The ARET Challenge
Evaluating a Negotiated Voluntary Challenge Program
W ERNER A NTWEILER and K ATHRYN H ARRISON
University of British Columbia
February 16, 2006
Abstract
The Accelerated Reduction/Elimination of Toxins (ARET) Challenge was a voluntary environmental program initiated in 1994 by the
Government of Canada in collaboration with various industry partners
to promote reduction of toxic releases. Although hailed as a success,
the achievement of ARET’s reduction goals was merely self-reported.
We attempt to validate this claim of success for the subset of ARET
substances reported to Canada’s National Pollutant Release Inventory
(NPRI). We model and test empirically which facilities participate in
the ARET program. We employ treatment effects regressions to determine the true impact of ARET on emissions, thus controlling for the
self-selection bias associated with the voluntariness of program participation. Many of the highest priority ARET substances were not added
to NPRI until 2002, after the ARET program concluded. We find not
only that releases of these substances were not “virtually eliminated”
as sought by ARET, but that a significant disparity between voluntary
ARET reporting prior to 2001 and mandatory NPRI reports thereafter
is indicative of significant under-reporting of releases of these substances by ARET members. For the 17 other ARET substances that
were covered by NPRI during the ARET reporting period, we find evidence that ARET accelerated emission reductions in 5 cases at most.
The launch of this voluntary program appears to have come at the cost
of too much flexibility with respect to program requirements and a lack
of verifiability.
Antweiler: Sauder School of Business, UBC, 2053 Main Mall, Vancouver, BC, V6T 1Z2,
Canada. Phone: 604-822-8484. E-mail: werner.antweiler@ubc.ca. Harrison: UBC, Department of Political Science, 1866 Main Mall, Vancouver, BC, V6T 1Z1, Canada. Phone: 604822-4922. E-mail: harrison@politics.ubc.ca. The authors thank John Horowitz and Donna
Ramirez, as well as participants of the APPAM 2005 conference for helpful comments. We
are also thankful for valuable feedback from Kathy Baylis, Jim Brander, Keith Head, John
Ries, and Tom Ross at a workshop presentation at UBC. The authors also gratefully acknowledge the financial support of the UBC Hampton Fund, the Social Sciences and Humanities Research Council of Canada, and the UBC Weyerhaeuser Program in US Natural
Resources and Environmental Policy, research assistance from Manny Villanueva, and the
support and cooperation of Brad Fisher from Environment Canada’s ARET Program office.
1
Introduction
The ARET (Accelerated Reduction/Elimination of Toxics) challenge was issued in early 1994 by the Government of Canada and a group of industry representatives with the intention of promoting voluntary reductions
of releases of toxic substances across Canada. Firms were challenged to
reduce their discharges of 30 chemicals considered to be toxic, persistent,
and bioaccumulative by 90% by the year 2000, and of 87 others that met
some but not all of those criteria by 50% by the same year. ARET was a relatively early example of the wave of state-sponsored voluntary challenge
programs that have been adopted throughout the world in the last decade.
Such programs are purported to offer a win-win opportunity to meet environmental objectives at a lower cost to both taxpayers and the private sector. Governments can avoid the administrative burden of developing and
enforcing formal regulations, while industry can reduce its costs by taking advantage of the greater flexibility typically afforded by voluntary programs, for instance with respect to timing and control methods. Although
voluntary programs have been embraced enthusiastically by governments
and business—though less so by the environmental community—nagging
questions remain about whether they actually deliver the goods. In particular, does industry’s consent to reduce its environmental impacts voluntarily
necessarily come at a price of reduced expectations?
Evaluation of voluntary programs is complicated by a number of factors
(Harrison 2002). First, data on participating firms’ performance often are
not publicly available, as they typically are in the case of mandatory regulations. Second, when self-reported data are made available, they may not
have been subject to independent verification. Third, voluntary programs
are fraught with self-selection bias, the problem being that the firms that intend to reduce their releases anyway, whether in response to market forces
or concurrent regulations, will be the ones that have the greatest incentive
to sign on as participants, since they can gain any available credit for making voluntary reductions without actually assuming any additional costs.
A critical question in evaluation of any voluntary program is thus to what
degree the reported environmental benefits are in fact attributable to the
program in question.
On the face of it, the results of the ARET program have been quite
promising. By the program’s end in 2000, 351 facilities across Canada had
voluntarily committed to reduce their releases of the substances on the
ARET list.1 ARET reports that more than 90% of facilities in nine industrial
1
Because the ARET program did not maintain records of the year facilities joined the
program, we constructed a list of ARET members with their year joined based on membership data published in appendices of the ARET program’s annual reports. On the advice of
the ARET secretariat, we interpreted facilities owned by companies included in the Leaders 1 and Leaders 1 Update reports as having joined in 1994, facilities in Leaders 2 as joined
in 1995, facilities in Leaders 2 update as joined in 1996, facilities in Leaders 3 as joined in
1
sectors participated in the program (OECD 2003b), though, as discussed
below, our own tallies are considerably lower. The 50% reduction target
for 87 substances was met by ARET program participants in 1997, three
years ahead of schedule, and a reduction of 72% was achieved by the end of
2000 (ARET Secretariat 2003). ARET participants did not achieve the goal of
90% reduction of the 30 toxic, persistent, and bioaccumulative substances by
2000, but nonetheless did reduce their releases of those substances by 61%.
These are encouraging outcomes, particularly when one considers that the
ARET program targeted these 117 substances at a cost of only about $2 million to the federal government (OECD 2003b), a small fraction of what has
been spent on management of a handful of toxic substances on a case-bycase basis under the Canadian Environmental Protection Act.
This article provides a critical review of these outcomes, however. We
find that the benefits of the ARET program have been significantly overstated. 46% of the reductions reported to ARET “relative to base years”
were actually achieved before the program was even announced. The figure is 83% in the case of class of toxic, persistent, and bioaccumulative substances. Moreover, since some participating facilities selectively reported
their releases of only some ARET chemicals even if they also released others (OECD 2003b), the reported reductions by ARET members even after the
program’s launch are undoubtedly overstated. Consistent with the problem
of limited data availability noted above, the ARET program, even though
it is administered and funded by the federal government, is not willing to
publicly release annual reports on individual substances submitted by participating companies without the consent of individual firms. However, facilities were required by law to report releases of some 28 ARET substances
covered by the National Pollutant Release Inventory (NPRI) throughout the
period of interest.2 Based on analysis of this subset of ARET chemicals, we
find that ARET participants generally outperformed their non-ARET counterparts. However, when we control for self-selection into ARET participation, we find few cases where ARET appears to have accelerated emission
reductions.
