Delinking and environmental Kuznets curves for waste indicators in Europe

Abstract

The paper provides preliminary empirical evidence on delinking trends concerning waste indicators in Europe. First, methodological issues regarding the analysis of delinking are discussed, and the related Environmental Kuznets Curves literature is critically examined. Then, European-level data on municipal and packaging waste are investigated by panel data approaches, examining different EKC specifications. For packaging and municipal waste flows, decoupling seems to occur, at best, only on a relative basis. No significant evidence in favour of an inverted U-shape Kuznets curve is found. Europe as a whole seems to be still lagging behind in reaching the critical turning point of the relationship between waste and household consumption. Our results suggest some possible advantages and limitations of panel data for a set of relatively homogenous countries as well as the possible added value of further delinking analyses for specific waste materials and/or single countries. Although preliminary, depending on available data, results suggest that waste prevention, claimed as a priority by EU waste policy, has not been successfully pursued by European and national policies in practice.

Keywords:

• Decoupling
• waste indicators
• economic drivers
• Kuznets curves
• environmental efficiency

1. Introduction

Indicators of ‘decoupling’ or ‘delinking’, that is improvements in environmental/resource indicators with respect to economic activity indicators, are increasingly used to evaluate progresses in the use of natural and environmental resources. OECD is carrying out extensive work on decoupling indicators for reporting and policy evaluation purposes (OECD 2002).

Various decoupling or resource efficiency indicators are included in the European Environment Agency's state-of-the-environment reports (EEA 2003c). A few European countries started to include delinking-oriented indicators in official analyses of environmental performance (DEFRA/DTI 2003). Some countries are considering delinking-based targets for major environmental policies, and the US adopted an ‘emission-intensity’ target for their climate policy.

Delinking trends have been under scrutiny for decades for industrial materials and energy in advanced countries. 1 In the 1990s, research on delinking extended to air pollution and GHG emissions, also proposing ‘stylized facts’ on the relationship between pollution and economic growth named as the ‘Environmental Kuznets Curve’ (EKC), due to their similarity with Kuznets's (1955) suggestions on long-run income distribution paths.2 The EKC hypothesis is the natural extension of delinking analysis. The hypothesis is shortly that for many pollutants, an inverted U-shaped relationship between per capita income and pollution is documented. The hypothesis does not originally stem from a theoretical model, but it has followed conceptual intuition, though recent contributions explore the extent to which the Environmental Kuznets hypothesis may be included in formalized economic models.3 Despite increasing applied research efforts, empirical evidence from EKC on emissions, however, is still ambiguous. Some pollutants, mainly associated with a regional/local impact, seem to show a ‘turning point’ at certain levels of income, but it is a shared view that some critical externalities, like CO₂ emissions and waste flows, are monotonically rising with income. At best, a ‘relative delinking’ may take place (Stern 2004).4

Research on delinking and EKC for materials and waste is less well developed compared with pollution and GHG emissions. Although works by Wuppertal Institut and Eurostat are filling the gap for material flows indicators, 5 the still limited research results for the waste sector represent a problem in a policy perspective. The EU ‘thematic strategies’ on both resources and waste entail the reference to ‘absolute’ and ‘relative’ delinking-based indicators (European Commission 2003a,b). Since a decreasing ratio of a material input with respect to an economic driver would suggest a decreasing future production of waste, delinking across the material-to-waste chain can be interpreted in terms of ‘prevention’, which represents the stated priority of the EU waste policy strategy, also transposed in national waste legislations. Following OECD (2002, 2003), waste-prevention activities and policies could be monitored and evaluated through ‘absolute’ or ‘relative’ decoupling indicators by addressing, in particular, the trends towards reduction at source and reuse.6

Our contribution is a preliminary attempt to fill the gap by developing a quantitative analysis of delinking for two major waste flows, municipal and packaging waste, at the European level. Results indicate that European countries, characterized by high income levels and by a relatively long history on waste policies, are at best experiencing relative delinking, with waste indicators increasing only slightly less than economic drivers. The elasticity of two major non-hazardous waste flows considered with respect to consumption is, at the European level, not significantly different from unity.

The paper is structured as follows. Section 2 discusses some conceptual and methodological issues of delinking and EKC analysis, and the suggestions from the EKC empirical literature that are useful for analyses on waste. Section 3 presents the datasets on municipal and packaging waste and discusses key issues regarding EKC econometric analysis. Section 4 presents main empirical results deriving from panel data analysis of different EKC specifications for waste. We finally draw some conclusions and policy implications.

