The impacts of population change on carbon emissions in China during 1978–2008
Qin Zhu ⁎,Xizhe Peng 1
The State Innovative Institute for Public Management and Public Policy Studies,Fudan University,Shanghai 200433,China
a b s t r a c t
a r t i c l e i n f o Article history:
Received 29December 2011
Received in revid form 15March 2012Accepted 27March 2012Available online 7May 2012Keywords:
Carbon emission Population growth STIRPAT model Ridge regression
This study examines the impacts of population size,population structure,and consumption level on carbon emissions in China from 1978to 2008.To this end,we expanded the stochastic impacts by reg
ression on population,af fluence,and technology model and ud the ridge regression method,which overcomes the negative in fluences of multicollinearity among independent variables under acceptable bias.Results reveal that changes in consumption level and population structure were the major impact factors,not changes in population size.Consumption level and carbon emissions were highly correlated.In terms of population structure,urbanization,population age,and houhold size had distinct effects on carbon emissions.Urbanization incread carbon emissions,while the effect of age acted primarily through the expansion of the labor force and conquent overall economic growth.Shrinking houhold size incread residential consumption,resulting in higher carbon emissions.Houholds,rather than individuals,are a more reasonable explanation for the demographic impact on carbon emissions.Potential social policies for low carbon development are also discusd.
©2012Elvier Inc.All rights rerved.
1.Introduction
During the past 200years,global population,global income (gross domestic product),and carbon emissions have incread 6,70,and 20times,respectively (Jiang and Hardee,2009).The history of mo
st developed countries shows that in the development process,industry accounts for the largest proportion of carbon emissions.However,recent statistics reveal that since the 1990s,the contribution of residential energy consumption in some developed countries to carbon emissions has exceeded that of industrial ctors.Therefore,the impacts of population growth and associated residential consump-tion on carbon emissions have attracted increasing rearch interest (Bin and Dowlatabadi,2005;Druckman and Jackson,2009;Weber and Adriaan,2000).
Clearly identifying the relationship between population and carbon emissions is highly challenging primarily becau of the wide-ranging effects of population on carbon emissions.The effects usually exert indirect in fluence over consumption,production,technology,and trade,among others.In terms of population characteristics,almost all important demographic factors,including population size,structure,quality,distribution,and migration,constantly change,thereby impos-ing complicated and variable effects on carbon emissions.Studies have thus far concentrated on the relationship between population growth and emission increa,as well as on the impacts of population structure,including age structure,urbanization level,regional distribution,and houhold composition,on carbon emissions.
The approaches to studying the relationship between population and carbon emissions can be categ
orized into two:investigating the causalities and mechanisms of interaction between population and carbon emissions,and quantitatively evaluating the impacts of population growth on carbon emission increa.Birdsall (1992)summarized two principal mechanisms through which population growth in developing countries contributes to greenhou gas emissions.The first is the effect of large populations on fossil fuel consumption —an effect that stems from the incread energy demand for power generation,industry,and transport.The cond mechanism is the effect of population growth-related emissions on deforestation.The author concluded that reductions in population growth matter,but are not the key factor in leveling off carbon emissions.Knapp and Mookerjee (1996)discusd the nature of the relationship between global population growth and CO 2emissions by conducting a Granger causality test on annual data for 1880–1989.The results suggest no long-term equilibrium relationship,but imply a short-term dynamic relationship between CO 2emissions and population growth.
The IPAT identity (Ehrlish and Holdren,1971)has been extensively ud in the quantitative evaluation of the effects of population growth on carbon emission increa.According to the principle of the formula and its stochastic form,the stochastic impacts by regression on population,af fluence,and technology (STIRPAT)model,the main driving forces behind environmental impact (I)are population (
P),af fluence (A),and technology (T).Rearchers typically asss the impact of population on carbon emissions by altering population size while keeping other variables constant.Shi (2003)examined 1975–1996data on 93countries using the IPAT model and found that the impact of population change on carbon emissions is considerably more pronounced in developing countries than in developed nations.The author also determined that
Environmental Impact Asssment Review 36(2012)1–8
⁎Corresponding author.Tel.:+862155665490;fax:+862155665211.
