Decomposition of toxicity emission changes on the demand and supply sides: empirical study of the US industrial sector

Decomposition analysis is widely applied in energy and environmental research fields to clarify the driving forces or determinants underlying these changes (Wang et al 2017). According to Hoekstra and van den Bergh (2003), two major techniques that can identify changes in decomposition indicators at the sector level are structural decomposition analysis (SDA) and index decomposition analysis (IDA).

SDA was originally developed as an input–output analysis for understanding the change in consumption-based emissions (e.g. carbon footprint). One of the main advantages of SDA is that it clearly considers the economic system, including trade patterns, the combination of intermediate material inputs, and the scale of final demand.

IDA was developed in energy and environmental research fields to understand the intensity change caused by technological development, such as that of emission intensity due to resource efficiency (Wang et al 2017). IDA has several estimation options, including the Laspeyres index and logarithmic mean Divisia index (LMDI) (Ang 2015).

Notably, our decomposition analysis is divided into two steps: SDA is used for demand-side effects analysis (the first step), and IDA is used for supply-side effects analysis (the second step). We primarily use two different methods to distinguish the change in toxic intensity (the technology factor) and demand-side effects (the economic system factor).

In addition, IDA is advantageous in terms of decomposing emission intensity change. According to Wang et al (2017), SDA is beneficial when analysing demand-side effects and trade-related issues. By contrast, IDA is preferred for analysing the change in emission intensity because it is more flexible in modelling and easier to implement and understand compared with SDA.

The toxicity of emitted chemical substances in the US induced by the final demand of country k in year t, Q_t,k^US, is expressed in equation (1):

$$\begin{equation} Q_{t,k}^{{\rm{US}}} = {{\bf{T}}_t} \cdot {{\bf{L}}_t} \cdot {{\bf{f}}_{t,k}} \end{equation} \tag{ 1 }$$

where T_t is the row vector consisting of the toxic emission intensity (i.e. toxicity of chemical emissions (unit: tons) per industrial output (unit: one thousand dollars) in the US), L_t = (I − A_t)⁻¹ is the Leontief inverse matrix, A_t is the input coefficient matrix in the multiregional input–output table for year t, I is the identity matrix, and f_t,k is the column vector consisting of the final demand (unit: one thousand dollars) of country k in year t (Miller and Blair 2009). Based on the number of industries M and countries N, T_t,A_t and f_t,k can be expressed as follows:

$$\begin{equation} {\bf T}_{t} = [0 \quad \cdots \quad 0 \quad T^{\rm US}_{1,t} \quad \cdots \quad T^{\rm US}_{M,t} \quad 0 \quad \cdots \quad 0], \end{equation}$$

$$\begin{equation} {\bf A}_{t} = \left[\begin{array}{l@{}l@{}l@{}} {\bf A}_{t}^{11} \quad {\bf A}_{t}^{12} \quad \cdots \quad {\bf A}_{t}^{1N}\\ \\ {\bf A} _{t}^{21} \quad {\bf A}_{t}^{22} \quad \cdots \quad {\bf A}_{t}^{2N}\\ \\ {\vdots} \quad \quad \quad {\vdots} \quad {\ddots} \quad \quad {\vdots}\\ \\ {\bf A}_{t}^{N1} \quad {\bf A}_{t}^{N2} \quad \cdots \quad {\bf A}_{t}^{NN} \end{array} \right], {\bf r}_{r,k}=\left[\begin{array}{l@{}l@{}l@{}} {\bf f}^{1k}_{t}\\ \\ {\bf f}^{2k}_{t}\\ \\ \,\, \vdots\\ \\ {\bf f}^{Nk}_{t}\\ \end{array}\right] \end{equation}$$

where T_i,t^US(i = 1, ⋯ ,M) is the toxic chemical emissions per industrial gross output in each industrial sector i in the US in year t, A_t^rs is the input coefficient matrix from country r(r = 1, ⋯ , N) to country s(s = 1, ⋯,N) in year t, and f_t^rk(r = 1, ⋯, N) is the vector of final demand from country r to country k.

