Environment-adjusted regional energy efficiency in Taiwan

(1)

Environment-adjusted regional energy efﬁciency in Taiwan

Jin-Li Hu

a,⇑

, Mon-Chi Lio

b

, Fang-Yu Yeh

c

, Cheng-Hsun Lin

d

a

Institute of Business and Management, National Chiao Tung University, Taiwan

b

Department of Political Economy, National Sun Yat-Sen University, Taiwan

c

Science & Technology Policy Research and Information Center, National Applied Research Laboratories, Taiwan

d

Research Division 1, Taiwan Institute of Economic Research, Taiwan

a r t i c l e

i n f o

Article history: Received 20 April 2010

Received in revised form 9 November 2010 Accepted 28 January 2011

Available online 25 February 2011 Keywords:

Data envelopment analysis (DEA) Regional energy efﬁciency Environment-adjusted efﬁciency

a b s t r a c t

This study applies the four-stage DEA procedure to calculate the energy efficiency of 23 regions in Taiwan from 1998 to 2007. After controlling for the effects of external environments, only Taipei City, Chiayi City, and Kaohsiung City are energy efficient. Note that Kaohsiung City reaches the efficiency frontier due to the adjustment via partial environmental factors such as higher education attainment and transport vehi-cles. We also find a worsening trend for Taiwan’s energy efficiency. Not only is there a gap of energy effi-ciency between Taiwan’s metropolitan areas and the other regions, but the gap has also widened in recent years. Those inefficient counties should be given priority and the savings potential. Except for road density, the evidence indicates that each environmental factor has partial incremental effects on input slacks. As more cars and motorcycles are unfavorable externalities affecting partial energy efficiency, the central government should help local governments retire inefficient old motor vehicles, encourage energy-saving vehicle models, and provide convenient mass transportation systems. Besides, people with higher education cause industrial energy inefficient in Taiwan. The conscious of effective energy saving is necessary to schools, communities, and employee accordingly.

1. Introduction

With the growing concern over global warming and sustainable energy usage, every region needs to improve its energy efﬁciency. In 2009, Taiwan’s imported energy accounts for 99.7% of total en-ergy consumption[1]and its energy consumption growth is much higher than economic growth in recent years. From 1999 to 2007, Taiwan’s value of energy imports to GDP ratio rose from 2.69 to 11.31, and per capita energy consumption went from 3,894.74 (barrel of oil equivalent) to 5,301.66. At this time, however, energy productivity fell from 111.48 (NT$/oil quantities) in 1999 to 108.20 in 2007[1].

In 1998, 2005, and 2009, Taiwan held three National Energy Councils to address energy-related policy issues, in which efforts to improve energy efﬁciency were of priority concern. Following the 2005 council, the Bureau of Energy[2]published the Energy White Paper, which declares six policy guidelines that focus on sta-bilizing energy supply, improving energy efﬁciency, opening up energy industries, emphasizing environmental protection, strengthening research and development capabilities, and acceler-ating educational dissemination. In 2008, the Ministry of Economic

Affairs[3]announced the Framework of Taiwan’s Sustainable En-ergy Policy, which states that sustainable enEn-ergy policies should increase the efﬁciency of using limited energy resources to create a win–win–win solution for energy, environment, and economy. Thus, energy efﬁciency has become one of the most important is-sues in Taiwan’s public policy.

An economy’s energy efficiency improvement has to be based on regional energy efficiency promotion. For instance, in 2007, the per capita electricity consumption was 6382 kW h in Taipei City, Taiwan’s capital and largest city, while Taitung County, a typ-ical rural region in eastern Taiwan, consumed only 3589 kW h[4]. Understanding how administrative regions differ in energy effi-ciency would greatly facilitate efforts to coordinate energy policies and set action agendas.

According to the mentions of Patterson[5]and Ang[6], energy efficiency is a relative concept. Different people may have different definitions of energy efficiency. Three indicators are commonly used to measure it, namely, monetary-based, physical-based, and thermodynamic indicators. The former refers to the energy requirement per unit currency output (e.g., per unit US dollar out-put). Relatively, monetary-based and physical-based indicators are often used at the macro-level such as in making regional, econ-omy-wide and sectoral energy policy[6]. At the macro-economy level, data envelopment analysis (DEA) has recently become popu-lar in measuring energy efficiency. It provides a simple assessment method to deal with multiple inputs and outputs in explaining

⇑ Corresponding author. Address: Institute of Business and Management, National Chiao Tung University, 118, Chung-Hsiao W. Rd., Sec. 1, Taipei City 100, Taiwan. Fax: +886 2 23494922.

E-mail address:jinlihu@yahoo.com(J.-L. Hu).

Contents lists available atScienceDirect

Applied Energy

(2)

relative energy efﬁciency. DEA is addressed by Charnes et al.[7,8]

and extended by Banker et al.[9]. To overcoming the limitations of traditional ratio analysis, this method also provides an alternative framework for the estimation of the production frontier involving multiple inputs and outputs.