2 Methodological framework

2.1 Defining a proper use of delinking and EKC analyses

Relationships between ‘delinking’ and EKC approaches, and some limitations of both, can be discussed in the framework of a simple IPAT model. 7 The latter defines total impact I (waste production) as the (multiplicative) result of the impacts of population level (P), ‘affluence’ (A), measured by GDP per capita, and the impact per unit of economic activity, i.e. I/GDP, representing the ‘technology’ of the system (T), so I = P • A • T. This is an accounting identity suitable for decomposition exercises aimed at identifying the relative role of A, P, and T for the observed change of I over time and/or across countries.8

While the meaning of P and A as drivers of I is clear, the exact meaning of T deserves attention. This is an ‘intensity’ indicator, which measures how many units of Impact (natural-resource consumption) are required by an economic system for ‘producing’ one unit (one dollar) of GDP. As a technical coefficient representing the ‘resource-use efficiency’ of the system (or, if its reciprocal GDP/I is taken, its ‘resource productivity’ in terms of GDP), it is the most aggregate way for representing the average ‘state of the technology’ of an economy in terms of the Impact variable. Changes of T, for a given GDP, reflect a combination of shifts towards sectors with a different resource intensity (from manufacturing to services) and the adoption/diffusion, in a given economic structure, of techniques with a different resource requirements (e.g. inter-fuel substitution in manufacturing). If T decreases over time, there is a gain in environmental efficiency, or resource productivity, and T can be directly looked at for delinking analysis. By being responsive to changes in ‘state of technology’ and influences by markets and policy actions, T is the main ‘control variable’ of the system. In a cross-country setting, T has a less clear-cut interpretation, but delinking can emerge again as a negative relationship between I and the level of GDP or GDP/P.

In an IPAT framework, three aspects of ‘delinking analysis’ and ‘EKC analysis’ emerge more easily. First, delinking analysis or the observation of T alone can provide ambiguous suggestions. A decrease in the variable I over time is commonly defined as ‘absolute decoupling’, even though a decrease in I does not, in itself, represent a delinking process as it says nothing about the role of economic drivers. An environmental impact growing less (or diminishing more) than economic drivers, i.e. a decrease in T, is generally defined as ‘relative delinking’. Therefore, ‘relative delinking’ might be strong, while ‘absolute delinking’ might not take place (i.e. I is stable or increasing) if the increasing efficiency is not sufficient to compensate for the ‘scale effect’ of other drivers. Even the opposite can be true in certain phases. Assuming a stable Population, a negative GDP growth may push down the Impact variable (‘absolute delinking’), as it has been the case with most ex-communist European countries in the 1990s, while T, the Impact per unit of GDP, might be non-diminishing (no ‘relative delinking’) or might increase (‘relative re-linking’) because the I-specific ‘state of technology’ is stationary or worsening. Therefore, a diminishing I is always a positive environmental signal, but it might not be the result of structural gains in the I-relevant technological efficiency, whereas delinking, i.e. a diminishing T, is always the result of structural or technological change but it cannot allow us to claim the environment is improving unless the change in I is considered. Only a joint analysis of I and T dynamics can provide sound suggestions on environmental performances.

Second, a delinking process, i.e. a decreasing T, suggests that the economy is more efficient but, in itself, does not provide explanations on what is driving the process. In its basic accounting formulation, the IPAT framework implicitly assumes that the drivers are all independent variables. However, the evidence on dynamics of economic systems suggests that each driver as well as the Impact can be reciprocally interdependent through a network of direct/indirect causation. For example, evidence suggests that population dynamics (P) depends on GDP per capita (A), and vice versa to some extent. 9 Similar relationships or inverse-causation effects are also relevant for T. Theory and evidence suggest that T can, in general, depend on GDP or GDP/P, and vice versa if T refers to a key resource as energy, but also a relationship between changes of P and I and T dynamics can be highlighted (Zoboli 1996). In particular, in a dynamic setting, I can be a driver of T as the emergence of natural resource/environmental scarcity stimulates invention, innovation, and diffusion of more efficient technologies through market mechanism (changes of relative prices) and policy actions, including price- and quantity-based ‘economic instruments’. The re-discovery of the Hicksian ‘induced innovation’ hypothesis represents the attempt to capture the channels by which I influences T, while models including ‘endogenous technological change’ can capture some influences of both I and GDP on T. In fact, improvements of T for a specific I can also stem from general techno-economic changes, e.g. ‘dematerialization’ associated with ICT diffusion, which are not captured by resource-specific ‘induced innovations’ mechanisms and can be very different for given levels of GDP/P because of the different innovativeness of similar countries. Then, a decrease in T can summarize micro and macro non-deterministic processes also involving dynamic feedbacks, on which economics has provided a still open set of interpretations.10