The remainder of this paper is structured as follows. Section 2 provides
an overview of the ARET challenge. Section 3 discusses the defining char1997, and facilities in Leaders 3 update as joined in 1998. A handful of facilities that joined
after 1998 were added based on personal communication with the ARET program. The resulting total of 351 facilities does not correspond to the figure of 318 published in the final
Leaders 4 report (ARET Secretariat 2003). We reviewed the discrepancies with the ARET
secretariat and believe that our list is the more accurate figure.
2
See table 1. By 2002, 49 ARET substances are required to be reported to the NPRI, including 18 of the 30 A-1 ARET priority substances. The ARET program reports a larger
number of overlapping substances prior to 2000 since they include cases where a more
specific chemical definition adopted by the ARET program is included within a more general category for NPRI. We have not analyzed these cases, however, since we cannot be
certain that any trends reported to the broader NPRI category correspond to those for the
particular ARET substances.
2
acteristics of Public Voluntary Programs and reviews theoretical insights
and empirical findings from previous studies of the latter. In section 4 we
describe the coverage of ARET substances through the NPRI. In section 5
we consider preliminary evidence concerning whether ARET indeed succeed in lowering emissions, and in section 6 we identify characteristics that
help predict which facilities participated in ARET. Sections 7 and 8 present
the methodology and results of the treatment effect analysis for determining ARET’s contribution to accelerating emission reductions for individual
chemical substances. Section 9 concludes.
2
Overview of ARET
ARET was inspired by the US EPA’s 33/50 program, which was launched in
1991. The 33/50 program challenged facilities releasing any of 17 chemicals,
all reported to the US Toxics Release Inventory, to reduce their releases of
those substances by 33% by the end of 1992 and 50% by the end of 1995.
Building on the work of the New Directions Group, a voluntary committee of industry and environmental leaders seeking non-adversarial solutions to environmental problems, in 1992 the Canadian Environment Minister struck a committee of stakeholders to advise him on a toxic chemical
reduction strategy for Canada. The committee, comprising representatives
from federal and provincial governments, industry, the environmental community, and labour was asked which substances should be a priority for
reductions and how those reductions should be pursued. The ARET stakeholders committee was able to achieve broad consensus on the criteria of
toxicity, persistence, and bioaccumulative capacity, and on a resulting list
of 117 substances (though there was no agreement on the classification of
2 of those substances).3 However, as the initiative moved toward a goal of
reduction rather than elimination, and voluntarism rather than mandatory
regulation, environmental and labour representatives withdrew from the
process in protest (VanNijnatten 1998).
The remaining committee of government and industry representatives
forged ahead and issued the “ARET Challenge” in March 1994, calling on
all firms to voluntarily reduce their releases of 30 substances considered
toxic, persistent, and bioaccumulative (the A-1 group) by 90% by the year
2000, 85 substances considered to have one or two of those characteristics
(8 deemed toxic and bioaccumulative in group B-1, 33 deemed persistent
and toxic in group B-2, and 44 deemed toxic in group B-3) by 50%, and
to employ “best efforts” (presumably at least achieving 50% reduction) of
two other substances considered by some, but not all, of the original stake3
Most of the substances in the A-1 group are polycyclic aromatic hydrocarbons (PAHs),
nitro PAHs, and chlorinated organics. More of these substances are found on the B lists.
The B-2 list also adds several metal compounds, and the B-3 list also adds a number of
aromatics and nitrosamines.
3
holder committee to be persistent, toxic, and bioaccumulative (group A-2).
Although 117 substances were covered by the ARET program, participants
eventually reported releases for only 93 of the 117 substances.
Several features of the ARET program negotiated by government and industry partners are noteworthy. First, it afforded participants considerable
flexibility, for instance to choose which substances to report, what targets
to adopt for those substances, and the means to reduce their releases. Participants were not required to commit to the ARET program’s overall targets of 90% for A-1 substances and 50% for others. Second, consistent with
its voluntary basis, the program did not threaten any consequences should
participants fail to meet even their own targets.
The program has been lauded as a success by Environment Canada
(ARET 1997, 1999, 2003) and especially by Industry Canada (1998).4 The
latter in particular stressed that the high levels of participation in key manufacturing sectors dispelled the fear that voluntary programs like ARET inevitably will be plagued by low participation rates as non-joiners attempt
to free ride on any reputational or regulatory benefits to the entire sector
as a result of industry leaders’ efforts. However, several weaknesses of the
program have also been noted from the outset.
First, the absence of third party verification of participants’ claims has
been described as an “Achilles Heel” of the program (Darier and VanNijnatten 1996, 2001), calling into question the validity of reported benefits.
This problem is mitigated somewhat for the subset of ARET chemicals covered by NPRI. These substances are the focus of our analysis.
Second, ARET allowed program participants to choose their own “base
year,” with flexibility to choose a base year up to six years before the 1994
launch of the program (Gallon 1998). Not surprisingly, different ARET participants chose different base years, presumably with the intent of maximizing the reductions they could claim. As a result, roughly half (46%) of all
reductions claimed by the ARET program had occurred before the program
was even launched, and the fraction was even higher (83%) for the group A
substances.
Third, detailed reports by individual ARET members are not publicly
available. This is consistent with the voluntary nature of the program. Since
private firms, which typically are reluctant to divulge confidential information, did not authorize public release of all ARET reports when the program
was negotiated, Environment Canada staff are not authorized to release
participants’ detailed reports after the fact without consent from individual ARET members.5 Indeed, even though ARET was initiated, staffed, and
funded by the federal government, it was described to us by one Environment Canada official as an “industry program” controlled by its members
4
See also Dauvergne (1998) for Industry Canada’s perspective.
Given the large number of participants, it would have been very costly for us to seek
permission individually, and it likely would have introduced a non-response bias into our
analysis.
5
4
rather than a “government program.” As noted above, the closed nature of
the program limits our analysis to the subset of ARET facilities that report
releases of ARET substances to the NPRI program.