So far, fewer studies focus on the quantitative analyses of en-ergy efficiency of a region. Wei et al.[10]proposes an energy effi-ciency index based on the DEA analysis approach to examine energy efficiency on 29 provinces in China. They document that there exist remarkable differences in energy efficiency among provinces and regions. Most of the eastern provinces are close to benchmark, while the underdeveloped western area has the lowest energy efficiency levels. Shi et al.[11]consider desirable and unde-sirable outputs in measuring Chinese industrial energy efficiency and explore the maximum energy-saving potential in 28 adminis-trative regions in China. From the scale efficiency perspective, they show that the industrial structure in the east area does not depend on the consumption of a large amount of energy to gain benefits as they have constant returns to scale or decreasing returns to scale. On the contrary, other regions have increasing returns to scale, which largely causes these regions to expand their industrial econ-omy at the expense of energy saving.

Comparing in the analysis of regional energy efficiency, previ-ous studies concentrated mainly on total-factor energy efficiency such as electricity use or the amount of gasoline consumed. Hu and Wang[12], Chien and Hu[13], and Honma and Hu[14]utilize a total-factor method to observe energy efficiency in China, Orga-nization for Economic Corporation and Development (OECD) and non-OECD countries, and Japan, respectively. Lee[15] used DEA and regression methods to evaluate the energy efficiency of Taiwan government office and buildings. Yang and Pollitt[16] evaluated the performance of Chinese coal-fired power plants. They find that some plants with relatively low efficiency scores were inefficient partially due to their relatively unfavorable operating environ-ments. Lee and Lee [17] classified energy performance into the scale and management factors to analyze the energy efficiency of 47 government office buildings in Taiwan. By using CCR and BCC models to calculate the technical, pure technical and scale efficien-cies for farmers categories-wise and zone-wise in India, Nassiri and Singh[18]show that small farmers had higher energy-ratio and low specific energy requirement.

However, these studies with measuring specific-state energy efficiency by DEA method may still have some limitations: First, they did not take into account the environmental effects. The abil-ity of a producer to transform the inputs into outputs is not only affected by controllable inputs, also by uncontrollable external operating environments. For example, this could be transporta-tions, location characteristics, human capital, and regional size. Second, some producers may operate under a favorable external operating environment, while others operate under an unfavorable external operating environment. At this time, an unfavorable external environment makes firms’ efficiency under-estimated. An unfavorable external operating environment refers to addi-tional inputs are required to produce the same level of output in order to overcome the external disadvantage. If we ignore this cir-cumstance, efficiency assessment resembles penalizing good pro-ducers who operate in an unfavorable external operating environment and rewarding poor producers who operate in a favorable external operating environment. This will bias the effi-ciency assessment results and lead to misleading conclusions

[19]. For instance, we may wrongly blame a local government for poor management in energy use, but in fact the local government’s management is good. Lower energy efﬁciency found in the admin-istration region is due to unfavorable environmental factors.

In order to comprise the inﬂuences of external environmental factors, this study applies the four-stage DEA procedure developed

by Fried et al.[19] to evaluate the pure managerial energy effi-ciency of the 23 administrative regions in Taiwan. Each adminis-trative region in Taiwan can be treated as a decision-making unit (DMU). The managerial efficiency of DMUs can be evaluated in terms of their ability to either minimize input usage under the pro-duction of given output or to maximize output propro-duction with gi-ven inputs. The four-stage DEA model rests on the premise that DMUs operating in relatively unfavorable environments are at a disadvantage in the traditional DEA model. In other words, a rela-tively efficient DMU may be wrongly labeled as inefficient if the impacts of unfavorable environments are not removed in the effi-ciency-estimation approach. To the best of our knowledge, exactly how the various administrative regions in Taiwan differ in energy efficiency has not yet been addressed. In particular, the environ-ment-adjusted energy efficiency of a region is left to be first exam-ined by this study.

The paper is organized as follows. Section2provides a brief dis-cussion of the methodology. Section3interprets data sources and descriptive statistics. Section4 presents and discusses empirical results. The ﬁnal section gives the conclusion and some policy implications.

2. Methodology

According to the rationale of Fried et al.[19], the four-stage DEA approach can be introduced as follows. The ﬁrst stage calculates a DEA frontier by using the selected inputs and outputs according to standard production theory while the environmental variables are excluded. The piecewise linear input requirement set under vari-able returns to scale is deﬁned as follows:

LðyÞ ¼ x : Yz P y; Xz 6 x; Iz ¼ 1; z 2 RKþ

n o

; ð1Þ

where y is an (M 1) vector of M outputs, x is an (N 1) vector of N inputs used to produce output y, Y is an (M K) matrix of outputs, X is an (N K) matrix of inputs, z is a (K 1) vector of activities or weights, I is a (1 K) vector of ones, K is the number of DMUs, M is the number of outputs, and N is the number of inputs. Given out-put vector y, all inout-put vectors that are feasible for producing outout-put vector y are in the input requirement set. All convex combinations of input vectors, which are less than or equal to the input bundle x and are feasible to produce at least output vector y, establish the isoquant or reference frontier for output y and is the basis for calcu-lating the Farrell[20]technical efﬁciency.

