外文翻譯--基于三階段dea模型中國區(qū)域能源效率分析（節(jié)選）

1、<p>　　中文5100字，2700英文單詞，1.6萬英文字符</p><p>　　文獻出處：Huang D, Dong Y, Zhang C, et al. Regional Energy Efficiency in China Based on a Three-Stage DEA Model[J]. 資源與生態(tài)學(xué)報(英文版), 2014, 5(2).</p><p><

2、b>　　原文：</b></p><p>　　Regional Energy Efficiency in China Based on a Three-Stage DEA Model</p><p>　　HUANG Dechun, DONG Yuyi, LIU Bingsheng</p><p>　　Introduction</p>&

3、lt;p>　　China achieved rapid economic growth through reform and opening up over the last three decades. However, such achievement was made with a cost of energy and to the environment including natural resource shortag

4、es, energy depletion and environmental deterioration. The Chinese Government stated in the Eleventh Five-year Plan that “GDP energy consumption shall be reduced by about 20% over the end of the Tenth Five-year Plan perio

5、d, while major pollutants emission shall be reduced by 10% on the basi</p><p>　　Existing research literature on this topic involves two aspects of energy efficiency: energy efficiency evaluation and analysis

6、 on energy efficiency influencing factors. Both single-factor and total-factor energy efficiency indicators have been used to evaluate energy efficiency. For total factor evaluation methods, this paper summarizes the dim

7、ensions of input indicator, output indicator, and research method and research time (Table 1).</p><p>　　Gao and Wang (2006), adopting clustering methodology, thought economic development level, industrial st

8、ructure, investment and energy price were factors influencing energy efficiency. Qiu and Shen (2008), adopting clustering methodology and the Theil method, conducted quantitative analysis on energy efficiency influencing

9、 factors using panel data. Li and Wang (2008), through analyzing data from 30 provinces and cities from 1995 to 2005 with a generalized Fisher index, thought regional structural</p><p>　　Energy efficiency ev

10、aluation analyses were mainly conducted on total-factor energy efficiency indicators, which offset the defect of a single-factor energy efficiency indicator; however, most studies have adopted DEA or super-efficiency DEA

11、 models which could not avoid the influence of environment and errors on the efficiency value. Therefore, analysis with data from different times and regions might be concluded with deviated results. No unified standard

12、system had been formed for analyzing ene</p><p>　　Principles of three-stage DEA model</p><p>　　The first stage: traditional DEA model</p><p>　　Banker et al. (1984) first built the D

13、EA-CCR model in 1978; however, CCR analyzed input and output and calculated efficiency value under the condition of returns to scale remaining constant, which was contrary to the practical situation. Coelli et al. (1998)

14、 introduced the DEABCC model which decomposed the technical efficiency (TE) in the CCR model into pure technical efficiency (PTE) and scale efficiency (SE), i.e. TE=PTE×SE. This more accurately reflects operations a

15、nd the management level of </p><p>　　Suppose there are n decision making units (DMU), each with m input and s output; xik (i=1, 2, …, m) is the ith input variable of the kth decision making unit; yjk (j=1, 2

16、, …, s) is the jth output variable of the kth decision making unit. Then, calculation of total efficiency of the pth decision-making unit is converted into a linear programming problem:</p><p>　　where, X1=(x

17、11, x22, …, xm1), Y1=(y11, y21, …, ys1) and the model is called a BCC model. θ is the total efficiency value of investigated decision-making unit, and 0≤θ≤1. When θ=1, the investigated decision-making unit is a point on

18、efficient frontier plane, hence it is effective. For ineffective unit of θ<1, 1–θ is the proportion of redundant input by investigated decision-making unit.</p><p>　　The second stage: adjusting input indi

19、cator variable</p><p>　　The slack variables of input and output analyzed in the first stage will be influenced by external environmental factors, random error and internal management factors. Traditional DEA

20、 models, instead of accurately reflecting whether the influence on efficiency comes from internal management or external environment and random error, attributes all influencing factors to internal management. Therefore,

21、 Timmer (1971) introduced stochastic frontier analysis (SFA) to consider the influence of externa</p><p>　　where, Sni is the slack variable of the nth input of the ith decision-making unit, Z1=(z1i, z2i, …,

