤xio ; i ¼ 1; …; M;
k¼1 n¼1 TGI ¼ GEE  ð1−MTRÞ: ð4Þ
XK X Nk
k k  m k
λn yrn ≥ 1 þ β yro ; r ¼ 1; …; S; The GMI represents the inefficiency of the DMUo in group-specific
k¼1 n¼1 ð1Þ
XK X Nk frontiers originating from input and undesirable output excess and
k k  m k
λn bfn ¼ 1−β bfo ; f ¼ 1; …; F; desirable output shortfall. The reason for the inefficiency is attributed
k¼1 n¼1 to managerial failure of DMUo, which is formulated as follows:
XK X Nk
k
λn ¼ 1; k ¼ 1; …; K; k

k¼1 n¼1 GMI ¼ 1−GEE ¼ β0 : ð5Þ


k
λn ≥0; n ¼ 1; …; Nk ;
Thus, the environmental inefficiency measured on the basis of the
→k  k  meta-frontier can also be expressed as follows:
Max D xio ; ykro ; bkfo ¼β k

X
Nk MTI ¼ TGI þ GMI: ð6Þ
k k k
s:t: μ n xin ≤xio ; i ¼ 1; …; M;
n¼1 Consider three groups as displayed in Fig. 1 with a single desirable
X
Nk  
k k k k output, and a single undesirable output. Group 1 consists of DMU M
μ n yrn ≥ 1 þ β yro ; r ¼ 1; …; S;
n¼1 ð2Þ and M′, Group 2 consists of DMU N and N′, and Group 3 consists of
X
Nk   DMU A, P, and P′. The group frontiers of Groups 1, 2, and 3 are labeled
k k k k
μ n bfn ¼ 1−β bfo ; f ¼ 1; …; F; M–M′, N–N′, and P–P′. The meta-frontier encompasses all group fron-
n¼1
tiers, implying that the meta-frontier is a nonconvex piecewise fron-
XNk
k
μn ¼ 1; tier, labeled M–N′. For example, consider DMU A from Group 3 in
n¼1 Fig. 1. The measures of environmental efficiency in meta-frontiers
μ kn ≥0; n ¼ 1; …; Nk ; and group frontiers are MEEðAÞ ¼ OF=OD and GEEðAÞ ¼ OE=OD, and
the MTR can be described as MTRg ¼ OF=OE. The technology gap in-
where λnk and μnk represent the intensity variables corresponding to the efficiency and managerial inefficiency of the DMU A are TGIðAÞ ¼
production process of the meta-technology set and the group technology GEEðAÞ  ð1−MTRðAÞÞ ¼ FE=OD and GMI ðAÞ ¼ 1−GEEðAÞ ¼ ED=OD.
C-R. Chiu et al. / Energy Economics 34 (2012) 1392–1399 1395

Desirable output, y Table 1


Data set covering 90 countries.

