INCOME GROWTH AND ATMOSPHERIC POLLUTION IN SPAIN: AN INPUT-OUTPUT APPROACH

Correspondence address: Jordi Roca; Departament de Teoria Econòmica; Universitat de Barcelona; Av. Diagonal, 690; 08034 Barcelona. Spain. Phone: (34) 934021942. Fax: (34) 934039082. E-mail: jordiroca@ub.edu Abstract The relationships between economic growth and environmental pressures are complex. Since the early nineties, the debate on these relationships has been strongly influenced by the Environmental Kuznets Curve hypothesis, which states that during the first stage of economic development environmental pressures increase as per capita income increases, but once a critical turning-point has been reached these pressures diminish as income levels continue to increase. However, to date such a delinking between economic growth and emission levels has not happened for most atmospheric pollutants in Spain. The aim of this paper is to analyse the relationship between income growth and nine atmospheric pollutants in Spain. In order to obtain empirical outcomes for this analysis, we adopt an input-output approach and use NAMEA data for the nine pollutants. First, we undertake a structural decomposition analysis for the period 1995-2000 to estimate the contribution of various factors to changes in the levels of atmospheric emissions. And second, we estimate the emissions associated with the consumption patterns of different groups of households classified according to their level of expenditure


Introduction
The relationships between economic growth and environmental pressures are undoubtedly complex. Economies are in constant evolution as the relative weight of their different economic sectors shift and new technologies are introduced. We cannot, therefore, automatically assume that a given degree of economic growth will result in an equivalent increase in environmental pressures.
Since the early nineties, the debate on the environmental effects of economic growth has been strongly influenced by the Environmental Kuznets Curve (EKC) hypothesis, which states that an inverted U relationship can be found between environmental pressures and per capita income: economic growth initially has negative environmental effects, but once a critical level of per capita income has been reached when environmental degradation affects other individuals -in other countries or those belonging to other generations -the consumer preferences over consumption of private commodities or environmental quality are no longer the main factor. In fact, the more environmental problems affect other individuals, the less likelihood there is of economic growth leading to political decisions that reduce environmental pressures. It is hardly surprising then that the majority of the environmental pressures that contribute to global and long-term problems -such as greenhouse gas emissions -correlate positively with per capita income, even at very high income levels.
The aim of this paper is to analyse the relationship between income growth and atmospheric pollution in Spain. In so doing it examines the emission of nine gases: the six greenhouse gases -carbon dioxide (CO2), methane (CH4), nitrous oxide (N2O), sulphur hexafluoride (SF6), hydrofluorocarbons (HFCs) and perfluorocarbons (PFCs)and three gases associated with local and regional environmental problems -sulphur oxides (SOx measured in units of SO2 equivalent), nitrogen oxides (NOx) and ammonia (NH3). We adopt two distinct, but complementary, perspectives: a longitudinal study together with a cross-section analysis. First, we conduct a structural decomposition analysis for the period 1995-2000 in order to examine the contribution of a range of factors involved in economic growth to the evolution of atmospheric emissions.
Unfortunately, we do not have at ours disposal data to undertake this analysis for a longer period. Second, we analyse the emissions associated with the consumption patterns of different household groups based on their levels of expenditure. Both approaches are particularly pertinent to the EKC debate as they include significant elements for estimating the dimension of key factors in the EKC hypothesis. Having said that though, this paper does not seek to test the existence of an EKC in Spain since the SDA is conducted over a very short period of time and our study of household emissions is a comparative static analysis.
The importance of this study lies in the fact that, as far as our knowledge, this is the first analysis of environmental pressures and household consumption patterns and the first structural decomposition analysis to be undertaken with economic and environmental data from Spain. Similarly, while the structural decomposition analysis reported for other countries have tended to examine CO2 emissions only, here we consider several gases.
The rest of the paper is organised as follows. In section 2, we present a brief overview of the recent evolution in atmospheric pollution in Spain. In section 3, we describe the National Accounting Matrix including Environmental Accounts framework and the input-output approach. In section 4, we present the structural decomposition analysis outcome and, in section 5, we analyse the emissions associated with household consumption patterns. Finally, in section 6 we present and discuss our conclusions.