Third, the EKC analysis addresses exactly the above relationship, i.e. between I and GDP or between T and GDP/P. It presents ‘benefits’ and ‘costs’. Even though it may provide empirical regularities having great heuristic value, it may not provide satisfactory economic explanations. 11 We recall that the EKC hypothesis is that the concentration/emission of a pollutant first increases with the economic driver, as a ‘scale effect’ prevails, then starts to decrease more or less proportionally, thus delinking itself from income due to a steady improvement of T. More specifically, the hypothesis predicts that the ‘environmental income elasticity’ decreases monotonically with income, and that it eventually changes its sign from positive to negative, thus defining a turning point for the inverted U-shaped relationship. We do not address here the very different meaning of the various formulations of the EKC hypothesis, which range from a relationship between I and GDP to a relationship between T (I/GDP) and GDP/P. Let us simply note that if the relationship is between I and GDP, the EKC provides the same information as the analysis of T. Furthermore, if there is an EKC between I and GDP, there should also be one between T and GDP because both P and GDP are, with some exceptions, increasing over the long run, and delinking must have occurred at some level of GDP. Instead, if there is an EKC between T and GDP or GDP/P, there is not necessarily also one between I and GDP, because GDP and P might have pushed I more than the ‘relative decoupling’, i.e. decreasing T, has been able to compensate for. The latter is the case of global CO₂ emissions in the very long run (see, for illustrative purposes, Figures 1 and 2). When relying on GDP or GDP/P as the only explaining variable, EKC suffers from the same issues highlighted above for delinking analysis, but with an additional risk. The existence of an EKC could give the wrong deterministic suggestion that a rapid growth towards high levels of GDP/P automatically drives to environmental efficiency, i.e. ‘absolute’ or ‘relative’ delinking, and then it can be the ‘best policy strategy’ to reduce environmental Impact. But, from the IPAT framework, it is clear that GDP or GDP/P growth by itself also implies a ‘scale effect’ on I, i.e. a growth of the Impact at each level of T (and P). In general, only if the negative effect of GDP/P on T is steadily higher than the positive effect of GDP/P on I can the process of economic growth lead to a negative change in I, leaving aside the effect of population.12 Inter alia, this is particularly relevant for global energy consumption and GHG emissions with a rapid economic and population growth in developing countries. The negative elasticity of T to GDP/P growth must be extremely high in developing countries in the future, given their stationary or still increasing T, in order to avoid a possible ‘scale catastrophe’ from GDP/P and P growth.

Figure 1. World CO₂ emissions (million tons of C) and world GDP level (million constant 1990 international Geary-Khamis dollars).

Figure 2. CO₂ emission intensity (fossil fuels) of world GDP and world GDP level (million tons of C per million international constant 1990 Geary-Khamis dollars, select years indicated). Sources: CO₂ emissions from fossil fuels from CDIAC, www.cdiac.org; data on GDP from www.theworldeconomy.org. Available data on total world real GDP are Angus Maddison's point estimates for 1870, 1900, 1913, and time series for 1950 – 2000. Data in the intervals 1871 – 1899 and 1901 – 1912 are our extrapolations assuming a constant average growth rate between the two available years. Data for world GDP in 1914 – 1949 are our estimates. We assumed the world GDP to be proportional, in each year of the interval, to the total GDP of a set of 44 countries in the Maddison's dataset representing 68% of world GDP in 1913 and 71% in 1950 (the same countries represent between 68 – 71% of world GDP also during the whole 1950 – 2000 period).

2.2 Estimating EKCs: Key issues

At the econometric level, the aim of EKC analysis is to estimate a vector of coefficients, each linked to a single variable entering as driver of the environmental index, by using a simple reduced form equation as conceptual model. EKC issues will be briefly commented on, with reference to the extensive literature developed over the last decade. 13 The focus is twofold: first, we suggest that the EKC framework is, under certain circumstances, a necessary step forward from the simpler decoupling analysis. Multivariate investigations add robustness to results. Nevertheless, the potential weaknesses of the EKC analysis will be thoroughly highlighted.

The EKC framework extends the basic decoupling reasoning, modelling a multivariate analysis of the environment – income relationship. Even if EKC does not rely on a specific economic model, many theoretical assumptions, on the consumption and production sides, are implicitly tested within the EKC empirical context. The main economic hypotheses revolving around the EKC setting are: (i) among the ‘negative effects’ of income increase, a typical scale effect, and (ii) among the ‘positive effects’, a composition effect concerning economic activities, a technological effect, a preference-drive effect (environment being a normal/luxury good), a market-instruments driven effect (which is integrated with the wider policy effect).

Knowing the benefits of an EKC multivariate econometric-based analysis, we have to be fully aware of the costs and then try pragmatic ways for mitigating them. It is necessary to draw out what the main EKC deficiencies and weaknesses are. Thus, extending the reasoning to a more complex setting has, quite obviously, costs and benefits.

We do not specifically focus on the more statistically oriented key issues (potential weaknesses), like (i) differences in estimated coefficients between parametric and non parametric models (Baiocchi and Di Falco 2001, Galeotti et al. 2001, Millimet et al. 2003); (ii) the degree of the polynomial used to proxy the environment – income relationship (Borghesi 1999, Markandya and Golub 2004); (iii) the econometric model specification used (Bradford et al. 2005). Less technical but possible critical issues are: (i) the environmental performance index and economic drivers investigated; (ii) the nature and quality of data.