No. Countries Group No. Countries Group

1 Japan HH 46 Oman LM
N’
2 Australia HH 47 Russian Federation LM
3 Korea, Republic HH 48 Jordan LM
M’ 4 United Arab Emirates HH 49 Greece LM
5 Belgium HH 50 Latvia LM
N P’
C 6 United Kingdom HH 51 Colombia LM
7 Sweden HH 52 Argentina LM
g = ( gy , gb ) B 8 Ireland HH 53 Panama LM
9 Iceland HH 54 Mexico LM
A
M P 10 France HH 55 Dominican Republic LM
11 Germany HH 56 El Salvador LM
12 Norway HH 57 Uruguay LM
O F E D Undesirable output, b
13 Switzerland HH 58 Peru LM
14 Austria HH 59 Venezuela, RB LM
Fig. 1. Decomposition of the environmental inefficiency of the meta-frontier. 15 Finland HH 60 Botswana LM
16 Denmark HH 61 Gabon LM
17 Netherlands HH 62 South Africa LM
Therefore, the environmental inefficiency measured on the basis of the 18 United States HH 63 Bangladesh LL
meta-frontier can also be described as MTIðAÞ ¼ TGIðAÞþ GMIðAÞ ¼ 19 Canada HH 64 Philippines LL
FD=OD. 20 Malaysia HM 65 Iran, Islamic Republic LL
21 New Zealand HM 66 Uzbekistan LL
22 China HM 67 Pakistan LL
3. Empirical analysis
23 Bahrain HM 68 Kazakhstan LL
24 Saudi Arabia HM 69 Syrian Arab Republic LL
3.1. Data resources and variable choice 25 Poland HM 70 Belarus LL
26 Slovenia HM 71 Ukraine LL
27 Italy HM 72 Haiti LL
This study evaluated the relative environmental efficiency of
28 Croatia HM 73 Honduras LL
several countries, including those in UNFCCC Annex I and Annex B of 29 Czech Republic HM 74 Ecuador LL
the Kyoto Protocol. The research period is from 2003 to 2007. The data 30 Hungary HM 75 Paraguay LL
were obtained from the World Development Indicators of the World 31 Portugal HM 76 Nicaragua LL
Bank, the International Labour Organization, and the Climate Analysis 32 Spain HM 77 Guatemala LL
33 Slovak Republic HM 78 Algeria LL
Indicators Tool of the World Resource Institute (WRI). The research
34 Estonia HM 79 Benin LL
sample included 90 countries (see Table 1). In addition, the cumulative 35 Chile HM 80 Cote d'Ivoire LL
CO2 emissions of these countries in 2007 accounts for 98.7% of the 36 Brazil HM 81 Egypt, Arab Republic LL
total CO2 emissions (WRI, 2011). Therefore, the CO2 emissions reduc- 37 Costa Rica HM 82 Ethiopia LL
38 Tunisia HM 83 Ghana LL
tions promised by these countries should have generated significant
39 India LM 84 Kenya LL
results. In terms of the choice of variables, we set the inputs of environ- 40 Vietnam LM 85 Morocco LL
mental efficiency to be labor force, real capital formation, and energy 41 Thailand LM 86 Mozambique LL
consumption and the outputs to be the real GDP and CO2 emissions 42 Indonesia LM 87 Nigeria LL
(Lozano and Gutiérrez, 2008; Zhou et al., 2010). The variables of real 43 Brunei Darussalam LM 88 Senegal LL
44 Turkey LM 89 Sudan LL
capital formation and real GDP are deflated by the consumer price
45 Lebanon LM 90 Zambia LL
index in the year 2000 to eliminate the price effect.
HH = high technological competitiveness and income countries.
HM = upper-middle technological competitiveness and income countries.
3.2. The choice of the clustering indicator LM = lower-middle technological competitiveness and income countries.
LL = low technological competitiveness and income countries.
In practice, countries around the world should face different pro-
duction technologies that force them to choose between different
sets of feasible input–output combinations depending on their produc- Table 2 lists the descriptive statistics of the input and output vari-
tion environments, geographical locations, and resource endowments ables of the research sample from 2003 through 2007. It shows that
(Huang et al., 2010). Therefore, some articles employ geographical loca- (1) the mean real capital formation (x2), energy consumption (x3),
tion to classify countries into different groups (Battese et al., 2004; real GDP (y1), and CO2 emissions (y2) of the HH countries are higher
O'Donnell et al., 2008; Oh and Lee, 2010). However, this indicator is than those of the HM, LM, and LL countries in all 5 years; (2) the HM
not directly linked to the technology level of a country. Learning and countries invest excessively in labor force; and (3) every year, the
absorbing new technological knowledge, spending on research and de- mean of the input and output variables increases for the total re-
velopment, and expanding the technological knowledge domestically search sample.
are major factors that affect the technology development potential of a
country (World Economic Forum, 2010). Accordingly, we choose the 3.3. Results and analysis
technological competitiveness indicator, created by the technological
readiness and innovation metrics from the newest edition of World The meta-technology ratio and environmental efficiency of the
Economic Forum (2010), and the average annual per capita income meta-frontier and group frontiers are estimated for the 90 countries
(Iyer et al., 2006) to categorize technology level of group-frontiers. in each of the 5 years (O'Donnell et al., 2008). LINGO 9.0 is used to
Thereafter, we use two-stage cluster analysis to create four groups of solve the linear programming problem of the directional distance
countries (see Table 1): high technological competitiveness and income function. The empirical results of the efficiencies, MTR, and inefficien-
countries (HH, 19 countries), upper-middle technological competitive- cies of the meta-frontier and group frontiers from 2003 to 2007 are
ness and income countries (HM, 19 countries), lower-middle technolog- presented in Table 3 (the efficiencies are the average score of five ob-
ical competitiveness and income countries (LM, 24 countries), and low served time periods). The second and ninth columns in Table 3 present
technological competitiveness and income countries (LL, 28 countries). the environmental efficiencies of the countries in the meta-technology
1396 C-R. Chiu et al. / Energy Economics 34 (2012) 1392–1399

Table 2
The descriptive statistics of the input and output variables during 2003 to 2007.

