2.1. The state of municipal solid waste in Japan

In Japan, waste is generally separated into two categories: industrial waste and domestic waste. The Waste Disposal and Public Cleansing Law defines certain types of waste generated by industrial activity as industrial waste and the rest as domestic waste. A typical example of industrial waste is waste generated by a factory, whereas a typical example of domestic waste is waste generated by households, small businesses, restaurants, convenience stores or office buildings. Thus, in Japan, domestic waste corresponds to MSW, as is typically understood in studies of the WKC hypothesis. In what follows, we refer to domestic waste as MSW to remain consistent with the previous literature.

Japanese MSW can be classified into two types: MSW from households and MSW from business activities. As defined in the previous section, the former type of waste is classified as household MSW and the latter as business MSW.Footnote ⁶ The Japanese Ministry of the Environment (2008) reported that a total of approximately 49.7 million tons of MSW was disposed of in Japan in 2005, of which, 33.5 million tons (67 per cent) was household MSW, and 16.2 million tons (33 per cent) was business MSW.

Although there are differences in the waste disposal systems used across municipalities, as noted below, the waste generator associated with each type of MSW is in many cases obliged to purchase disposal bags as designated by the municipality and to bring the MSW to the appointed location on the designated day. In Japan, most collected waste is incinerated at an intermediate waste treatment facility before it is landfilled. For example, in 2005, 77.4 per cent of disposed waste was directly incinerated, 5.1 per cent was directly recycled, and 2.9 per cent was directly landfilled. The remaining 14.6 per cent underwent intermediate processing with the aim of recycling the waste or reducing its weight. Overall, 14.7 per cent ended up in landfills, and 14.1 per cent was eventually recycled.Footnote ⁷ The amount of waste that is landfilled greatly depends on the method of intermediate waste disposal employed by each municipality and thus is not decided at the household level. Because the volume of landfilled waste reflects a different type of result in our empirical analysis based on the per capita income level of households, we decided not to include this amount in this paper.

In the past, the priority of the MSW management policy in Japan was to provide acceptable levels of sanitary waste disposal. For this purpose, incineration has become widely used as the waste disposal method in Japan. However, because of the increased amount of MSW in recent years, waste reduction is now a major aim of the waste management policy. In fact, the Japanese government promotes what is called the ‘3Rs’ principle and aims to build a sustainable society. 3Rs is an acronym for ‘Reduce, Reuse, Recycle’, and the order of words indicates the hierarchy of waste management strategies.Footnote ⁸ In accordance with this principle, several types of waste management legislation, such as the Basic Act for the Promotion of the Recycling-Oriented Society, the Act on Promoting Green Purchasing and the Containers/Packaging Recycling Act have been introduced in Japan. Whereas national legislation establishes the national strategy for waste management, the practical operation of MSW disposal services is planned and conducted by each individual municipality. In fact, chapter four of the Waste Disposal and Public Cleansing Law stipulates that each municipality is responsible for creating its own plan for disposing of the MSW generated in its region. Thus, the waste management policies for MSW differ widely across municipalities.Footnote ⁹ For example, some municipalities collect plastics as combustible waste, whereas other municipalities collect plastics as incombustible refuse. When processing recyclable waste, some municipalities pick up only packaging materials, whereas others collect waste paper or used textiles in addition to packaging materials. Waste collection systems vary across municipalities. For example, some municipalities have set up waste-collection points, whereas others have introduced door-to-door collection schemes that are similar to the curbside collection systems used in Europe and the United States.

Moreover, there are municipalities that simultaneously use both types of collection systems. A number of municipalities have adopted the waste-collection points system for household MSW while introducing a door-to-door collection system for business MSW. Because the Waste Disposal and Public Cleansing Law permits municipalities to outsource waste collection to the private sector, the operating body responsible for waste collection is not the same across municipalities. Although there are several municipalities that provide waste collection services themselves, a number of municipalities outsource all or a part of their operations. The same is true of waste disposal operations.

Thus, a municipality can be considered an independent decision-making entity within the context of waste management. Consequently, aggregating the data (at the national level, for example) is problematic because it may obscure the effects of disparate waste management policies that differ across municipalities. We therefore use data at the municipality level in our empirical analysis.