22、zki) presents k environmental variables; f(Zi,βn) represents the influence of environmental variable on input slack variable Sni; generally, f(Zi, βn)=Zi, βn; Vni+Uni is the combined error term; suppose Vni.N(0, σ2 vn) r

23、eflects random error item, Uni reflects management inefficiency and obeys a truncated normal distribution, i.e. Uni.N(μ, σ2 vn) and Vni is independent from and uncorrelated to Uni.</p><p>　　Through adjusting

24、 input variable data of the nth decision making unit with the result of SFA model regression and eliminating the influences of environmental factor and random error, an efficiency value purely reflecting management level

25、 can be calculated. The adjustment formula is as follows:</p><p>　　where, X*ni is input after adjustment; Xni is original input value. The first square bracket in Formula (3) indicates all decision-making un

26、its are adjusted to be under the same external environment and the second square bracket indicates random errors of all decision-making units are adjusted to the same to make all decision-making units under the same exte

27、rnal environment and fortune.</p><p>　　The third stage: DEA model after adjustment</p><p>　　In this stage, input the input data after adjustment in the second stage and original output data into

28、 the DEA model to calculate the relative efficiency value; then the result is the efficiency value reflecting management level after eliminating the influence of environmental factor and random factor and reflects effici

29、ency value of the management level.</p><p>　　Variable selection and data source</p><p>　　Selection of input and output variables</p><p>　　Energy, labour force and capital were selec

30、ted in this paper as three input factors. Wherein, energy input factor is represented by energy consumption, which has been converted into 10 000 ton standard coal; labour force input is mainly represented by employees a

31、t the end of current year; and capital input is represented by capital stock, which is generally estimated with the perpetual inventory method at the end of each year in case the data of capital stock cannot be obtained

32、directly. This pa</p><p>　　The selection of output variable takes into account that energy input results in rapid economic and industrial development. Therefore, in this paper, total industrial output value

33、and regional GDP are selected as output variables.</p><p>　　Selection of environmental variable</p><p>　　Generally, environmental variable means the factor which may influence energy utilization

34、 efficiency but is beyond the scope of subjective control. In consideration of the factors influencing energy utilization, in this paper, technological progress and industrial structure are adopted as environmental varia

35、bles.</p><p>　　Technological progress becomes increasingly influential to energy utilization efficiency. Advanced energy conversion technologies may reduce energy waste and energy-saving technologies may dir

36、ectly reduce energy consumption per unit product. Under the current condition with increasingly less traditional energy, every country increases the input in scientific and technological research and development to find

37、new energy and technologies. This paper adopts R&D funding as the measurement indicator </p><p>　　Among the three industries, the secondary industry has high energy consumption. Therefore, the secondary

38、industry development in each region is directly relevant to energy utilization and industrial structure is represented by the proportion of the secondary industry in GDP.</p><p>　　Data source</p><

39、p>　　In consideration of data integrity and availability, this paper selects data for 29 provinces, municipalities and autonomous regions in 2009 (Tibet was not selected due to the incompleteness and data of Chongqing

40、is merged into that of Sichuan). Except the R&D fund from the China Statistical Yearbook on Science and Technology in 2010, all others are from the China Statistical Yearbook in 2010.</p><p>　　Empirical

41、analysis</p><p>　　Traditional DEA empirical result in the first stage</p><p>　　Adopting the input oriented BCC model, technical efficiency (TE), pure technical efficiency (PTE) and scale efficie

42、ncy (SE) are obtained in the first stage (Table 2). In addition, the difference value, i.e. slack value, between the ideal value and actual value of input variable can be obtained, which will be applied in the next stage

43、.</p><p>　　According to Table 2, taking no account of the influences of environmental factor and random error, the average technical efficiency, average pure technical efficiency and average scale efficiency

44、 of these provinces and cities in China are 0.863, 0.917 and 0.943 respectively. Five provinces have technical efficiencies reaching 1, i.e. the technology frontier, which are Beijing, Tianjin, Shanghai, Guangdong and Ga

45、nsu; however, the other 24 provinces are under an inefficient state with large deman</p><p>　　According to results obtained in the first stage, pure technical inefficiency is the major factor restricting the

