To introduce recycling or not: A panel data analysis in Japan

Resources, Conservation and Recycling 101 (2015) 84–95

Contents lists available at ScienceDirect

Resources, Conservation and Recycling journal homepage: www.elsevier.com/locate/resconrec

To introduce recycling or not: A panel data analysis in Japan Takehiro Usui a,∗ , Kazuhiko Kakamu b , Mitsuko Chikasada a a b

Faculty of Economics, Soka University, Tokyo, Japan Graduate School of Business Administration, Kobe University, Japan

a r t i c l e

i n f o

Article history: Received 22 December 2014 Received in revised form 27 April 2015 Accepted 6 May 2015 JEL classification: C11 C23 C25 Q53 Keywords: Cost minimization of waste management Prolonging a life span of landfill Utilizing energy in incineration facility Municipal decision on recycling Bayesian panel spatial autoregressive probit model

a b s t r a c t Why do municipalities choose recyclables collection although it is more expensive than garbage collection? This study analyzes the determinants of a municipality’s decision-making on the collection and separation of recyclables. For this analysis, we use municipal-level panel data on whether municipalities in Japan recycle glass bottles, plastic containers, and paper containers. Applying a Bayesian panel spatial autoregressive probit model, we find that municipalities with refuse-derived fuel (RDF) facilities are less likely to collect and separate plastic containers for the purpose of recycling. Instead, they use plastic containers for generating energy. Furthermore, we find that municipalities holding a self-owning landfill site are more likely to introduce recyclables collection in order to save landfill capacity. We further find that inter-municipal cooperation involving spatial interaction with neighboring municipalities influences the decision-making and collaborative action of municipalities to implement recyclables collection and reduce costs, and that such collaborative action strengthens every year. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Residential waste recycling is widespread, having increased in many parts of the world over the past quarter-century. Over the past decade, Japan has promoted the idea of a recycling-based society to prolong the lifespan of landfill sites and to encourage recycling efforts. The Japanese government has encouraged the 3Rs of reducing, reusing, and recycling residential waste under the Basic Law for Establishing a Recycling-based Society. This law has promoted the recycling of plastic containers and packaging materials, which has resulted in resource conservation, energy-input reduction in incineration facilities, and the lengthening of the lifespan of landfill sites. Against the background of this increasing trend toward recycling, this study empirically investigates the determinants of a municipality’s decision making on the collection and separation of recyclables in Japan. Many studies have investigated the recycling motivation of households, such as Kinnaman and Fullerton (2000), Suwa and Usui (2007), and Allers and Hoeben (2010). However, few have explored

∗ Corresponding author at: Faculty of Economics, Soka University, 1-236 Tangi-cho Hachioji, Tokyo 192-8577, Japan. Tel.: +81 42 691 4016; fax: +81 42 691 8232. E-mail address: [email protected] (T. Usui). http://dx.doi.org/10.1016/j.resconrec.2015.05.006 0921-3449/© 2015 Elsevier B.V. All rights reserved.

the motivation for municipalities implementing recyclables collection services (see Keeler and Renkow, 1994; Kinnaman, 2005). Furthermore, no research has focused on how these motivations vary according to the types of recyclable materials collected. Our study clarifies the determinants of a municipality’s decision making on the collection and separation of recyclables and makes two main contributions. The first contribution is that we investigate municipalities’ decision to provide collection services for all types of recyclable containers and packaging such as glass bottles, paper containers, and plastic containers, whereas other studies have focused on recyclables as a whole. These different materials seem to have different impacts on the environment. For example, for municipalities that have incineration facilities, paper and plastic containers improve combustion and serve as fuel to produce electricity, which might prevent municipalities from collecting and separating these items for recycling. In fact, we find that the probability of initiating collection and separation services for plastic containers in municipalities using refuse-derived fuel (RDF) facilities is lower. This finding implies that having RDF facilities creates a disincentive to recycle valuable resources such as plastics. The second contribution of this study is that we analyze the spatial dependency of recyclables collection services based on scale economies and explain their economic background. The benefit of scale economies, which is beyond the scope of small municipalities

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

operating individually, may be one of the main factors driving this cooperation between municipalities. This is because sharing a recyclables collection facility may lead to scale advantages for both municipalities. In order to capture this phenomenon, we consider spatial dependency in recyclables collection services. To our knowledge, this is one of the first attempts to consider spatial dependency in recyclables collection services. For overall municipal solid waste management, some studies have explored spatial dependency problems, such as Bivand and Szymanski (2000) and Bel et al. (2011). However, as mentioned previously, no studies have examined the spatial dependency problem in the field of recyclables collection. The remainder of this paper is organized as follows. In Section 2, we describe the research background and clarify the municipal and national recycling goals. In Section 3, we introduce the econometric model. In Section 4, we explain the data used in this study. In Section 5, we present the estimation results of the Bayesian panel spatial autoregressive probit model. Finally, in Section 6, we draw conclusions and offer policy suggestions for future research. 2. Research background All municipalities are confronted with budget constraints. Therefore, we need to understand the cost structure of recyclables collection in terms of total waste disposal, which is closely related to the introduction of municipal recyclables collection. In this section, we describe the theoretical background of introducing recyclables collection. We explain the background of the Japanese municipal solid waste services programs and highlight the diversity of waste management initiatives, particularly how plastic and paper containers are handled. In this respect, we clarify that municipalities may start recyclables collection from the perspective of cost minimization. 2.1. Institutional background in Japan Japan faces a shortage of landfill capacity owing to its relatively small geographical size. The country’s direct landfill rate was only 11% in 1995, the lowest among OECD countries, compared with the rate of 57% in the United States and 83% in the United Kingdom.1 Since a country’s waste generation and gross domestic product (GDP) are closely related (Daskalopoulos et al., 1998), disposable goods, plastic bottles, and paper containers have proliferated in Japan since its rapid economic growth during the 1970s. The Japanese pose too large a challenge for landfill space, because containers and wrappings account for about 60% of their total waste by volume (METI, 2003). Therefore, the Japanese government drafted the Containers and Packaging Recycling Law based on the principle of extended producer responsibility (OECD, 2001). The Japanese Containers and Packaging Recycling Law (hereafter abbreviated as the Recycling Law) went into effect in 1997. The basic principle of the Recycling Law is that every stakeholder has a role to play in recycling. For example, consumers should separate their waste by category, and municipalities, who are financially responsible for waste collection and disposal costs, should collect the separated waste. Businesses should recycle what has been collected into new products. Concretely speaking, government-designated organizations operate recycling businesses on behalf of specified business entities.2 By paying recycling fees to these government-designated organizations, business