A fourth and final flaw of the program has been inattention to whether
reported reductions are in fact attributable to the ARET challenge. This has
been most obvious with respect to the issue of early and varying base years,
where inattention to causation has at times bordered on willful misrepresentation. For instance, in 1997 the ARET program reported that reductions
of 49% had been achieved relative to participants’ base years by the end of
1995 and asserted that “for the most part” those reductions were “directly
attributable to the commitment of ARET participants to this voluntary initiative” (ARET Secretariat 1997, p. 1), even though the data presented in the
same report clearly indicated that a majority of those reductions predated
the ARET program. Subsequent ARET reports were more cautious in their
claims, but continued to emphasize reductions relative to base years rather
than the launch of the program. Even with respect to reductions made by
program participants after the launch of the ARET program, limited effort
has been made to determine whether facilities changed their behaviour in
response to the ARET challenge or, rather, were claiming credit for reductions they would have made anyway, whether in response to market or regulatory pressures (Harrison 2001). The program made no effort to confirm
which reductions claimed by participants were in fact voluntary, even after a survey commissioned by Environment Canada concluded that only a
minority of ARET participants had changed their behaviour in response to
the ARET challenge (Gunn et al. 1999). Anecdotal evidence suggested that
reductions prompted by concurrent regulation may have been quite significant (Gallon 1998). The question of whether reductions reported to ARET
are in fact attributable to this voluntary program is a central focus of the
analysis that follows.
3
3.1
Voluntary Challenge Programs
Typology
A typology of voluntary programs proposed by the OECD (2000, 2003a) has
been widely accepted in the scholarly literature (Lyon and Maxwell 2004,
Alberini and Segerson 2002, Khanna and Damon 1999, Harrison 2002). The
OECD distinguishes between four classes of voluntary programs: (a) unilateral programs adopted by individual firms or trade associations; (b) private voluntary agreements between polluters and third parties (e.g., certification organizations, such as the International Organization for Standardization, or environmental groups); (c) negotiated voluntary agreements between private firms and governments; and (d) public voluntary programs
5
or “voluntary challenges” that are issued unilaterally by the state.6 Our focus will be on the last two. Negotiated voluntary agreements are exemplified by the Dutch “covenants” between the state and industrial sectors, the
US EPA’s environmental performance agreements, and a recent agreement
between the government of Canada and the automobile sector concerning
greenhouse gas emissions. A prime example of a voluntary challenge program (VCP)—and a predecessor of ARET—is the 33/50 challenge unilaterally instituted by the U.S. Environmental Protection Agency (Arora and
Cason, 1995, 1999; Khanna et al., 1998; Khanna and Damon, 1999).
Construction of a simple typology is a necessarily perilous exercise,
since it entails privileging one or two characteristics of programs that invariably differ along many other dimensions as well. The OECD’s typology
implicitly focuses on two dimensions: the mix of public and private participants, and whether or not voluntary programs are negotiated. According to
the OECD typology, the ARET program would qualify as a negotiated voluntary agreement rather than a voluntary challenge, since the terms of the
challenge were negotiated by the state and representatives of various trade
associations. However, we believe that the case of ARET reveals that the
typology’s emphasis on negotiation is misplaced. Two more central distinctions between negotiated voluntary agreements and voluntary challenge
programs lie in the degree to which the program is tailored to particular
firms and the degree of flexibility afforded participants.
Voluntary agreements are designed to evoke contractual commitments
by particular firms or groups of firms to achieve quite specific goals.7 In
contrast, voluntary challenge programs, whether or not their terms are negotiated, involve an open challenge to a relatively large number of potential
participants and also typically offer participants considerable flexibility to
choose their own targets.8 The issue is not so much whether the terms of the
program are negotiated as what is being negotiated: a firm commitment or
a flexible commitment, and the consequences of noncompliance. With this
amendment to the definition of a voluntary challenge program, we assert
that ARET, although negotiated, remains at its core a voluntary challenge
6
Many of the authors noted above have not analyzed the private agreements category
and have thus offered only a three-way typology.
7
The fact that most voluntary agreements are not legally binding indicates that there is,
in fact, a continuum in degree of contractual commitment. Our focus here is on ideal cases
at opposite ends of the spectrum. It is also noteworthy that voluntary programs differ on
many other dimensions as well. Of particular interest is whether the program is concerned
with effort, which is the focus of environmental management systems such as ISO 14001
(Prakash and Potoski 2005) and the chemical industry’s Responsible Care program (King
and Lenox 2000), or with outcomes such as emissions.
8
The implication of our amendment to the typology is that there is another category
of either government-sponsored or private voluntary programs. Like voluntary challenge
programs, ecolabelling programs entail an open challenge to potential participants but they
depart from voluntary challenge programs like 33/50 and ARET in specifying detailed
terms of participation.
6
program since it entailed (a) an open challenge to all firms, (b) very flexible
terms of participation with no penalty upon non-compliance.
3.2
Motives for Participation
Why do firms participate in VCPs such as the ARET Challenge? Voluntary
challenge programs entail decisions by both the state and private firms to
participate. As such, both the government issuing the challenge and private sector participants must perceive that the benefits they receive will
outweigh the costs. We will consider each in turn.
The theoretical underpinning for defining the benefits of participation in
voluntary challenge programs or negotiated voluntary agreements has been
developed in a series of research contributions. Prominent examples of this
literature include Maxwell et al. (2000), Lyon and Maxwell (2003, 2004), and
Segerson and Miceli (1998) who point to the importance of corporate strategy and government-business interaction. Videras and Alberini (2000) have
summarized important participation incentives for voluntary programs,
and Alberini and Segerson (2002) identify key mechanisms through which
VCPs affect (or fail to affect) environmental outcomes.
Firms may unilaterally or collectively participate in VCPs in order to
obtain various kinds of benefits, some of them tangible, some of them anticipatory. We enumerate several possibilities below.
1. Hypotheses related to lowering the costs from regulation:
(a) VCP participation may influence future regulation. By demonstrating a commitment to reducing releases via participation in
a voluntary program, firms may achieve less intrusive regulatory interventions, and, at the limit, may avoid regulation altogether. A voluntary approach may be attractive because it provides firms with greater flexibility to choose their own strategies
to abate emissions than prescriptive regulations. It is noteworthy, though, that firms have incentives to respond voluntarily to
a regulatory threat regardless of the opportunity to participate
in a formal voluntary program (Antweiler 2003). At issue from
the firm’s perspective is the marginal impact of the voluntary
challenge program: either because its existence sends a signal
concerning regulators’ priorities, or because participation lends
credibility to firms’ claims to regulators concerning their voluntary actions.