E-mail address:zhuqin@fudan.edu (Q.Zhu),xzpeng@fudan.edu (X.Peng).1
Tel.:+862155664676;fax:+8621
55665211.
0195-9255/$–e front matter ©2012Elvier Inc.All rights rerved.doi:
10.1016/j.eiar.2012.03.003
Contents lists available at SciVer ScienceDirect
六年级下册英语单词Environmental Impact Asssment Review
j o u r n a l h o me p a g e :w ww.e l s e v i e r.c om /l oc a t e /e i a r
the elasticity of emissions with respect to global population change was1.42.Cole and Neumayer(20
04)and Rosa et al.(2004)also measured the impact of population on carbon emissions using the IPAT model,and found that the elasticities of emissions in relation to population were0.98and1.02,respectively.Wei(2011)discusd the role of technology in the STIRPAT model,and argued that the different functional forms of STIRPAT can explain the differences among estimates in studies on the environmental impacts of population and affluence.
The effects of population on carbon emissions are commonly embodied in production and consumption behaviors,which are cloly tied to population size and population structure.Satterthwaite (2009)investigated the CO2emission levels in various nations for the periods1950–1980and1980–2005.The results show little association between rapid population growth and high emission increa becau nations with very low emissions per capita are mostly tho with the highest population growth rates.Jiang and Hardee(2009)argued that consumption and production patterns among various population groups differ.In almost all climate models,however,population size is the only demographic variable considered.The assumption behind this treatment is that each individual in a population shares the same production and consumption behavior,but this assumption may be inaccurate and misleading.Hence,paying more attention to the variables of population structure is necessary in investigating the impact of population on carbon emissions.
Rearchers have cloly monitored urbanization levels becau the are highly relevant to residential consumption scale and consumption structure.Urbanization generally affects carbon emissions in three ways. First,the u of energy in production is concentrated primarily in cities, and residential consumption level increas in line with urbanization. Both situations increa energy demand,resulting in carbon emission increa,given that the energy structure remains the same.Second,the requirements for infrastructure and dwelling hous grow along with urbanization,increasing the demand for building materials(especially cement products),which are important sources of carbon emissions. Third,urbanization involves the conversion of grasslands and woodlands, the land-u changes increa carbon emissions.Poumanyvong and Kaneko(2010)empirically investigated the effects of urbanization on energy u and CO2emissions.In the investigation,the authors considered different development stages using the STIRPAT model, as well as a balanced panel datat that covers1975–2005and includes 99countries.Thefindings suggest that the impact of urbanization on carbon emissions is positive for all income groups,but that this effect is more pronounced in the middle-income group than in the other income groups.Pachauri and Jiang(2008)compared the houhold energy transitions in China and India since the1980s by analyzing aggregate statistics and nationally reprentative houhold surveys.The authors revealed that compared with rural houholds,the urban houholds in both nat
ions consumed a disproportionately large share of commercial energy and were much further along in the transition to modern energy.Satterthwaite(2009)considered the implications of population growth and urbanization for climate change between1980and2005. The author concluded that the increasing number of urban consumers and their consumption levels,not population growth,drive the increa in greenhou gas emissions.
Studies on the relationship between age structure and carbon emissions focus on the accelerated global aging process.Rearch in this area is still at its infancy.Fan et al.(2006)analyzed the impact of population,affluence,and technology on the total CO2emissions of countries at different income levels at the global scale over the period 1975–2000.The results show that population age(15–64years)has less impact on CO2emissions than do population size,affluence,and technology.Dalton et al.(2008)incorporated population age structure into an energy–economic growth model with multiple dynasties of heterogeneous houholds to estimate and compare the effects of aging populations and technical change on the baline paths of US energy u and CO2emissions.The authors showed that an aging population reduces long-term emissions by almost40%in a low-population scenario,and that the effects of the aging process on emissions can be as large as,or larger than,tho of technical change in some cas, given a clod economy,fixed substitution elasticity,andfixed labor supply over time.