85

entities are deemed to have fulfilled their recycling obligations. Thus, the Recycling Law defines the shared responsibilities of each stakeholder, namely citizens, producers, recyclers, and municipalities. The Japanese government cannot use the Recycling Law to force municipalities to provide recyclables collection services. Therefore, the Recycling Law allows municipalities to pursue a diversity of recyclables collection approaches, freely choosing from among many waste treatment options, especially with respect to plastic and paper containers, given the lack of regulatory power and order of priority for the 3Rs. In order to consider a municipality’s choices from these options, we focus on its possession of waste facilities such as incinerators and RDF facilities, because if municipalities have such facilities, they can reduce their waste volume through combustion. Municipalities can choose either recyclables collection or unsorted-waste collection (e.g., mixed waste). If a municipality chooses to collect paper or plastic containers, its landfill waste (i.e., that which goes directly or via incineration into a landfill) would be reduced. Therefore, they save the lifespan of landfill sites and might reduce waste disposal costs. However, the municipality incurs additional collection costs (Central Environment Council, 2005). For example, an incinerator has lower heating value when it reduces a combustible feedstock volume comprising plastic and paper containers.3 When municipalities face a shortfall of feedstock, they may use oil additionally, and this increases their incineration costs. Further, a municipality may decide to build a recyclable waste sorting facility jointly with adjacent municipalities, giving rise to the possibility of economies of scale by lowering the average cost of solid waste management below the average cost of building a sorting facility alone. From the above argument, we point out that municipal disposal management is mainly confronted with the following three cost–benefit budget trade-offs: (a) increase in the benefit of saving landfill sites by getting rid of certain waste categories and collecting them as recyclable waste vs. increase in the costs of recyclables collection, (b) increase in the cost of additional fuel owing to the shortfall of the feedstock of paper or plastic containers vs. decrease in the cost of saving landfill sites by reducing the volume of incineration waste, and (c) cooperation with adjacent municipal entities to build recyclable sorting facilities vs. building a facility alone. These three budget trade-offs are all closely related to cost minimization. 2.2. Theoretical model We describe our theoretical model for a municipality to introduce recyclables collection. Previous studies have clarified the cost structure of municipal waste collection; for example, see Stevens (1977), Carroll (1995), Callan and Thomas (2001), and Dijkgraaf et al. (2003). We focus on the municipal behavior of implementing recyclables collection because we cannot directly observe its cost structure. Here, we assume that municipalities are rational and that they initiate collection if the net benefit of recyclables collection is positive. Since the net benefit when a municipality starts recyclables collection is positive, we can indirectly evaluate the cost structure of recyclables collection through municipal behavior. We describe the cost structure as follows:

 y∗ =

B( · ) − C( · ) ≥ 0 → y = 1 B( · ) − C( · ) < 0 → y = 0

(1)

1

See OECD (2008, p. 16). Business entities are defined as follows: (1) manufacturers who use containers and wrappings for shipping their products, (2) retailers and wholesalers who use containers and wrappings for selling merchandise, (3) manufacturers of containers, 2

and (4) importers who import and sell merchandise in containers and wrappings. For further details of the Recycling Law, refer to METI (2007). 3 For further technical details, refer to Consonni et al. (2005).

86

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

Table 1 Descriptive statistics. Variables

Mean

SD

Min

Glass collection dummya Plastic collection dummya Paper collection dummya Dummy: possessing incinerationb Dummy: possessing waste power generationb Dummy: possessing RDFb Dummy: possessing landfill solelyb Dummy: sharing landfillb Landfill capacity (year)b Waste generation (tons)b Wage rate (per capita · year: 1000 yen)b Population density (per km2 )b Outsourced proportion (%)b

0.676 0.268 0.521 0.916 0.089 0.018 0.290 0.505 28.530 8650 9039 470 0.258

0.468 0.443 0.500 0.277 0.285 0.131 0.454 0.500 256.762 19,692 828 808 0.398

0 0 0 0 0 0 0 0 0 29 5366 1 0

Max 1 1 1 1 1 1 1 1 8345 292,735 12,632 6196 1

Note: 1 Euro = 100 yen (Jan 2012). a Data period from 2000 to 2002. b Data period from 1999 to 2001.

where y* is a latent variable; if y = 1, the net benefit is positive and the municipality starts collecting recyclables; if y = 0, the net benefit is negative and the municipality does not start collecting recyclables. Using econometric techniques, we clarify why municipalities introduce recyclables collection.

W is called the spatial weight matrix (see, e.g., Anselin, 1988), where the element of W in row i and column j is denoted by wij , as defined above. Then, the likelihood function of the model (2) is given by L(y|, ˛, ˇ, z, X, W) =

T

f (yt |t , ˛, ˇ, zt , Xt , W),

(3)

t=1

where

3. Panel spatial autoregressive probit model In this section, we introduce the panel spatial autoregressive model proposed by Kakamu et al. (2010); this can be used to estimate spatial interactions, as suggested by LeSage and Pace (2009), in the spatial autoregressive model. Let yit be binary data and xit be a 1 × k vector of exogenous variables, where i indicates the region (i = 1, 2, . . ., n) and t represents the time period (t = 1, 2, . . ., T). Let wij denote the spatial weight of the jth region in the ith region, which is given by (i) wij = 0 for all i = j and (ii) wij = 1/mi when the jth region is contiguous with the ith region and wij = 0 otherwise, where mi denotes the number of regions contiguous with the ith region. Note that we have  n w = 1 for all i. In the panel model, we consider that the unobj=1 ij servable component—which is specific to the ith region—affects the dependent variable (e.g., Kakamu and Wago, 2008). ˛i is denoted by the unobservable component. Then, the panel spatial autoregressive probit model with parameters t , ˛i , and ˇ is written as follows:

yit

zit

=

=

⎧ 1, if zit ≥ 0, ⎪ ⎨ ⎪ ⎩ t

0,

if zit < 0, (2)

n 

wij zjt + ˛i + xit ˇ + it ,

it ∼N(0, 1),

j=1

where zit is a latent variable (see Tanner and Wong, 1987). In (2), t represents the spatial interaction at time t, but we assume that t is independent across time. Let us define:  = (1 , 2 , . . ., T ) ,

˛ = (˛1 , ˛2 , . . ., ˛n ) ,

zt = (z1t , z2t , . . ., znt ) ,

z = (z1 , z2 , . . ., zT ) ,

yt = (y1t , y2t , . . ., ynt ) ,

y = (y1 , y2 , . . ., yT ) ,

Xt = (x1t , x2t , . . ., xnt ) ,

X = (X1 , X2 , . . ., XT ) .



f (yt |t , ˛, ˇ, zt , Xt , W) = (2)

×

n −2

n

|In − t W| exp



e et − t 2



yit 1[0,∞) (zit ) + (1 − yit )1(−∞,0] (zit ) .

i=1

In indicates the n × n unit matrix, et is given by et = zt − t Wzt − Xt ˇ − ˛, and 1(a,b) (x) denotes the indicator function, which takes 1 when x lies on the interval between a and b. 4. Data and variables Our empirical study is carried on a panel dataset of 2951 municipalities in Japan (after excluding outliers and missing values) during the period 2000 to 2002, which coincides with the introduction of the Recycling Law. Importantly, this study period is before municipalities began to be merged in 2003. Although using a long and up-to-date panel dataset would be ideal, doing so would raise three problems, which should be addressed in future research. First, it would cause sample selection bias because the municipal mergers undertaken since 2003 are strategic as opposed to random. Second, using a long panel dataset would restrict spatial econometrics because the weight matrix is constant over time. Third, because a longer panel dataset expands over time, the calculation burden rises. Indeed, such a calculation is almost technologically impossible given current computing limits. Based on the foregoing, we use a short but robust panel dataset in this study. The descriptive statistics based on the dependent and independent variables are presented in Table 1. 4.1. Dependent variables We use three sets of binary variables in our estimations. The binary variables pertain to whether a municipality collects

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

87

Table 2 Sample mean of recyclables collection rate, sorted by year: glass bottle, paper containers, and plastic containers. Year

Percentage Glass collection

Percentage Paper collection

Percentage Plastic collection

2000 2001 2002

0.657 0.678 0.693

0.489 0.517 0.556

0.220 0.269 0.317

Note: The number of sample municipalities is 2951.