(b) VCP participation may deflect enforcement of existing regulations. If facilities can convince regulators of their good intentions, they may be subject to relaxed regulatory enforcement or
reduced penalties for violations. A related approach involves
7
grandfathering clauses that protect existing production facilities
for a limited duration.
2. Voluntary participation may invite new regulations. Leading firms
may have an incentive to demonstrate advanced abatement technologies in order to invite regulators to set more stringent standards that
will place their competitors at a disadvantage (Lutz et al. 2000, Arora
and Cason 1995).9
3. Firms may gain productivity advantages due to the installation of
newer generations of equipment that are simultaneously more efficient and cleaner. Firms face incentives to reduce their production
costs regardless of the existence of voluntary programs. However,
VCPs may assist in this effort if the program alerts firms to unrealized opportunities for savings, if the state provides technological
assistance, or if the voluntary program creates opportunities for participants to share information about existing technologies or pool resources for developing new and cleaner technologies.
4. Firms may also gain a competitive advantage over rivals in the market place by signaling environmental responsibility. As in the case of
productivity advantages and regulatory incentives, if firms face green
market pressures, whether from consumers,10 investors, or lenders,
they will have incentives to signal their environmentally friendly behaviour regardless of the existence of government-sponsored voluntary programs. However, a VCP may offer a valuable service in providing a more credible signal to consumers, whether for individual
firms or an entire sector where there are economies of scale in marketing environmental friendliness. Alternatively, and more troubling,
a VCP may act as a coordination device among participants to limit
rivalry among group members in response to green market pressures.
The decision by a firm to participate in a voluntary program will depend
not only on the benefits but also on the costs of participation. The cost to
private firms will depend to a large extent on what participation demands
and what consequences will ensue if a participant fails to reach one or more
of the emission reduction targets. From a firm’s perspective, the appeal of a
voluntary challenge program relative to a negotiated voluntary agreement
or binding regulation may lie in both greater flexibility with respect to targets and the absence of contractual sanctions, though there may nonetheless
9
Alternatively, environmental laggards may invite stricter regulations as a barrier to
entry if their existing technology is grandfathered for exetended periods. However, this
objective is hardly furthered by participating in a voluntary program for reducing current
emissions. Likely, environmental laggards would pursue more direct channels such as the
strategic options process under CEPA.
10
For example, see the empirical analysis by Antweiler and Harrison (2003) of green
consumerism in Canada.
8
be reputational costs for noncompliance. All else being equal, the costs of
participation will be lower when lower demands are placed on participants.
However, even very demanding voluntary programs may not trigger extraordinary costs to achieve emission reduction targets depending on what
a firm is already doing independent of the program. If a firm is undertaking scheduled replacement of outdated technology with newer vintages,
this newer technology often combines greater productive efficiency with
greater environmental friendliness at no extra cost. Similarly, if a firm can
claim credit for reductions they were going to make anyway in response to
the demands of concurrent regulation, green market pressures, or opportunities to reduce production costs, the costs of participation will be negligible. There is thus a wide range of possible costs and potential benefits,
and we would expect firms to divide into participants and non-participants
along the lines where potential benefits exceed costs.
From the state’s perspective, a voluntary approach to environmental
harm reduction may generate public benefits aside from the direct environmental benefits. Direct regulatory intervention may be costly not only
for industry but also for governments, which face the cost of monitoring
and enforcing environmental regulations. The societal costs of a voluntary
approach, however, will depend on how effective they are at achieving environmental benefits. If voluntary approaches turn out to be less effective
than regulation, the private and public cost savings may be offset by increased environmental harm.
3.3
Empirical Strategies
The foregoing discussion suggests five hypotheses with respect to firms’
participation in VCPs. Firms may be motivated by benefits in the form
of reduced production costs, green market pressures, reduced regulatory
compliance costs or greater compliance costs for competitors. On the other
hand, firms may participate simply because the costs of doing so relative to
a business-as-usual baseline are negligible. The challenge for researchers is
to identify indicators of each motivation in the absence of access to data on
firms’ decision-making. Unfortunately, many of the available indicators are
consistent with multiple hypotheses.
There is empirical evidence that threatened regulation is less effective
than actual regulation (Harrison and Antweiler 2003). However, regulatory
threats may nonetheless contribute to VCP participation and some voluntary reductions. Khanna and Damon (1999) and Gamper-Rabindran (forthcoming) find that regulatory pressure and threats of future regulation were
important motives for firms’ participation in the 33/50 program. Similarly,
Videras and Alberini (2000), deCanio and Watkins (1998), and Welch et
al. (2000) report that firms facing greater regulatory pressures were more
likely to participate in the US EPA’s WasteWise and Green Lights programs,
and the US Department of Energy’s Climate Challenge Program. Although
9
we do not have comparable direct measures of regulatory pressure, we anticipate that the threat of regulation will loom larger the greater a facility’s
emissions and the more people there are in the surrounding community
who are affected by its toxic emissions. The former is consistent with other
studies that have reported that facilities with larger emissions or environmental impacts were significantly more likely to participate in the 33/50
program (Arora and Cason 1995, 1996, Khanna and Damon, 1999, GamperRabindran forthcoming), and the US Climate Challenge Program (Welch et
al. 2000). In addition, Antweiler (2003) finds that the efficacy of threatened
regulation hinges on a number of firm characteristics relative to other firms
in a given industry, that is, a firm’s rung on the “abatement ladder.” Firms
with relatively high emission intensities thus may find it most worthwhile
to participate in a VCP in order to preempt or influence future regulation.11
Finally, given the propensity of regulators to adopt sectoral regulations, we
would also anticipate regulatory benefits of collective action: that is, facilities will be more likely to escape punitive regulations if the entire sector can
demonstrate a commitment to voluntary reduction. As such, one might expect coordinated participation—though such participation may be difficult
to achieve given incentives to free ride.
In contrast to the incentives to avoid regulatory costs, the motives for
firms seeking a first-innovator advantage by inviting regulation would be
quite different. There we would expect firms to be more likely to participate if they are less pollution-intensive and that the least pollution-intensive
firms would be least inclined to cooperate with competitors in the same sector.