Years Groups x1 x2 (unit: million NT$) x3 (unit: kWh) y1 (unit: million NT$) y2 (unit: tons)

2003 HH Mean 21,280 251,197 240,387 1,249,358 601,095


Std. Dev. 34,752 460,021 495,406 2,381,622 1,280,732
HM Mean 50,109 71,278 121,924 262,793 353,621
Std. Dev. 165,752 142,733 293,026 420,451 936,416
LM Mean 34,186 26,705 83,434 127,525 203,775
Std. Dev. 81,845 37,168 153,482 161,529 369,298
LL Mean 15,435 6964 30,695 30,690 64,568
Std. Dev. 15,948 9078 38,130 32,937 95,502
Total Mean 28,989 77,366 108,287 362,787 275,979
Std. Dev. 89,953 240,798 287,114 1,207,966 780,696
2004 HH Mean 21,384 265,203 244,929 1,288,472 608,821
Std. Dev. 34,909 491,628 505,794 2,464,434 1,302,824
HM Mean 50,682 78,018 135,032 277,671 398,700
Std. Dev. 167,109 160,149 341,125 450,321 1,108,467
LM Mean 34,821 30,656 86,713 136,322 211,000
Std. Dev. 83,233 42,443 157,119 171,930 375,075
LL Mean 15,859 7482 31,852 32,652 66,586
Std. Dev. 16,407 9550 39,481 34,769 98,268
Total Mean 29,433 82,960 113,247 377,141 289,681
Std. Dev. 90,873 257,812 302,741 1,250,219 836,294
2005 HH Mean 21,625 276,523 245,713 1,320,387 609,995
Std. Dev. 35,240 512,795 508,324 2,532,369 1,307,315
HM Mean 51,242 82,730 143,754 292,353 432,095
Std. Dev. 168,475 173,344 369,923 482,941 1,232,594
LM Mean 35,473 34,581 89,290 145,224 217,113
Std. Dev. 84,895 48,461 161,307 183,132 383,645
LL Mean 16,309 7986 32,736 34,387 69,129
Std. Dev. 16,902 10,006 40,703 36,510 101,392
Total Mean 29,916 87,549 116,216 389,892 299,400
Std. Dev. 91,877 269,770 311,375 1,285,497 875,429
2006 HH Mean 21,855 285,760 244,815 1,357,214 606,353
Std. Dev. 35,666 521,108 503,714 2,599,176 1,288,297
HM Mean 51,771 90,546 152,954 310,755 467,884
Std. Dev. 169,783 191,888 404,182 525,586 1,372,408
LM Mean 36,148 38,959 92,525 155,449 227,458
Std. Dev. 86,628 54,631 167,246 196,624 404,131
LL Mean 16,739 8689 33,783 36,510 72,846
Std. Dev. 17,382 10,761 41,916 38,668 109,079
Total Mean 30,391 92,535 119,157 404,939 310,102
Std. Dev. 92,887 277,071 319,515 1,321,908 915,633
2007 HH Mean 22,071 288,189 246,142 1,390,625 612,942
Std. Dev. 35,934 505,096 511,372 2,652,285 1,306,343
HM Mean 52,147 99,236 160,057 332,797 496,358
Std. Dev. 170,797 212,511 428,514 579,477 1,479,526
LM Mean 36,828 43,747 96,039 165,638 237,121
Std. Dev. 88,376 61,784 171,820 209,382 416,000
LL Mean 17,124 9876 35,269 38,859 75,911
Std. Dev. 17,768 12,035 43,903 41,314 114,033
Total Mean 30,817 96,528 122,336 420,093 321,034
Std. Dev. 93,767 274,393 329,427 1,353,197 957,931

x1: labor force; x2: real capital formation; x3: energy consumption; y1: real GDP; y2: CO2 emissions.