recyclable materials such as glass bottles, plastic containers, and wrappings as defined by the Recycling Law. These dependent variables are based on data obtained from the Ministry of the Environment, 2001–2003); these data are issued every year after a survey of all the municipalities in Japan. The dummy variable takes a value of 1 if the municipality collects certain types of presorted or commingled recyclables and then sorts them by type; otherwise, the dummy variable takes a value of 0. As mentioned in Section 2.1, the decisions on whether to collect and store recyclable containers and packaging as well as the types of recyclables to be collected are left to the discretion of municipalities. The factors motivating a municipality’s decision to implement recyclables collection also vary according to the types of recyclables. For example, as discussed in Section 2, plastic or paper containers, which are combustible materials, can be handled in two ways, that is, recycling and incineration. Therefore, the presence of incineration facilities may adversely affect a municipality’s decision to implement recyclables collection. On the contrary, because glass bottles are not combustible, the existence of incineration facilities would not prevent municipalities from implementing such a recyclables collection service. Therefore, we create dependent variables for each type of material and assume that the municipality’s decision to select bottles is mutually independent. Table 2 shows the proportion of municipalities that collect or separate each type of waste. The proportion and number of municipalities seem to have grown year by year across all types of recyclables. The collection of glass bottles already seems to be widely distributed, because this is a traditional recyclables collection service. Meanwhile, the collection of paper and plastic containers seems to have been rarely introduced, since attempts at recycling such waste items are rather rare for municipalities. Table 3 presents the mean, standard deviation, minimum, and maximum values of the three dependent variables. By decomposing the standard deviation into between and within components, we found that the within standard deviation for each of the three dependent variables was non-zero. These results show that all of the dependent variables vary over time. In other words,

Fig. 1. Spatial distribution related to introduction of glass bottle collections, 2002.

municipalities seem to have changed their recyclables collection behavior over time. Therefore, we consider it to be important to conduct panel data analysis. Figs. 1–3 outline the spatial distribution of the introduction of each type of recyclables collection in 2002. Because of space limitations, we only show the maps of the municipalities that introduced bottle, paper, and plastic container collections in 2002 instead of showing the maps from 2000 to 2002. 4.2. Independent variables Next, we explain the independent variables used in our models, which can be divided into three categories: (1) incineration and RDF, (2) landfill, and (3) others. All the independent variables except wage rate and population density are obtained from a national census of waste management conducted by the Ministry of the Environment, 1999–2002). Wage rate and population density are drawn from an appraisal of the municipalities’ accounts and the Basic Resident Register (nationwide census), respectively.

Table 3 Descriptive statistics (only dependent variables). Variables

Mean

Glass collection dummy 0.676 Overall Between Within Plastic collection dummy 0.268 Overall Between Within Paper collection dummy Overall 0.521 Between Within

SD

Min

Max

0.468 0.452 0.122

0 0 0.009

1 1 1.343

0.443 0.411 0.166

0 0 −0.398

1 1 0.935

0.500 0.474 0.159

0 0 −0.146

1 1 1.187

Note: The standard deviations of the dependent variables were decomposed into between and within components.

Fig. 2. Spatial distribution related to introduction of paper container collections, 2002.

88

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

landfill sites inside a single municipality and the other representing the sharing of landfill sites with adjacent municipalities. In the estimation, to avoid multicollinearity, we drop the dummy variables when no landfill site is owned. The incentive to conserve a landfill differs with the type of landfill possession, because solely owning a landfill site is like owning a private good, whereas sharing a site with adjacent municipalities is like sharing a public good that cannot be used exclusively. Therefore, the expected sign of the coefficient of landfill sites within a municipality is positive if there is no other possession and negative if the landfill site is shared. We add a variable for landfill capacity (year), that is, the number of years a municipality can bury waste at the current rate of landfill waste production in order to quantitatively consider the nature of land scarcity. The expected sign for all equations is negative.

Fig. 3. Spatial distribution related to introduction of plastic container collections, 2002.

We interpret the coefficients in terms of costs and benefits: if a municipality starts recyclables collection, its benefits increase through a reduction in the collection costs of combustible and incombustible waste (or mixed waste). It also saves the opportunity cost of landfill siting. By contrast, some costs increase, such as those related to vehicles used for recyclables collection, labor, and the creation of separating facilities for handling recyclable waste (Porter, ` 2002). To omit simultaneous decisions vis-a-vis recyclables collection and facility choices, we use a one-year lag for the independent variables. 4.2.1. Incineration and RDF We include three dummy variables relating to waste-to-energy facilities, namely possessing an incineration facility, waste power generation, and RDF facilities. We define a dummy variable for possessing an incineration facility inside the municipality. The expected sign of the estimated coefficient for plastic and paper containers is negative. This is because a municipality may have a negative attitude toward initiating the collection of paper or plastic containers to increase the energy efficiency of an incineration facility. Otherwise, a municipality may need to add heating oil to burn wet waste. We also define a dummy variable for waste power generation. This variable is defined in terms of whether a municipality has an incineration plant supplemented with an electric generator.4 The expected sign of the coefficient is negative for paper and plastic containers, because these materials have such a high caloric content that the facility can produce a substantial amount of electricity for sale. The equation includes a dummy variable for the possession of RDF facilities. This coefficient is expected to be negative, because RDF facilities process combustible recyclables such as paper and plastic containers. 4.2.2. Landfill Here, we address landfill scarcity qualitatively. We define two landfill dummy variables, one representing the possession of

4 There were no private landfill sites or incineration plants in these data periods. However, nowadays, landfill management or incineration plants can be carried out by private contractors under private finance initiatives. Because our data are from before this privatization trend, we need not consider private landfill companies.

4.2.3. Other variables Finally, we define the other five control variables that affect the costs and benefits of recyclables collection. First, we introduce the annual per-capita wage rate of municipal employees. This variable proxies for collection cost, which mainly relates to the labor costs for collection. Here, the total labor cost in a municipality is divided by the number of municipal employees. Second, the outsourced proportion is defined by the proportion of waste collection consigned to a private collector (in tons). Third, we define the amount of total waste generated per year as a variable used as a proxy of scale. Fourth, population density (person per km2 ) is a proxy of land price for waste separation facilities.5 Finally, we add a dummy, namely year t (benchmark year = 2000), which absorbs the macro-level shock of all municipalities. 4.3. Spatial weight matrix We use a contiguity-based spatial weight matrix for this study, defining two areas as adjacent if they share a border or vertex. We estimate equations for each type of recyclable material, using balanced panel data on the 2951 municipalities for × 3 years, as already mentioned, and we then apply the spatial autoregressive panel probit model. 5. Results In this section, we show the estimation results obtained by using the spatial autoregressive panel probit model. We ran the Markov chain Monte Carlo (MCMC) algorithm, applying 50,000 iterations and discarding the first 10,000 iterations.6 For the prior distributions, we set the hyper-parameters as follows: ˇ0 = 0, 0 = 100 × Ik , 0 = 0, 02 = 100, 0 = 2, 0 =0.01. (4) The estimation results are presented in Tables 4–6. 5.1. Spatial interaction The spatial interaction at time t, t , was positive and did not include zero in the 95% credible intervals for all years and equations. A municipality’s implementation of recyclables collection or separation might have a spatial interaction with its neighboring

5 6

We use this proxy because we are unable to acquire any land price variable data. As a rule of thumb, we discard the first 10,000 iterations.

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

89

Table 4 Estimation result: Glass bottles.

Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Waste generation (tons) Wage rate Population density Outsourced proportion Year 01 Year 02 ˛ 2 0 1 2

Mean

SD

2.5%CI

97.5%CI

−0.159 0.181 0.439 1.009 −0.007 0.128 5.661 −1.754 −0.694 0.089 0.141 0.259 2.009 10.306 0.386 0.409 0.433

0.214 0.228 0.443 0.208 0.167 0.238 0.845 1.273 0.167 0.126 0.070 0.075 1.257 1.049 0.022 0.022 0.021

−0.587 −0.260 −0.386 0.597 −0.349 −0.331 3.962 −4.355 −1.025 −0.163 0.003 0.113 0.355 8.445 0.341 0.361 0.390

0.253 0.641 1.358 1.425 0.316 0.607 7.239 −0.01 −0.368 0.329 0.278 0.409 4.583 12.432 0.429 0.451 0.472

Note: Mean = the posterior mean; Std = the posterior standard deviation; CI = the posterior credible interval. Table 5 Estimation result: Paper containers.

Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Waste generation (tons) Wage rate Population density Outsourced proportion Year 01 Year 02 ˛ 2 0 1 2

Mean

SD

2.5%CI

97.5%CI

0.006 0.169 −0.713 1.047 −0.243 −0.129 3.154 −1.571 −0.261 0.534 0.240 0.585 0.619 13.648 0.293 0.318 0.367

0.223 0.243 0.541 0.212 0.177 0.202 0.624 1.318 0.177 0.132 0.062 0.069 1.295 1.253 0.023 0.022 0.021

−0.436 −0.317 −1.773 0.621 −0.605 −0.532 1.955 −4.246 −0.615 0.271 0.119 0.451 −1.092 11.395 0.249 0.273 0.324

0.440 0.655 0.359 1.467 0.096 0.260 4.418 0.260 0.084 0.790 0.363 0.720 3.231 16.365 0.338 0.362 0.407

Note: Mean = the posterior mean; Std = the posterior standard deviation; CI = the posterior credible interval. Table 6 Estimation result: Plastic containers.

Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Waste generation (tons) Wage rate Population density Outsourced proportion Year 01 Year 02 ˛ 2 0 1 2

Mean

SD

2.5%CI

97.5%CI

−0.193 −0.145 −2.298 1.142 0.063 0.123 0.690 −1.893 −0.244 0.244 0.475 0.994 −0.633 10.268 0.317 0.312 0.388

0.232 0.229 0.693 0.207 0.174 0.205 0.432 1.236 0.168 0.105 0.08 0.093 1.226 1.080 0.025 0.024 0.022

−0.652 −0.603 −3.673 0.738 −0.290 −0.283 −0.173 −4.395 −0.580 0.025 0.322 0.814 −2.213 8.339 0.269 0.265 0.345

0.260 0.315 −0.973 1.551 0.397 0.529 1.532 −0.208 0.082 0.446 0.634 1.175 1.855 12.475 0.364 0.358 0.433

Note: Mean = the posterior mean; Std = the posterior standard deviation; CI = the posterior credible interval.

municipalities. A municipality may construct a recyclable waste sorting facility jointly with its adjacent municipality if both of them can gain from its cost-effectiveness. The coefficients of glass bottles are the largest among the equations. Glass bottle collection programs have traditionally been implemented by many municipalities prior to the Recycling Law (once again, see Table 2). Therefore, the estimated results can be interpreted thus: recycling

facilities are often shared among adjacent municipalities in order to reduce separation costs, making use of scale economies.7

7 We additionally estimate the other weight matrices, such as adjacent income and population matrices. The results are quite similar to those presented here, indicating the robustness of our model in terms of variable selection. Those results are available from the authors upon request.

90

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

5.2. Interpretation of direct and indirect effects First, we describe the direct, indirect, and total effects. Since we estimated  as a variable in time, we can decompose the direct and indirect effects over the years (see Tables 7–9 ). The direct effect means only the effect within the municipality, whereas the indirect effect represents the spillover effect from adjacent municipalities. The sum of the direct and indirect effects is the total effect, which is calculated as shown below. We base our interpretation of the total, direct, and indirect effects on the discussion of LeSage and Pace (2009) and LeSage and Dominguez (2012) and interpret the parameters from ˇ and the Bayesian panel spatial autoregressive probit model in Equation (2). In order to interpret the total, direct, and indirect effects including the spatial spillover, we use the coefficient estimates of the rth explanatory variables to summarize the average (cross-sample) impact of changing the rth explanatory variable on the dependent variable vector z. To see this, we rewrite the model in (2) as a matrix form into Eq. (5).

95% credible intervals.8 On the other hand, the estimated coefficient of the dummy variable for RDF facilities did not include zero in the 95% credible intervals in the equation for plastic containers, and as expected, the sign is negative (see Table 6). Plastic containers can be used as solid waste fuel, and some municipalities may consider them to be important for use in RDF facilities and choose not to recycle plastics in order to reduce their RDF facility costs. The cost of saving landfill sites was controlled for in the equation, and thus it does not have to be considered here. If a municipality with an RDF facility recycles plastics, it increases not only its explicit costs (e.g., the collection and separation costs for recycling) but also its implicit costs (e.g., the benefits from supplying fuels that are forsaken to pursue recyclables collection).

5.3. Estimation results

5.3.2. Landfill We found evidence that the degree of landfill site ownership of a municipality affects the probability of initiating collection and separation services. The estimated coefficient of a dummy variable for being the sole owner of a landfill site was positive and did not include zero in the 95% credible intervals for all equations, including glass bottles and plastic and paper containers (Tables 7–9). In other words, municipalities that have their own landfill sites show a higher probability of offering collection or separation services for recycling than those that do not. To interpret the introduction of recyclables collection programs from an economic perspective, if the net present value of the benefits of such a program is positive, a municipality initiates it. The following assumptions are inherent in this assertion: the benefits (calculated at the present value) of recyclables collection mitigate the opportunity costs associated with the landfill by removing recyclable waste from combustible or incombustible waste; combustible or incombustible waste collection costs drop (we exclude labor costs because we use its proxy variable); and costs (calculated at the present value) include additional collection costs associated with programs for recyclables collection. We investigate why municipalities without landfill sites are less likely to collect recyclables than those solely owning landfill sites. Municipalities that do not possess landfill sites pay tipping fees by volume of waste to municipalities that do. The tipping fees paid by municipalities without landfill sites are often set equal to the average cost of landfill disposal for municipalities solely owning landfill sites (according to an interview with municipality personnel conducted on July 7, 2014 about the costs related to landfill disposal). The costs related to landfill disposal for those solely possessing landfill sites explicitly include the cost of building landfill sites (fixed cost) and the operating cost (variable cost). The average cost of landfill disposal for municipalities solely owning landfill sites implies the amount of these explicit total costs divided by the quantity of waste to which the tipping fees are set equal. However, the costs related to landfill disposal implicitly include more than these. For example, before they build their own landfill sites, municipalities are quite likely to have faced many difficulties such as opposition from the inhabitants around the planned sites. If such implicit costs are also taken into consideration, the average cost of landfill disposal for municipalities solely owning landfill sites exceeds the tipping fees for those not owning landfill sites. Therefore, municipalities that have their own landfill sites are more highly motivated to initiate recyclables collection, which can reduce landfill waste and save the lifespan of landfill sites.

5.3.1. Incineration and RDF The estimated coefficients of all variables related to incineration facilities, such as the dummy variables for possessing incineration facilities and those for waste power generation, included zero in the

8 These results are estimated using the Bayesian method. Therefore, we interpret our coefficients in the Bayesian style.

z = (IT −  ⊗ W)

−1

[iT ⊗ ˛ + Xˇ + ]

(5)

In the period t, (In − t W)zt = ˛ + Xt ˇ + t , zt =

k 

Srt (W)X,tr + V (W)˛ + V (W)t ,

(6)

(7)

r=1

where X,tr is the rth row of Xt , Srt (W) = ˇr V(W), V (W) = (In − t W)−1 = In + t W + t2 W2 + t3 W3 + . . .. In this situation, the derivative of zit with respect to xjtr is

∂zit = Srt (W)ij , ∂xjtr

(8)

and the derivative of zit with respect to xitr is

∂zit = Srt (W)ii . ∂xitr

(9)

Thus, the direct and indirect effects are defined by Mt (r)direct = n−1 tr(Srt (W)), Mt (r)total =

n−1 in Srt (W)in ,

Mt (r)indirect = Mt (r)total − Mt (r)direct .