Atmospheric pollution in Spain. A global perspective
In this section, we describe atmospheric pollutant emission trends in Spain, extending the analysis undertaken by Roca et al. (2001) and Roca and Padilla (2003).2 Figure 1 shows the overall trend in the emission of the six gases (measured in CO2-equivalent tonnes) regulated by the Kyoto protocol for the period 1990 to 2004. 2 The data used in this section include process emissions as well as energy emissions. Data for 1990Data for -2004 are drawn from the Banco Público de Indicadores Ambientales of the Spanish Ministry of Environment.
For the period 1980-1990 we use data provided directly by the Spanish Ministry of Environment. The latter have not been officially revised. Our data series (in which 1980 acts as base=100) establishes a link between the 1980-1990 and 1990-2004 series. The European Union (EU) as a "bubble" has undertaken to keep the 2008-2012 average emission of these six greenhouse gases to a level that is 8% lower than that of the base year considered, i.e. 1990.3 Yet, Spain, with per capita emissions lower than the EU average, was granted permission to increase its emissions by 15%, while other countries found themselves in a position of having to achieve reductions that greatly exceeded 8%. However, Spain has greatly exceeded this accepted level.
In the EKC debate, though, it might well be argued that the data to be analysed should not be those describing total emissions but rather those that describe per capita emissions. Moreover, the EKC hypothesis argues that the supposed reduction in environmental pressures is accounted for by the changes involved in economic growth rather than by the simple passing of time. In Figure 2 we can see the relationship for the years 1980 to 2003 between "real" per capita income and per capita emissions4 of the three main greenhouse gases -CO2, CH4 and N2O -5 and also of three other atmospheric 3 The undertaking refers to the aggregation of the six gases, which are summed in accordance with their global warming potential values as established by the IPCC. The conversion factors are: 1 for CO2, 21 for CH4, 310 for N2O and 23,900 for SF6. For the PFC group, values oscillate between 6,500 and 9,200 depending on the gas in question, while for the HFC group, values range between 140 and 11,700. The chlorofluorocarbons (CFCs) were not included in the Kyoto protocol as they had been regulated by an earlier international agreement, the Montreal Protocol. For HFCs, PFCs and SF6, 1995 can be taken as the base year. pollutants -SO2, NOx and NH3 -associated with local and regional environmental problems including acidification, eutrophication and troposphere ozone concentration.
Figure 2 reveals quite distinct trends. Only SO2 per capita emissions fell as the EKC hypothesis would lead us to expect. The trend in N2O per capita emissions, meanwhile, is unclear. For the remaining gases, emissions increased considerably and there was no evidence of any change in this trend. NOx and NH3 emissions increased significantly but at a rate that was lower than that of GDP, i.e. indicating relative delinking but not absolute delinking. Throughout most of the period, CO2 emissions increased roughly in line with GDP or even at a faster rate. The one exception to this was during the eighties, when the use of nuclear energy increased in Spain. This is a good example, perhaps, of how one environmental indicator improves to the detriment of another -in this case, increased nuclear risks. CH4 emissions increased the most, in particular during the eighties and this is mostly due to emissions from waste management.
We are therefore drawn to the conclusion that the evolution in the emission of gases follows a range of different paths and that these, in general, are not an invitation for optimism. Clearly this issue need to be analysed more fully. But at this aggregate level of analysis, it is not possible to further our understanding of the factors that might account for these differences. Consequently, the relationship between income growth and atmospheric pollution in Spain needs to be examined in greater depth.
In the sections that follow, we adopt two distinct, but complementary, approaches that should contribute to the EKC debate. First, we conduct a structural decomposition analysis to estimate the contribution of several factors to the evolution of atmospheric emissions. Second, we analyse how the consumption patterns of different household groups classified according to their levels of expenditure contribute to these atmospheric emissions.

Data base and methodology
This section describes the National Accounting Matrix including Environmental Accounts (NAMEA) system and the methodological approach that is adopted in the rest of this study, namely, that of input-output (IO) analysis. It also explains the procedures and data preparation required in applying these approaches.

NAMEA system
In the early nineties, Statistics Netherlands (CBS) developed the NAMEA, which was subsequently adopted by EU countries within the EUROSTAT environmental accounting project (Keuning et al., 1999). In this framework, environmental information is compiled so that it is compatible with the presentation of economic activities in national accounts. In this way, the national accounting matrix (NAM) can be extended to include environmental accounts (EA), usually expressed in physical units.