set. The third and tenth columns present the environmental efficiencies are relatively efficient in the context of the group technology set,
of the countries in the group technology set, divided into HH, HM, LM, technology heterogeneity exists in these countries between the
and LL. The fourth and eleventh columns present the MTR of countries meta-frontier and group frontiers. In terms of the TGI, we find that
based on evaluations made using Eq. (3). Finally, the TGI, GMI, and country 68 (Kazakhstan) has the largest TGI (0.883), which indicates
MTI reported in the fifth to seventh and twelfth to fourteenth columns large technology heterogeneity between the meta-frontier and group
are based on evaluations made using Eqs. (4)–(6). frontier.
Of the 90 countries, 22—countries 1, 6, 7, 8, 9, 12, 13, 18, 23, 36, 43, The box plots provide more insight into the distribution of envi-
47, 57, 60, 61, 62, 66, 72, 79, 80, 87, and 90—are relatively environ- ronmental efficiency among the HH, HM, LM, and LL countries when
mentally efficient (MEE and GEE are equal to 1) when the meta- the meta-technology set is used as the evaluation basis. Fig. 2 shows
technology set and group technology set are regarded as the basis that HH countries yield the best average and variance of environmen-
of evaluation. Of these 22 efficient countries, 8 are in HH and 2 are tal efficiency and HM countries perform worse than the other three
in HM. Among the remaining 12 countries, 6 are in LM and 6 are in groups. Australia and Korea have the worst environmental perfor-
LL. This suggests that HH countries allocate resources efficiently and mance among the HH countries. The distributions of the environmen-
that LM and LL countries do not use resources more excessively tal efficiency of LM and LL countries are skewed to the left.
than do HM countries. Additionally, when using the group technology The source of the environmental inefficiency of the meta-frontier
set as the basis of evaluation, 26 countries have an environmental can be decomposed into the TGI and GMI. The GMI must be zero
efficiency of 1 (GEE is equal to 1), which indicates that while they when the GEE is unity. The MTI originates primarily from the TGI,
C-R. Chiu et al. / Energy Economics 34 (2012) 1392–1399 1397

Table 3
The efficiencies, MTR, and inefficiencies of meta-frontier and group frontiers.