As far as the environmental performance index is concerned, we may note that careful attention should be paid to deriving policy implications. In fact, EKC studies often use different environmental indexes (absolute, per capita, output based, input based, per unit of GDP). There is no general consensus as to what indicators to use. Different measures have nevertheless different implications and interpretation. For example, if a measure on per capita basis in OECD countries faces few problems of understanding, and absolute measures could be avoided, if we measure the intensity in the vertical axis, the presence of a lower bound implies that total emissions are growing at the same rate of income, in a sort of ‘steady state’ equilibrium. The measures on the vertical and horizontal axis should of course be compatible with each other. 14 We also note that there is no consensus on the type and number of explanatory factors introduced as potential drivers of the environmental performance. Some studies use income variables only. Other studies include many socio-economic variables with the (correct) aim of extending the conceptual setting behind the EKC empirics (Harbaugh et al. 2000); a few include policy drivers (Markandya et al. 2004). The choice obviously depends on both data availability and research objectives.

The nature and quality of data are also crucial issues. In fact, for reasons linked to existing data availability, the first wave of the EKC literature has witnessed a large majority of contributions which focused on the analysis of cross-country datasets, generally taken from official OECD and World Bank sources. Nevertheless, first the quality of macro data for some regions (non-OECD countries) has been questioned, and second, even the exploitation of panel datasets does not allow the researcher to calculate specific country-level coefficients for the income – environment relationship. 15 The conceptual key fact is that not a single relationship but many different relationships may apply to different categories of countries. In other words, the policy relevance of worldwide cross-country analyses seems limited. European countries, if compared with international datasets usually exploited for EKC analyses, represent a homogeneous set of statistical units. Although the limited data variability is an intrinsic feature of such a dataset, the relevancy for policy-making purposes is higher. Future research, as will be stressed in the conclusions, should then focus on delinking analysis that exploit datasets regarding environmental and economic indicators at a provincial/regional level (at national/European level). It follows that a higher added value is going to be found in studies based on national/regional rather than international datasets. The more evidence is micro-based (regionally/locally disaggregated), the better it is for statistical and policy aims. A European-level analysis thus presents some added value and some weaknesses.

There are few EKC analyses on waste and material flows. Cole et al. (1997) find no evidence for an inverted U-shape EKC curve concerning municipal waste. 16 Leigh (2004) presents evidence for EKC concerning a waste/consumption indicator deriving from the environmental sustainability indexes (ESI). The analysis faces two potential problems: data exist only for 2001 – 2002, and the index is based on a comparative rather than on an absolute scale. Wang et al. (1998) also find evidence in favour of a negative elasticity, by focusing on US stock of hazardous waste as an environmental impact indicator and exploiting a county-based cross-sectional dataset. The nature of the pollution effect (stock/flow, hazardous/non-hazardous) seems to matter: non-hazardous and flow externalities appear to be less likely associated with a negative elasticity, even in industrialized countries. Some authors have recently suggested that for stock pollution externalities, the pollution – income relationship changed into an EKC-shaped curve with difficulty, with pollution stocks monotonically rising with income (Lieb 2004). Another structural motivation concerning the lack of evidence for waste may be that the change in sign of the income elasticity of the environment/income function should occur at relatively lower income levels for pollutants whose production and consumption can be easily spatially separated, e.g. by exporting associated pollution or by relocating activities (Khanna and Plassmann 2004). Although strict EKC evidence has been rare, the literature underlines that waste indicators generally tend to increase with income or other economic drivers, and in general, an inverted U-shape curve is still not fitting data.17

The European waste sector thus emerges as an area for further exploration of the EKC hypothesis. However, at present, waste-data availability does not allow either nationally focused studies or EKC analysis including waste-policy changes, since most EU waste policies have been implemented from the 1990s. Nevertheless, panel datasets for municipal and packaging waste can be used for preliminary multivariate analysis. Given (i) the relative homogeneity across those countries in terms of structural characteristics and (ii) the panel framing which helps dropping off non-observed fixed factors. Our results, though preliminary, could be considered robust and of policy interest for the European framework.

3 Waste indicators and delinking: Empirical evidence for Europe

European waste policy is composed of an extensive and evolving system of Directives and regulations, which is the main source of the Member States' waste legislation and policy framework. General principles and procedures have been established by ‘horizontal legislation’ for waste management, i.e. the ‘waste framework directive’, the hazardous waste directive, and the waste shipment regulation. They provide the basic elements on which the whole policy framework is shaped, i.e. the ‘waste management hierarchy’, which puts ‘prevention’ at the highest priority level, the ‘polluter pays principle’, which paves the way for ‘economic instruments’, and the requirement that waste management may not adversely impact human health and the environment. The horizontal waste legislation is complemented by more specific sets of policies: (i) legislations on waste treatment and disposal operations, such as the landfill and incineration directives; (ii) legislations on the management of specific waste streams. Waste treatment has been addressed generally through legislative measures, including three recently adopted directives: the directive on ‘integrated pollution prevention and control’ (IPPC), the ‘landfill directive’, and the ‘incineration directive’. In the area of specific waste streams, important hazardous wastes have been addressed such as waste oils, PCBs/PCTs, and batteries. Recycling and recovery targets have been set for some key complex waste flows, i.e. packaging, end-of-life vehicles (ELVs), and waste electrical and electronic equipment (WEEE). Both the ELV directive and the WEEE directive explicitly include elements of ‘producer responsibility’, on which also the experience of recovey/recycling schemes for packaging and packaging waste in EU countries are, de facto, based. 18 The Commission launched in 2003 a ‘thematic strategy on prevention and recycling of waste’, which also propose innovative instruments, e.g. ‘pay-as-you-throw’ mechanisms and marketable permits systems (European Commission 2003a).