The System of Economic and Environmental Accounts (SEEA) considers two types of NAMEA accounts: hybrid supply and use tables (HSUT) and hybrid inputoutput tables (HIOT).6 The former consist of a pair of tables, one showing those 6 The term hybrid accounts indicates that monetary and physical data are included in the same accounting framework, and at the same time differentiates them from the physical IO accounts (see Hoekstra and Van den Bergh, 2006). Elsewhere, this term is sometimes applied to "energy IO tables" in which certain flows between economic units are expressed in energy units rather than in monetary units (Casler and Willbur, 1984). Moreover, in the literature the HSUT are also referred to as hybrid make and use tables.
industries that supply commodities (supply table), the other showing economic units that use them (use table). In this case, two different classifications are used for industries (NACE) and commodities (CPA) respectively. In the second method, a symmetric IO table results from the transformation of the supply and use tables so that each industry represents one particular homogeneous type of good or service. However, neither the new SEEA nor the EUROSTAT makes any explicit recommendation as to which NAMEA type accounts should be used, rather EU countries are merely encouraged to do the most they can given the IO framework available for domestic use.
In the case of Spain, the NAMEA system is organised in accordance with the In the Spanish HSUT, emissions are allocated to heterogeneous industries, since they need to be attributed in a way that is consistent with economic data. This has significant consequences for the interpretation of environmental information. For 7 Transport emissions are allocated to households only when they are emissions from private cars and motorbikes. Heating emissions are allocated to households in the case of fuels used domestically.
instance, emissions associated with electricity production as an ancillary or secondary activity are, nevertheless, allocated to the particular industry that undertakes this production according its principal activity and not to NACE 40.1 (Production and distribution of electricity). The same principle holds true for transport emissions, which are allocated to the economic agents that perform the activities that generate the emissions.
In addition, in line with the NAMEA framework and national accounting principles, air emissions due to incineration and the decomposition of waste in landfills (principally CH4) are included within NACE 90 (Sewage and refuse disposal, sanitation and similar activities). However, such emissions might be considered separately from industry and household emissions. In this paper, in line with the Dutch NAMEA experience (Keuning et al., 1999), we distinguish three sources of atmospheric pollutants: "industries", "households" and "other sources", and include CH4 emissions from waste management in this final category.9 Although the supply and use table is more readily associated with data from other areas, such as environmental information, the symmetric IO table offers greater analytical power. Both the current European System of Accounts 1995 and the new SEEA agree that the HIOT is more appropriate than the HSUT for conducting analytical research, particularly when it is used to calculate indirect effects. Therefore, in order to exploit to the full the advantages offered by NAMEA and IO analysis, we first need to make a number of transformations to the Spanish NAMEA. These we describe below. 9 In the Spanish NAMEA the emissions of the NACE 90 are aggregated together with NACE 91 (Activities of membership organization), 92 (Recreational, cultural and sporting activities) and 93 (Other service activities) under the heading "Other community, social and personal service activities". Since no more disaggregated data are available and the majority are NACE 90 emissions, all the CH4 emissions from these four sectors have been classified as "other sources".

Input-Output analysis
IO analysis provides a framework for considering specific questions about the relationship between economic structure and economic activity and so opens up a path for the study not only of economic production but also of the effects of production and consumption on the physical environment. In the early 1970s, Wassily Leontief himself and other authors extended the IO model to consider some links between the economy and the environment (Leontief, 1970), in particular atmospheric pollution (Leontief and Ford, 1972).
Formally, for an economy of n sectors the standard IO model is represented by the following expression: or as a function of final demand as: It should be noted that in expression (3)  to which all products from one industry are assumed to be produced with the same technology.12 In common with the standard IO model, the environmental IO model helps identify those that contribute directly to the emission of pollutants, while highlighting the indirect role played by intermediate consumption. Therefore, the more an industry uses products whose production is pollution intensive, the greater will be the pollution generated indirectly to satisfy final demand. Tables 1 and 2 present the Spanish industries with the greatest total emission intensities in 2000. More specifically, we show only those that have an emission multiplier which is more than twice the mean of the economy. The ranking of sectors for greenhouse gas emissions are presented in Table 1, while Table 2 shows the rankings for the other three gases considered in this article.