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

No. MEE GEE MTR TGI GMI MTI No. MEE GEE MTR TGI GMI MTI

1 1.000 1.000 1.000 0.000 0.000 0.000 46 0.567 1.000 0.567 0.433 0.000 0.433
2 0.564 0.564 1.000 0.000 0.436 0.436 47 1.000 1.000 1.000 0.000 0.000 0.000
3 0.535 0.535 1.000 0.000 0.465 0.465 48 0.279 0.309 0.904 0.030 0.691 0.721
4 0.608 0.734 0.828 0.126 0.266 0.392 49 0.653 1.000 0.653 0.347 0.000 0.347
5 0.838 0.850 0.985 0.013 0.150 0.162 50 0.501 0.667 0.752 0.166 0.333 0.499
6 1.000 1.000 1.000 0.000 0.000 0.000 51 0.699 0.925 0.756 0.226 0.075 0.301
7 1.000 1.000 1.000 0.000 0.000 0.000 52 0.947 1.000 0.947 0.053 0.000 0.053
8 1.000 1.000 1.000 0.000 0.000 0.000 53 0.795 0.819 0.971 0.024 0.181 0.205
9 1.000 1.000 1.000 0.000 0.000 0.000 54 0.783 1.000 0.783 0.217 0.000 0.217
10 0.999 0.999 1.000 0.000 0.001 0.001 55 0.786 0.841 0.935 0.055 0.159 0.214
11 0.989 0.991 0.998 0.002 0.009 0.011 56 0.795 0.874 0.910 0.079 0.126 0.205
12 1.000 1.000 1.000 0.000 0.000 0.000 57 1.000 1.000 1.000 0.000 0.000 0.000
13 1.000 1.000 1.000 0.000 0.000 0.000 58 0.756 0.959 0.789 0.202 0.041 0.244
14 0.825 0.831 0.994 0.005 0.169 0.175 59 0.535 0.686 0.780 0.151 0.314 0.465
15 0.871 0.906 0.961 0.035 0.094 0.129 60 1.000 1.000 1.000 0.000 0.000 0.000
16 0.985 0.986 0.999 0.001 0.014 0.015 61 1.000 1.000 1.000 0.000 0.000 0.000
17 0.964 0.967 0.997 0.003 0.033 0.036 62 1.000 1.000 1.000 0.000 0.000 0.000
18 1.000 1.000 1.000 0.000 0.000 0.000 63 0.527 1.000 0.527 0.473 0.000 0.473
19 0.769 0.778 0.988 0.009 0.222 0.231 64 0.880 1.000 0.880 0.120 0.000 0.120
20 0.635 0.858 0.740 0.223 0.142 0.365 65 0.120 1.000 0.120 0.880 0.000 0.880
21 0.551 0.871 0.632 0.321 0.129 0.449 66 1.000 1.000 1.000 0.000 0.000 0.000
22 0.180 1.000 0.180 0.820 0.000 0.820 67 0.716 0.949 0.754 0.233 0.051 0.284
23 1.000 1.000 1.000 0.000 0.000 0.000 68 0.117 1.000 0.117 0.883 0.000 0.883
24 0.819 1.000 0.819 0.181 0.000 0.181 69 0.537 0.908 0.591 0.371 0.092 0.463
25 0.718 1.000 0.718 0.282 0.000 0.282 70 0.124 0.842 0.148 0.717 0.158 0.876
26 0.507 1.000 0.507 0.493 0.000 0.493 71 0.672 1.000 0.672 0.328 0.000 0.328
27 0.866 1.000 0.866 0.134 0.000 0.134 72 1.000 1.000 1.000 0.000 0.000 0.000
28 0.431 0.736 0.586 0.305 0.264 0.569 73 0.523 1.000 0.523 0.477 0.000 0.477
29 0.253 0.468 0.541 0.215 0.532 0.747 74 0.296 0.738 0.401 0.442 0.262 0.704
30 0.581 0.816 0.712 0.235 0.184 0.419 75 0.915 1.000 0.915 0.085 0.000 0.085
31 0.656 0.869 0.755 0.213 0.131 0.344 76 0.781 1.000 0.781 0.219 0.000 0.219
32 0.636 0.855 0.744 0.219 0.145 0.364 77 0.670 1.000 0.670 0.330 0.000 0.330
33 0.364 0.661 0.551 0.297 0.339 0.636 78 0.263 1.000 0.263 0.737 0.000 0.737
34 0.227 1.000 0.227 0.773 0.000 0.773 79 1.000 1.000 1.000 0.000 0.000 0.000
35 0.532 0.768 0.693 0.235 0.232 0.468 80 1.000 1.000 1.000 0.000 0.000 0.000
36 1.000 1.000 1.000 0.000 0.000 0.000 81 0.873 1.000 0.873 0.127 0.000 0.127
37 0.846 1.000 0.846 0.154 0.000 0.154 82 0.661 0.922 0.717 0.261 0.078 0.339
38 0.445 0.861 0.516 0.417 0.139 0.555 83 0.327 0.483 0.676 0.157 0.517 0.673
39 0.292 1.000 0.292 0.708 0.000 0.708 84 0.584 0.887 0.658 0.303 0.113 0.416
40 0.176 0.348 0.505 0.172 0.652 0.824 85 0.381 1.000 0.381 0.619 0.000 0.619
41 0.397 0.590 0.674 0.192 0.410 0.603 86 0.998 1.000 0.998 0.002 0.000 0.002
42 0.465 0.695 0.669 0.230 0.305 0.535 87 1.000 1.000 1.000 0.000 0.000 0.000
43 1.000 1.000 1.000 0.000 0.000 0.000 88 0.541 1.000 0.541 0.459 0.000 0.459
44 0.716 0.908 0.789 0.192 0.092 0.284 89 0.572 0.912 0.627 0.340 0.088 0.428
45 0.532 0.796 0.668 0.264 0.204 0.468 90 1.000 1.000 1.000 0.000 0.000 0.000

1.0 for example, country 22 (China). In other words, the GEE must be
equal to the MEE when the TGI is zero. This suggests that the ineffi-
ciency may be derived from input and undesirable output excess or
0.8 desirable output shortfall. For example, the GEE of country 2 (Australia)
must be equal to the MEE because the TGI is zero.
Because the distribution of the efficiency score is unknown in the
0.6 DEA model, the differences between the efficiencies of the meta-
MEE