The effect of changes in houhold size on carbon emissions is another rearch focus.Given afixed population size,a change in the number of houholds due to a change in houhold size can influence consumption scale and consumption structure,thereby significantly affecting carbon emissions.Thus far,there is no commonly accepted standard for defining houhold types in terms of environmental influence,and the effect of changes in houhold size on carbon emissions remains uncertain.Dalton et al.(2007)incorporated houhold size into the population–environment–technology model to simulate economic growth,as well as changes in the consumption of various goods,direct and indirect energy demand,and carbon emissions over the next 100years.Jiang and Hardee(2009)discusd the impact of shrinking houhold size on carbon emissions and argued that houholds,rather than individuals in a population,should be ud as the variable in analyzing demographic impact on emissions.This approach is favorable considering that houholds are the units of consumption,and possibly also the units of production in developing societies.
China is currently at a demographic turning ,changing from an agricultural into an urban society,from a young society to an old one,and from a society attached to land to a morefloating one (Peng,2011).Population dynamics and changes in consumption patterns have influenced and will undoubtedly continue to influence China's energy u and conquent carbon emissions.Examining
the issues will facilitate improvements in decision making for low carbon development.In this study,therefore,we incorporate population structure(age structure,urbanization level,and houhold size)into the STIRPAT model to examine the impacts of population size,population structure,and consumption level on carbon emissions.By doing so, we hope to more completely and accurately reflect the impacts of population change on carbon emissions.To overcome the negative influences of multicollinearity among independent variables,we u the ridge regression method to estimate the coefficients of the model. As an empirical ca study,the impacts of population and consumption on emissions in China from1978to2008are quantitatively assd and analyzed.Corresponding policy suggestions for energy conrva-tion and emission reduction in China are propod.
2.Model
The IPAT identity(Ehrlish and Holdren,1971)is an equation that is commonly ud to analyze the impacts of human behavior on environmental pressure.The equation is expresd as
I¼PAT;ð1Þwhere I reprents environmental impact,P reprents population,A stands for affluence,and T denotes technology.
The IPAT identity is an accounting model,in which one term is derived from the values of the three ot
her terms.The model requires data on only any three of the four variables for one or a few obrvational units,and it can only be ud to measure the constant proportional impacts of the independent variables on the dependent variable.To overcome this limitation,Dietz and Rosa(1994)established the STIRPAT model by reformulating the IPAT identity into stochastic form:
I¼aP b A c T d e;ð2Þwhere I,P,A,and T have the same definitions as in the IPAT identity;a,b, c,and d are coefficients;and e is a residual term.In this reformulation, data on I,P,A,and T can be ud to estimate a,b,c,d,and e with statistical
2Q.Zhu,X.Peng/Environmental Impact Asssment Review36(2012)1–8
regression methods.The reformulated version can convert the IPAT accounting model into a general linear model,in which statistical methods can be applied to test hypothes and asss the non-proportionate importance of each in fluencing factor.As a special ca,the stochastic version can be converted back to the original model given that a =b =c =d =e =1.
York et al.(2003)developed an additive regression model in which all variables are in logarithmic form,facilitating estimation and hypothesis testing.York et al.(2003)and Wei (2011)argued that in the typical application of the STIRPAT model,T should be included in the error term,rather than paratel
y estimated,for consistency with the IPAT model,where T is solved to balance I ,P ,and A .The modi fied STIRPAT model is expresd as follows:ln I ¼ln a þb ln P ðÞþc ln A ðÞþe :
ð3Þ
According to the concept of ecological elasticity (York et al.,2003),
coef ficients b and c from Eq.(3)are the population and af fluence elasticities,respectively.The elasticities refer to the responsiveness or nsitivity of environmental impacts to changes in corresponding impact factors.For instance,coef ficient b indicates percentage change in I in respon to a 1%change in population,with other factors held constant.