Tables 1 and 2 show how the expenditure of one monetary unit in the purchase of a range of different goods and services -classified by sectors or industries -can have very different implications in terms of the quantity and type of emissions. For instance, one euro spent in the "Electricity, gas, steam and hot water supply" sector (mainly electricity production) was found to generate much higher emissions of CO2 and SO2 than the same euro spent in any other sector. For these two pollutants, the manufacture of mineral and refined petroleum products also gave rise to a high intensity of emissions. The emission intensity of NOx was also considerable in the "Electricity, gas, steam and hot water supply" sector and even higher in fishing and water transport. As expected, the activities that generated the highest levels of CH4, N2O and NH3 emissions were those related to the agricultural sector and the manufacture of food products. Meanwhile, SF6, HFCs and PFCs emissions were closely linked with specific manufacturing activities: SF6 with the manufacture of machinery, electrical and optical equipment; HFCs with the manufacture of chemicals, rubber and plastic products; and PFCs with aluminium production (included in the manufacture of basic metals).

A longitudinal perspective: structural decomposition analysis
Since the late seventies, energy and environmental analyses have increasingly used decomposition analysis techniques to study the contribution of a range of factors to energy use and environmental pressures. Indeed, Ang and Zhang (2000) listed more than a hundred studies, some of which adopted an IO methodology; in this latter case, the name commonly used to refer to the decomposition analysis we undertake is structural decomposition analysis (SDA) (Hoekstra, 2005).
The purpose of SDA is to break down the variation of an aggregate variable to reveal the contribution of different effects. In this section, we conduct an SDA in order to analyse the evolution in Spain of the atmospheric pollutants considered in section 2i.e., six greenhouse gases (CO2, CH4, N2O, SF6, HFCs and PFCs) and three other air pollutants (SO2, NOx and NH3).13 Given i' as a row vector of n ones, the total emissions of an economy in any period t can be expressed as:14 In this expression, final demand is divided into a structure component ( s t y ) and a volume component ( v t y ). Thereby, the decomposition of the change in emissions between two periods is given by: In (5)  ). This decomposition enables us to analyse whether the two main factors which underpin the EKC hypothesis -technological change and final demand structuretends to reduce emissions or not. If they do, this analysis should also enable us to determine whether they were of sufficient weight to counteract the effect of economic growth.
As discussed in the SDA literature, several techniques might be adopted for decomposing the total emission variation into its different factors. Arguably, the most "intuitive" method involves calculating the contribution of each factor by simulating the effects resulting from changes in one factor while the initial values of the other factors are held invariable. However, this approach -"Laspeyres" approach -does not provide us with a complete decomposition as the total change in emissions does not coincide with the sum of the different effects. This difference is known as the residual or interaction term, which might be very high when the different factors change considerably. For this reason many studies choose to apply other decomposition methods in an attempt at reducing or eliminating this interaction term. One alternative is to calculate the effects as the average of the "Laspeyres" approach and the "Paasche" approach (see e.g. Wier and Hasler, 1999); however, while this method reduces the interaction effect, it is not entirely eliminated. Another alternative involves, first, calculating the effects with the Laspeyres approach and then sharing out the interaction term among the different effects in line with the "jointly created and equally 17 In order to compare this variation with that reported elsewhere we computed the relative weight of the changes in V and in the Leontief inverse. For the majority of gases the changes in the Leontief inverse were significant but still much smaller than the changes in V.
The latter, however, is not the only possibility for obtaining a complete decomposition. Dietzenbacher and Los (1998) show that when three effects are considered, there might be six different exact decomposition forms (3! or in general n! where n is the number of effects considered). They claim that all these possible forms are "equivalent, in the sense that no form is to be preferred on theoretical grounds to the others" (p. 314). They also show that outcomes of the different forms can differ greatly.