In spite of this significant policy experience, there is currently no empirical evidence concerning the delinking even for major waste streams, such as municipal and packaging waste. 19

Trends for packaging waste, household consumption, and GDP, which are homogeneously available for the EU15 countries over the period 1997 – 2001, show waste increases lower than economic indicators (7.1% versus 10.1% of GDP, both in per-capita terms). Correlations between packaging waste per capita and GDP/household consumption per capita are 0.36 and 0.46, respectively. The correlation between municipal waste per capita and household consumption per capita is 0.74 (from 1995 to 2000, considering EU₂₅, waste increased by 13.2%, while consumption increased by 14%, in per capita terms). Correlations are positive and significant.

In order to provide preliminary evidence on the shape of waste generation-economic drivers relationship, we have produced two panel databases on packaging waste and municipal waste, respectively, in European countries. As far as packaging waste generation is concerned, the current available information for EU15 countries (over 1997 – 2001) is exploited. Although limited in time, the panel dataset may provide preliminary evidence on the existence of delinking and on the current shape of the EKC for this waste indicator, and it helps dealing with fixed country-specific effects. For municipal waste, a dataset regarding 28 Eastern and Western European countries over 1995 – 2000 is set up, since data before 1995 are available only for some countries, and the quality/reliability is generally low. 20 The two panel data matrixes were thus set in order to minimize missing values and the presence of observations associated to a lower reliability.

Descriptively speaking, the generation of packaging waste per capita ranges from 88 to 214 kg, across the observed countries ( Figure 3): Greece and Finland are the lowest countries in the ranking, while Ireland and France are at the top, considering year 2001. The mean value is 171 kg per capita in 2001. As far as municipal waste is concerned (Figure 4), top ranking countries are Cyprus, Norway, Iceland, and Switzerland, while at the bottom we find Slovakia, Poland, Latvia, and Greece, considering 2000. The range is between 245 – 742 kg of waste generation per capita in 2000.

Figure 3. Packaging waste generation per capita (EU15).

Figure 4. Municipal waste generation per capita (EU28).

The first methodological problem for the applied analysis is how to specify the EKC functional relationship. There is no consensus on this point. Some authors adopt a second-order polynomial; others have estimated third- and even fourth-order polynomials, comparing different specifications for relative robustness. It is worth noting that neither the quadratic nor the cubic function can be considered a full realistic representation of the income – environment relationship. The cubic implies that environmental degradation will tend to plus or minus infinity as income increases, and the quadratic implies that environmental degradation could eventually tend to zero. The issue is thus unresolved. The third- or fourth-level polynomial could also lead to N- rather than U-shaped curves, opening new problematic issues in understanding the income – environment relationship for policymaking. This N shape is justified by a nonlinear effect by the scale of economic activity on the environment, which is difficult to prove. 21 Finally, the use of the income factor only, without quadratic and cubic terms, would collapse the EKC analysis to the basic decoupling analysis.

Here, we test the hypothesis by specifying a proper reduced form usual in the EKC field (Stern 2004):

(1) log(waste) = β_0i  + α _t  + β₁ log(consumption/GDP) _it  + β₂ log(consumption/GDP)² _it  + β₃ log(consumption/GDP)³ _it   +  e _it

where the first two terms are intercept parameters, which vary across countries and years. Different polynomial specifications are tested by including (i) as dependent variable waste per capita and waste in absolute terms, and (ii) as independent variables either house-hold consumption or GDP per capita, thus testing the hypothesis which indicates that consumption is a more appropriate driver for waste. In fact, recent studies (Rothman 1998, Jacobsen et al. 2004) indicate that for municipal and packaging waste, the proper economic driver is household consumption and not GDP. This is a key issue on both conceptual and statistical grounds.

Thus, for each combination of the dependent and independent variable listed above, different specifications are estimated, including: the linear regressors only (delinking baseline case), linear and squared terms (EKC most usual case), and finally a specification with linear, squared and cubic terms. Given the panel data framework, the relative fit of fixed effect and random effect models is compared by the Hausman statistic.

4 Empirical results

This section sums up and comments on the results presented in Tables I and II. It is worth noting that, according to our ex ante expectations, the economic driver which better fits with (both) waste indicators is household consumption. Regressions concerning GDP are less robust in terms of both coefficient significance and relative signs, and they are not presented. This is a first result testing a crucial hypothesis on drivers, which is relevant for future applied studies on waste delinking and EKCs.