To comprehensively obrve the impact of population on carbon emissions,we incorporate the indicators of population structure,including urbanization level,age structure,and houhold size,into the STIRPAT model to come up with the following expanded form:ln I ¼ln a þb s ln Ps ðÞþb c ln Pu ðÞþb a ln Pw ðÞþb f ln Ph ðÞþc ln A ðÞþe ;
ð4Þ
为什么会长老年斑where
–I refers to carbon emissions;–Ps denotes population size;
–Pu ,Pw ,and Ph are the three factors that indicate population structure;that is,Pu for urbanization rate,Pw for the proportion of working age (16–64years old)population,and Ph for houhold size,which is indicated by the average number of houhold members;–A reprents per capita annual expenditure;–e is a residual term.
3.Data description and data testing 3.1.Data description
The population,consumption,and carbon emissions in China from 1978to 2008are summarized in Table 1.Data on carbon emissions from fossil fuels and cement come from the data center of the Carbon Dioxide Information Analysis Center of Oak Ridge National Laboratory,USA (CDIAC,2011).Population and consumption data are obtained from the China Statistical Yearbook,relead by China's National Bureau of Statistics.Expenditure data are adjusted to fit the fixed prices in 2000.
Fig.1shows the changing rates of all the variables,with 1978as the ba year.Almost all the variables were non-stationary,with a continuous uptrend or downtrend during the period.Among all the variables,per capita expenditure prented the fastest growth at 8.17times,followed by carbon emissi
ons (3.72times)and urbanization rate (1.55times).Population size and proportion of working age population incread by 37.96%and 22.35%,respectively.Average houhold size showed a continuous shrinking trend,decreasing by 32.24%over the period.
Taking the logarithm of data can reduce non-stationarity,as well as linearize variables,so that the disadvantage prented by variables having different measurement units is eliminated;thus,all the data ud in the current work are transformed into natural logarithmic ries.
3.2.Stationarity test
The acceptability of a regression result is commonly bad on the premi that the ries ud in the regression model are stationary or co-integrated if the ries are non-stationary;otherwi inauthentic regression may occur.Furthermore,multicollinearity among indepen-dent variables can cau large variances in estimated coef ficients and decrea the accuracy of estimated equations;a multicollinearity test should be performed on independent variables.
The augmented Dickey –Fuller (ADF)unit root test is typically ud to examine the stationarity of time ries,in which a high-order autoregressive model with an intercept term is established (Maddala and Kim,1998).Taking the ADF test on ries ln I as an example,we express the test equati
on with the constant term,as well as the trend and intercept terms,as follows:Δln I t ¼αþβt þδln I t −1þ
X k i ¼1
βi Δln I t −i þεt ;ð5Þ
where α,β,and δare coef ficients;εis a residual term;and k is the lag length,which turns the residual term into a stochastic variable.
The null hypothesis H 0is δ=,at least one unit root exists,causing the non-stationarity of the ries.The test is conducted with three formulations:(α≠0,β≠0),(α=0,β≠0),and (α=0,β=0).As long as one of the three models rejects the null hypothesis,the ries are considered stationary.However,when the results of all the three models do not reject the null hypothesis,the ries are regarded as non-stationary.
The results of the stationary test on all the ries are summarized in Table 2.
According to the results,ries ln Pu ,ln Pw ,ln Ph ,and ln A are I (0)or stationary.Series ln Ps and ln I are I (1),indicating that they are first-order integrated ries.Hence,the co-integration between the tw
o ries must be examined to determine whether they satisfy the precondition of regression analysis.3.3.Co-integration test
Series ln Ps and ln I are both I (1);thus,they satisfy the precondition of the same integrated order for conducting a bivariate co-integration test.On the basis of the Engle –Granger test method (Engle and Granger,1987),we express the co-integration regression equation as ln I t ¼αþβln Ps t þεt :
ð6Þ
Denoting the estimated regression coef ficients of Eq.(8)as ^α大染坊剧情介绍
and ^β,the estimated residual ries is then expresd as follows:^ε¼ln I t −^α−^βln Ps t
:ð7Þ
If ^ε
is I (0),then ln I and ln Ps are co-integrated.Coef ficients ^α
and ^βare estimated by ordinary least squares (OLS),and then the unit root test is performed on estimated residual ries ^ε
using the ADF test method.The results are shown in Table 3.