2.2. Data

In the following empirical analysis, we developed a municipality-level, cross-sectional data set for Japan in 2005. The waste-related data were obtained from the Japanese Ministry of the Environment (2008), whereas other socio-economic data, such as information regarding income per capita and population density, were obtained from the Ministry of Internal Affairs and Communications (2008).

There are two main reasons why we use cross-sectional data rather than panel data. The most important reason is the large number of municipality mergers led by the national government from the late 1990s to the mid-2000s. In fact, the total number of municipalities was reduced by more than half as a result. The mergers caused an attrition problem that seriously weakens the reliability of the related panel data.Footnote ¹⁰ The other reason we use cross-sectional data is the availability of socio-economic data. Because crucial socio-economic data (e.g., information regarding household composition) are only released every five years, we could not develop a panel data set that included sufficient longitudinal information. For these reasons, we employ the latest available cross-sectional data, those from 2005, in the following analysis.

Table 1 presents descriptive statistics for the variables we use. In table 1, waste is the total MSW generation per capita (unit: grams per day per capita) in a municipality. These data can be disaggregated into two classes: wasteh and wasteb. The former is the household MSW, and the latter is the business MSW. As described above, because the relationship between income level and the amount of waste discharged will be different for households and businesses, it is important for us to separate waste into household MSW and business MSW when analyzing the WKC hypothesis.

Table 1. Summary statistics

Notes: The first three variables are dependent variables, all of which are on a per capita basis. See the text for sources and units.

The most important non-waste variable in this study is income. This variable is defined as the total taxable gains (unit: million yen) in a municipality. perinc (million yen per capita) is simply calculated by dividing income by the number of income tax payers. Thus, perinc is considered the income per household rather than the income per capita. perinc2 is the square of perinc.

In analyzing the effect of waste disposal policies, we define hprice, bprice and sorting. hprice(bprice) is a dummy variable that equals one if a municipality introduces unit pricing for the disposal of household MSW (business MSW). To avoid the endogeneity problem with regard to waste policy, a one-year lag is adopted for both hprice and bprice.Footnote ¹¹ However, this approach cannot be directly employed with the new municipalities that came into being as a result of the mergers in 2005. In these cases, we use the weighted average of the one-year lagged policy variables for each municipality in the pre-merger period.Footnote ¹² Under the assumption that both household and small-business behavior are rational, we expect less waste generation if unit pricing is introduced. Although studies such as Kinnaman and Fullerton (Reference Kinnaman and Fullerton2000) and Eyckmans et al. (Reference Eyckmans, De Jaeger and Verbeke2009) considered the effect of charges for waste collection, they examined charges for total MSW rather than distinguishing between household MSW and business MSW.

sorting is the number of categories of waste that the municipality sets and requires the waste discharger to separate. Each municipality can use its own discretion in setting this number. For instance, one municipality might separately collect combustible waste, noncombustible waste, used paper, used plastics and metals, whereas another municipality might collect most of these waste types (or recyclables) together. Like hprice and bprice, to avoid the endogeneity problem, we calculate the weighted average of the policy variables for those municipalities that constitute a new post-merger municipality.Footnote ¹³ To our knowledge, among the studies of WKC, none has used sorting as an explanatory variable. We hypothesize that sorting has a negative sign because those who employ more time-consuming sorting practices (namely, with greater sorting) will be aware of the reduction in waste generation. Note that MSW policies pricing policies and sorting practices are quite different across municipalities.

We also use other socio-economic variables that affect waste generation. shousehold is the ratio of single-person households to total households. We expect that there will be less per capita MSW generated if there are more than two people in a household. elderly is the ratio of households composed of elderly couples to the total,Footnote ¹⁴ and we expect that an elderly household will generate less per capita MSW than a younger household because the amount of goods consumed by elderly people is less than that consumed by younger people. The effect of household size was also considered in Kinnaman and Fullerton (Reference Kinnaman and Fullerton2000), Mazzanti and Zoboli (Reference Mazzanti and Zoboli2009) and Abrate and Ferraris (Reference Abrate and Ferraris2010). Kinnaman and Fullerton (Reference Kinnaman and Fullerton2000), Mazzanti and Zoboli (Reference Mazzanti and Zoboli2009) and Eyckmans et al. (Reference Eyckmans, De Jaeger and Verbeke2009) examined the effect of elderly households.