Given the piecewise linear input requirement set in(1), the DEA model used to compute the Farrell technical efﬁciency for unit k, k = 1, . . . , K, is formulated as the following linear programming problem: TEk¼ min z;k k s:t: Yz P yk_; Xz 6 kxk_; Iz ¼ 1 z 2 RKþ; ð2Þ

where TE is a measure of efﬁciency under the restriction that a lin-ear combination of efﬁcient units produces the same or more of all outputs and that the reduction in inputs is equi-proportionate; yk

and xk_{are output and input vectors for unit k, respectively; k is a}

scalar value representing a proportional contraction of all inputs, holding input ratios and output level constant. The minimum value of k that satisﬁes all constraints is the Farrell radial technical efﬁ-ciency measure.

The radial measure computed in the ﬁrst stage evaluates the performance of a DMU relative to best practice, predicated upon

(3)

the inputs and the outputs included in the model. Other variables, however, influence the managerial ability to transform the inputs into outputs, but which are outside the managerial control. These variables refer to the external environment. Unfavorable external conditions mean that additional inputs are required to produce the same level of output in order to overcome the external disad-vantage. In other words, the radial efficiency score generated by the initial model in the first stage overstates the efficiency of DMUs operating under favorable conditions and understates the effi-ciency of DMUs operating under unfavorable conditions.

The second stage is to estimates the N input slack equations using an appropriate econometric method such as the Tobit regres-sion in this study. The dependent variables are radial plus non-ra-dial input slack, while the independent variables are the measures of external conditions applicable to the particular input. The pur-pose is to quantify how the external environment affects the exces-sive use of inputs. The N equations are speciﬁed as follows:

ITSk

j ¼ fj Qkj;bj;ukj

; j ¼ 1; . . . ; N k ¼ 1; . . . ; K; ð3Þ

where ITSkj is unit k’s total slack for input j based on the DEA results

from the ﬁrst stage, Qk

j is a vector of variables characterizing the

external environment for unit k that may affect the utilization of in-put j, bj is a vector of coefﬁcients, and ukj is a disturbance term.

These equations explain the variation in total by-variable consider-ations of inefﬁciency. The independent variables characterizing the operating environment in(3)are not limited to be the same across equations, need not have a linear relationship with the dependent variables and can be a mixture of continuous and categorical variables.

The third stage uses the estimated parameters from the second stage to predict total input slack (I^TSk

j) for each input and each unit

based on its external environmental variables:

I^TSkj ¼ fj Qkj; ^bj

; j ¼ 1; . . . ; N k ¼ 1; . . . ; K: ð4Þ

These predictions can be used to adjust the primary input data for each unit according to the difference between maximum pre-dicted slack and prepre-dicted slack:

xkadjj ¼ x k j þ Max k fI^TSKjg I^TS K j h i j ¼ 1; . . . ; N k ¼ 1; . . . ; K: ð5Þ

The region with maximum input slack is the one faces the least favorable environmental conditions. The maximum predicted input slack thus serves as a benchmark for the least favorable set of external conditions. A region with external variables corresponding to this benchmark level would not have its input vector adjusted. A region with external variables generating a low-er level of predicted slack would have its input vector adjusted upward to put it on an equal footing with the country having the least favorable external environment. Eq.(5)creates a new pseudo dataset where the inputs are adjusted for the effect of external environment.

The fourth stage re-runs the DEA model under the initial input– output specification and generates new radial measure of efficiency by using the adjusted dataset. These new radial scores measure the efficiency that is attributable to management. The re-sult excludes any influence from environmental variables and, therefore, can reflect efficiency more accurately.

3. Data and variables’ descriptions

This paper examines 23 administrative regions in Taiwan. As Golany and Roll[21]illustrated, the number of evaluated DMUs should be more than ﬁve times the total selected number of input and output; otherwise, the validity and credibility of study’s results

will be seriously compromised. Hence, this study selects six factors as the inputs and outputs.

The relationship among output, energy use, and employment are build upon econometric framework. From a policy viewpoint, the direction of causality between these variables has important implications[22,23]. If unidirectional causality runs from electric-ity consumption to income or employment reducing electricelectric-ity consumption could lead to a fall in income and/or employment

[24]. Lean and Smyth[25]find that there is unidirectional Granger causality running from electricity consumption and emissions to economic growth for five ASEAN, implying that these countries are energy dependent countries. The direct effect of energy con-sumed for commercial use which generates higher rates of eco-nomic growth, higher electricity consumption results in an increase in energy production, which has the indirect effect of gen-erating employment and infrastructure in energy services. Narayan and Wong[26]find that oil consumption, oil prices, and state in-come are panel cointegrated in Australia. Moreover, estimated long-run elasticities reveal that oil prices have had a statistically insignificant impact on oil consumption while income has had a positive and statistically significant effect on oil consumption. Uti-lizing the generalized forecast error variance decomposition tech-nique, Sari and Soytas [27] document that the total energy consumption explains 21% of forecast error variance of GDP in the case of Turkey. Besides, energy consumption appears to be al-most as important as employment.