Equation (1) calculates the toxicity of emitted chemical substances in the US through the global supply chains induced by the final demand of country k. It should be noted that T_t is negatively affected by management efforts focused on reducing toxic chemical substances in US industrial sectors only, whereas A_t and f_t,k are affected as entire supply-chain networks by changes in production technology and lifestyles of the different countries, respectively.

We set the change in the consumption-based emissions of chemical substances during year t and year t+1 as ΔQ_k^US = Q_t+1,k^US−Q_t,k^US, and these factors were decomposed into the three effects shown in equation (S1) in supplementary method 1 available at stacks.iop.org/ERL/12/124008/mmedia: the changes in toxic emission intensity (i.e. ΔIntensity), in production structure (ΔSTR), and in the final demand scale (ΔScale) (Sun 1998).

Changes in the toxic emissions intensity effects (ΔIntensity) are attributed to changes in the toxic emissions per industrial gross output in the US. The structural change effects (ΔSTR) are attributed to changes in the structure of industrial production at a national scale. Notably, this structural effect includes the changes in the international trade patterns of intermediate products or national production technologies. Change in the final demand effects (ΔScale) are directly affected by changes in the scale of the final demand in each country.

The final demand effects and the production structural effects can be considered demand-side effects, whereas the toxic emissions intensity effects can be considered supply-side effects. To further clarify the effect of toxic chemical management by US industries, the changes in toxic emissions intensities are decomposed further in the following section.

A decomposition framework for the chemical emissions in industrial production processes has been proposed by Fujii and Managi (2013). By referring to their framework, the change in the toxic emissions intensity of industry i in the US, T_i,t^US, can be further decomposed into the following three factors: (1) end-of-pipe (EOP) treatment, (2) cleaner production (CP) and (3) the toxicity of emitted chemicals (TEC).

To apply this decomposition approach to the US, this study used two variables for the toxic chemicals: total release by industry i in the US in year t (E_i,t^US) and total waste management by industry i in the US in year t (M_i,t^US). These data are obtained from the US toxic release inventory. Using these two variables, we calculated the total amount of toxic chemical substances generated by industry i in the US in year t (G_i,t^US), and we herein define G_i,t^US = E_i,t^US + M_i,t^US. Y_i,t^US as the gross output in industry i of the US in year t deflated to the 1995 price.

First, the industrial CP indicator in the US, which is defined as $\frac{{G_{i,t}^{\rm US}}}{{Y_{i,t}^{\rm US}}}$, represents the generation of toxic chemicals per industrial gross output. This indicator can be decreased by reducing the generation of toxic chemical substances while maintaining industrial production. The reduction in CP can be achieved by improved production processes and product designs that reduce the input of intermediate chemical materials.

Second, the industrial EOP indicator in the US, which is defined as $\frac{{E_{i,t}^{\rm US}}}{{G_{i,t}^{\rm US}}}$, represents the share of emissions out of the total amount of toxic chemical substances generated. This indicator can be reduced by increasing the waste management of total toxic chemicals. Finally, the industrial TEC indicator in the US, defined as $\frac{{Z_{i,t}^{\rm US}}}{{E_{i,t}^{\rm US}}},$ represents the toxicity intensity of emitted chemical substances. Z_i,t^US is the toxic emissions in industry i of the US in year t. Here, the toxicity per gross output of industry i in the US ($T_{i,t}^{\rm US} = \frac{{Z_{i,t}^{\rm US}}}{{Y_{i,t}^{\rm US}}}$) was decomposed in equation (2).