Hence, the best option seems to involve calculating the average effects of all these six exact decomposition forms. This average results in exactly the same outcome as Sun's proposal (Hoekstra, 2005;p.141). In this section, we adopt this "refined Laspeyres method", which can be expressed as:  Alcántara and Roca (1995) examines energy use and CO2 emissions in Spain between 1980 and 1990 using energy balances and an IO perspective to approximate the primary energy required -and the associated emissions of CO2 -to provide the different forms of final energy and to distribute the primary energy into three uses: "economic sectors", transport and residential use. An extension of this analysis was undertaken in Alcántara and Roca (2004). While other IO studies analyse energy and CO2 emissions in Spain, all of them have other approaches (see, e.g. Manresa et al., 1998;Labandeira and Labeaga, 2002;and Alcántara and Padilla, 2003). that our analysis only takes into account domestic and imported "productive" emissions and that neither direct emissions from households nor CH4 emissions from waste management are considered.19 As can be seen in Table 3, the emissions we do consider are the most significant, accounting for over 95 per cent of the total emissions of the economy for most of the gases with the exceptions of CO2, NOx and CH4.20 Table 4 shows the outcome of the SDA for Spain and for the period 1995-2000 using the method proposed by Sun (1998). In order to avoid the influence of price variations when analysing changes over time, IO tables must be expressed in constant prices. However, as the Spanish NAMEA data are only provided in current prices, we were obliged to deflate the 2000 IO table to 1995 constant prices applying the biproportional projection method proposed by Dietzenbacher and Hoen (1998).21 Obviously, in all cases the volume effect acted in the same direction and resulted in increased emissions. For the majority of gases, we can concludein line with other studies -that the technological effect was also significant, causing emissions to fall.
However, this effect was only strong enough to counteract that of volume in the case of SO2. In the cases of CO2, CH4, N2O and NOx the technological effect was significant but much less so than the volume effect. We also found exceptions to the beneficial effects of technology: the cases of NH3 and, in particular, of the group of synthetic 19 See section 3.1. In the case of CH4 more than 90% of "direct" emissions are due to waste management. 20 One could expect a share of direct CO2 and NOx emissions in total emissions larger than the share that Table 3 reports. One of the reasons that could explain these low relative values is that "emissions from industries" include emissions linked to the production of imported commodities.

A cross-section perspective: household consumption pattern analysis
In recent years, there has been an increasing interest in measuring the environmental effects of household consumption patterns. This involves studying the relative responsibility of different household-types for generating certain environmental pressures. Herendeen and Tanaka (1976) and Herendeen et al. (1981) are seminal works examining the "energy cost of living" for different types of household in the USA.
These studies take into account not only the direct demand for energy products but, We are concerned here with breaking down total emissions from household consumption by different household in order to study a question posed directly by the ECK debate: how do emissions change as households become wealthier and spend more money.28 Households are, therefore, classified according to their level of expenditure.
However, we should point out two aspects concerning such a classification. First, it might be argued that it would be more appropriate to consider the income rather than the expenditure variable; nevertheless, we have chosen to use the latter for two reasons.
The first reason is that the source we have used -i.e. ECPF -provides more complete and reliable data on expenditure than on income. The second reason is that linking income and emissions taking into account only consumption expenditures could be interpreted as supposing that savings do not result in emissions when in fact investment can be as environmentally problematic as consumption, or even more so.
Second, household are different in size and composition. Thus, a decision has to be taken as to whether it is better to work with total household expenditure or to apply some type of transformation in order to calculate the "equivalent expenditure". In this paper we adjust the data from each household in accordance with the "modified OECD scale"29 (Wier et al., 2001).
28Clearly, income (and expenditure) levels are not the only factors influencing lifestyle. In order to consider other factors, alternative perspectives need to be adopted such as the multivariate econometric approach (Lenzen et al., 2006) or household classifications compiled on the basis of several characteristics, e.g. Duchin (1998) classifies United States households using 40 "geo-demographic lifestyle clusters".
Our main findings for the different gases are organized as follows. We include graphs showing: 1) average equivalent emissions for the different household types ordered by equivalent expenditure deciles (Figures 3 and 5); and 2) average emissions intensity -i.e., total emissions divided by total expenditure -for the different household types ordered by equivalent expenditure deciles (Figures 4 and 6). Furthermore, as a synthetic quantitative indicator we show the expenditure elasticity of emissions using microdata of 9,628 different households (Tables 5 and 6); in this case, we also present outcomes using non-corrected expenditure data, i.e. total expenditure elasticity.30 The graphs and the quantitative indicator are directly connected: an increasing function in Graph 1) means a positive elasticity; moreover, the elasticity will be higher or lower respectively than one if the function in Graph 2) is increasing or decreasing.