Based on the aforementioned literature, real income can be treated as the function of electricity consumption and oil sales. Five factors to serve as the inputs: total employment (Employ), house-hold and commercial electricity consumption (Houelec), industrial electricity consumption (Indelec), gasoline sales (Gas), and diesel sales (Diesel). Note that Taipower Company[4]has two categories in accordance with two different fare rates in Taipower Company’s statistics, called ‘the electric lamp’ and ‘the electric power’. The ‘electric lamp’ fare rate applies to the household and commercial sectors, whereas the ‘electric power’ fare rate applies to industries. Therefore, we use the regional figure of the electric lamp tion to measure household and commercial electricity consump-tion, and the regional figure of the electric power consumption is used to measure industrial electricity consumption. The nature candidate of output factor used in the DEA model is the total real income (Income) of each region, as measured in millions New Tai-wan dollars (NTD) of constant 2001. The data are provided annu-ally from 1998 to 2007. All nominal variables are transformed into real variables at the 2001 price level by Taiwan’s GDP defla-tors. The research data are compiled from Bureau of Energy[28], Taipower Company[4]and Taiwan National Statistics[29].

Due to locating in different environments, administrative re-gions have different performance form each other. A part of this difference is caused by nondiscretionary environmental factors such as location characteristics and educational attainment. Hence, the environmental factors used in the second stage include the number of proﬁt organizations (Proﬁt), the ratio of higher educa-tion attainment of populaeduca-tion aged 15 and over (Highedu), road density (Roaden), the number of cars (Car), the number of motor-cycles (Bike), and the number of local phones (Phone). Table 1

summarizes the deﬁnitions of variables used in this study.Table 2

presents the descriptive statistics for each variable.

4. Results and discussion 4.1. Stage one: initial DEA

Efﬁciency scores for all administrative regions are ﬁrst calcu-lated using the DEA model with the output and seven inputs.

(4)

Table 3reports those results. According to this table, urban areas appear to be more energy efficient than rural areas in Taiwan. For instance, Penghu County, Keelung City, Chiayi City, and Taipei City perform the best with efficiency scores of 1, indicating that these regions create more value-added by using the same re-sources and energy inputs. Taipei City is the largest city in Taiwan owning the most public service and commercial activities. Keelung City is an urban area neighboring Taipei City in northern Taiwan. However, Kaohsiung City and Taichung City appear to be less en-ergy efficient than other large cities. They are the largest city in Taiwan’s central area and the largest city in southern Taiwan, respectively. As for the rural areas, Miaoli County, Yunlin County, Chiayi County, and Hualien County (i.e. the typical rural regions in Taiwan) appear to be less energy efficient. The efficiency score of Chunghua County is as low as 0.670 in 1998 and Taichung County is as low as 0.561 in 2007. These results are perhaps that the efficiency of regions in unfavorable circumstances will be un-der-estimated, and vice versa. Therefore, we have to quantify the effects of the external environment on inputs, and the efficiency scores should be re-estimated.

4.2. Stage two: quantifying the effects of the operating environment The second stage of analysis adopts Tobit regressions to quantify the environmental effects embedded in the input slacks

computed using the DEA analysis. The Tobit model is a regression model dealing with a censored dependent variable. There are ﬁve regressions, one for each input. The dependent variables are the to-tal input slacks of administrative regions. Single Tobit equations are estimated since the independent variables are the same across the ﬁve input slack equations. For further details for Tobit regres-sion model, see Greene[30]. The parameter estimates and standard errors are summarized inTable 4.

The positive (negative) coefficients suggest that environment is unfavorable (favorable) owing to it’s associated with greater (less) excess use of the input.Table 4 exhibits the estimated results of five Tobit regressions, one for each input. The regressors include road density, the number of profit organizations, the ratio of higher education attainment of population aged 15 and over, the number of cars, the number of motorcycles, and the number of local phones.

The results ofTable 4suggest that a higher road density is an unfavorable environment for the efficiency of household electricity consumption and gasoline use. Besides, a larger number of profit organizations offer a favorable environment for gasoline use, but an unfavorable environment for employment. Similarly, a larger share of higher education is a favorable environment for the effi-ciency of employment and gas use, while an unfavorable environ-ment for household electricity consumption. More cars produce an unfavorable environment for energy efficiency in terms of gas use. The number of motorcycles is positively associated with the slack in gas use, household and industrial electricity consumption. Inter-estingly, the number of cars is negatively associated with the slack in household and industrial electricity consumption, implying that more transportation vehicle can offer a favorable environment for electricity use. Ultimately, more telephones produce a favorable environment for employment and diesel use, indicating that the use of communication technology can substitute for the use of transportation facilities.

The above-mentioned results exhibit that promoting efficiency of household and industrial electricity consumption could be a beneficial influenced by car quantity, while a disadvantage influ-enced by partial transport facilities (i.e., motorcycles), education attainment and road density. A region with a ratio of higher educa-tion enjoys an unfavorable environment in industrial energy effi-ciency. This finding is possibly owing to that, although individuals with a high education and income may be more con-scious of environmental protection issues, they also tend to use products and services that consume more energy (e.g., large cars, TVs, and refrigerators). This will lead to produce negative external-ities, worsening energy efficiency. On the other hand, higher trans-port facilities have an unfavorable environment in efficiency of gasoline use, while higher communication facilities are to be a favorable environment in efficiency of diesel use. More transport facilities use will harm the efficiency of gasoline use. On the one hand, this outcome could be caused by lacking of energy-saving facilities on vehicles. On the other hand, lower energy prices let people to ignore the importance of energy-savings. The energy market in Taiwan generally belongs to monopolies due to the pol-icy regulations and law restrictions. Compared to neighboring East Asian economies, energy prices in Taiwan are relatively much low-er, worsening its energy efficiency accordingly.