Table I. EKC analysis for packaging waste.

Dependent variable	Log WASTE/POP	Log WASTE/POP	Log WASTE/POP	Log WASTE/POP	Log WASTE/POP
Model	Fixed effect panel data	Random effect panel data	Fixed effect panel data	Random effect panel data	Fixed effect panel data
Constant	…	3.10☆☆☆	…	3.13☆☆☆	…
Log C	0.90☆☆☆	0.78☆☆☆	0.89☆☆☆	0.76☆☆☆	7.88☆☆☆
Log C ^2	…	…	0.77	0.29	5.75☆☆☆
Log C ^3	…	…	…	…	−2.72☆☆☆
Turning point	…	…	No estimated turning point	No estimated turning point	No estimated turning point
N (15 countries, 5 years)	75	75	75	75	75
Model F test (probability value)	0.000	0.000	0.000	0.000	0.000
☆☆☆Significant at 1%. ☆☆Significant at 5%. ☆Significant at 10%. R ^2 values are not shown as they are not as meaningful as the fit measure in the panel settings. C: household consumption per capita; log waste/pop is waste per capita.

Moving to regressions exploiting consumption as driver, cubic forms do appear to perform worse than linear and squared forms, in terms of plausibility and economic significance of coefficients. We point out that the N-shape eventually associated with cubic forms has been raised as a theoretically plausible case mainly for emission externalities; in addition, testing the N-shape is maybe more meaningful for externalities which have already presented a turning point.

4.1 Packaging waste

For packaging waste, the basic specification with the linear term only shows a significant and positive coefficient for consumption. The Hausman test favours the random effect model: elasticities with respect to consumption are 0.78 for the random effect and 0.90 for the fixed effect model, respectively. Since the linear specification tests delinking by using econometric evidence instead of simple correlation analysis, we may observe that a ‘relative delinking’ evidence emerges.

Second, the nonlinear form also shows a positive and significant coefficient for the linear term and a positive, but highly not significant, squared term. The elasticity value is 0.89, in the fixed effect model, for both the basic case and when correcting for heteroskedasticity. Further, adding time effects to the baseline specification does not affect those results: the estimated elasticity is 0.85. Nevertheless, all these elasticities are not statistically different from unity: also, relative delinking is questioned. Finally, using absolute packaging waste production as an indicator does not change the results (absolute waste and waste per capita present a correlation of 0.35).

Additional statistical points are worth mentioning. Estimated autocorrelations across specifications lie in a range between 0.2 and 0.4. Given those values and the limited number of years, this should not represent a serious problem. The Hausman test signals a better fit for the random model: nevertheless, results are only slightly different between the two models, with elasticities less than one, and non-significant squared terms.

The analysis suggests that the first phase of the Packaging waste policy (most national policies started being operative in 1996 – 1997, though some were still operative in the early 1990s) has probably been effective in increasing environmental efficiency (a slight relative delinking), although no evidence appears to support an inverted U-shape EKC curve. This evidence confirms a shared statement, mostly qualitative, on the effectiveness of the first wave of packaging waste policies. More effort is then needed to reverse the relationship between environmental impact and economic drivers.

4.2 Municipal waste

As far as municipal waste is concerned, a preliminary econometric analysis on the full set of the 28 western and eastern European countries shows a limited statistical robustness. This is not unexpected, since we observe that the flaw may revolve around the lower quality/reliability of Eastern European data. Thus, we proceed considering only the subset of 18 western countries; EKC regressions ( Table II) show that: (i) in the linear case, elasticity is 0.60; (ii) in the squared term regression, positive and negative signs are respectively associated with the linear and nonlinear consumption factors; statistical significance is nevertheless associated only with the linear, with an estimated elasticity at 0.83. As for packaging, results do not change when correcting for heteroskedasticity and adding time effects. The cubic relationship is again not plausible looking at signs and coefficient levels. In all regressions, the Hausman test here favours the fixed effect model.

Table II. EKC analysis for municipal waste.

Dependent variable	Log WASTE/POP	Log WASTE/POP	Log WASTE/POP	Log WASTE/POP	Log WASTE/POP
Model	Fixed effect panel data	Fixed effect panel data	Fixed effect panel data
Constant	…	…	…	…	3,12☆☆☆
Log C	0,60☆☆☆	0,83☆☆☆	0,83☆☆☆	0,30	2,20☆☆☆
Log C ^2	…	−0,57
(t ratio−0.74)	−0,57
(t ratio−1.172)	0,57☆	−0,57☆☆☆
Log C ^3	…	…	…	−0,14☆☆	…
Turning point	…	No estimated turning point	No estimated turning point	No estimated turning point	Turning point
N (18 countries, 6 years)	108	108	108	108	96
Model F test (probability value)	0.000	0.000	0.000	0.000	0.000
☆☆☆Significant at 1%. ☆☆Significant at 5%. ☆Significant at 10%. R ^2 values are not shown as they are not as meaningful as the fit measure in the panel settings. C: household consumption per capita; log waste/pop is waste per capita.