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} T_{i,t}^{\rm US} = \frac{{Z_{i,t}^{\rm US}}}{{Y_{i,t}^{\rm US}}} = \frac{{Z_{i,t}^{\rm US}}}{{E_{i,t}^{\rm US}}} \times \frac{{E_{i,t}^{\rm US}}}{{G_{i,t}^{\rm US}}} \times \frac{{G_{i,t}^{\rm US}}}{{Y_{i,t}^{\rm US}}}\\ = TEC_{i,t}^{\rm US} \times \\ EOP_{i,t}^{\rm US} \times CP_{i,t}^{\rm US}\quad \end{array} \end{equation} \tag{ 2 }$$

Next, we define the change in the toxic emissions intensity as ΔT_i^US(= T_i,t+1^US−T_i,t^US) and obtain equation (3).

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {\rm{\Delta }}T_i^{{\rm{US}}}\, = T_{i,t + 1}^{\rm US} - T_{i,t}^{\rm US} = TEC_{i,t + 1}^{\rm US} \times EOP_{i,t + 1}^{\rm US} \times \\ \\ CP_{i,t + 1}^{\rm US} - TEC_{i,t}^{\rm US} \times EOP_{i,t}^{\rm US} \times CP_{i,t}^{\rm US} \end{array} \end{equation} \tag{ 3 }$$

Similar to the decomposition approach to the induced effects in equation (1), we apply the Laspeyres decomposition analysis to equation (3) and decompose ΔT_i^US into three effects: the changes in EOP treatment, the changes in CP, and the changes in TEC (see equation (S4) in supplementary method 2).

The World input-output database is conventionally compiled through non-survey methods; therefore, the obtained input coefficient matrix tends to contain measurement error problems in the observed transaction tables (Roland-Holst 1989).

As in Roland-Holst (1989) and Dietzenbacher (2006), we conducted a Monte Carlo experiment to examine whether the measurement error in the input coefficient matrix yields a significant input-output multiplier bias and whether such a multiplier bias consequently leads to substantial uncertainty regarding consumption-based chemical toxicity. Monte Carlo analysis has the advantage of applicability to simulations based on a variety of actual transaction tables (Roland-Holst 1989).

As in Dietzenbacher (2006), we see that the 'observed' intermediate input matrix D = (d_ij), the 'observed' row vector of toxic emissions Z = (Z_j^US), and the 'observed' column vector of final demand f = (f_i) are assumed to be biased as follows:

$$\begin{equation} d_{ij}^k = {d_{ij}} + \varepsilon _{ij}^k,{\rm Z}_j^{{\rm US},k} = {\rm Z}_j^{\rm US} + \delta _j^k,{\rm{and}}\, f_i^k = {f_i} + \varphi _i^k \end{equation}$$

with

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} \varepsilon _{ij}^k \sim N(0,\left[ {\rho \cdot d_{ij} } \right]^2 ),\;\;\delta _j^k \sim N(0,\left[ {\rho \cdot Z_j^{US} } \right]^2 ),\;\rm{and}\\ \varphi _i^k \sim N(0,[\rho \cdot f_i ]^2 ) \end{array} \end{equation}$$

where k = 1,2,..., K. Here, k is the sample size of the Monte Carlo method, and we set K = 100 and ρ = 0.1, as in Dietzenbacher (2006). Notably, the sample size of the Monte Carlo method must be set at the relatively small number of 100 due to the larger data size of the World Input–Output Database. The error terms of ε_ij^k,δ_j^k, and ε_i^k follow normal distributions. In this multi-regional input–output table including data noise, the gross outputs include data noise as $Y_i^k = \mathop \sum \limits_j d_{ij}^k + f_i^k$ and therefore the obtained intermediate input coefficient matrix also includes it as A^k = (a_ij^k) = (d_ij^k/Y_j^k).

Then, the Leontief inverse matrix (i.e. direct and indirect input requirement coefficients) and the row vector of toxic emission intensity, including data noise, are calculated as L^k = (l_ij^k) = (I − A^k)⁻¹ and T^k = (T_j^US,k = (Z_j^US,k/Y_j^US,k), respectively where Y_j^US,k is the gross industrial outputs of the US including data noise.