46、 energy utilization rate; however, we still have to further analyze if pure technical efficiency is underestimated or scale efficiency over estimated without considering the influences of the external environment and ran

47、dom error.</p><p>　　SFA regression result in the second stage</p><p>　　Let slack value of each input variable analyzed in the first stage be the dependent variable and environmental variable R&a

48、mp;D funds and number of industrial enterprises be independent variables to analyze if external environmental variable influences the difference between ideal and actual input variables. If the analysis shows that enviro

49、nmental variables will influence input variable difference, Formula (3) will be adopted to eliminate external environmental factors, thereby to obtain the input </p><p>　　According to Table 3, the slack vari

50、ables of R&D funding and proportion of the secondary industry in GDP to capital stock and energy consumption passes the test with 1% significance and the slack variable of proportion of the second industry in GDP als

51、o passes the test, indicating that environmental factors have a significant influence on input redundancy. Formula (3) shall be used to eliminate external environment variables and random factor and finally make all prov

52、inces the same external envi</p><p>　　Input slack variable means possible reduced input through improving operation and management levels; therefore, if the environmental variable is positively correlated wi

53、th input slack variable, increasing the environmental variable input will go against improving energy utilization rate. It can be seen from Table 3 that regression coefficients of R&D funding input to two slack varia

54、bles of employees and capital stock are both negative and both pass the test at a 1% significance level, indicating</p><p>　　DEA empirical results after adjusting input in the third stage</p><p>

55、;　　After adjusting input variables of energy efficiencies of 29 provinces and municipalities in China in 2009, input the adjusted variable (this variable is the value obtained after eliminating environmental variable and

56、 random factor with Formula (3)) and the original output variable into DEAP2.1, then the technical efficiency taking no account of the external environmental factor and random error can be obtained (Table 4). </p>

57、<p>　　Through comparing results of the first stage, it can be seen that the energy efficiency values before and after adjustment are different to some extent. The average technical efficiency is reduced from 0.863

58、to 0.801 and scale efficiency value declines from 0.943 to 0.802 and pure technical efficiency shows great growth from 0.917 to 0.998. According to further research, scale efficiencies of all provinces and regions all re

59、duce before and after adjustment, which indicates that diseconomy of sca</p><p>　　Without eliminating the external environmental factor and random factor, all provinces have overestimated scale efficiencies

60、and underestimated pure technical efficiency and the overestimation extent is higher than underestimation extent, resulting in overestimation of the technical efficiency value.</p><p>　　Analysis on overall a

61、nd regional differences of energy efficiencies in </p><p>　　provinces and municipalities in China</p><p>　　Eliminating the influence of external environmental factor and random factor, analysis

62、in the third stage may reflect practical operating conditions in energy utilization. Therefore, integrating analysis results in the third stage with the practical condition allows for deeper analysis.</p><p>

63、;　　Overall analysis</p><p>　　According to analysis results in the third stage, the comprehensive technical efficiency value is 0.801 with a low overall level and pure technical efficiency value of 0.998 with

64、 high level and good performance, indicating that most enterprises have matured decision-making and management levels regarding energy utilization. On the other hand, low comprehensive energy efficiency has mainly result

65、ed from low scale efficiency in each province or municipality, which in reality is mainly reflected in</p><p>　　基于三階段DEA模型中國區(qū)域能源效率分析</p><p>　　HUANG Dechun, DONG Yuyi, LIU Bingsheng</p>&l

66、t;p><b>　　引言</b></p><p>　　中國在過去三十年間，通過改革開放實現(xiàn)了經(jīng)濟的快速增長。然而，這種成就卻也在能源環(huán)境問題上付出了沉重的代價，包括自然資源短缺，能源消耗和環(huán)境惡化。中國政府在“十一五”規(guī)劃中申明，“國內(nèi)生產(chǎn)總值能源消耗比“十五”期末降低20%左右，主要污染物排放量在2005年基礎(chǔ)上下降10%”。因此，在這種背景下，研究和提高區(qū)域能源效率將有助于提升中國