The notion that participation in VCPs may help firms discover productivity advantages is similar to the well-known Porter and van der Linde
(1995a, 1995b) argument that tightening environmental standards may spur
productivity improvements, in part through accelerated introduction of
newer and cleaner technology. If the productivity gains story applies, it
should be the laggards with both low productivity and low environmental
efficiency that participate. This again suggests that participation in ARET
should increase with emission intensity. To the extent that members of a sector seek to pool pollution reduction expertise or all benefit from technological assistance from the state, one would also expect to see greater benefits
of participation the more competitors within a sector are also participating.
Population density should not be relevant to direct cost savings.
Other studies have found evidence of consumer pressure as a motive for
participation in public voluntary programs (Arora and Cason 1996, Khanna
and Damon 1999, Gamper-Rabindran, forthcoming). We do not have com11
Although in the context of business sponsored, rather than public voluntary, programs,
King and Lenox (2000) and Anton et al. (2004) report that firms with higher pollution intensities were more likely to participate in Responsible Care and other environmental management systems. However, as discussed below, Anton et al. (2004) offer an alternate
explanation for this finding.
10
parable data on advertising expenditures that those studies used to construct a measure of consumer proximity. However, as in the case of regulatory threat, we would expect market pressures, including those mobilized by watchdog environmental groups, to be greater the larger a facility’s emissions and the more people affected in a neighbouring community. It is unclear whether the scale of a facility’s production would matter to market pressures; the surrounding community’s greatest concern is
presumably the magnitude of emissions rather than production, but larger
facilities may attract greater attention simply by virtue of their size, regardless of their emissions. This is consistent with previous studies that have
found that larger firms are more likely to participate in voluntary challenges
(Arora and Cason 1995, 1996, DeCanio and Watkins, 1998, Videras and Alberini 2000, Gamper-Rabindran forthcoming). However, the impact of firm
size may not reflect greater benefits of participation so much as lower costs
by virtue of economies of scale in pollution abatement. To the extent that
firms look to VCPs to capture economies of scale in signaling environmental friendliness to consumers, we should see a high participation rate for
industries that participate in a VCP. Participation may also increase with
firm size because participation by large firms will make it more likely that
overall emission reductions can be met. This in turn may lower the threat
of regulatory intervention. Put another way, small firms have a greater incentive to free ride.
Although Khanna and Anton (2002) and Anton et al. (2004) have argued
that a correlation between pollution intensity and participation in the case
of industry-sponsored voluntary programs is indicative of community pressure, we remain skeptical. While facilities and regulators have information
and incentives to calculate pollution intensity relative to production, it is a
stretch to expect the average consumer to perform such a calculation. Citizens may personally experience negative impacts on environmental quality
directly or employ toxic release inventories like TRI and NPRI to determine
total releases from neighbouring facilities, but it seems to us unlikely that
they would calculate and compare emissions intensities of like facilities. Indeed, the US TRI (the basis for Khanna and colleagues’ conclusion) does
not include information on the number of employees or other proxies for
production that would even allow the average citizen to calculate emission intensity for particular facilities. Nor are we familiar with any cases
of environmental groups citing pollution intensity as opposed to total releases. Moreover, facilities’ neighbours who breathe contaminated air are
unlikely to be swayed by arguments that additional emissions are justified
by a larger scale of production. In contrast, the sector-specific, technologybased approach to regulation that has prevailed in Canada and the US historically directs both regulators’ and regulated firms’ attention to emission
intensity of like facilities.
Finally, we consider the hypothesis that firms would have made reductions anyway in two ways. First, we employ qualitative evidence concern11
ing regulatory changes affecting participating industrial sectors. Second, as
discussed below, we apply statistical methods to control for self-selection
and ask which reductions are in fact attributable to the ARET program. In
analyzing determinants of participation, we apply fixed effects for industry sectors, which will tend to control for both regulation and technological
change (though only to the extent that these effects are experienced equally
by all facilities of a sector). Findings concerning the impact of voluntary
challenge programs after controlling for self-selection have been mixed.
Hartman (1988) found a much-reduced but still significant impact of energy conservation programs after controlling for self-selection. However,
it is noteworthy in that case that the voluntary program was accompanied
by public subsidies. Khanna and Damon (1999) found that 33/50 had a
significant impact in reducing chemical firms’ releases based on data from
the early years of the program, albeit an effect considerable less than one
might assume without controlling for self-selection. However, analyzing
data from all years of the program and for six industrial sectors, including
the chemical industry, Gamper-Rabindran (forthcoming) found significant
effects of 33/50 in only the metals sector (though emissions reductions there
were offset by increases in offsite transfers) and the pulp and paper sector.
In the latter case, tightening of state effluent permits and anticipation of
the US EPA’s “cluster rule” may offer an alternative regulatory explanation.
Similarly, Welch et al. (2000) found no impact of the Department of Energy’s
Climate Challenge after controlling for self-selection. 12
4
ARET and NPRI
Canadian plants started reporting to the National Pollutant Release Inventory (NPRI) in 1993. Table 1 provides an overview of the substance categories established through ARET and links the ARET substances to those
covered by NPRI. Before the ARET program’s conclusion in the year 2000
virtually none of the A-1 and A-2 substances were required to be reported
to the NPRI, and only about one-third of the B-1, B-2, and B-3 substances.
Significant changes to the reporting requirements occurred in 2000, however, when a reporting requirement was added for micropollutants, that is,
highly toxic substances, such as dioxins and furans, that are released in very
small quantities.
The insufficient overlap between NPRI and ARET substances is further complicated by the difficulty of aggregating substances using toxicity
12
In the case of industry-sponsored voluntary programs, King and Lenox (2000) find
no independent effect of the chemical industry’s Responsible Care program. Rivera et
al. (2005) find impacts of the ski industry’s Sustainable Slopes program only with respect
to energy conservation (which can save money) and highly visible recycling programs.
However, Prakash and Potoski (2005) do find a significant impact of the ISO 14001 environmental management system.
12
scores. Where we aggregate chemicals in our analysis we employ the U.S.