4.3. Stage three: data adjustment

The parameter estimates presented in Table 4are utilized to adjustment the dataset utilized in stage 1 according to Eq. (5). The adjusted data controls for the inﬂuence of the external operat-ing environment. The adjustment essentially amounts to penaliz-ing the region for its ability to use fewer inputs under favorable external conditions. By increasing the input quantity and leaving

Table 1

Deﬁnitions of variables.

Variables Deﬁnitions and units Output

Income Total real income of a region, in millions of constant 2001 NTD Inputs

Employ Employment, total

Houelec Household and commercial electricity consumption, in kilowatt hours

Indelec Industrial electricity consumption, in kilowatt hours Gas Gasoline sales, in thousand liters

Diesel Diesel sales, in thousand liters External environment variables

Roaden Road density (km per sq. km) Proﬁt Number of proﬁt organizations

Highedu Higher education attainment of population aged 15 and over (% of total people)

Car Number of cars, in thousand

Bike Number of motorcycles, in thousand Phone Number of local phones, in thousand

Table 2

Descriptive statistics (1998–2007).

Variables Mean Standard deviation Minimum Maximum

Output Income 345,800.6 378,915.2 17,221.4 1,621,601.2 Inputs Employ 420,478.3 356,875.0 31,000.0 1,753,000.0 Houelec 2,250,735.7 2,223,870.6 149,971.9 10,742,804.0 Indelec 4,555,612.5 4,389,452.3 93,099.0 21,417,491.2 Gas 424,703.9 334,255.8 19,644.0 1,559,655.0 Diesel 146,116.9 105,659.2 3,121.0 483,538.0

External environment variables

Roaden 2.6 2.6 0.3 8.8 Proﬁt 47,911.4 46,990.7 4,072.0 197,914.0 Highedu 24.2 9.4 8.0 55.3 Car 222,020.8 180,300.8 10,713.0 781,974.0 Bike 531,174.8 411,324.0 49,709.0 2,155,791.0 Phone 561,306.3 610,303.4 35,175.0 2,470,616.0

Note: Data sources are compiled from the Bureau of Energy[28], Taipower Com-pany[4]and Taiwan National Statistics[29].

(5)

the output unchanged, the region’s external advantage over gener-ating income is removed. This makes it possible to isolate net man-agerial efﬁciency by re-running the DEA model on the adjusted pseudo dataset.

4.4. Stage four: re-computing efﬁciency measures

The final stage is to re-compute efficiency scores with the initial DEA model by using the adjusted data. The new efficiency scores incorporate the effects of the external environmental effects. The descriptive statistics of adjusted efficiency scores are shown in

Ta-bles 5 and 6compares the differences from stage 1 to stage 4.

In general, after controlling for the effects of environmental fac-tors, the average efficiency scores, minimum scores, and number of efficient regions all decrease. The decrease in the average scores implies that without controlling for the external environment, the benefit to regions operating under favorable environments is greater than the penalty to regions operating under unfavorable environments. The annual minimum efficiency scores obtained in stage 4 are much lower than in stage 1, indicating the region with

the worst managerial energy efficiency operates under favorable environments. The decrease in the number of efficient regions means that without controlling for the external environment, the number of inefficient regions operating under favorable circum-stances and misjudged to be efficient is larger than the number of efficient regions operating under unfavorable circumstances and misjudged to be inefficient.

Comparing to the results ofTables 3 and 5, the efficiency scores show that some efficient regions in the first stage such as Penghu County (efficient in 1 year) and Keelung City (efficient in 10 years) are inefficient regions operating under favorable environmental conditions. After controlling for the effects of external environ-ments, Keelung City is inefficient. Chiayi City and Kaohsiung City, inefficient inTable 3and efficient inTable 5, are both efficient re-gions operating under unfavorable environmental conditions. After controlling for the effects of external environment, only one in metropolitan areas, one in North Taiwan (Taipei City) and the other in South Taiwan (Kaohsiung City and Chiayi City) are efficient over the sample period. The average values of efficiency scores present a declining pattern from 0.852 to 0.697 since 1998 to 2007, implying

Table 3

Technology efﬁciency scores of the stage one in years 1998–2007.