Table 3shows that the calculated ADF t -statistic of ries ^ε
was −1.8455,which is less than the critical value at the 10%signi ficance level.Hence,the result rejects the null hypothesis,indicating that
ries ^ε
without a unit root is ,^εis I (0).Therefore,ries ln I and ln Ps are co-integrated.
We examine the Granger causality between ries ln I and ln Ps .The bivariant regression models for the Granger causality test are expresd as follows:ln I t ¼α0þ
X k i ¼1
αi ln I t −i þ
wash的过去式X k i ¼1
βi ln Ps t −i ;ð8Þ
3
Q.Zhu,X.Peng /Environmental Impact Asssment Review 36(2012)1–8
ln Ps t ¼α0þ
X k i ¼1
αi ln Ps t −i þ
X k i ¼1
βi ln I t −i :ð9Þ
The null hypothesis is β1=β2=…=βk =0given that the maximal lag length is k =2.The test results are shown in Table 4.
The first hypothesis states that ries ln Ps is not the Granger cau of ries ln I ;the concomitant signi ficance of this hypothesis was 0.1619,suggesting that ln Ps is the Granger cau of ln I ,with 83.81%signi ficance.The concomitant signi ficance for the cond hypothesis was 0.8752,indicating that ln I is not the Granger cau of ln Ps .
3.4.Multicollinearity test
Multicollinearity refers to a situation in which two or more
independent variables in a multiple regression model are highly linearly related (Donald and Robert,1967).In this situation,the standard errors of the affected coef ficients tend to be large,and the coef ficient estimates may change erratically in respon to small changes in data.Such erratic changes result in the possible failure of the regression model to provide valid results on individual variables.
The multicollinearity of the independent variables in the model is examined by OLS regression and by valuing the variance in flation factors (VIFs)of the variables.Taking the test on multicollinearity among ln Ps and the other variables as an example,we u the OLS method to regress ln Ps on the other independent variables.As shown in Table 5,the estimated coef ficient of determination (R 2)of the model was 0.9803and the F -test was highly signi ficant,with an F -statistic of 323.8751at the 0.1%signi ficance level.The VIFs of the variables ranged from 29.6551to 173.5764,which are considerably greater than 10.Given that Marquardt (1970)ud a VIF greater than 10as a guideline for vere multicollinearity,we can conclude that a high degree of
-
100%
0%100%200%300%400%500%600%700%
800%900%19
7819
80
19
821984198619881990199219941996199820002002200420062008
Year
C h a n g i n g R a t e
Fig.1.Changing rates of population,consumption,and carbon emissions in China (1978–2008).Sources:same as in Table 1.
Table 1
Population,consumption,and carbon emissions in China (1978–2008).Year Carbon emissions (MtC)a Population size (104)Urbanization rate (%)Proportion of working age population (%)Houhold size
(person/houhold)Per capita
expenditure (CNY)197840,768.996,25917.9259.50 4.66740197941,648.997,54218.96%60.00% 4.65791198040,698.698,70519.39%60.50% 4.61862198140,292.5100,07220.16%61.00% 4.54934198243,122.8101,65421.13%61.50% 4.51997198345,468.6103,00821.62%62.37% 4.461079198449,433.6104,35723.01%63.24% 4.411207198553,587.3105,85123.71%64.12% 4.331370198656,348.0107,50724.52%64.99% 4.241435198760,123.0109,30025.32%65.86% 4.151520198864,445.3111,02625.81%66.15% 4.0516********,473.6112,70426.21%66.45% 3.971635199065,855.4114,33326.4166.74 3.931695199169,147.7115,82326.9466.30 3.891842199272,143.5117,17127.4666.20 3.852*********,019.8118,51727.9966.70 3.812262199481,807.1119,85028.5166.60 3.782367199588,471.7121,12129.0467.20 3.74255319
9692,597.1122,38930.4867.20 3.722793199791,486.8123,62631.9167.50 3.642919199886,614.1124,76133.3567.60 3.633091199990,501.7125,78634.7867.70 3.583346200092,886.8126,74336.2270.15 3.443632200195,144.0127,62737.6670.40 3.4238552002100,957.7128,45339.0970.30 3.3941252003118,724.4129,22740.5370.40 3.3844152004139,067.5129,98841.7670.92 3.3147732005153,424.4130,75642.9972.04 3.2451422006166,458.9131,44843.9072.32 3.1756362007180,165.9132,12944.9472.53 3.1762392008
192,268.7
132,802
45.68
72.80
3.16
6782
Sources:The carbon emission data are obtained from the CDIAC (2011);the data on population and consumption are from the China Statistical Yearbook,with some interpolation for the missing data on working age population for veral years in the 1980s;the expenditure data are adjusted to fit the fixed prices in 2000.a
MtC refers to million-ton carbon
Table 2
Results of the stationary test using the ADF test.