2 frontier and group frontiers for various groups must be examined


3
using non-parametric statistics (Cooper et al., 2000). We applied the
0.4 Kruskal Wallis test to examine the technology frontier differences
between the HH, HM, LM, and LL countries. The result shows that
the Kruskal Wallis test value is 13.22 and represents the four groups
0.2 that have distinct technology frontiers in the environmental efficiency
of the meta-frontier. This indicates that the four samples are formed
by different populations and that technology heterogeneity exists be-
tween them.
HH HM LM LL
In Fig. 3, the average environmental efficiency of the HH countries
Group (0.892) is greater than those of the LM, LL, and HM countries (0.695,
0.646, and 0.592, respectively) when the meta-technology set is used as
Fig. 2. The box plots for the environmental efficiency of the meta-frontier in different
groups. Notes: MEE present the environmental efficiency of the meta-frontier and
the evaluation basis. This means that in terms of measurement using
the median for each group is indicated by the black center line. O2 (Australia) and the meta-frontier, the HH countries have the best relative environmental
O3 (Korea) are outliers of HH countries. performance. The MTI originates from the input and undesirable output
1398 C-R. Chiu et al. / Energy Economics 34 (2012) 1392–1399

100%
80%
60%
40%
20%
0%
MEE GEE MTR TGI GMI MTI
LL 0.646 0.952 0.673 0.306 0.048 0.354
LM 0.695 0.851 0.806 0.156 0.149 0.305
HM 0.592 0.882 0.665 0.290 0.118 0.408
HH 0.892 0.902 0.987 0.010 0.098 0.108

Fig. 3. The stack straight bar chart for the average efficiencies and inefficiency of the different groups.

excess and output shortfall of the production process for the HH countries HM countries is the worst. Australia and Korea have the worst envi-
because the TGI is lower than the GMI. However, the MTI stems from the ronmental performance among the HH countries. The distributions
technical gap between the meta-frontier and group-specific frontiers of of the environmental efficiency of LM and LL countries are skewed
the production process for the HM, LM, and LL countries when the TGI to the left. Second, the technology frontiers of the HH, HM, LM, and
is higher than the GMI. In terms of the TGI, HH countries have the smallest LL countries differ significantly according to the Kruskal Wallis test.
technical gap (the closer the TGI is to 0) between the meta-frontier and The test results demonstrate dissimilar levels of technology among
group frontiers, followed by the LM and HM countries, and LL countries the four groups. Third, the environmental performance ranks are in
have the largest technology heterogeneity. the order of HH, LM, LL, and HM countries when the meta-technology
set is used as the evaluation basis. In addition, higher technological
4. Conclusions competitiveness and average per capita annual income are accompa-
nied by higher CO2 emissions (see the descriptive statistics in Table 2).
Most environmental studies have found a significant positive rela- Consequently, HM countries perform worse than LM and LL countries
tionship between CO2 emissions and economic activity (Kim et al., because of their excessive labor force usage and CO2 emissions. Fourth,
2010). The CO2 produced as a by-product of economic development the MTI of the HH countries originated from managerial failure, whereas
is regarded as undesirable output, i.e., a negative externality produced the MTI of the HM, LM, and LL countries stemmed from technological
as a by-product in the production process. The directional distance differences. From the TGI aspect, countries with the highest technolog-
function considers the impact of undesirable outputs on environmental ical competitiveness and income have the lowest technology hetero-
efficiency, but it does not take into account the possible technology geneity and countries with the lowest technological competitiveness