Regions 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Taipei County 0.770 0.783 0.724 0.788 0.850 0.843 0.811 0.846 0.856 0.832 Yilan County 0.842 0.860 0.771 0.705 0.679 0.753 0.698 0.781 0.715 0.675 Taoyuan County 0.777 0.765 0.751 0.751 0.754 0.739 0.783 0.746 0.734 0.723 Hsinchu County 0.864 0.884 0.866 0.800 0.799 0.815 0.829 0.878 0.760 0.700 Miaoli County 0.760 0.734 0.691 0.714 0.677 0.698 0.659 0.676 0.725 0.685 Taichung County 0.675 0.697 0.616 0.587 0.615 0.586 0.588 0.566 0.565 0.561 Chunghua County 0.670 0.612 0.594 0.575 0.631 0.647 0.635 0.610 0.638 0.595 Nantou County 0.799 0.917 0.794 0.749 0.723 0.722 0.762 0.784 0.815 0.723 Yunlin County 0.673 0.724 0.626 0.688 0.624 0.625 0.654 0.612 0.700 0.604 Chiayi County 0.669 0.701 0.750 0.696 0.630 0.647 0.669 0.698 0.694 0.628 Tainan County 0.719 0.716 0.688 0.657 0.642 0.687 0.673 0.624 0.652 0.627 Kaohsiung County 0.687 0.700 0.652 0.608 0.636 0.613 0.659 0.666 0.601 0.654 Pingtung County 0.742 0.698 0.689 0.647 0.704 0.705 0.644 0.692 0.681 0.714 Taitung County 0.844 0.830 0.909 0.875 0.895 0.845 0.744 0.699 0.804 0.891 Hualien County 0.776 0.743 0.757 0.758 0.673 0.702 0.681 0.620 0.651 0.655 Penghu County 0.858 0.941 1.000 0.876 0.850 0.816 0.899 0.809 0.907 0.829 Keelung City 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Hsinchu City 0.829 0.845 0.864 0.885 0.821 0.828 0.901 0.826 0.874 0.889 Taichung City 0.812 0.800 0.698 0.812 0.702 0.755 0.671 0.654 0.705 0.758 Chiayi City 1.000 0.878 0.922 0.949 1.000 0.932 0.742 0.895 0.923 0.852 Tainan City 0.733 0.683 0.708 0.682 0.702 0.716 0.714 0.667 0.682 0.729 Taipei City 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Kaohsiung City 0.911 0.955 0.906 0.940 0.853 0.892 0.904 0.944 0.918 0.926 Mean 0.800 0.803 0.781 0.771 0.759 0.764 0.753 0.752 0.765 0.750 Standard deviation 0.105 0.111 0.127 0.128 0.127 0.117 0.118 0.128 0.125 0.129 Minimum 0.669 0.612 0.594 0.575 0.615 0.586 0.588 0.566 0.565 0.561 Maximum 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Table 4

Tobit regression results (N = 230).

Variables Slacks

Employ Houelec Indelec Gas Diesel

Intercept 66213.18***_(12351.96) _1019560***_(70055.39) _{1845479 (1137997)} _37580.15**_(15936.64) _102235.4***_(27865.66) Roaden 3818.12 (3697.92) 60508.5**_(23656.86) _{225797.2 (524146.8)} _9298.02**_(4419.07) _{9052.76 (8416.90)} Proﬁt 1.52***_(0.50) _{4.05 (5.93)} _{68.02 (56.39)} _3.87***_(0.65) _{0.30 (0.73)} Highedu 1304.61** (519.14) 1119.51 (3994.06) 117525.7** (53181.5) 1608.96*** (590.33) 207.05 (857.22) Car 0.01 (0.11) 4.34***_(1.56) 44.09***_(11.63) _0.71***_(0.17) _{0.38 (0.22)} Bike 0.003 (0.04) 2.25***_(0.42) _8.08**_(3.33) _0.13***_(0.04) 0.09 (0.05) Phone 0.14***_(0.03) _{0.18 (0.21)} _{3.27 (2.49)} _{0.04 (0.04)} _0.21***_(0.04) Log-likelihood 2120.68 510.37 2572.01 2303.14 2179.64 Sigma 39025.33***_(13036.59) _544073.5***_(128391.7) _4933675***_(1699160) _49513.07***_(8981.67) _107665.4***_(28624.5)

Note: Numbers in parentheses are standard errors.

** _{Signiﬁcance at the 5% level.} ***_{Signiﬁcance at the 1% level.}

(6)

that regional energy efficiency in Taiwan is worsen. It is worth to note that Kaohsiung City is inefficient in stage 1, but is efficient in all stages when environmental factors are considered. The rea-son is due to the ratio of higher education grows from 24.99% in 1998 to 38.64% in 2007 and car quantities apparently grows from 310,000 in 1998 to 370,000 in 2007. Besides, the actual values of higher education and cars multiplying the estimated coefficients will increase the slacks, resulting in new inputs promotion. Com-pared to other county, the increments of adjusted inputs in Kaoh-siung City are most significant. This outcome signifies that higher education attainment and transport vehicle are important favorable externalities for Kaohsiung City to improve it energy efficiency.

We further investigate the difference of regional energy effi-ciency in the analysis of stages 1 and 4, as shown inTable 6. The Mann–Whitney U statistics reject the null of equality of two-stage efficiency scores. This implies that there indeed exists difference of unadjusted and adjusted regional energy efficiency at 1% signifi-cance level. Besides, this evidence also confirms the importance of controlling for the external environmental factors. Under such

circumstance, we may reﬁne the true outcomes of regional energy efﬁciency.

5. Conclusions

Given Taiwan’s declining energy productivity in recent years, energy efficiency is a critical policy issue. This study applies the four-stage DEA procedure developed by Fried et al.[19], which can purge away the effects of external environments, to calculate the pure managerial efficiency of energy for the 23 administrative regions in Taiwan. Empirical evidence indicate that, even com-pared with Taiwan’s own benchmark on energy efficiency, most re-gions are inefficient in energy use. After controlling for the effects of external operating environments, only Taipei County, Chiayi City, and Kaohsiung City (forming the major metropolitan areas in North and South Taiwan) reach efficiency scores of 1. The evi-dence further reveals a worsening trend in Taiwan’s energy effi-ciency over the period of 1998–2007. Taiwan relies heavily on imported energy and should be concerned about energy efficiency issue more seriously. As such, more public efforts and better policy coordination are necessary.