Variable Difference order Exogenous (α,β,k )t -Statistic Signi ficance level Test critical value Verdict
ln Ps
人的弱点
1(α,β,1)−3.71955%−3.5806I (1)ln Pu 0(0,0,1)−2.94401%−2.6471I (0)ln Pw 0(0,0,1)−3.25631%−2.6471I (0)ln Ph 0(0,0,1)−4.05991% 2.6471I (0)ln A 0(α,β,1)−3.298310%−3.2217I (0)ln I
1
(α,0,4)
−3.2972
5%
−2.9862
I (1)
4Q.Zhu,X.Peng /Environmental Impact Asssment Review 36(2012)1–8
multicollinearity exists among ln Ps and the other independent variables in Eq.(4).
The same multicollinearity test was performed on the other independent variables;all the results indicate a high degree of multicollinearity among the variables.
4.Regression estimation
4.1.Ridge regression
The danger of multicollinearity primarily stems from its generation of
large standard errors among related independent variables;the errors are characterized by large variances in model parameters,making the model unstable.Given that the standard errors are significantly reduced using a curtain method,the negative conquences of such errors can be effectively eliminated even when multicollinearity remains in the model.Ridge regression,which can obtain acceptably biad estimates with smaller mean square errors in independent variables through tradeoffs in bias–variance,is one of the most effective solutions for multicollinearity.
Hoerl and Kennard(1970)explicitly specified the estimation procedure for ridge regression as an improved substitute for traditional OLS regression.Consider the standard model for multiple linear regression,
Y¼Xβþε;ð10Þwhere X is(n×p)and is of rank p,βis(p×1)and unknown,E[ε]=0,
and E[εε′]=δ2I.The unbiad estimate ofβis normally given by
^β¼X′X
ðÞ−1X′Y:ð11Þ
When a high degree of multicollinearity exists among X,the X′X matrix is ,the value
of its determinant|X′X|≈0, and attempts to calculate the(X′X)−1matrix may be highly nsitive to slight variations in data.In controlling the inflation and general instability associated with least squares estimates,as well as in estimatingβ,the ridge regression that incorporates small positive quantity k to the diagonal of normalized independent variable matrix X′X us
^βüX′XþkI
ðÞ−1X′Y:ð12Þ
This equation creates a variance in parameter estimates that is less than that estimated by OLS regression under the condition k≥0.
Therefore,choosing an appropriate k,accepting minimal bias,and substantially reducing variance are possible,thereby remarkably improving estimation.Ridge regression can be converted back to OLS regression as a special ca given that k=0(Hoerl and Kennard,1970).
Considering that the relationship of a ridge estimate to an ordinary estimate is given as
^βüIþk X′X
ðÞ−1
h i−1
^β;ð13Þ
风景速写we can derive the expression for estimating the bias introduced when ^βÃis ud rather than^βas follows:
bias¼Iþk X′X
ðÞ−1
h i−1
:ð14Þ4.2.Estimation results
The ridge traces estimated for the expanded STIRPAT model are shown in Fig.2.The results for all the estimated normalized coefficients are summarized in Table6.