Environmental Protection Agency (2002) Chronic Human Health Indicator
(CHHI), which was highest ranked in a review of 13 different toxicity scoring approaches by Toffel and Marshall (2004). Consistent with the ARET
program’s focus on direct releases to the environment, we also focus only
on two emission streams: releases into air and water. We employ the CHHI
toxicity for oral intake in the case of releases to water and that for inhalation
to adjust for releases to air. We exclude underground injections, releases to
land, and transfers to offsite locations for storage or processing.
In our comparison we do not consider a number of metallic toxins in the
ARET B-2 group. Even though several of these substances are reported to
the NPRI, their definition varies considerably. For example, whereas NPRI
requires reporting of “mercury and its compounds” and “copper and its
compounds”, ARET reports “elemental, inorganic mercury” and “inorganic
salts of copper,” respectively. Because of the possibility of comparing apples
with oranges, we do not evaluate these substances.
The NPRI database provides little ancillary information about the reporting facilities. Other than the emission data, we have information that
identifies to which industry a facility belongs, and we know the facility’s
location and employment level. From the location we use Census data to
infer how many people live in the surrounding area within a radius of 4km.
Drawing on the ARET program’s annual reports, we have constructed a
database of facilities participating in ARET and the year they joined. This
information was coded to match the NPRI facility identifiers.
5
Did ARET succeed in lowering emissions?
Table 3 presents releases of NPRI-recorded substances that were also reported to ARET from 1993 through 2002, with the primary ARET phase between 1994 and 2000. Because class A-1 toxins (the primary target of ARET)
were not reported to the NPRI during this period, this table shows only
class B substances except for a single A-2 substance. The table also reports
the number of reporting facilities, the percentage share of emissions from
ARET facilities, and changes over time for ARET and non-ARET facilities.
Comparing the time period 1999-2002 to 1994-1996 as the base period,13
ARET members outperformed non-ARET members for seventeen of twenty
group B substances for which data is available for both groups. Given the
small number of facilities reporting some substances, in some cases the difference is a function of only one or two members or non-members. However, in other cases there are real and quite dramatic differences between
13
The ARET program’s own objectives would imply comparison between releases in
1993, the year before the program began in early 1994, and 2000, the target year. However, since substances reports can vary substantially from year to year, we report multiyear
averages but do not claim that they are equivalent to ARET’s own reporting period.
13
members and non-members. For instance, ARET members significantly
reduced their releases of dichloromethane, formaldehyde, tetrachloroethylene, and methyl isobutyl ketone, even while non-ARET facilities increased
their emissions of the same substances. While both ARET members and
non-members reduced their releases of some substances, for substances like
trichloroethylene and bis-(2-ethylhexyl)-phthalate, ARET members made
deeper reductions. In other cases, however, both ARET members and nonmembers increased their releases, but ARET members’ releases typically
increased by less, as they did for anthracene and acetaldehyde.
ARET targeted a 90% reduction in emissions of class A-1 substances.
While we cannot ascertain whether or not the 90% target has been achieved
because there is no public record of emissions during the 1993-2000 period,
it is still possible to ascertain whether or not the long-term goal of “virtual elimination” of these toxins has been achieved. Table 4 presents the
average annual emission of class A-1 substances as reported to the NPRI
between 2001 (when the reporting requirement was introduced) and 2003
(the last year for which data are available at this point). The table shows the
average releases in tonnes (sorted in descending order). There are several
polycyclic aromatic hydrocarbons (PAHs) that are still emitted in relatively
large quantities in the order of 70-160 tonnes per year during the 3-year period, such as phenanthrene, pyrene, and fluoranthene.14 However, several
other substances such as hexachlorobenzene15 have seen significant reductions. ARET participants continue to account for the bulk of emissions of
A-1 substances, roughly 80% of total emissions.
Our table 5 shows substances reported both to NPRI and the ARET secretariat by ARET facilities. ARET participants reported emission levels in a
(arbitrary) base years of their own choosing and then annually during the
ARET reporting phase. In addition to the total in the (varying) base years
we show releases in 1993 and 2000, and corresponding percentage changes
relative to the base years. ARET reports significant reductions all across the
list of NPRI-matched substances; this is the basis for the claim of success of
ARET. However, it is noteworthy that in several cases a very high fraction
of the reductions reported to ARET had occurred before the program even
began.
How do our findings based on NPRI data compare with information reported to the ARET secretariat? To allow for volatility in annual emissions,
we base our ARET/NPRI comparison on 1995-2000 average annual emissions. It is peculiar that the same facilities often reported different numbers
to the NPRI. There are several cases where facilities inexplicably reported
substantially lower emissions to the NPRI than the ARET secretariat. In
14
PAHs are used to conduct research and to make dyes, plastics, pesticides, explosives,
and drugs. Risks to humans is primarily through breathing contaminated air.
15
Hexachlorobenzene (HCB) is a by-product in the production of organic compounds
such as rubber, dyes, and wood preservatives. Hexachlorobenzene will bioconcentrate in
fish and enter into the food chain.
14
some other cases higher emissions were reported to NPRI. The differences
are sufficiently large to leave the impression that quite different reporting
procedures must have been followed.
In tables 6 and 7 we also show the self-reported ARET results for groupA and group-B substances, respectively, that were not reported to NPRI during the period of ARET reporting. Again, the picture is one of significant
environmental progress across the entire spectrum of substances. However,
when we compare the ARET-reported emissions of the group-A substances
(1995-2000) in table 6 with the NPRI-reported emissions of the same substances in table 4, we find significant increases, particularly in the highest
volume substances. For example, phenanthrene, pyrene and fluoranthene
had annual average emissions of 62.2, 62.9 and 48 tonnes/year during 20012003 as reported to the NPRI, compared to 10.8, 7.7, and 9.9 tonnes/year as
reported to the ARET secretariat in 2000. This amounts to roughly quadrupled emission levels relative to releases of the same substances voluntarily
reported to ARET in 2000. Logically, either there was significant underreporting of releases to the ARET program (recall that ARET members were
not required to report their releases of all ARET substances), or emission
levels have increased quite dramatically. Either conclusion is troubling.
6
Who participated in ARET?
One of the fundamental questions about any voluntary challenge program
is: who participates in the program? What are the characteristics of companies or their production facilities that determine which participate and
which do not?
Understanding participation is critical from two points of view. First, as
participation is potentially endogenous, linking the participation decision
to observables allows researchers to analyze the efficacy of the program
by correcting for the self-selection bias caused by voluntary participation.