In general, the analysis on municipal waste does not show any evidence of a bell-shaped curve, suggesting only a ‘relative delinking’ path. The squared specification presents a negative sign on the nonlinear term, which never overcomes the statistical threshold (the t ratio is 1.172 for the regression corrected for heteroskedasticity). By adding time period effects, the squared term coefficient shows a (maximum) t ratio of  –1.54.

There is a need for a further investigation using larger and, more important, longer datasets, as confirmed by a final applied exercise carried out after dropping off two small ‘outliers’, countries like Malta and Iceland, thus reducing the sample to 16 countries. The two-terms nonlinear regression shows both terms significant, with expected signs, when implementing the Random Effect model, though we note that for municipal waste, Hausman statistics favour the fixed effect model. Nevertheless, this outcome is worth noting since the EKC non-stability concerning results when different factors change is a key issue we have to deal with and has been raised in the literature. This opens up the way to more detailed analyses at the country level using material-specific and regional-based datasets.

5 Conclusions

The paper provides a methodological perspective and econometric estimates on delinking for waste indicators in Europe. EKCs are addressed as a natural extension of delinking analysis. They confirm the hypothesis that even European countries, characterized by high income levels and by a relatively long history on waste policies, are at best experiencing relative delinking, with waste indicators increasing slightly less than economic drivers. The elasticity of two major non-hazardous waste flows considered with respect to consumption is, at the European level, not significantly different from unity. We are far from reversing the waste – consumption relationship.

It is worth noting that a panel data analysis focusing on a homogenous set of countries is associated with fewer flaws and is more policy-informative, if compared with international cross-section/panel analysis. The relative homogeneity, which characterises the European framework, in addition to the European framing of most waste policies, adds informative value to applied investigations, even if they generate mean estimates concerning the defined sample (the income elasticity is assumed being the same in all countries at a given income level).

As far as packaging waste is concerned, the absence of delinking specifically indicates that the task of the second age of European policies is to achieve and increase delinking from now to 2010, at least for aggregate packaging waste if not for all specific waste materials.

Obviously, delinking and waste-efficiency gains are the statistical counterpart of prevention on policy grounds because, in our case, delinking would signal a reduction at source of municipal waste production and fewer packaging materials put on the market. Therefore, it seems that waste policies implemented over the 1990s have had a low effectiveness in terms of prevention. Directives have generally quantified targets in terms of shifts between waste disposal technologies and recovery/recycling, while waste prevention, i.e. the first priority in the EU waste hierarchy, remained a policy objective without targets. No country has introduced an explicit and binding policy target on prevention, even for one material. Prevention targets remain a moot point also in the new revised 2004 Packaging Directive, which sets more ambitious global and material-specific targets for recycling/recovery. Waste policies did not create so far incentives to reduction of waste at source and innovative processes in the packaging system. In the case of packaging and packaging waste, ‘producer responsibility’ policies were actually effective for recycling and recovery, but the packaging system was not pushed to pursue the ‘first priority’ of prevention (EEA 2003a). Mechanisms of incremental recycling/recovery costs coverage prevailed over economic incentive instruments to reduce packaging put on the market, so much so that a trade-off between recovery/recycling objectives and prevention objectives can be envisaged. This conclusion applies also to single country policies.

Policy evaluation opens up a new complex arena for further research on delinking where two are critical issues: (i) the possibility to assess a policy-driven structural break over a short period of time; (ii) the choice over which policy proxies are more sound as an object of analysis: main Directives/national polices, single instruments in the policy mix, and indirect costs caused by policies. The ex post effectiveness for waste reduction of policies such as the Landfill tax is being debated. This instrument is probably too far from waste production to have any effects in terms of waste reduction at source. For these reasons, in order to increase the probability of achieving a turning point in the waste/income/consumption relationship, a stronger and more prevention-focused policy effort along the resource-to-waste chain is needed.

Further analysis concerning (i) specific packaging waste materials for Europe and/or (ii) single countries will also be worthwhile as soon as a sufficient set of country/material data is available: different delinking processes could arise by considering specific countries and materials. Nevertheless, although preliminary, the applied exercise presented confirms and reinforces the shared view that a U-inverted EKC curve is still not generally characterizing waste generation in Europe.

Notes

1. For extensive evidence up to the early 1990s, see Tilton (1988, 1991) on metals/materials, Martin (1990) on energy, and Zoboli (1995) for a selective review and discussion. For recent thorough analyses of the long run trends for energy, see Gruebler et al. (1999), Ayres et al. (2004) and many other works by IIASA, www.iiasa.ac.at.

2. Among the early works on pollution, see Holtz-Eakin and Selden (1992), Ten Kate (1993), Grossman and Krueger (1994), and Selden and Song (1994).