As shown in Fig.2,when k=0.20,the coefficients of the indepen-dent variables tend to be stable.In this situation,the model exhibited a high goodness-of-fit,with an adjusted coefficient of determination
(R2)of0.9454.The F-test of the model was highly significant,with an F-statistic of104.8277at the0.1%significance level.All the estimated coefficients pasd the significance tests with t-statistic at the0.1% significance level.The VIFs of the estimated coefficients ranged from 0.1704to0.4791,all much lower than10.The bias introduced was
Table4
Results of the Granger causality test on ln I and ln Ps.
Null hypothesis:Obs F-statistic Probability
ln Ps is not the Granger cau of ln I29 1.96630.1619
ln I is not the Granger cau of ln Ps0.13400.8752Table5
Multicollinearity test on ln Ps and other independent variables by OLS.
Adjusted R20.9803 Standard error0.0155
F-statistic323.8751⁎⁎⁎ln Ps Coefficient t-Statistic VIF
ln Pu−0.2706⁎−2.4252120.7909
(0.1116)
ln Pw0.3378 1.274929.6551
(0.2650)
ln Ph−0.5381⁎−2.3791100.8291
(0.2262)
ln A0.1383⁎ 2.4129173.5764
(0.0573)
Constant11.1249⁎⁎⁎15.3781
(0.7234)
Standard errors are in parenthes.
⁎⁎⁎p b0.001(two-tailed test).
⁎p b0.05(two-tailed test).
涅盘是什么意思
-1
大盘鸡做法1
2
0.000.050.100.150.200.250.300.350.400.450.50
k
N
o
r
m
a
l
i
z
e
d
c
o
e
f
f
i
c
i
e
n
t
Fig.2.Ridge trace estimated for Eq.(4).
Table3
Results of the unit root test on^ε.
Significance t-Statistic Probability
ADF test statistic−1.84550.0626 Test critical values:1%level−2.6471
5%level−1.9529
10%level−1.61005
Q.Zhu,X.Peng/Environmental Impact Asssment Review36(2012)1–8
0.0979,which is acceptable for the estimates.Thus,the estimation is considered satisfactory,with a robust explanatory power for Eq.(4).
5.Discussion
Table6lists the contributions of the impact factors on carbon emissions in terms of the absolute value of the normalized coefficients. The impact factors are ranked in descending order as follows:per capi
ta expenditure,with a contribution ratio of23.39%;houhold size,with 21.46%;urbanization rate,with20.29%;proportion of working age population,with16.78%;and population size,with12.45%.On the basis of the results,we conclude that the effects of changes in residential consumption and population structure on carbon emissions in China over the studied period exceeded tho of population size.
5.1.Residential consumption
According to the estimated equation,the impact of changes in per capita expenditure on carbon emissions in China was higher than that of the other factors considered in the model.
Table1and Fig.1illustrate that the residential consumption level in China maintained continuous growth from1978to2008.In terms of thefixed prices in2000,the per capita expenditure ro8.17times from CNY740to CNY6782,with an annual average growth rate of7.67%. This rate was higher than tho of the other variables investigated.
As an important indicator of the affluence of residents,consump-tion level affects carbon emissions through two main channels.The first is through direct emissions from houhold energy require-ments,including cooking,hot water u,and heating.The cond is via indirect emissions from non-en
ergy residential consumption goods and rvices,which emit carbon during,rather than after,the production process.The impacts of human behavior on carbon emissions are primarily manifested in production and consumption behaviors.This obrvation indicates that to satisfy consumption demands,people create wealth for society by participating in production activities,leading to inevitable emissions in a specific stage restricted to the level of productivity and resource endowment.In this n,carbon emissions can be considered an indicator of social and economic development in a particular historical period.Hence,a high correlation between consumption level and carbon emissions is expected.