67、的整體經(jīng)濟的競爭力和發(fā)展的可持續(xù)性，使經(jīng)濟健康地發(fā)展。從傳統(tǒng)角度來看，能源效率評估涉及分析省際和區(qū)域間能源差異或工業(yè)部門的能源消耗強度。選定的評價指標主要分為單因素和全要素能效指標。然而，單因素能效指標只考慮到單一能源投入的有效產(chǎn)出，而沒有考慮資本和勞動力等其他影響因素，具有很大的缺陷。近年來，許多國家采用了全要素指標來評估能源效率，而數(shù)據(jù)包絡(luò)分析（DEA）的方法則更是被普遍采用。在能源效率影響因素分析上，研究者將產(chǎn)業(yè)結(jié)構(gòu)和技術(shù)進步視為

68、主要因素；然而，由于在數(shù)據(jù)選擇的時間跨度，采取方法以及指標選取上的差異，具體選取什么影響因素依然存在分歧。在此，作者試圖克服傳統(tǒng)研究方法的缺陷，采用非參數(shù)三階段DEA方法，對中國29個?。ú缓鞑?，將重慶的數(shù)據(jù)并入四川）進行能</p><p>　　現(xiàn)階段，關(guān)于能源效率這一主題的研究文獻主要涉及兩個方面：能效評估和能源效率影響因素分析。 單因素和全要素能效指標都被學(xué)者們用于評估能源效率。 對于全要素評估方法，本文匯

69、總了投入指標、產(chǎn)出指標、研究方法、研究時間四個維度，見表1。</p><p>　　Gao 和 Wang（2006）采用聚類方法，將經(jīng)濟發(fā)展水平，產(chǎn)業(yè)結(jié)構(gòu)，投資及能源價格作為影響能源效率的因素。Qiu 和 Shen（2008）采用聚類方法和Theil方法，采用面板數(shù)據(jù)對能效影響因素進行定量分析。Li 和 Wang（2008）通過對1995年至2005年30個省市的數(shù)據(jù)進行廣義Fisher指數(shù)分析，認為區(qū)域結(jié)構(gòu)因素是

70、能源強度變化的主要因素。Qu（2009）以能源效率，技術(shù)進步，能源價格，產(chǎn)業(yè)結(jié)構(gòu)，行業(yè)結(jié)構(gòu)，制度因素等變量構(gòu)建區(qū)域能源效率面板模型，認為這些因素對提高東部能源效率有很大影響，而中國中西部地區(qū)影響較弱。Yang（2009）對1986年至2005年29個省的面板數(shù)據(jù)進行了多元回歸分析，發(fā)現(xiàn)結(jié)構(gòu)因素對能源效率的影響最大，資本深化和開放對能源效率有負面影響。</p><p>　　目前，對能源效率評估主要針對全要素能效指標

71、分析，彌補了單因素能效指標的缺陷； 然而，大多數(shù)研究者采用DEA或超效率DEA模型，該方法無法避免環(huán)境和誤差對效率值的影響。 因此，不同時期和地區(qū)的數(shù)據(jù)分析結(jié)果可能會有偏差。目前還沒有形成分析能效影響因素的統(tǒng)一標準體系。 因此，作者采用Fried等（2002）提出的三階DEA模型。評估中國區(qū)域能源效率，目的是獲得更準確的結(jié)果，并提出更合理的建議。</p><p>　　三階段DEA模型原理</p>&

72、lt;p>　　第一階段-傳統(tǒng)DEA模型</p><p>　　Banker等（1984）于1978年首先建立了DEA-CCR模型;然而，CCR是在規(guī)模報酬不變的情況下對投入產(chǎn)出進行分析計算出效率值，這并不符合規(guī)模報酬可變的實際情況。 Coelli等人（1998）將CCR模型中的技術(shù)效率（TE）分解為純技術(shù)效率（PTE）和規(guī)模效率（SE）的DEA-BCC模型，即TE = PTE×SE。這更準確地反映了

73、決策單位的運作和管理水平。 BCC模型可以分為投入導(dǎo)向型和產(chǎn)出導(dǎo)向性。前者是在產(chǎn)出保持不變的條件下，最大限度地減少資源投入，提高效率，而后者是在投入保持不變的情況下提高產(chǎn)出。關(guān)于能效評估，控制投入是容易的，但控制產(chǎn)出相對困難。因此，本文采用投入導(dǎo)向型BCC模型。</p><p>　　假設(shè)有n個決策單元（DMU），每個都有m個輸入和s個輸出; xik（i = 1,2，...，m）是第k個決策單元的第i個輸入變量;