commutein indicates the ratio derived by dividing the number of commuters from areas outside the municipality by the number of people who commute from the municipality to elsewhere. We believe this variable indicates the level of economic activity because economically growing municipalities provide employment opportunities for more people and, thus attract people who live outside the municipality. Although studies such as Eyckmans et al. (Reference Eyckmans, De Jaeger and Verbeke2009) and Mazzanti et al. (Reference Mazzanti, Montini and Nicolli2011, Reference Mazzanti, Montini and Nicolli2012) considered the effect of population inflow as defined by the inflow of tourists, no study has focused on the influence of commuters from areas outside the municipality. We expect commutein to be positively related to the amount of waste discharged and the volume of landfill waste. Finally, popden denotes the population density (the population per 1,000 m²). This indicator was used in a number of studies, including Hage et al. (Reference Hage, Sandberg, Soderhölm and Berglund2008), Mazzanti and Zoboli (Reference Mazzanti and Zoboli2009) and Abrate and Ferraris (Reference Abrate and Ferraris2010). Because population density tends to be high in economically significant municipalities, we expect popden to be positively related to the amount of waste discharged.

3.1. The need for spatial consideration

One caveat here regards our choice of econometric methods. Our initial conjecture that the behavior of Japanese municipal governments tends to ‘mimic’ the waste-collection policy employed by their neighbors requires that we assume spatial correlation in our econometric model.Footnote ¹⁵ Thus, we test the WKC hypothesis by adopting two different spatial econometric models: a spatial autoregressive model (SAR) and a spatial error model (SEM).Footnote ¹⁶

The SAR model assumes that the dependent variable, i.e., the amount of waste, is spatially correlated across municipalities, whereas the SEM model assumes that the errors are spatially correlated. In Japan, each municipality belongs to one of 47 prefectures. In this two-tiered system, municipalities in the same prefecture tend to have the same information and regulations, which are provided by their prefectural government. They also tend to implement similar waste management policies. Thus, municipal governments in Japan tend to mimic each other more with regard to their waste-reduction efforts than in their actual waste reduction. Given that these efforts are latent and act as omitted variables in our analysis, we suspect that the errors rather than the amount of waste are spatially correlated.

To reflect this two-tiered regional government system, we design the spatial weight matrix such that before row standardization, its ij element is one if municipality j (≠i) is in the same prefecture as the municipality i and zero otherwise.Footnote ¹⁷ In what follows, we refer to this matrix as spatial weight matrix I (SWM1). This matrix captures the administrative relationship between municipalities rather than focusing on more orthodox geographic relationships (with the elements of the matrix as the inverse of the distance between the municipalities squared.Footnote ¹⁸ ) We refer to the latter matrix as SWM2, and we use these two matrices alternatively in estimating all the models.

3.2. OLS and the Lagrange multiplier tests

We begin by specifying the following quadratic relationship between waste generation and per capita income, which is standard in the EKC literature:

(1) $$Y=\beta_{0}+X\beta_{1}+X^{2}\beta_{2}+Z\gamma+\epsilon$$

where Y is the n × 1 vector of waste generation per capita; X is the n × 1 vector of the per capita income of the municipalities; and Z is the n × k matrix of the exogenous variables, where β₀, β₁, β₂ and γ are the corresponding parameters. In our model, parameters that satisfy β₁ > 0 and β₂ < 0 imply evidence for the WKC hypothesis.

The above simple OLS model is considered the model under the null hypothesis such that ρ = λ= 0. In a more general model,

$$\eqalign{& \quad y=\beta_{0}+\rho W^{L}Y+X\beta_{1}+X^{2}\beta_{2}+Z\gamma+\mu \cr & \quad \mu = \lambda W^{E}+\epsilon}$$

where the parameters ρ and λ are spatial autocorrelation coefficients; W ^L (W ^E ) as the spatial weight matrix defined above, and ε is the error, which is assumed to be independent and identically distributed (iid) with mean 0. We compute the Lagrange multipliers for both lags and errors under the null of ρ = λ= 0 and the robust Lagrange multipliers for lags and errors under the null of ρ = 0 and λ = 0 without any restrictions on the values of λ and ρ.Footnote ¹⁹ Comparing these figures indicates whether it is more appropriate to assume SAR or SEM. We will elaborate on these findings in the results section.