For each gas, the elasticity is defined according to the equation: where E means total household emissions and K means household expenditure. The subsequent estimation is based on an application of the ordinary least-squares method to: For all the pollutants, emissions increased monotonically with household expenditure (Figures 3 and 5), and no turning point was recorded. However, if we analyse the evolution in emission intensity (Figures 4 and 6), we observe that in general the amount of pollutants emitted per unit of household consumption decreased with 30 To date most studies have not corrected their data so as to take into account the demographic characteristics of the households. This presentation, therefore, ensures that our outcomes can be more easily compared with those of other studies. expenditure level; the exception to this was the greenhouse synthetic gases: SF6, HFCs and PFCs. The most significant -albeit moderate -decrease was reported for those pollutants closely associated with agriculture and cattle raising -CH4, N2O and NH3, which is unsurprising given that one of the few consumption "laws" is that the proportion of money spent on food decreases with the level of expenditure.
The elasticity values31 oscillated from 0.71 to more than 1 when using equivalent data according to the modified OECD scale. These values were even higher when using uncorrected data, with the exception of the synthetic greenhouse gases (Tables 5 and 6).
Technical changes (autonomous or induced by environmental policy) could act in the opposite direction but always these outcomes suggest that further increases in income and expenditure levels should lead to a rise in the pollution generated by private consumption.
As discussed above, studies conducted in other countries have similarly analysed expenditure elasticity for energy or CO2 emissions; however, to our knowledge this is the first to examine other atmospheric pollutants. Given the strong relationship between energy requirements and associated CO2 emissions we can compare our elasticity outcome for CO2 -i.e., 0.91 -with the expenditure elasticity for energy obtained for several countries in a recent work by Lenzen et al. (2006)32. They report values which 31 We are assuming that one euro spent on one type of product will result in the same amount and type of pollution as another euro spent on the same type of product. Yet, Vringer and Blok (1995) stand: "However, it is conceivable that households with a higher income (or a higher expenditure level) systematically buy products that cost more per physical unit. The consequence of this is that the real elasticity of the energy requirement related to income (or expenditure level) can be smaller than the value computed here" (p. 901). range from 0.64 (Japan) to 1 (Brazil) with values of 0.78 for Australia and 0.86 for Denmark and India. Thus, our result lies within the range of those reported in this work, and the increase in emissions with increasing expenditure can be considered particularly high.

Conclusion
In this paper we have analysed atmospheric pollution in Spain, taking into consideration nine different gases. Our main interest has been in determining whether any evidence can be found for a delinking between income growth and atmospheric emissions, as the EKC hypothesis implies for rich countries. Using Spanish NAMEA data and an IO analysis, we have adopted two approaches. Firstly, we have undertaken an SDA for the period 1995-2000; secondly, we have conducted a cross-section study for 2000 to evaluate the atmospheric pollution associated with the consumption of different household types, classified according to their equivalent levels of expenditure.
For some pollutants -NH3, CH4 and N2O -both approaches provide some evidence that income growth is associated with a reduction in the intensity of gas emissions. This reflects changes in final demand structure and, in particular, a relative decrease in demand for food products. However, this trend only accounts for a weak relative delinking between economic growth and emissions and can by no means be interpreted as an absolute delinking. By contrast, the "new" greenhouse gases are mainly associated with manufactured products with an increasing weight in the consumption of wealthier households. For other pollutants -including the main greenhouse gas, CO2 -we did not find in our SDA any change in the final demand we consider the expenditure elasticity of CO2 emissions directly, our estimate is even higher, i.e. 0.99 (see Table 5).
structure that might lead to a reduction in emissions and, moreover, our estimate of household expenditure elasticity presented a value very near to one.
Both the cross section analysis and the estimation of the effect of final demand structure on SDA are alternative methods for determining the role of one of the factors that explains the relationship between income growth and emissions. We would expect similar outcomes from both approaches, but the outcomes will not be identical for

Manufacture of machinery and equipment 235
Source: Own elaboration from2000 Spanish NAMEA.