74、yjk（j = 1,2，...，s）是第k個決策單元的第j個輸出變量。 然后，將第p個決策單元的總效率的計算轉(zhuǎn)換為線性規(guī)劃問題：</p><p>　　其中，X1=(x11, x22, …, xm1)，Y1=(y11, y21, …, ys1)，該模型稱為BCC模型。 θ是被考查決策單位的總效率值，0≤θ≤1。 當(dāng)θ= 1時，表示該決策單元是有效邊界平面上的一個點，因此是有效的。 對于θ<1的無效單位，1-θ

75、是被考查決策單位冗余輸入的比例。</p><p>　　第二階段-調(diào)整投入指標變量</p><p>　　第一階段分析的投入和產(chǎn)出的松弛變量受外部環(huán)境因素、隨機誤差和內(nèi)</p><p>　　部管理因素的影響。 傳統(tǒng)的DEA模型不能準確地反映對效率的影響是來自內(nèi)部管理還是外部環(huán)境和隨機誤差，而是將所有影響因素歸因于內(nèi)部管理。 因此，Timmer（1971）引入了隨機前沿分

76、析（SFA）來考慮外部環(huán)境因素對相對效率的影響。 假設(shè)第i個DMU的第n個輸入值為Xni，松弛變量為Sni，則Sni=Xni–Xnλ>0。 </p><p>　　根據(jù)Batese和Coelli（1995）的俺就結(jié)果，松弛變量與環(huán)境變量之間的關(guān)系模型是：</p><p>　　其中，Sni是第i個決策單元的第n個輸入的松弛變量，ZI=(z1i, z2i, …, zki)為k個

77、環(huán)境變量; f(Zi,βn)表示環(huán)境變量對輸入松弛變量Sni的影響; 通常，f(Zi, βn)=Zi, βn; Vni+Uni是組合誤差項; 假設(shè)Vni.N(0, σ2 vn)反映隨機誤差項，Uni反映管理效率低下，并服從截斷正態(tài)分布，即Uni.N(μ, σ2 vn)，并且Vni獨立于Uni且不相關(guān)。 當(dāng)γ=σ2 vn/(σ2 vn+σi2)接近1時，管理因素的影響占主導(dǎo)地位; 當(dāng)γ=σ2 vn/(σ2 vn+σ2 vn)接近0時，隨機

78、誤差的影響占主導(dǎo)地位。</p><p>　　通過SFA模型回歸結(jié)果調(diào)整第n決策單位的輸入變量數(shù)據(jù)，消除環(huán)境因素和隨機誤差的影響，可以計算純粹反映管理水平的效率值。 調(diào)整公式如下：</p><p>　　其中，X*ni為調(diào)整后的輸入量; Xni是原始輸入值。 公式（3）中的第一個方括號表示所有決策單位被調(diào)整為相同的外部環(huán)境，第二個方括號表示所有決策單位的隨機錯誤調(diào)整為相同，以使所有決策單位 在

79、相同的外部環(huán)境和運氣下。</p><p>　　第三階段-調(diào)整后的DEA模型</p><p>　　在此階段，將第二階段調(diào)整后的投入數(shù)據(jù)輸入到DEA模型中，計算相對效率值; 那么結(jié)果是消除環(huán)境因素和隨機因素影響后反映管理水平的效率值，反映了管理水平的效率值。</p><p><b>　　變量選取和數(shù)據(jù)來源</b></p><p&

80、gt;　　投入和產(chǎn)出變量的選取</p><p>　　本文選擇能源，勞動力和資本三個投入因素。其中能源投入因子以能耗來表示，已轉(zhuǎn)為萬噸標準煤;勞動力投入主要由本年末職工來表示;資本投入以資本存量來表示，一般采用永續(xù)庫存法來估計每年末無法直接獲得資本存量資料。本文采用Zhang等（2004年），其中經(jīng)濟折舊率為每個省份固定資產(chǎn)總額的9.6％作為折舊率。計算方法：Ki,t=Ii,t +(1–δi)Ki.t-1，其中Ki