When we set the unknown 'true' values of l_ij, T_j^US, and f_i^*, their t-statistics are obtained as follows:

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} t_{ij}^l = \frac{{\overline l _{ij} - i_{ij}^* }} {{s_{ij}^l /\sqrt K }},\;t_j^T = \frac{{\overline T _j^{US} - T_j^{US*} }} {{s_j^T /\sqrt K }},\;\;t_i^f = \frac{{\overline f _i - f_i^* }} {{s_i^f /\sqrt K }} \end{array} \end{equation}$$

respectively, where we have

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {\bar l_{ij}} & = & \frac{{\mathop \sum ^K_{k = 1} l_{ij}^k}}{K},\bar T_j^{\rm US} = \frac{{\mathop \sum ^K_{k = 1} T_j^{{\rm US},k}}}{K},\,{\bar f_i} = \frac{{\mathop \sum ^K_{k = 1} f_i^k}}{K}, \end{array} \end{equation}$$

with

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {\left( {s_{ij}^l} \right)^2} & = & \frac{{\mathop \sum ^K_{k = 1} {{\left( {l_{ij}^k - {{\bar l}_{ij}}} \right)}^2}}}{{K - 1}}\\ {\left( {s_j^T} \right)^2} & = & \frac{{\mathop \sum ^K_{k = 1} {{\left( {T_j^{{\rm{US}},k} - \bar T_j^{{\rm{US}}}} \right)}^2}}}{{K - 1}},\,{\rm{and}} \end{array} \end{equation} \tag{ 4 }$$

$$\begin{equation} {\left( {s_i^f} \right)^2} = \frac{{\mathop \sum \limits_{k = 1}^K {{\left( {f_i^k - {{\bar f}_i}} \right)}^2}}}{{K - 1}} \end{equation}$$

The t-statistics above can provide 95% confidence intervals for the direct and indirect input requirement coefficients, toxic emission intensities, and final demand; therefore, these upper values can be used to estimate the upper value of consumption-based toxic emissions with equation (1), whereas the lower values provide the lower value of consumption-based toxic emissions.

Toxic chemical emissions data from the US industrial sector are obtained from the toxic release inventory (TRI) database published by the US Environmental Protection Agency (US EPA 2017). The TRI was created by Section 313 of the Emergency Planning and Community Right-to-Know Act in 1986. It is a publicly available database that contains information on toxic chemical releases and waste management activities, which is reported annually by certain industries and federal facilities.

In addition, chemical data on seven industries, which were added to the TRI programme in 1997, only became available starting in 1998. These newly added sectors included the mining sector and the electric power sector, which produce large amounts of chemical emissions. Therefore, we set the research periods from 1998–2009 to ensure that we considered these sectors as well.

We used US emissions data for 588 chemical substances with risk–screening environmental indicators (RSEI) as proxies for the toxicities of chemical substances (US EPA 2015). The emissions data cover toxic chemical releases to all media. Horvath et al (1995) introduced the TRI as the database with the most comprehensive and widely reported information on hazardous discharges into the environment in the United States.

Additionally, Toffel and Marshall (2004) investigated the comparative analysis of toxic weighting methods for the TRI database. They compared 13 methods including RSEI, human toxicity potential, the Indiana Relative Chemical Hazard Score, and so on. Toffel and Marshall (2004) concluded that RSEI is the recommended weighting method for integration of TRI data because of greater coverage of TRI chemical substances. Many previous studies have used the TRI and RSEI to evaluate toxic chemical substances in environmental indicators (Velagapudi et al 2017) and their human health impact (Lewis and Bennett 2013) and as parts of accounting research (Cong et al 2014).