heterogeneity among different DMUs because it is assumed that the and income have the highest technology heterogeneity.
DMUs being evaluated have similar levels of technology. However,
this assumption is unrealistic and would lead to an inappropriate effi-
ciency frontier due to the nature of differences in socioeconomic condi- Acknowledgments
tions and geographical locations. Although the MTR, proposed by
O'Donnell et al. (2008), can estimate the technology heterogeneity be- The authors are very grateful to two anonymous referees for de-
tween the meta-frontier and group-specific frontiers, the source of the tailed comments and helpful suggestions that greatly enhanced the
environmental inefficiency measured on the basis of the meta-frontier quality of the paper.
cannot be identified. Therefore, this paper applies the MTR concept to
decompose the meta-frontier inefficiency into technology gap ineffi-
References
ciency (TGI) and managerial inefficiency of group-specific frontiers
(GMI). Abbott, M., 2006. The productivity and efficiency of the Australian electricity supply
Moreover, classifying DMUs by using an appropriate indicator is industry. Energy Econ. 28, 444–454.
Arcelus, F., Arocena, P., 2005. Productivity differences across OECD countries in the
one of the major issues in the meta-frontier and group-frontier presence of environmental constraints. J. Oper. Res. Soc. 56, 1352–1362.
framework. Previous research using geographical locations to catego- Barros, C.P., 2008. Efficiency analysis of hydroelectric generating plants: a case study
rize DMUs assumes that countries with similar geographical locations for Portugal. Energy Econ. 30, 59–75.
Barros, C.P., Peypoch, N., 2008. Technical efficiency of thermoelectric power plants.
have similar technology levels. However, technology levels may not Energy Econ. 30, 3118–3127.
correspond to geographical locations as closely as expected. As such, Battese, G.E., Rao, D.S.P., 2002. Technology gap, efficiency, and a stochastic metafrontier
the use of geographical locations to classify DMUs might not be rele- function. Int. J. Bus. Econ. 1, 87–93.
Battese, G.E., Rao, D.S.P., O'Donnell, C.J., 2004. A metafrontier production function for
vant to the output generated at a certain technology level. It is neces- estimation of technical efficiencies and technology gaps for firms operating
sary to choose indicators that are directly related to technology. We under different technologies. J. Product. Anal. 21, 91–103.
therefore divide 90 countries into 4 groups using both the technolog- Boyd, G.A., Tolley, G., Pang, J., 2002. Plant level productivity, efficiency, and environ-
mental performance of the container glass industry. Environ. Resour. Econ. 23,
ical competitiveness indicator and average per capita annual income. 29–43.
All the sample countries are classified into HH, HM, LM, and LL Chung, Y.H., Färe, R., Grosskopf, S., 1997. Productivity and undesirable outputs: a direc-
countries. tional distance function approach. J. Environ. Manage. 51, 229–240.
Cooper, W.W., Seiford, L.M., Tone, K., 2000. Data envelopment analysis: a comprehen-
We chose the labor force, real capital formation, and energy con-
sive text with models, applications, references and DEA-solver software. Kluwer,
sumption as the inputs and the real GDP and CO2 emissions as the Boston.
desirable and undesirable outputs to measure the environmental effi- Färe, R., Grosskopf, S., Tyteca, D., 1996. An activity analysis model of the environmental
ciency of various nations. The empirical results are as follows. First, performance of firms—application to fossil-fuel-fired electric utilities. Ecol. Econ.
18, 161–175.
HH countries outperform the other three groups in terms of the aver- Färe, R., Grosskopf, S., Noh, D.W., Weber, W., 2005. Characteristics of a polluting tech-
age and variance of environmental efficiency, and the performance of nology: theory and practice. J. Econ. 126, 469–492.
C-R. Chiu et al. / Energy Economics 34 (2012) 1392–1399 1399