Second, understanding the determinants of the participation decision may
enable policy makers to improve participation.
Table 2 provides a breakdown of ARET members by 4-digit NAICS industry classification. Participants are heavily concentrated in a relatively
small number of industries. They fall largely into four groups: the pulp and
paper industry; the metal mining and processing industry; the chemical
products industry; and the oil and gas extraction and processing industries.
This relatively strong concentration of ARET membership supports the notion that there are potential intra-industry spillover effects or industry-level
economies of scale. This observation is in turn consistent with the hypotheses discussed above concerning regulatory threats, market pressures, and
direct cost savings, though not with firms seeking first-innovator advantage. It is also consistent with the origins of the ARET program, in that it
was negotiated by representatives of the trade associations for nine indus15
trial sectors, which subsequently encouraged participation by their members (ARET Secretariat 2003).
Our figure 1 reveals that facilities participating in ARET are smaller in
size than the population of NPRI reporting facilities overall. A significant
share of ARET facilities employs fewer than 100 people, whereas midsize
and large facilities are clearly underrepresented. Contrastingly, the last two
columns of table 2 reveal that it is mostly the larger facilities in the participating industries who enrol in ARET. Thus the apparent smaller average
size of ARET facilities relative to the NPRI population is a result of industry selection. Conditional on industry, larger facilities are more likely to
join ARET. As discussed above, this may be indicative of market benefits or
simply lower costs of abatement.
The need to deflect enforcement of existing regulations depends necessarily on the level of existing regulation, with industries facing significant
regulation being prime candidates for this hypothesis. As it turns out (see
table 2), industries that already are quite heavily regulated—such as the
pulp and paper industry—provide a significant number of ARET participants.
Through linking ARET membership to the NPRI data we are able to
learn some highly stylized facts about the participation decision. The NPRI
database provides information about employment Lijt of facility j in industry i in year t. From longitude/latitude information of facilities we measure
the population Nijt in a 4km radius around each facility. We also calculate
the toxicity-weighted sum Eijt of all direct emissions into air and water from
each facility. For this purpose we apply the CHHI scores discussed above.
We observe membership in ARET by linking information provided to us by
the ARET office to the NPRI database. Let ARETijt denote a binary indicator for facility j’s participation in ARET in year t. To capture the degree of
ARET pervasiveness for a particular industry, we calculate the percentage
share of the ARET facilities in that industry weighted by size (i.e., number
of employees). Concretely, we calculate
P
Higt = 100% ·
j
Lijt · ARETijt
P
j Lijt
(1)
at the NAICS 4-digit level (indexed g); thus our measure Higt retains intraindustry variation at the NAICS-2 level (indexed i) when we apply industry
fixed effects in our analysis. The distribution of Higt is shown in figure 2. For
most NAICS-4 industries that had at least one ARET participant the intraindustry participation rates were relatively low.
Using a Logit model of ARET participation, we estimate the log odds ratio of the probability pijt of facility j in industry i participating in the ARET
16
program in year t:16
"
#
pijt
= β1 ln(Lijt ) + β2 ln(Eijt ) + β3 ln(Nijt ) + β4 Higt + µt + νi + ijt (2)
ln
1 − pijt
In estimating equation (2) we express all firm characteristics in logarithmic
form as all three regressors are scaled over several orders of magnitude.
Conveniently, this expresses almost all estimates as odds-ratio elasticities.
We allow for an idiosyncratic error ijt , industry-specific effects νi , and timespecific effects µt .
In column (A) of table 8 we only include the main regressors. In columns
(B) through (D) we also allow for (2-digit NAICS) industry fixed effects and
year fixed effects. The results are extremely robust across specifications and
are highly significant. Clearly the strongest influence is exerted by the size
of the facility. Larger facilities are much more likely to participate in ARET
than smaller facilities. This is consistent with previous work on 33/50 and
Responsible Care (Arora and Cason, 1995, 1996, King and Lennox, 2000).
The odds-ratio elasticity of 0.53 suggests that doubling the size of a facility
increases the odds ratio of participation by about 44%.17 The significance of
facility size in determining ARET participation is perhaps an indication that
benefits from such participation are proportionally larger for large firms,
though this is consistent with several of the motives for participation discussed above.
There is also a small positive effect from the size of a facility’s total emissions. Facilities that generate more overall emissions—although not necessarily emissions covered by the ARET program—are slightly more likely to
enroll in the program. Again, this confirms previous work on other voluntary programs (Arora and Cason 1995, 1996, Khanna and Damon 1999) and
may be indicative of regulatory threat, consumer pressure, or opportunities
for productivity gains.
An initially surprising result is that ARET participation appears to fall
with the population of the area surrounding the facility as seen in columns
(A) and (B). This population density effect seems to indicate that facilities in
urban areas are less likely to participate in ARET. However, when we control for industry effects at the 2-digit NAICS level, this population density
effect becomes insignificant. We speculate that this may reflect geographic
idiosyncracies of the Canadian economy, in which pollution-intensive sectors such as pulp and paper and mining and smelting, both of which are
active in ARET, are located in rural areas in proximity to the resources they
16
For our data probit yields very similar results to logit.
What does it mean when the odds ratio increases by x? If a facility had even odds to
participate originally, i.e., p0 = 50%, then its new probability would be p1 = (1 + x)/(2 + x).
If x is 44%, then the new participation probability is 59%. More generally, for any initial
participation probability p0 > 0 and odds ratio increase x, the new participation probability
is given by p1 = (1 + x)/(1/p0 + x).
17
17
exploit. Once we control for industry-specific effects, we implicitly control
for location and population density.
Consistent with the regulatory threat, market pressure, and production
cost hypotheses, the pervasiveness of ARET participation in a given industry is a very strong positive predictor of a facility’s participation. The more
facilities participate (as measured by their cumulative percentage share of
employment Higt ), the greater the likelihood that an individual facility participates. For a 10%-point increase in ARET pervasiveness in an industry,
an individual facility’s odds ratio of participating in ARET increases by
49%.18 This is consistent with the origins of the ARET program; the ARET
challenge was negotiated by representatives of nine trade associations and
there was considerable encouragement, both within sectoral associations
and from Industry Canada and Environment Canada, for facilities in those
sectors to participate. These findings offer clear evidence for the importance of intra-industry coordination benefits and intra-industry spillovers,
lending support to our corresponding hypotheses above.