To integrate the toxicities of different chemical substances, we applied the RSEI published by the US EPA (2015). The integrated toxicity score (Z_i,t^US) for chemical substances generated by a specific sector i in the US in year t was estimated by equation (5):

$$\begin{equation} Z_{i,t}^{{\rm{US}}}\, = \,\mathop \sum \limits_j \left( {{\rm{emission \, }}_{i,j,t}^{{\rm{US}}}\, \times \,{\rm{toxicity}}\,{\rm{weigh}}{{\rm{t}}_j}} \right)\, \end{equation} \tag{ 5 }$$

where j denotes a specific chemical substance. The toxic emissions intensity of 16 sectors between 1998 and 2009 in the US were estimated by dividing the sectoral integrated toxicity score by the sectoral gross output (table S1 for the sector classification). According to the US EPA (2015), the toxicities estimated by equation (5) are hazard-based results, which differ from toxic 'risks' to human health that clearly consider the population density of the region of emissions. Therefore, the toxicity scores in this study focus on the 'toxicity hazard emission' or 'toxicity hazard embedded in production processes'. Using the hazard-based toxic emissions data, we can evaluate how the US sector applied toxic chemical management in their production processes considering the toxicity of chemicals.

The list of toxic chemical substances in the TRI changed with time. To analyse the change in toxicity using time-series data, we applied the toxic chemical data listed in the 1998 core chemical substances published by the US EPA (see table S2). Therefore, the toxic scores in different years are estimated using the same coverage of chemical substances. Thus, we can compare the toxicity data directly between different years.

It should be noted that TRI data are subject to limitations and uncertainty. The US EPA (2009) investigated the uncertainty involved with seven major environmental indicators, including TRI. The US EPA (2009) noted that 'TRI data reflect only 'reported' chemicals, not all chemicals with the potential to affect human health and the environment. TRI does not cover all toxic chemicals or all industry sectors. The following are not included in this indicator: (1) toxic chemicals that are not on the list of approximately 650 toxic chemicals and toxic chemical categories, (2) wastes from facilities within industrial categories that are not required to report to TRI, and (3) releases from small facilities with fewer than 10 employees or that manufactured or processed less than the threshold amounts of chemicals.'

In addition to these limitations, TRI data are subject to uncertainty due to facility-level errors caused by the diverse arrays of methods and information sources used by facilities to prepare their TRI reports (US EPA 2009). The uncertainty involved with self-reported environmental data has also been noted by de Marchi and Hamilton (2006). They find that large decreases in air emissions reported by firms in the TRI are not always matched by similar reductions in measured concentrations by EPA monitors.

We used the World Input–Output Database (WIOD), which covers 41 countries and regions and 17 industries between 1998 and 2009 (see table S1) (Dietzenbacher et al 2013, Timmer et al 2015). Following Temurshoev et al (2013) and Los et al (2014), the World input–output tables for 1998–2009 were deflated to the tables based on 1995 US dollars using price deflators and the generalized RAS method.

We should note our limitation related to data availability. The WIOD released data from its 2016 version, which covers 2000 to 2014. However, the WIOD did not release the socio-economic data, including price index data. Thus, we find translating the database into real terms, which is needed for time-series analysis, difficult. Additionally, the WIOD released data from its 2013 version, which covers 1995 to 2011. Moreover, the WIOD only releases price index data from 1995 to 2009 from the 2013 version of its socio-economic accounting database. Based on these available data, we create an input–output table data in real terms for 1995 to 2009.

Although the WIOD includes the time-series data from 1995 to 2009, we used the data from 1998 because of the consistency of the data for toxic chemical substances in the TRI in the US. As a result, we compiled the environmentally extended world input–output tables from 1998–2009. Because industries are clarified differently between the TRI and WIOD, we integrated the TRI industry using North American Industry Classification System (NAICS) codes (see table S3).

Table 1 shows the toxic emissions from the five largest industries associated with the final demand of countries around the World in 1998 and 2009 (table 1). First, we examined the magnitude of chemical pollution in 1998 and 2009. In 1998, the total amount of toxic chemicals generated by the five industrial sectors in the US accounted for 5743 billion tons (table 1). Interestingly, the induced chemical toxicity (i.e. consumption-based chemical toxicity) decreased by 4779 billion tons during the decade from 1998–2009, implying that US industries have contributed to the rapid decrease in chemical toxicity observed in the US.