(10) (11) (12)

In econometrics, Eq. (5) is referred to as the data-generating process. If we take a partial derivative of z in Eq. (5) with respect to a change in the rth variable xitr , we can obtain the Eqs. (10), (11), and (12). We briefly explain all the estimation results in Tables 4–6. First, all the direct effects are constant over time, whereas the indirect effects vary, changing particularly dramatically in 2002. Second, the direct and indirect effects have the same sign for every estimated value. Third, the size of the posterior mean of the indirect effects ranges from one-third to half compared with the size of the direct effects. We next explain the total effects in Section 5.3 and present the estimation results in Tables 7–9.

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

91

Table 7 Direct, indirect, and total effect: Glass bottles. Year 00

Year 01

Year 02

Mean

SD

2.5%CI

97.5%CI

Mean

SD

2.5%CI

97.5%CI

Mean

SD

2.5%CI

97.5%CI

Direct effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.194 0.141 0.364 5.257 0.973 −0.026 0.153 −2.306 −0.710 0.127 – –

0.193 0.227 0.394 0.892 0.184 0.192 0.245 0.310 0.152 0.105 – –

−0.541 −0.416 −0.345 3.661 0.556 −0.433 −0.336 −2.962 −1.000 −0.083 – –

0.188 0.545 1.136 6.872 1.356 0.347 0.669 −1.679 −0.409 0.326 – –

−0.195 0.142 0.366 5.286 0.979 −0.027 0.154 −2.318 −0.714 0.128 – –

0.194 0.228 0.396 0.899 0.186 0.194 0.247 0.306 0.153 0.105 – –

−0.541 −0.421 −0.348 3.706 0.561 −0.436 −0.336 −2.961 −1.007 −0.083 – –

0.190 0.547 1.147 6.934 1.368 0.349 0.678 −1.699 −0.408 0.328 – –

−0.196 0.143 0.369 5.321 0.985 −0.027 0.155 −2.333 −0.718 0.129 – –

0.196 0.230 0.399 0.907 0.186 0.195 0.248 0.309 0.154 0.106 – –

−0.546 −0.422 −0.349 3.695 0.563 −0.435 −0.340 −2.980 −1.018 −0.084 – –

0.191 0.552 1.152 6.974 1.369 0.349 0.681 −1.709 −0.413 0.331 – –

Indirect effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.109 0.076 0.200 2.959 0.545 −0.015 0.084 −1.291 −0.399 0.070 – –

0.110 0.127 0.220 0.637 0.119 0.114 0.138 0.213 0.100 0.058 – –

−0.315 −0.247 −0.194 1.868 0.337 −0.289 −0.198 −1.844 −0.611 −0.047 – –

0.102 0.290 0.640 4.314 0.813 0.204 0.369 −0.947 −0.215 0.179 – –

−0.122 0.083 0.222 3.262 0.601 −0.017 0.094 −1.415 −0.440 0.076 – –

0.125 0.142 0.246 0.761 0.142 0.126 0.155 0.206 0.114 0.064 – –

−0.357 −0.295 −0.208 1.980 0.379 −0.354 −0.206 −1.851 −0.662 −0.056 – –

0.111 0.328 0.745 4.863 0.908 0.217 0.445 −1.030 −0.236 0.197 – –

−0.134 0.094 0.249 3.625 0.666 −0.020 0.104 −1.573 −0.490 0.086 – –

0.137 0.158 0.271 0.781 0.136 0.136 0.169 0.197 0.124 0.072 – –

−0.390 −0.309 −0.235 2.265 0.419 −0.347 −0.233 −1.967 −0.749 −0.060 – –

0.127 0.374 0.798 5.167 0.965 0.234 0.463 −1.230 −0.265 0.224 – –

Total effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.303 0.217 0.564 8.216 1.519 −0.041 0.237 −3.597 −1.109 0.197 0.231 0.418

0.303 0.354 0.612 1.488 0.296 0.306 0.382 0.500 0.248 0.162 0.116 0.114

−0.861 −0.669 −0.538 5.562 0.906 −0.735 −0.531 −4.718 −1.600 −0.130 −0.002 0.178

0.289 0.834 1.765 10.944 2.161 0.548 1.040 −2.717 −0.635 0.501 0.459 0.650

−0.317 0.225 0.588 8.548 1.580 −0.044 0.248 −3.733 −1.154 0.204 0.237 0.433

0.319 0.369 0.638 1.601 0.317 0.319 0.401 0.467 0.260 0.168 0.115 0.116

−0.904 −0.719 −0.549 5.871 0.940 −0.793 −0.548 −4.726 −1.657 −0.141 −0.002 0.191

0.299 0.872 1.847 11.716 2.263 0.568 1.124 −2.916 −0.657 0.522 0.456 0.668

−0.331 0.237 0.617 8.946 1.651 −0.047 0.259 −3.906 −1.208 0.214 0.250 0.452

0.332 0.387 0.668 1.659 0.316 0.330 0.417 0.482 0.276 0.178 0.122 0.115

−0.932 −0.729 −0.586 6.035 0.990 −0.783 −0.574 −4.927 −1.764 −0.145 −0.003 0.206

0.322 0.922 1.948 12.103 2.313 0.580 1.149 −2.976 −0.676 0.549 0.483 0.672

Note: Mean = the posterior mean; Std = the posterior standard deviation; CI = the posterior credible interval.

5.3.3. Other variables The estimated coefficients of the wage rate variable did not include zero in the 95% credible intervals and were found to be negative for all equations. These results imply that increasing labor costs for collection and separation make municipalities more reluctant to implement recyclables collection and separation programs. More specifically, the marginal collection costs of recyclables are more expensive than the marginal collection costs of mixed waste.9 According to the empirical evidence presented by Stevens (1994) and Porter (2002), the increase in the costs of collecting recyclables is much greater in absolute value than the reduction in the cost of collecting combustible and incombustible waste. In other words, recyclables collection becomes expensive despite a decline in the volume of combustible and incombustible waste. Hence, the wage rate is positively correlated with the cost of introducing recycling. The coefficient of the outsourced percentage was negative and did not include zero in the 95% credible intervals for the glass bottle equation. This result indicates that the municipal consignment ratio is negatively correlated with the probability of municipalities implementing collection or separation programs for glass bottles. The reasons for this finding may be related to financial problems or

9 METI (2005) provides evidence of this in a cost–benefit analysis based on the Recycling Law undertaken through a bottom-up survey of some municipalities.

business judgment. In terms of the former reason, because a municipality is in a fragile financial condition, it outsources its collection and separation services. Such consigned private company-run recyclables collection programs incur lower costs than those directly run by municipalities. In fact, most consigned private companies already have their own separation facilities and can thus provide collection services at lower costs. If outsourcing is a sign of a municipality’s financial stringency, the municipality is less likely to recycle the heaviest (i.e., most expensive) material, namely glass bottles. In terms of the latter reason (business judgment), a municipality outsources its collection and separation services to minimize costs. Thus, again, it is less likely to recycle glass bottles, a process that incurs the highest cost among these three materials. The estimated coefficients of the total waste generation variables had positive signs and did not include zero in the 95% credible intervals for glass bottles or paper containers, which is consistent with our expectation. The implication here is that total waste generation in the previous year is positively correlated with the probability of a municipality introducing recyclables collection or separation programs. This result provides evidence of scale economies in the establishment of recycling facilities. 5.3.4. Random effect Figs. 4–6 are graphic representations of the spatial patterns of the estimated random effects and posterior means of ˛i for each equation. The darker the shade of blue for a municipality in this

92

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

Table 8 Direct, indirect, and total effect: Paper containers. Year 00

Year 01

Year 02

Mean

SD

2.5%CI

97.5%CI

Mean

SD

2.5%CI

97.5%CI

Mean

SD

2.5%CI

97.5%CI

Direct effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.036 0.129 −0.796 3.122 1.035 −0.276 −0.131 −2.305 −0.273 0.549 – –