81、,t是區(qū)域i的第t年的資本存量，Ii,t是第i個區(qū)域的投資年，δi為區(qū)域i的固定資產(chǎn)折舊率。</p><p>　　產(chǎn)出變量的選擇考慮到能源投入導(dǎo)致經(jīng)濟和工業(yè)的快速發(fā)展。因此，本文選</p><p>　　取工業(yè)總產(chǎn)值和區(qū)域GDP作為輸出值。</p><p><b>　　環(huán)境變量的選取</b></p><p>　　一般來說，

82、環(huán)境變量是指可能影響能源利用效率但超出主觀控制范圍的因素。 考慮到影響能源利用的因素，本文將技術(shù)進步和產(chǎn)業(yè)結(jié)構(gòu)作為環(huán)境變量。</p><p>　　技術(shù)進步對能源利用效率的影響越來越大。 先進的能源轉(zhuǎn)換技術(shù)可能減少能源浪費，節(jié)能技術(shù)可能直接降低單位產(chǎn)品的能耗。 在當(dāng)前日益減少的傳統(tǒng)能源條件下，各國增加科技研發(fā)投入，尋找新能源和技術(shù)。 本文采用研發(fā)資金作為技術(shù)進步的指標。</p><p>　　

83、在三大行業(yè)中，第二產(chǎn)業(yè)能源消耗高。 因此，各地區(qū)二次產(chǎn)業(yè)發(fā)展與能源利用直接相關(guān)，產(chǎn)業(yè)結(jié)構(gòu)以第二產(chǎn)業(yè)占GDP的比重為代表。</p><p><b>　　數(shù)據(jù)來源</b></p><p>　　考慮到數(shù)據(jù)的完整性和可用性，本文選擇2009年29個省，自治區(qū)的數(shù)據(jù)（由于不完整而沒有選定西藏，重慶的數(shù)據(jù)并入四川）。本文數(shù)據(jù)除研發(fā)經(jīng)費支出一項來自《中國科技統(tǒng)計年鑒2010》外，其

84、余數(shù)據(jù)均來自《中國統(tǒng)計年鑒2010》。 </p><p><b>　　實證分析</b></p><p>　　第一階段傳統(tǒng)DEA實證結(jié)果</p><p>　　采用投入導(dǎo)向的BCC模型，在第一階段獲得技術(shù)效率（TE），純技術(shù)效率（PTE）和規(guī)模效率（SE）（表2）。此外，可以獲得投入變量的理想值和實際值之間的差值，即松弛值，這將在下一階段中應(yīng)用。&

85、lt;/p><p>　　根據(jù)表2，不考慮環(huán)境因素和隨機誤差的影響，中國這些省市的平均技術(shù)效率，平均純技術(shù)效率和平均規(guī)模效率分別為0.863,0.917和0.943。技術(shù)效率達到1即處于技術(shù)前沿面上，共有5個省份：北京，天津，上海，廣東，甘肅。然而，其他24個省份處于低效狀態(tài)，都需要提高效率。大多數(shù)省份的規(guī)模效率大于純技術(shù)效率，這意味著大多數(shù)省份的技術(shù)效率低下是由于技術(shù)效率低下而不是規(guī)模效率低下。根據(jù)第一階段取得的成果

86、，純技術(shù)效率低下是限制能源利用率的主要因素;然而，如果不考慮外部環(huán)境和隨機誤差的影響，我們?nèi)匀恍枰M一步分析純粹的技術(shù)效率是否被低估或規(guī)模效率過高估計。</p><p>　　第二階段SFA回歸結(jié)果</p><p>　　讓第一階段分析的每個輸入變量的松弛值為因變量，環(huán)境變量R＆D資金和工業(yè)企業(yè)數(shù)量為自變量，分析外部環(huán)境變量是否影響理想和實際輸入變量之間的差異。 如果分析顯示環(huán)境變量將影響輸入