Zoom In Zoom Out Reset image size

Next, we discuss the regional distribution of consumption-based toxic emissions by examining several countries and regions that induce significant toxic emissions in the US. Table 1 shows that most of the toxic emissions in the US industrial sector were triggered by US domestic final demand. The share of US final demand increased slightly from 85.2%–87.0% from 1998–2009. Thus, 13.0% of the toxic emissions from the US industrial sector were triggered by the final demand of other countries outside the US in 2009.

The contribution of the final demand of BRIC countries (Brazil, Russia, India and China) to US toxic emissions increased from 1.2%–2.0% from 1998–2009, whereas that of other countries decreased or remained steady over the same period (see figures S2 and S3 in the supplementary information for the geographical distribution).

The data in table 1 indicate that the contribution of each industry to the final demand differed among countries and regions because different countries have their own relative industrial strengths in the global market as well as different objectives and characteristics that drive the importation of US products as intermediate goods or final products.

The driving forces underlying changes in consumption-based chemical toxicity and chemical toxicity transferral can be captured using economic factors, such as the industrial structure, international trade structure, final demand structure, and production technology.

Figure 2 shows the trends for consumption-based toxicity in the US. The plotted line shows the changes in the consumption-based toxicity of the US as a ratio of that in the base year 1998, and the bar chart shows the effects of economic factors on the consumption-based toxicity of the US. The summation of the bar chart score is equivalent to the value of the plotted line.

From figure 2, consumption-based toxic emissions in the US decreased by approximately 83% from 1998–2009, and an 80% reduction was achieved by 2004. The main drivers for this rapid reduction were the observed decreases in TEC and EOP treatment, which together accounted for a 66% reduction. Thus, pollution abatement technologies and initiatives in the US industrial sector have markedly contributed to the reduction in toxic emissions in the US. In addition, changes in the production structure (STR) and CP methods contributed to decreasing the toxicity by 24% and 17%, respectively.

Next, we consider the changes in induced toxicity in each industrial sector. Table 2 shows the changes in the induced toxic emissions in the US industrial sector related to increased product demand in 41 countries and regions. The trading-factor results were estimated by the environmentally extended multi-region input–output analysis model, and the breakdown of toxic emission intensity change (ΔT) was calculated by a further decomposition analysis using the Laspeyras index decomposition approach. The LMDI is an alternative approach for further decomposition analysis (see supplementary method 3) (Lan et al 2016). To confirm the robustness of the results obtained by further decomposition, we compared the estimation scores obtained using the Laspeyres index and LMDI approaches. Table S4 in the SI describes the results of these further decomposition approaches, and the results appear to corroborate the obtained results.

Table 2 shows that the induced toxic emissions of the US industrial sector decreased by 83% from 1998–2009. The main driver of this dramatic reduction is a decrease in the toxic emissions intensity (i.e. toxic emissions per industrial gross output). These findings imply that the reduction in induced toxic emissions was achieved by decreasing the toxic emission intensity in the US industrial sector. Compared with the demand-side effects, we found that the supply-side effects exert a markedly stronger influence on the changes in induced toxic emissions in most industries and countries (see table 2 and table S5 in SI). This trend differs from that reported in previous studies that focused on consumption-based increases in CO₂ emissions and concluded that consumption scale is the main driver of changes in emissions (Kagawa et al 2015). One possible reason for the observed disparity is that toxic chemicals can be managed by EOP, which is a difficult approach to inducing reductions in CO₂ emissions. Another reason is the marked difference in toxicity among chemicals, which means that substituting a highly toxic chemical with a less toxic chemical can cause a marked reduction in toxic emissions. The finding that supply-side effects lead to a greater reduction in induced toxic emissions relative to demand-side effects is explained by the simultaneous application of supply-side and demand-side approaches.

Zoom In Zoom Out Reset image size

Table 2 shows that the toxicity induced by the mining and transportation equipment industries decreased markedly because of a reduction in the toxic emission intensity. However, the main determining factors for the observed change in toxic emission intensity differed between the two industries.