3. See Andreoni and Levinson (2001), Chimeli and Braden (2005), and Kelly (2003), who finds that the EKC shape depends on the dynamic interplay between marginal costs and benefits of abatement.

4. The empirical literature is too extended to be surveyed here. Summing up the main results, water pollution seems to present a turning point between $5000 and $17 000 of per capita income depending on the specific pollutant. For air emissions, all main externalities, except CO₂ and transport-related emissions, appear to have a turning point in the range of $10 000 – $20 000 (Yandle et al. 2002).

5. See Moll et al. (1999), Eurostat (2001, 2002), Femia et al. (2001), and Bringezu et al. (2003).

6. For an extensive discussion on the prevention meaning of delinking indicators for materials and waste and their possible adoption in policy evaluation, see Jacobsen et al. (2004).

7. Since its formulation by Ehrlich (1971), many variants of the model have been applied to global resource dynamics, in particular by addressing the role of population growth.

8. The equation can be derived by a model like I = f(P; A; T) by assuming ∂I/∂P = AT; ∂I/∂A = PT; ∂I/∂T = PA. Decomposition of the equation can easily be done on the logarithms of the variables, which makes the relationship an additive one: ln I = ln P + ln T + ln A, in which elasticities with respect to each element, ∂ln I/∂ln(•), are all equal to 1.

9. See Zoboli (1996) for an account about positive/negative effects of population on economic growth.

10. See Jaffe et al. (2003), for an overview on environmental technology issues, and Zoboli (1995), for a discussion on the role of natural resource price and price-based policies in the line of induced innovation hypothesis. See Carraro et al. (2003) for the present state of research on ‘endogenous technological change’ in macro-models of energy. For the long-run structural-change perspective to technological innovation and natural resources, see Rosenberg (1996); the analysis of energy systems by Gruebler et al. (1999) and Rosenberg (1994); and the analysis on innovation and the global environment by Quadrio Curzio et al. (1994) and Quadrio Curzio and Zoboli (1995, 1997). It must be noted that, outside the neoclassical approach, the relationship between resource scarcity, economic growth, and technological change has been clearly defined since the 1960s by some theoretical models of rent, growth, and income distribution (Quadrio Curzio and Pellizzari 1999).

11. For a simple presentation of EKC with a discussion of core hypotheses and major empirical evidence, see De Bruyn et al. (1998).

12. If I = f(P, A, T), where A is an indicator of economic development, with ∂I/∂A, ∂I/∂P, ∂I/∂T > 0, and T = g(A), with dT/dA < 0, then the total derivative of I with respect to A will be negative if ∂I/∂A < – ∂I/∂T • dT/dA, or the direct positive effect of A on I is lower than the negative effect of A on T, given the effect of T on I.

13. Good critical surveys are Dasgupta et al. (2002), Dinda (2004), and Stern (2004).

14. Some argue that the choice over the dependent variable could depend on the issue considered. The per capita option is probably more compatible with situations where the degradation derives from overexploitation linked to population growth, whereas emission intensity is more compatible with scenarios with externalities caused by industries.

15. In fact, econometric panel studies usually provide information on mean-value coefficients since they usually rely on the assumption of different constant terms but equal coefficients across units (fixed effects model). We note that the superiority of heterogeneous panel data models is questioned. Baltagi et al. (2002) claim:

	Depending on the extent of cross sectional heterogeneity in parameters, researchers may prefer these heterogeneous estimators to the traditional pooled homogenous parameter estimators […] underlying the poolability of the data is the assumption of homogeneity of the parameters across states […]
	Our results show that when the data is used to estimate heterogeneous models across states, individual estimates offer the worst out-of-sample forecasts. Admittedly, this is another case study using US data, but it does add to the evidence that simplicity and parsimony in model estimation offer better forecasts. (pp 376 – 381).

Thus, added value may be found in the usual ‘homogenous panel analysis’ but concerning national/regional datasets.

16. They use municipal-waste data for the period 1975 – 1990 in 13 OECD countries, finding no turning point, with environmental indicators (per capita municipal waste) monotonically increasing with income over the observed range.

17. Rothman (1998) argues that delinking is less likely to occur when we tackle ‘consumption-based’ measures.

18. For the analysis of economic instruments based on ‘producer responsibility principles’ in European ELV policies, see Mazzanti and Zoboli (2005).

19. We have only scattered pieces of evidence. Among the others, Martin and Scott (2003) claim that waste production continues to have a positive relationship with increased wealth.

20. Municipal waste data derive from the EUROSTAT/OECD joint questionnaire, national sources, EEA; reliability is not homogenous across countries (Eurostat 2003). For packaging waste, data are from Member States' reports to the DG Environment in pursuance of Directive 94/62/EC on Packaging and Packaging Waste. For an assessment on waste data issues and methodology, see EEA (2003b).

21. Shobee (2004) suggests a third-order polynomial specification as a more realistic relationship between environmental degradation and income per capita. The issue still remains unresolved, with the EKC hypothesis relying mainly on empirical evidence.