87、變量差異，則采用公式（3）來消除外部環(huán)境因素，從而在消除外部環(huán)境變量后獲得輸入變量X* ni。 邊界4.1用于獲得回歸結(jié)果（表3）。</p><p>　　根據(jù)表3，研發(fā)資金和第二產(chǎn)業(yè)在國內(nèi)生產(chǎn)總值與資本存量和能源消耗量之間的松散變量通過了1％的顯著性測試，第二產(chǎn)業(yè)在GDP中的比例變化也通過了測試，表明 環(huán)境因素對輸入冗余有重要影響。 公式（3）用于消除外部環(huán)境變量和隨機因素，最終使所有省份在第三階段具有相同的外部

88、環(huán)境特征。</p><p>　　投入松弛變量指的是可以通過改善操作和管理水平來減少投入量;因此，如果環(huán)境變量與投入松弛變量呈正相關(guān)，則增加環(huán)境變量的投入將不利于提高能源利用率。從表3可以看出，R＆D資金投入到員工和資本存量兩個松弛變量的回歸系數(shù)均為負數(shù)，均通過了1％的顯著性水平，表明研發(fā)資金投入增加將提高能源利用率。同樣，第二產(chǎn)業(yè)主要是能源消耗大，行業(yè)廣泛，能源消耗大，污染嚴重的問題，第二產(chǎn)業(yè)占GDP的比重將會限

89、制能源利用率的提高，也符合實際情況。合理調(diào)整產(chǎn)業(yè)結(jié)構(gòu)，加大科技研發(fā)投入，開發(fā)新技術(shù)，是提高能源利用的完美途徑。</p><p>　　第三階段調(diào)整投入后DEA實證結(jié)果</p><p>　　在調(diào)整2009年中國29個省市能效投入變量后，將調(diào)整后的投入變量（該變量是消除環(huán)境變量和公式（3）后的隨機因子后得到的值）和原始輸出變量輸入DEAP2.1軟件，則可以獲得不考慮外部環(huán)境因素和隨機誤差的技術(shù)效

90、率（表4）。</p><p>　　通過比較第一階段的結(jié)果，可以看出調(diào)整前后的能效值在一定程度上是不同的。平均技術(shù)效率從0.863下降到0.801，規(guī)模效率值從0.943下降到0.802，純技術(shù)效率從0.917增長到0.998。根據(jù)進一步研究，各省區(qū)規(guī)模效益在調(diào)整前后都有所減少，這表明規(guī)模不經(jīng)濟是造成能源效率低下的原因，與第一階段的結(jié)果不符，表明純技術(shù)能源效率低效率低下消除環(huán)境因素和隨機因素后，技術(shù)效率值下降了15

91、個省份，表明由于環(huán)境因素較好或運氣好，估計被高估了; 11個省份由于外部環(huán)境較差或財富較差，技術(shù)水平低下，效率下降。</p><p>　　不消除外部環(huán)境因素和隨機因素，所有省份都高估了規(guī)模效率，低估了純技術(shù)效率，過高估計程度高于低估程度，導(dǎo)致技術(shù)效率價值高估。</p><p>　　我國省市能源效率總體與區(qū)域差異分析</p><p>　　消除外部環(huán)境因素和隨機因素的影

92、響，第三階段的分析可能反映了能源利用中的實際運行狀況。 因此，本文將對第三階段的綜合分析結(jié)果與實際情況做較深入的分析。</p><p><b>　　總體分析</b></p><p>　　根據(jù)第三階段分析結(jié)果，綜合技術(shù)效率值為0.801，總體水平較低，純技術(shù)效率值為0.998，水平高，性能好，表明大多數(shù)企業(yè)對能源利用的決策和管理水平已經(jīng)成熟。另一方面，低綜合能源效率主要

93、是由于各省或市的規(guī)模效率低下，實際上主要體現(xiàn)在大多數(shù)企業(yè)對能源利用的重視程度不高，在決策中占有的地位不高。但隨著能源被過度開發(fā)，能源利用成本越來越高而可用的能源越來越少，能源利用效率低下，成為制約企業(yè)發(fā)展的越來越重要的因素。要提高綜合能源利用效率，必須提高能源規(guī)模利用率。</p><p>　　表1. 能源效率評價文獻分析</p><p>　　表2. 2009年我國29個省份能源效率比較&l