Taiwan’s cities are generally more efficient in energy use than rural areas. However, after controlling for the external environ-ments, we find that partial cities appearing to be efficient in the first stage of initial DEA, including Keelung City, Penghu County, are actually inefficient regions operating under favorable environ-mental conditions. On the contrary, Kaohsiung City, which appears to be inefficient in the first stage, is an energy-efficient region oper-ating under favorable environmental conditions due to the promo-tion of higher educapromo-tion attainment and transport facilities. The annual average efficiency scores obtained in stage four and stage one exists significantly discrepancy and become much lower than without considering external noise factors. These results indicate that the input data must be adjusted for environmental effects since the four-stage DEA procedure can avoid misleading results. For the effects of environmental factors, we find that a higher population density, a higher ratio of population with a higher

Table 5

Technology efﬁciency scores with environmental-adjusted scores of stage four in years 1998–2007.

Regions 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Taipei County 0.928 0.919 0.853 0.884 0.895 0.906 0.947 0.947 0.951 0.941 Yilan County 0.550 0.590 0.555 0.503 0.491 0.554 0.515 0.559 0.506 0.454 Taoyuan County 0.720 0.700 0.692 0.676 0.697 0.686 0.722 0.694 0.683 0.685 Hsinchu County 0.499 0.540 0.562 0.536 0.558 0.574 0.606 0.648 0.579 0.512 Miaoli County 0.464 0.480 0.468 0.491 0.481 0.503 0.488 0.484 0.515 0.468 Taichung County 0.629 0.634 0.560 0.522 0.563 0.536 0.537 0.517 0.522 0.532 Chunghua County 0.629 0.540 0.532 0.496 0.574 0.579 0.560 0.518 0.549 0.511 Nantou County 0.502 0.549 0.452 0.442 0.453 0.448 0.482 0.481 0.500 0.415 Yunlin County 0.539 0.516 0.456 0.484 0.442 0.441 0.460 0.418 0.481 0.406 Chiayi County 0.443 0.426 0.456 0.415 0.389 0.395 0.423 0.424 0.428 0.369 Tainan County 0.606 0.572 0.576 0.539 0.547 0.581 0.571 0.516 0.555 0.531 Kaohsiung County 0.684 0.648 0.621 0.569 0.621 0.586 0.630 0.623 0.574 0.632 Pingtung County 0.757 0.677 0.673 0.623 0.670 0.679 0.618 0.637 0.639 0.682 Taitung County 0.405 0.419 0.439 0.440 0.457 0.442 0.464 0.411 0.429 0.408 Hualien County 0.523 0.515 0.538 0.554 0.498 0.520 0.525 0.455 0.472 0.442 Penghu County 0.453 0.483 0.520 0.431 0.413 0.413 0.498 0.430 0.452 0.368 Keelung City 0.802 0.827 0.775 0.720 0.663 0.677 0.823 0.797 0.722 0.757 Hsinchu City 0.610 0.642 0.674 0.708 0.687 0.693 0.775 0.716 0.790 0.796 Taichung City 0.785 0.771 0.667 0.770 0.659 0.713 0.645 0.623 0.671 0.665 Chiayi City 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Tainan City 0.855 0.763 0.821 0.780 0.794 0.802 0.808 0.734 0.771 0.838 Taipei City 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Kaohsiung City 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Mean 0.669 0.661 0.647 0.634 0.633 0.640 0.656 0.636 0.643 0.627 Standard deviation 0.191 0.183 0.181 0.191 0.191 0.190 0.190 0.197 0.191 0.216 Minimum 0.405 0.419 0.439 0.415 0.389 0.395 0.423 0.411 0.428 0.368 Maximum 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 Table 6

Differences of stage 1 and stage 4 efﬁciency scores.

Average efﬁciency Minimum scores M–U statistics

Stage 1 Stage 4 Stage 1 Stage 4

1998 0.800 0.669 0.669 0.405 2.63 (0.007)*** 1999 0.803 0.661 0.612 0.419 3.01 (0.003)*** 2000 0.781 0.647 0.594 0.439 2.97 (0.003)*** 2001 0.771 0.634 0.575 0.415 2.80 (0.003)*** 2002 0.759 0.633 0.615 0.389 2.85 (0.003)*** 2003 0.764 0.640 0.586 0.395 2.94 (0.003)*** 2004 0.753 0.656 0.588 0.423 2.45 (0.014)*** 2005 0.752 0.636 0.566 0.411 2.43 (0.015)*** 2006 0.765 0.643 0.565 0.428 2.67 (0.008)*** 2007 0.750 0.627 0.561 0.368 2.26 (0.024)***

Note: Numbers in parentheses are p-values. Data sources are compiled from the Bureau of Energy[28], Taipower Company[4]and Taiwan National Statistics[29].

(7)

education, more cars, and more motorcycles are unfavorable envi-ronmental conditions for efficient energy use. A higher road den-sity, more profit organizations, a larger share of total employment by industry, a larger share of total employment by service, and more telephones are favorable environmental condi-tions for efficient energy use.