Färe, R., Grosskopf, S., Weber, W.L., 2006. Shadow prices and pollution costs in US Pasurka, C.A., 2006. Decomposing electric power plant emissions within a joint produc-
agriculture. Ecol. Econ. 56, 89–103. tion framework. Energy Econ. 28, 26–43.
Färe, R., Grosskopf, S., Margaritis, D., 2008. Efficiency and productivity: Malmquist and Picazo-Tadeo, A.J., Reig-Martinez, E., Hernandez-Sancho, F., 2005. Directional distance
more. In: Fried, H.O., Lovell, C.A.K., Schmidt, S.S. (Eds.), The Measurement of Pro- functions and environmental regulation. Resour. Energy Econ. 27, 131–142.
ductive Efficiency and Productivity Growth. Oxford University Press, Oxford. Pombo, C., Taborda, R., 2006. Performance and efficiency in Colombia's power distribu-
Halicioglu, F., 2009. An econometric study of CO2 emissions, energy consumption, in- tion system: effects of the 1994 reform. Energy Econ. 28, 339–369.
come and foreign trade in Turkey. Energy Policy 37, 1156–1164. Ramos-Real, F.J., Tovar, B., Iootty, M., De Almeida, E.F., Pinto Jr., H.Q., 2009. The evolu-
Huang, Y.J., Chen, K.H., Yang, C.H., 2010. Cost efficiency and optimal scale of electricity tion and main determinants of productivity in Brazilian electricity distribution
distribution firms in Taiwan: an application of metafrontier analysis. Energy Econ. 1998–2005: an empirical analysis. Energy Econ. 31, 298–305.
32, 15–23. Sueyoshi, T., Goto, M., 2010. Should the US clean air act include CO2 emission control?:
Iyer, K., Rambaldi, A., Tang, K.K., 2006. Globalisation and the Technology Gap: Regional Examination by data envelopment analysis. Energy Policy 38, 5902–5911.
and Time Evidence, Leading Economic and Managerial Issues Involving Globalisa- Sueyoshi, T., Goto, M., 2011. Efficiency-based rank assessment for electric power indus-
tion. Nova Science New York, United States, pp. 213–227. try: a combined use of Data Envelopment Analysis (DEA) and DEA-Discriminant
Kim, K., Kim, Y., in press. International comparison of industrial CO2 emission trends and Analysis (DA). Energy Econ. 34, 634–644.
the energy efficiency paradox utilizing production-based decomposition. Energy Econ. Sueyoshi, T., Goto, M., Ueno, T., 2010. Performance analysis of US coal-fired power
Kim, S., Lee, K., Nam, K., 2010. The relationship between CO2 emissions and economic plants by measuring three DEA efficiencies. Energy Policy 38, 1675–1688.
growth: the case of Korea with nonlinear evidence. Energy Policy 38, 5938–5946. Vaninsky, A., 2006. Efficiency of electric power generation in the United States: analysis
Kraft, J., Kraft, A., 1978. On the relationship between energy and GNP. J. Energy Dev. 3, and forecast based on data envelopment analysis. Energy Econ. 28, 326–338.
401–403. World Economic Forum, 2010. The Global Competitiveness Report 2010–2011. World
Kumar, S., 2006. Environmentally sensitive productivity growth: a global analysis using Economic Forum, Geneva, Switzerland.
Malmquist–Luenberger index. Ecol. Econ. 56, 280–293. World Resource Institute, 2011. The Climate Analysis Indicator Tool Database. http://cait.
Lee, J.D., Park, J.B., Kim, T.Y., 2002. Estimation of the shadow prices of pollutants with wri.org/.
production/environment inefficiency taken into account: a nonparametric direc- Yörük, B.K., Zaim, O., 2005. Productivity growth in OECD countries: a comparison with
tional distance function approach. J. Environ. Manage. 64, 365–375. Malmquist indices. J. Comp. Econ. 33, 401–420.
Lozano, S., Gutiérrez, E., 2008. Non-parametric frontier approach to modelling the rela- Yu, E.S.H., Choi, J.Y., 1985. Causal relationship between energy and GNP: an interna-
tionships among population, GDP, energy consumption and CO2 emissions. Ecol. tional comparison. J. Energy Dev. 10, 249–272.
Econ. 66, 687–699. Zaim, O., Taskin, F., 2000. A Kuznets curve in environmental efficiency: an application
Lozano, S., Villa, G., Brännlund, R., 2009. Centralised reallocation of emission permits on OECD countries. Environ. Resour. Econ. 17, 21–36.
using DEA. Eur. J. Oper. Res. 193, 752–760. Zhou, P., Ang, B., 2008. Decomposition of aggregate CO2 emissions: a production-
Nag, B., 2006. Estimation of carbon baselines for power generation in India: the supply theoretical approach. Energy Econ. 30, 1054–1067.
side approach. Energy Policy 34, 1399–1410. Zhou, P., Ang, B., Poh, K., 2006. Slacks-based efficiency measures for modeling environ-
O'Donnell, C.J., Rao, D.S.P., Battese, G.E., 2008. Metafrontier frameworks for the study of mental performance. Ecol. Econ. 60, 111–118.
firm-level efficiencies and technology ratios. Empir. Econ. 34, 231–255. Zhou, P., Ang, B., Poh, K., 2008. Measuring environmental performance under different
Oh, D.H., 2010a. A global Malmquist–Luenberger productivity index. J. Product. Anal. environmental DEA technologies. Energy Econ. 30, 1–14.
34, 183–197. Zhou, P., Ang, B., Han, J., 2010. Total factor carbon emission performance: a Malmquist
Oh, D.H., 2010b. A metafrontier approach for measuring an environmentally sensitive index analysis. Energy Econ. 32, 194–201.
productivity growth index. Energy Econ. 32, 146–157. Zofio, J.L., Prieto, A.M., 2001. Environmental efficiency and regulatory standards: the
Oh, D.H., Lee, J.D., 2010. A metafrontier approach for measuring Malmquist productivity case of CO2 emissions from OECD industries. Resour. Energy Econ. 23, 63–83.
index. Empir. Econ. 38, 47–64.

You might also like