0.196 0.236 0.437 0.731 0.193 0.209 0.189 0.345 0.167 0.107 – –

−0.458 −0.467 −1.646 2.035 0.634 −0.710 −0.518 −2.964 −0.602 0.363 – –

0.300 0.527 0.062 4.713 1.457 0.142 0.225 −1.549 0.070 0.756 – –

−0.036 0.130 −0.799 3.132 1.039 −0.277 −0.132 −2.312 −0.274 0.551 – –

0.196 0.237 0.438 0.733 0.194 0.210 0.189 0.343 0.168 0.108 – –

−0.460 −0.471 −1.648 2.042 0.634 −0.710 −0.520 −2.978 −0.605 0.364 – –

0.300 0.526 0.062 4.731 1.464 0.142 0.226 −1.551 0.070 0.756 – –

−0.037 0.131 −0.806 3.162 1.048 −0.280 −0.133 −2.334 −0.277 0.556 – –

0.198 0.239 0.442 0.743 0.195 0.212 0.191 0.346 0.169 0.109 – –

−0.464 −0.475 −1.666 2.056 0.642 −0.718 −0.526 −3.000 −0.611 0.368 – –

0.304 0.532 0.063 4.800 1.474 0.143 0.228 −1.573 0.071 0.765 – –

Indirect effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.012 0.046 −0.299 1.155 0.384 −0.102 −0.051 −0.857 −0.101 0.205 – –

0.074 0.089 0.175 0.275 0.082 0.081 0.071 0.176 0.063 0.050 – –

−0.166 −0.182 −0.691 0.637 0.241 −0.294 −0.203 −1.266 −0.231 0.127 – –

0.119 0.191 0.021 1.726 0.581 0.054 0.079 −0.599 0.024 0.305 – –

−0.016 0.052 −0.329 1.276 0.425 −0.112 −0.055 −0.943 −0.113 0.226 – –

0.081 0.098 0.185 0.291 0.089 0.088 0.077 0.146 0.071 0.052 – –

−0.189 −0.197 −0.693 0.815 0.263 −0.332 −0.212 −1.289 −0.260 0.140 – –

0.123 0.215 0.025 1.928 0.622 0.060 0.088 −0.685 0.029 0.333 – –

−0.019 0.065 −0.411 1.616 0.533 −0.144 −0.068 −1.187 −0.142 0.285 – –

0.102 0.124 0.230 0.412 0.103 0.112 0.097 0.186 0.089 0.066 – –

−0.232 −0.255 −0.875 0.920 0.335 −0.405 −0.273 −1.574 −0.325 0.173 – –

0.153 0.284 0.031 2.630 0.760 0.071 0.113 −0.862 0.033 0.418 – –

Total effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.049 0.175 −1.095 4.277 1.419 −0.378 −0.182 −3.162 −0.374 0.754 0.350 0.840

0.269 0.325 0.609 0.984 0.266 0.290 0.259 0.502 0.229 0.153 0.089 0.091

−0.633 −0.645 −2.298 2.673 0.900 −0.995 −0.715 −4.187 −0.828 0.495 0.176 0.663

0.417 0.721 0.083 6.440 2.001 0.195 0.304 −2.209 0.095 1.060 0.522 1.021

−0.052 0.182 −1.127 4.407 1.464 −0.389 −0.186 −3.255 −0.387 0.778 0.360 0.866

0.277 0.335 0.621 1.008 0.278 0.297 0.266 0.472 0.238 0.157 0.087 0.089

−0.645 −0.670 −2.353 2.864 0.913 −1.041 −0.738 −4.237 −0.866 0.508 0.188 0.695

0.419 0.735 0.086 6.630 2.086 0.203 0.312 −2.274 0.100 1.087 0.530 1.035

−0.055 0.196 −1.218 4.778 1.582 −0.423 −0.201 −3.521 −0.419 0.841 0.390 0.936

0.301 0.363 0.670 1.139 0.290 0.324 0.288 0.512 0.258 0.173 0.096 0.093

−0.694 −0.728 −2.531 3.014 0.986 −1.114 −0.795 −4.548 −0.938 0.554 0.199 0.759

0.460 0.810 0.094 7.440 2.179 0.214 0.340 −2.472 0.106 1.171 0.580 1.118

Note: Mean = the posterior mean; Std = the posterior standard deviation; CI = the posterior credible interval.

Fig. 4. Spatial distribution of posterior mean of the random effects in glass bottle, ˛i .

Fig. 5. Spatial distribution of posterior mean of the random effects in paper container, ˛i .

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

93

Table 9 Direct, indirect, and total effect: Plastic containers. Year 00

Year 01

Year 02

Mean

SD

2.5%CI

97.5%CI

Mean

SD

2.5%CI

97.5%CI

Mean

SD

2.5%CI

97.5%CI

Direct effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.240 −0.178 −2.362 0.788 1.153 0.048 0.129 −2.744 −0.260 0.248 – –

0.209 0.224 0.575 0.433 0.193 0.188 0.198 0.357 0.164 0.094 – –

−0.676 −0.734 −3.377 −0.033 0.740 −0.360 −0.261 −3.394 −0.590 0.073 – –

0.131 0.222 −1.186 1.753 1.555 0.403 0.535 −1.876 0.064 0.434 – –

−0.240 −0.178 −2.356 0.786 1.150 0.047 0.129 −2.738 −0.259 0.248 – –

0.208 0.223 0.571 0.431 0.193 0.187 0.198 0.356 0.164 0.095 – –

−0.675 −0.728 −3.364 −0.033 0.739 −0.360 −0.260 −3.397 −0.589 0.072 – –

0.132 0.223 −1.188 1.730 1.556 0.401 0.534 −1.872 0.064 0.434 – –

−0.243 −0.181 −2.392 0.798 1.168 0.048 0.131 −2.780 −0.263 0.252 – –

0.211 0.227 0.578 0.437 0.195 0.190 0.200 0.363 0.166 0.096 – –

−0.684 −0.741 −3.412 −0.033 0.747 −0.364 −0.263 −3.448 −0.598 0.073 – –

0.134 0.228 −1.205 1.757 1.575 0.405 0.539 −1.887 0.065 0.440 – –

Indirect effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.101 −0.076 −0.998 0.331 0.482 0.019 0.053 −1.143 −0.108 0.104 – –

0.089 0.095 0.306 0.192 0.101 0.082 0.083 0.183 0.070 0.042 – –

−0.287 −0.325 −1.655 −0.013 0.322 −0.179 −0.110 −1.542 −0.253 0.028 – –

0.051 0.086 −0.444 0.809 0.674 0.174 0.221 −0.871 0.028 0.187 – –

−0.097 −0.070 −0.938 0.310 0.456 0.019 0.050 −1.082 −0.104 0.099 – –

0.084 0.090 0.255 0.164 0.090 0.078 0.077 0.142 0.068 0.042 – –

−0.276 −0.286 −1.415 −0.012 0.310 −0.164 −0.104 −1.349 −0.247 0.027 – –

0.052 0.091 −0.462 0.626 0.666 0.167 0.209 −0.833 0.026 0.187 – –

−0.135 −0.100 −1.329 0.439 0.647 0.027 0.072 −1.542 −0.146 0.141 – –

0.118 0.127 0.337 0.236 0.114 0.106 0.109 0.228 0.092 0.058 – –

−0.390 −0.412 −1.915 −0.020 0.426 −0.211 −0.141 −1.961 −0.332 0.040 – –

0.076 0.136 −0.644 0.922 0.883 0.224 0.281 −1.151 0.037 0.258 – –

Total effect Dummy: possessing incineration Dummy: possessing waste power generation Dummy: possessing RDF Waste generation (tons) Dummy: possessing landfill solely Dummy: sharing landfill Landfill capacity (year) Wage rate Outsourced proportion Population density Year 01 Year 02