3.3. Spatial autoregressive and spatial error models

Following Anselin (Reference Anselin and Baltagi2001) and others, we consider two alternative spatial econometric models: the SAR and the SEM. The spatial lag model is

(2) $$Y=\beta_{0}+\rho WY+X\beta_{1}+X^{2}\beta_{2}+Z\gamma+\epsilon$$

Again, the parameter ρ is a spatial autocorrelation coefficient, and W is the spatial weight matrix defined above. We assume here that ε is iid with mean 0.

Another econometric specification is the SEM, in which the spatial interdependence occurs through the error terms. Formally, this model is represented as follows:

(3) $$Y = \beta_{0}+X\beta_{1}+X^{2}\beta_{2}+Z\gamma+\mu$$

(4) $$\mu = \lambda W \mu+\epsilon$$

The behavioral assumption in the SAR model is that municipalities care about the actual amount of waste generated by their neighbors; we consider this assumption to be unlikely. Instead, we expect SEM to be more appropriate than SAR because municipalities in the same prefecture are expected to be consistent with regard to waste reduction efforts that are largely unobservable, such as educational programs and public advertisements encouraging waste reduction.Footnote ²⁰ Should the contiguity effect exist, it will appear in the error term as the omitted variable, which yields the SEM.

3.4. The estimation methods

The field of spatial econometrics has rapidly developed since the seminal textbook by Anselin (Reference Anselin1988) was written. This development enables us to expand the scope of spatial econometrics. It is inappropriate to estimate (2) using ordinary least squares (OLS) because of the endogeneity problem with regard to the spatial lag term. Given the large sample (n = 1, 798) in our work, it would be standard to use maximum likelihood estimation. However, the results of the Jaque–Bera test (reported in tables 2 and 3) cause us to reject the normality of the error distribution. To address this problem, we use the generalized spatial two-stage least squares approach developed by Kelejian and Prucha (Reference Kelejian and Prucha1998) for the estimation of SAR and GMM in the manner of Kelejian and Prucha (Reference Kelejian and Prucha1999) for SEM.

Table 2. Estimation results of wasteh (above) and wasteb (below) with SWM1

Notes: **1%, *5%, ^†10%.

Table 3. Estimation results of wasteh (above) and wasteb (below) with SWM2

Notes: **1%, *5%, ^†10%.

Another method to address relaxing the assumption of constant variance normal disturbances is to use Bayesian estimation. The introduction of Bayesian estimation has a very significant impact. One of the reasons for this impact is that ‘Bayesian models allow for the direct estimation of the influence of heteroskedasticity and outliers’ (Ross, Reference Ross2013: 458). With the help of the development of numerical computation techniques, the Bayesian method has been widely applied in the spatial econometrics literature.

In light of these advances, we have added Bayesian estimation results. For our purposes, being able to check the variance of the linear combination of the estimates without using the linear approximation (delta method) is an additional advantage of using Bayesian models. The difference between the conventional and Bayesian approaches is the use of prior information in Bayesian estimation.Footnote ²¹ Our assumption regarding the prior distribution (π(·)) is as follows.

(5) $$\pi({\bf \beta})\sim N(c,\sigma^{2}T)$$

(6) $$\pi(\sigma^{2})\sim NIG (a,b) ={b^{a}\over \Gamma (a)} (\sigma^{2})^{-(a+1)} \exp (-b/\sigma^{2})$$

(7) $$\pi(\rho)=U(\lambda^{-1}_{min}, \lambda^{-1}_{max} ),$$

where a, b, c, T are parameters. N(·) is the normal distribution, whereas NIG (·) denotes the normal inverse-gamma distribution. We assume non-informative priors for ρ and U(·), namely, that they are uniformly distributed, and that λ _min (λ _max ) denotes the minimum (maximum) eigenvalue of the spatial weight matrix. By combining this prior information into (2), (3) and (4), we can derive the posterior distribution of each spatial econometric model. To solve the models, we use Metropolis–Hastings sampling, which is one of the Markov Chain Monte Carlo (MCMC) approaches. We set the number of samplings to 150,000 and discard the first 5,000 as the ‘burn-in’.Footnote ²²