Figure 3 shows a comparison of the main driving factors affecting changes in induced toxic emissions in the mining and transportation equipment sectors. Toxic emissions in the mining sector mainly decreased because of CP factors. In the mining sector, compounds of copper, zinc and arsenic accounted 67% of the total toxic chemical release in 1998. However, the chemical release from these three metal compounds was reduced by 63% from 1998–2009, especially that from copper and arsenic compounds, which decreased by 87% and 83%, respectively. Copper and arsenic compounds rapidly decreased during this period for two main reasons.

First, copper and arsenic compounds were reduced due to scaled-down production in the US copper and gold mining sectors, respectively. The copper mining sector accounted more than 90% of the release from copper compounds in the US mining sector in 1998. Meanwhile, the production in US copper mines decreased from 1.86 million tons to 1.18 million tons from 1998–2009 (US Geological Survey 2017). Similarly, the gold mining sector accounted for more than 95% of the arsenic compound release in the US mining sector in 1998, as arsenic compounds are mainly released through the roasting of arsenious gold ores (Wang and Mulligan 2006). US gold mine production decreased from 366 tons in 1998 to 210 tons in 2009 (US Geological Survey 2017). Thus, the scaling down of copper and gold mine production contributed to the decrease in the total release of copper and arsenic compounds in the US mining sector, respectively.

Second, corporate pollution prevention activities are another contributing factor to toxic chemical release reduction in the mining sector. Especially in large-scale US mining companies, such activities were implemented to prevent environmental pollution in cyanidation practices and to enhance acid mine drainage control (Hilson and Murck 2001). Table S6 shows the corporate activities implemented for pollution prevention in the metal mining industry, as listed on the US EPA homepage (www3.epa.gov/enviro/facts/tri/p2.html). From this information, we can see that corporate pollution prevention activities also effectively contribute to reducing the sources of metal compound releases.

In the transportation equipment industry, ΔTEC strongly contributed to reducing the induced toxicity, whereas ΔCP did not have much of an effect on driving toxicity compared with that observed in the mining sector. These results imply that the transportation equipment industry improved its production processes using less toxic substances in its manufactured goods and these improvements manifested as a decrease in TEC substances, whereas the amount of chemical substances emission per industrial gross output did not change markedly. Specifically, we found that the reduction in induced toxicity in the transportation equipment industry was achieved by decreasing the release of asbestos. In the US transportation equipment sector, asbestos was responsible for 97% of the toxicity in terms of chemical releases in 1998. However, asbestos emissions were reduced from 1843 tons in 1998 to 54 tons in 2009, which decreased the share of asbestos in total chemical toxicity to 28.5% in 2009.

Asbestos was widely used in the production of high friction automotive parts (e.g. brake pads, clutch pads, etc.) (Cowan et al 2015). However, advances in materials have enabled the production of asbestos-free products (e.g. antimony compounds) (Iijima et al 2007). In the same period, the emissions of antimony compounds have increased three-fold (31 tons to 93 tons) in the transportation equipment industry. This increase implies that the transportation industry increasingly uses antimony compounds in its production processes. This substitution of chemicals and materials was the main reason for the reduction in the toxicity of chemical emissions induced by the transportation equipment sector from 1998–2009.

Here, we investigate how our estimation results are different from those in previous studies focusing on toxic chemical substance management but not on chemical toxicity. Fujii and Managi (2013) and Koh et al (2016) used decomposition analysis to analyse toxic chemical substance management in various industrial sectors in the US.

Both studies indicate that CP is a major contributor to emission intensity reduction in many industries. In this study, we reveal that TEC is a more important factor in reducing chemical toxicity in industrial sectors, especially the paper, fuel, metal, and transportation equipment sectors (see table 2). This information is useful when evaluating toxicity management activities by substituting the reduction of chemical substances or the input amount of chemical materials.