−0.342 −0.255 −3.360 1.119 1.635 0.067 0.182 −3.887 −0.368 0.352 0.694 1.474

0.297 0.319 0.866 0.622 0.284 0.269 0.281 0.502 0.233 0.135 0.102 0.104

−0.962 −1.053 −5.067 −0.045 1.089 −0.528 −0.369 −4.818 −0.839 0.101 0.498 1.285

0.185 0.308 −1.638 2.580 2.209 0.584 0.754 −2.854 0.094 0.620 0.895 1.671

−0.337 −0.248 −3.295 1.096 1.607 0.066 0.179 −3.820 −0.363 0.347 0.684 1.451

0.292 0.313 0.815 0.593 0.274 0.265 0.275 0.472 0.231 0.135 0.113 0.119

−0.950 −1.015 −4.755 −0.044 1.063 −0.520 −0.359 −4.746 −0.829 0.099 0.467 1.225

0.181 0.313 −1.634 2.323 2.217 0.569 0.733 −2.814 0.091 0.619 0.919 1.674

−0.379 −0.280 −3.720 1.237 1.815 0.075 0.203 −4.322 −0.409 0.392 0.773 1.642

0.329 0.353 0.904 0.671 0.301 0.296 0.309 0.571 0.257 0.153 0.126 0.156

−1.064 −1.150 −5.254 −0.053 1.173 −0.567 −0.406 −5.353 −0.923 0.113 0.530 1.356

0.208 0.363 −1.848 2.650 2.455 0.630 0.817 −3.105 0.103 0.697 1.020 1.973

Note: Mean = the posterior mean; Std = the posterior standard deviation; CI = the posterior credible interval.

seems to capture an unobserved effect among the adjacent municipalities, which could be distinct from the spatial interaction terms t (Kakamu and Wago, 2008). 6. Conclusion

Fig. 6. Spatial distribution of posterior mean of the random effects in plastic container, ˛i .

figure, the lower is the possibility that it introduces recyclables collection; conversely, the darker the shade of pink, the greater is the possibility of introducing recyclables collection. Each map

The purpose of this study was to estimate a municipality’s decision factors relating to the collection and separation of recyclable containers and packaging material under the Japanese Recycling Law. We used a spatial autoregressive panel probit model from a Bayesian perspective. This study focused on the various waste management objectives of local governments and the differences between the approaches of local and national governments. In particular, we investigated municipalities’ cost minimization behavior, such as inter-municipal cooperation in recyclables collection, incineration of recyclable materials, and saving scarce municipal landfill sites. We used municipal recyclables collection data, which relate directly to the issue of waste management, to show why some municipalities have recycling programs, while others do not. The municipality’s decision to collect recyclable material for incineration enables cannibalizing recyclables collections: municipalities with an RDF facility are less likely to collect and separate plastic containers because they can use these to generate energy. The decision of a municipality to implement recyclables collection also depends on whether it owns landfill sites, and not

94

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

on landfill capacity. In particular, the probability of a municipality introducing recyclables collection is higher when it owns landfill sites solely than when it has no possession whatsoever. Spatial interaction constitutes a topic of concern drawing increased attention in economics research. Our econometric model allows us to control for a municipality’s spatial interaction with adjacent municipalities and provides an accurate estimation with the use of available panel data. The coefficients of spatial correlation do not include zero in the 95% credible intervals and are positive for all years and all equations. This finding means that a municipality’s implementation of recyclables collection or separation programs shows spatial interaction with its neighboring municipalities owing to collaborative efforts to implement a recyclables facility and reduce costs; these collaborative efforts are seen to grow stronger year by year. With regard to policy recommendations, the national government may find it necessary to offer some sort of financial incentive to municipalities if it wants them to introduce recyclables collection programs. The estimated coefficients of the variable of total waste generation show positive signs for glass bottles and paper containers. This result provides evidence of scale economies with regard to the establishment of recycling facilities. This result further shows that small municipalities that do not benefit from scale economies when building sorting facilities may need financial support to implement recyclables collection services. Acknowledgements

sampler (e.g., see Gelfand and Smith, 1990), the random draws from the posterior distribution of {zt }Tt=1 , {t }Tt=1 , {˛i }ni=1 , ˇ, , and  2 are generated. A.1.1. Sampling {zt }Tt=1 In the case of the probit model, it is required to generate the latent variables {zt }Tt=1 . Let us define z−it = {z1t , . . ., zi−1,t , zi+1,t , . . ., znt }, where zit is excluded from zt . Tanner and Wong (1987) propose a data augmentation method to generate latent variables, and we make use of that here. The full conditional distribution of zit follows: zit |t , ˛i , ˇ, z−it , yt , xit , W∼Nait ≤zit ≤bit (ˆzit , 1), with zˆit = ˛i +

n

w y j=1 t ij jt

(14)

+ xit ˇ. Na
truncated normal distribution with mean  and variance 2 , where z is distributed between a and b (see LeSage and Pace, 2009). In this study, truncation is set to (ait , bit ) = (0, ∞) when yit = 1, and (ait , bit ) = (− ∞ , 0) when yit = 0. A.1.2. Sampling {t }Tt=1 Define −t = {1 , . . ., t−1 , t+1 , . . ., T }, where t is excluded from . From (13), the full conditional distribution of t is written as:



p(t |−t , ˛, ˇ, ,  2 , z, y, X, W) ∝ |In − t W| exp

−

et et 2



,

(15)

This study was partially supported by the Asahi Glass Foundation and by JSPS KAKENHI Grant Numbers 25245035 and 26380331. Any errors in this paper are the author’s responsibility.

with et = (In − t W)zt − Xt ˇ − ˛. Using the Metropolis algorithm (e.g., see Tierney, 1994), a random draw of t is sampled from the conditional distribution p(t |−t , ˛, ˇ, ,  2 , z, y, X, W). The following Metropolis step is used: (i) sample tnew from:

Appendix A. Joint posterior distribution

tnew = told + ct t ,

Since we utilize a Bayesian method for estimation, we adopt the following hierarchical prior: 2

(, ˛, ˇ, ,  ) =

 T

 n

(˛i |,  )

(t )

t=1

2

(ˇ)()( ),

(, ˛, ˇ, ,  2 |z, y, X, W) T

∝ (, ˛, ˇ, ,  2 )

f (yt |t , ˛, ˇ, zt , Xt , W).

(13)

t=1

We assume the prior distributions, i.e., (t ), (˛i |,  2 ), (ˇ), () and ( 2 ), as follows: t ∼U(−1, 1), ∼N(0 , 02 ),

˛i

|,  2 ∼N(,  2 ),

 2 ∼IG(

ω(told , tnew ) = min

i=1

which is used in Kakamu and Wago (2008); there,  and  2 indicate the mean and variance of ˛i , respectively. Given the prior density (, ˛, ˇ, ,  2 ) and the likelihood function (3), the joint posterior distribution can be expressed as:

ˇ∼N(ˇ0 , 0 ),

0 0 , ), 2 2

where IG(a, b) denotes the inverse gamma distribution with the scale parameter a and the shape parameter b. A.1. Posterior simulation From the joint posterior distribution (13), we can implement the MCMC method. The Markov chain sampling scheme can be constructed from the full conditional distributions of {zt }Tt=1 , {t }Tt=1 , {˛i }ni=1 , ˇ, , and  2 , which are shown as follows. Using the Gibbs

(16)

where ct is called the tuning parameter and told denotes the random draw previously sampled, (ii) evaluate the acceptance probability:



2

t ∼N(0, 1),

p(tnew |−t , ˛, ˇ, ,  2 , z, y, X, W) p(told |−t , ˛, ˇ, ,  2 , z, y, X, W)