Nevertheless,the impact of changes in consumption structure on carbon emissions should not be disregarded.Marked by an increasing proportion of rvice consumption and a decreasing proportion of product consumption,China exhibited a significantly improved residen-tial consumption pattern during the studied period(Wei et al.,2007). The relationship between consumption level and carbon emissions is not a simple linear correlation becau the carbon emission intensity of each type of residential good or rvice differs,and the improvement in production technology and energy structure constantly varies.The factors partly explain the growth rate of carbon emissions being lower than the consumption level in China during the reviewed period,with the elasticity of per capita expenditure at only0.16.
5.2.Population size
Table1and Fig.1show that from1978to2008,the population incread from0.963billion to1.328billion,which is equivalent to an increa of37.96%.The results of the Granger causality test(Table4) suggest that the logarithmic variable of population size is the Granger cau of carbon emissions,with83.81%significance.The regression estimation of the model shows that the elasticity of carbon emissions in relation to population size from1978to2008was0.55.Compared with the similar global-level elasticities assd by Shi(2003),Cole and Neumayer(2004),and Rosa et al.(2004),the impact of population size on carbon emissions in China during this period was considerably lower than the global average level,despite some differences among the variables and periods investigated in the studies.The lower impact of population size implies that during the period reviewed in the current work,population growth was not the major impact factor.
Population growth does not necessarily result in the inevitable intensification of environmental pressure.The conquences of popu-lation growth continue to be debated.In our opinion,the complexity of the relationship between population and the environment prents difficulties in resolving such controversial issues.First,a clo mutual relationship exists between population growth and environmental ,population growth influences natural resources and the ecos
ystem,and vice versa.Second,the two factors usually interact with each other indirectly through human production and consumption behaviors,which are in turn influenced by social and economic factors, including productive relations,industrial policy,and resource endow-ment.Third,even population growth itlf reflects different structural patterns,manifesting in varying age structures,gender compositions, and geographical distributions.Thus,comprehensively investigating the effects of the changes in population structure,rather than only tho in population size,is necessary.Such studies should also include an analysis of the social and economic factors that affect environmental pressure.The next ction describes the issues in detail.
5.3.Urbanization
Our results reveal that the urbanization of the population was key to the increa in carbon emissions in China,with an urbanization elasticity of0.33.
As shown in Table1and Fig.1,the urbanization rate of China's population ro from17.92%to45.68%in1978to2008,with a yearly average increa of nearly one percentage point.A significant disparity in production and consumption levels exists between urban and rural areas becau of the specialized urban–rural dualistic structure of China. Statistics show that for nearly a d
ecade,the per capita expenditure of urban residents remained3.5times higher than that of rural residents. The rising urbanization rate primarily reflects improving production and consumption levels;urban residents would have been responsible for the large impact on the carbon emissions in China during the studied period.
Nevertheless,urbanization partly alleviates environmental pressure. Through intensive development,urbanization can improve energy u efficiency and pollution treatment through asmbly and scale effects. The effects,in turn,mitigate the scarcity of energy resources and damage to the environment.According to the estimated urbanization
Table6
Ridge regression results of Eq.(4).
k0.2000 Adjusted R20.9454 Standard error0.1067
F-statistic104.8277⁎⁎⁎Bias0.0979
Variable Non-normalized
coefficient Normalized
coefficient
t-Statistic VIF
ln Ps0.5543⁎⁎⁎0.1245 4.53290.4144
(0.1223)
ln Pu0.3334⁎⁎⁎0.20299.53690.2486
(0.0350)
ln Pw 1.3210⁎⁎⁎0.1678 5.67960.4791
(0.2326)
ln Ph−0.7823⁎⁎⁎−0.2146−12.18080.1705
(0.0642)
ln A0.1646⁎⁎⁎0.233913.27840.1704
(0.0124)
Constant 5.5205⁎⁎⁎ 3.7809
(1.4601)
Standard errors are in parenthes.
⁎⁎⁎p b0.001(two-tailed test).
6Q.Zhu,X.Peng/Environmental Impact Asssment Review36(2012)1–8