Our findings have important policy implications. For energy efficiency, not only does a gap exist between Taiwan’s metropoli-tan areas and other regions, but the gap has also widened in recent years. More public efforts should be put forth on improving the en-ergy efficiency of non-metropolitan regions. Besides, the central government should make different energy policies across regions. Taking into consideration the discrepant energy efficiency and the savings potential, those inefficient counties or cities should be given priority and be allocated more energy saving quotas. To improve energy efficiency, as more cars and motorcycles are unfa-vorable externalities, the central government should help local governments retire inefficient old motor vehicles, encourage en-ergy-saving vehicle models, and provide convenient mass trans-portation systems. People with higher education cause industrial energy inefficient in Taiwan. The conscious of effective energy sav-ing is necessary to schools, communities, and employees.

References

[1] Bureau of Energy. Taiwan energy statistical yearbook; 2009<http://www. moeaboe.gov.tw/opengovinfo/Plan/all/energy_year/main/EnergyYearMain. aspx?PageId=default>.

[2] Bureau of Energy. Energy white paper, Taipei; 2005.

[3] Ministry of Economic Affairs. Framework of Taiwan’s sustainable energy policy, Taipei; 2008.

[4] Taipower Company. Statistical data; 2009 <http://www.taipower.com.tw/ left_bar/jing_ying_ji_xiao/statistical_data.htm>.

[5] Patterson MG. What is energy efﬁciency? Concepts, indicators and methodology issues. Energy Policy 1996;24:377–90.

[6] Ang BW. Monitoring changes in economy-wide energy efﬁciency: from

energy-GDP ratio to composite efﬁciency index. Energy Policy

2006;34:574–82.

[7] Charnes A, Cooper WW, Rhodes E. Measuring the efﬁciency of decision making units. Eur J Oper Res 1978;2:429–44.

[8] Charnes A, Cooper WW, Rhodes E. Evaluating program and managerial efﬁciency: an application of data envelopment analysis to program follow through. Manage Sci 1981;27:668–98.

[9] Banker RD, Charnes A, Cooper WW. Some models for estimating technology and scale inefﬁciencies in data envelopment analysis. Manage Sci 1984;30:1078–92.

[10] Wei C, Ni JL, Shen MH. Empirical analysis of provincial energy efﬁciency in China. China World Econ 2009;17:88–103.

[11] Shi GM, Bi J, Wang JN. Chinese regional industrial energy efﬁciency evaluation based on a DEA model of ﬁxing non-energy inputs. Energy Policy 2010;38:6172–9.

[12] Hu JL, Wang SC. Total-factor energy efﬁciency of regions in China. Energy Policy 2006;34:3206–17.

[13] Chien T, Hu JL. Renewable energy and macroeconomic efﬁciency of OECD and non-OECD economies. Energy Policy 2007;35:3606–15.

[14] Honma S, Hu JL. Total-factor energy efﬁciency of regions in Japan. Energy Policy 2008;36:821–33.

[15] Lee CC. Energy consumption and GDP in developing countries: a cointegrated panel analysis. Energy Econ 2005;27:415–27.

[16] Yang H, Pollitt M. Incorporating both undesirable outputs and uncontrollable variables into DEA: the performance of Chinese coal-ﬁred power plants. Eur J Oper Res 2009;197:1095–105.

[17] Lee WS, Lee KP. Benchmarking the performance of building energy

management using data envelopment analysis. Appl Therm Eng

2009;29:3269–73.

[18] Nassiri SM, Singh S. Study on energy use efﬁciency for paddy crop using data envelopment analysis (DEA) technique. Appl Energy 2009;86:1320–5. [19] Fried HO, Schmidt SS, Yaisawarng S. Incorporating the operating environment

into a nonparametric measure of technical efﬁciency. J Prod Anal 1999;12:249–67.

[20] Farrell MJ. The measurement of productivity efﬁciency. J Roy Stat Soc 1957;120:253–81.

[21] Golany B, Roll Y. An application procedure for DEA. Omega – Int J Manage Sci 1989;17:237–50.

[22] Asafu-Adjaye J. The relationship between energy consumption, energy prices and economic growth: time series evidence from Asian developing countries. Energy Econ 2000;22:615–25.

[23] Ghosh S. Electricity consumption and economic growth in India. Energy Policy 2002;30:125–9.

[24] Narayan PK, Smyth R. Electricity consumption, employment and real income in Australia evidence from multivariate Granger causality tests. Energy Policy 2005;33:1109–16.

[25] Lean HH, Smyth R. CO2 emissions, electricity consumption and output in

ASEAN. Appl Energy 2010;87:1858–64.

[26] Narayan PK, Wong P. A panel data analysis of the determinants of oil consumption: the case of Australia. Appl Energy 2009;86:2771–5.

[27] Sari R, Soytas U. Disaggregate energy consumption, employment and income in Turkey. Energy Econ 2004;26:335–44.

[28] Bureau of Energy. Regional statistics for gas and diesel sales; 2009.<http:// www.moeaboe.gov.tw/opengovinfo/Plan/oilgas>.

[29] Taiwan National Statistics. Key statistics for regions; 2009. <http:// 61.60.106.82/pxweb/Dialog/statﬁle9.asp>.

[30] Green WH. Econometric analysis. New York: Macmillan Publishing Company; 2000.