 ,1

,

and (iii) set t = tnew with probability ω(told , tnew ) and t = told otherwise. The proposed random draw of t generated from (16) takes the real value within the interval (− ∞ , ∞), although the prior distribution of t lies in the interval between −1 and 1. If the candidate of t does not lie in the interval (− 1, 1), the conditional posterior should be zero; accordingly, the proposed value would be rejected with a probability of one (see Chib and Greenberg, 1998). We need the tuning parameter ct in sampling t . In the numerical example discussed below, we choose the tuning parameter, such that the acceptance rate is between 0.4 and 0.6 (see Holloway et al., 2002). A.1.3. Sampling the other parameters The full conditional distribution of ˇ is: ˆ ), ˆ ˇ|, ˛, ,  2 , z, y, X, W∼N(ˇ,

(17)

ˆ = {X ˆ  (z − ( ⊗ W)z − ˛) + −1 ˇ0 },  = (1T ⊗ In ), 1T is a with ˇ 0 −1

ˆ = (X X + −1 ) . The element of  in T × 1 unit vector, and  0 row t and column s is given by t for t = s and zero for t = / s, i.e.,  = diag(). z, , and ˇ, (2) are written as: zit −

n  j=1

t wij zjt − xit ˇ = it ,

it ∼N(˛i , 1).

T. Usui et al. / Resources, Conservation and Recycling 101 (2015) 84–95

Let us define ˛−i = {˛1 , . . ., ˛i−1 , ˛i+1 , . . ., ˛n }, where ˛i is excluded from ˛. The full conditional distribution of ˛i is as follows: ˛i |, ˛−i , ˇ, ,  2 , z, y, X, W∼N(˛ ˆ i , ˆ 2 ),

(18)

T n with ˛ ˆ i = ˆ 2 { t=1 (zit −  w z − xit ˇ) +  −2 } and ˆ 2 = j=1 t ij jt −1

(T +  −2 ) . The full conditional distributions of  and  2 are given by: |, ˛, ˇ,  2 , z, y, X, W∼N(, ˆ ˆ 2 ),

(19)

ˆ ˆ   2 |, ˛, ˇ, , z, y, X, W∼IG( , ), 2 2

(20)

n −2

with  ˆ = ˆ 2 ( ˛ + 0−2 0 ), ˆ 2 = ( −2 n + 0−2 ) i=1 i  ˆ = (˛ − ) (˛ − ) + 0 . and 

−1

, ˆ = n + 0

Thus, from (14), (15), and (17)–(20), the random draws of (z, , ˛, ˇ, ,  2 ) are easily sampled from the Gibbs sampler (e.g., Gelfand and Smith, 1990). References Allers, M.A., Hoeben, C., 2010. Effects of unit-based garbage pricing: a differences-in-differences approach. Environ. Resour. Econ. 45, 405–428. Anselin, L., 1988. Spatial Econometrics: Methods and Models. Kluwer, London. Bel, G., Fageda, X., Mur, M., 2011. Why do municipalities cooperate to provide local public services? An empirical analysis. Local Gov. Stud. 39, 435–454. Bivand, R., Szymanski, S., 2000. Modelling the spatial competition of the introduction of compulsory competitive tendering. Reg. Sci. Urban Econ. 30, 203–219. Callan, S., Thomas, J.M., 2001. Economies of scale and scope: a cost analysis of municipal solid waste services. Land Econ. 77, 548–560. Carroll, W., 1995. The organization and efficiency of residential recycling services. Eastern Econ. J. 21, 215–225. Central Environment Council, 2005. Report from the 30th meeting of the Committee on the Waste Management and Recycling of the Central Environment Council , http://www.env.go.jp/council/03haiki/y030-30.html (in Japanese). Chib, S., Greenberg, E., 1998. Analysis of multivariate probit models. Biometrika 85, 347–361. Consonni, S., Giugliano, M., Grosso, M., 2005. Alternative strategies for energy recovery from municipal solid waste. Part A: Mass and energy balances. Waste Manage. 25, 123–135. Daskalopoulos, E., Badr, O., Probert, S.D., 1998. Municipal solid waste: a prediction methodology for the generation rate and composition in the European Union countries and the United States of America. Resour. Conserv. Recycl. 24, 155–166.

95

Dijkgraaf, E., Gradus, R.H.J.M., Melenberg, B., 2003. Contracting out refuse collection. Empirical Econ. 28, 553–570. Gelfand, A.E., Smith, A.F.M., 1990. Sampling-based approaches to calculating marginal densities. J. Am. Stat. Assoc. 85, 398–409. Holloway, G., Shankar, B., Rahman, S., 2002. Bayesian spatial probit estimation: a primer and an application to HYV rice adoption. Agric. Econ. 27, 383–402. Kakamu, K., Wago, H., 2008. Small sample properties of panel spatial autoregressive models: comparison of the Bayesian and maximum likelihood methods. Spatial Econ. Anal. 3, 305–319. Kakamu, K., Wago, H., Tanizaki, H., 2010. Estimation of regional business cycle in Japan using a panel spatial autoregressive model. In: Nolin, T.P. (Ed.), Handbook of Regional Economics. Nova Science Publishers, New York, pp. 555–571. Keeler, A.G., Renkow, M., 1994. Haul trash or haul ash: energy recovery as a component of local solid waste management. J. Environ. Econ. Manage. 27, 205–217. Kinnaman, T.C., 2005. Why do municipalities recycle? Topics Econ. Anal. Policy 5, 1–23. Kinnaman, T.C., Fullerton, D., 2000. Garbage and recycling with endogenous local policy. J. Urban Econ. 48, 419–442. LeSage, J.P., Dominguez, M., 2012. The importance of modeling spatial spillovers in public choice analysis. Public Choice 150, 525–545. LeSage, J.P., Pace, R.K., 2009. Introduction to Spatial Econometrics. CRC Press. Ministry of Economy, Trade and Industry, 2003. The Containers and Packaging Recycling Law - Make the Most of Our Resources!, Website: http://www.jcpra.or.jp/eng/cprl.html. Ministry of Economy, Trade and Industry, 2005. Youki-Housou Risaikuru Hou no Hiyou Kouka Bunseki (Effectiveness Analysis of the Containers and Packaging Recycling Law). http://www.meti.go.jp/committee/materials/downloadfiles/ g50616a40j.pdf (in Japanese). Ministry of Economy, Trade and Industry, 2007. Containers and Packaging Recycling Law: Outline of the Law. http://www.meti.go.jp/policy/recycle/ main/english/law/contain.html Ministry of the Environment, 1999–2002. Haikibutsu Syori Jigyo Jittai Chosa Toukei Siryou (National Survey of Municipal General Waste). http://www.env. go.jp/recycle/waste tech/ippan/index.html (in Japanese). Ministry of the Environment, 2001–2003. Results of Collection and Recycled Products under Containers and Packaging Recycling Law; 2000–02 Fiscal Year. OECD, 2001. Extended Producer Responsibility: A Guidance Manual for Governments. OECD Publishers, Paris. OECD, 2008. OECD Environmental Data COMPENDIUM 2006–2008. OECD, Washington, DC. Porter, R., 2002. The Economics of Waste. RFF Press. Stevens, B., 1977. Scale, market structure, and the cost of refuse collection. Rev. Econ. Stat. 60, 438–448. Stevens, B., 1994. Recycling collection costs by the numbers: a national survey. Resour. Recycl., November. Suwa, T., Usui, T., 2007. Estimation of garbage reduction and recycling promotion under the containers and packaging recycling law and garbage pricing. Environ. Econ. Policy Stud. 8, 239–254. Tanner, M.A., Wong, W.H., 1987. The calculation of posterior distributions by data augmentation. J. Am. Stat. Assoc. 82, 528–540. Tierney, L., 1994. Markov chains for exploring posterior distributions. Ann. Stat. 22, 1635–2134.