In columns (E) and (F) of table 8 we rearrange the estimating equation to
specifically identify emission intensity (i.e., our proxy variable ln(Eijt /Lijt )).
We wish to ascertain whether there is evidence in support of the productivity gains and regulatory threat hypotheses. Column (E) identifies emission
intensity without firm size, and column (F) identifies emission intensity together with firm size.19 The results confirm that there is a small but identifiable effect of emission intensity on ARET participation. While controlling
for industry effects, firms that are relatively more emission intensive are
more likely to participate in ARET. This gives support to the notion that environmental laggards are more likely to enroll in ARET than environmental
leaders. It refutes the notion that environmental leaders seek first-innovator
advantages by inviting regulation.
Table 8 also provides the Nagelkerke (1991) pseudo-R2 goodness-of-fit
statistic20 and the Akaike Information Criterion (AIC) statistic for the different logit model specifications. The reported pseudo-R2 s in table 8 suggest
that our model explains a significant portion of why facilities participate
in ARET. The AIC statistic aids in model selection; models with a lower
AIC are to be preferred. The regression results clearly prefer the models in
columns (D) and (F) with both time and industry fixed effects.21
18
For even odds of participation initially, this implies that the participation probability
rises to 60%, a 10%-point increase.
19
Note that columns (D) and (F) are linear transformations of each other because
β1 ln(L) + β2 ln(E) = (β1 + β2 ) ln(L) + β2 ln(E/L).
20
The Nagelkerke pseudo-R2 is calculated from two likelihood statistics: L(0) of the
model with intercept only, and L(b) of the model with
the full set of parameters
b. With n
observations, the pseudo-R2 is defined as the ratio 1 − {L(0)/L(b)}2/n / 1 − {L(0)}2/n .
21
The t-statistics reported in table 8 are based on ordinary standard errors, not clustered
standard errors. Because ARET participants may appear in several years in our panel,
there is the danger of reporting inflated t-statistics due to counting the same unit several
times. Because computing clustered standard errors in the presence of overlapping stratas
18
7
Evaluating ARET: Methodology
As discussed above, one of the more difficult issues in evaluating the efficacy of voluntary programs is the potential for self-selection. If firms that
are already inclined to take actions consistent with a program’s objectives
are more likely to sign on as participants, a straightforward evaluation of
the program’s impacts based on a comparison of participants and nonparticipants will tend to overstate the program’s marginal impact. We employ
an econometric technique to control for this “treatment effect” based on the
characteristics of “ARET joiners” compared to non-joiners.
In analyzing the merits of the ARET program, the starting point must
be the observation that ARET participation is voluntary. Participation is
an endogenous choice. Firms self-select into the ARET program. Consequently, determining whether or not companies experience environmental
performance improvements as a result of participating in ARET is subject to
self-selection bias. Thus ARET participation must be conditioned on determining factors. There is an extensive econometrics literature on treatment
effects, of which participation in ARET is a typical example.
The literature on estimating treatment effects is concerned with the evaluation of the performance of a particular program where the related explanatory variable (program participation) is binary. The basic idea is that
each firm or facility can have an outcome y1 with treatment and an outcome
y0 without treatment. However, we can only observe either y0 or y1 for a
given unit, not both. If p is a binary variable indicating participation (p = 1)
or non-participation (p = 0), then what is observed in practice is
y = (1 − p)y0 + py1 = y0 + p(y1 − y0 )
Researchers are interested in determining the average treatment effect
ATE = E{y1 − y0 } that describes the expected effect of treatment (program participation) on a randomly drawn company from the population.
Here, E indicates the expectation operator. If program participation was
randomized across companies, then the treatment effect could be estimated
consistently as the difference in means between the treated and non-treated
groups (similar to what is shown in table 7). However, program participation is not random as far as the ARET program is concerned. Concretely,
companies determine whether they participate in the program or not, and
their decision may be related to the benefits of participation, y1 − y0 . Then
the simple difference in means will overestimate the treatment effect. Correcting for self-selection complicates estimating the treatment effect.
(i.e., fixed effects) is very complicated and is not readily available in statistical software
packages, we have run separate logit regressions for each year to avoid clustering. The
resulting t-statistics are lower; they drop to about one-third of those reported in table 8.
However, all key estimates remain highly significant in table 8 except for the surrounding
population regressor, which is always insignificant.
19
A widely used method to control for treatment effects is described in
Greene (2003, pp. 787-788), which is very similar to the well-known Heckman estimator for dealing with selection bias.22 The binary decision pijt for
facility j in industry i to join the treatment program in year t is assumed to
be the outcome of a latent variable p∗ijt that is determined by a linear function of regressors zijt and noise µijt :
p∗ijt = zijt κ + µijt
(3)
The observed participation decision is pijt = 1 when p∗ijt > 0, and pijt = 0
otherwise. The equation of interest is
(4)
yijt = xijt β + γ pijt + ijt
where yijt is the performance measure, xijt are determining factors, and ijt
is noise. The parameter γ on the treatment variable pijt is of primary interest,
but is influenced by the participation decision. The disturbances and µ are
bivariate normal with zero mean and covariance matrix [[σ, ρ], [ρ, 1]]. The
two equations can be estimated jointly through maximum likelihood (ML)
or a two-step procedure. Conveniently, both estimators are implemented
in STATA through the treatreg command. We employ the ML estimator.
The difference in expected environmental performance between participants
and nonparticipants (ie, the average treatment effect) is given by
"
φ(zijt κ)
γ ≡ E(yijt |pijt = 1) − E(yijt |pijt = 0) = γ + ρσ
Φ(zijt κ){1 − Φ(zijt κ)}
∗
#
(5)
where φ(·) is the standard normal density function and Φ(·) is the standard
normal cumulative distribution function. If ρ is positive, OLS would have
overestimated the treatment effect; and if ρ is negative, OLS would have
underestimated the treatment effect.
To capture the effect of ARET participation, we estimate the change in
emissions Eijkt of chemical substance k facility j in industry i from year t to
year t + 1 through the model
ln(Eijk,t+1 ) − ln(Eijkt ) = β0 + β1 ln(Iijkt ) + β2 ln(Lijt ) + γpijkt + ijkt
22
(6)
We have also considered alternative estimation strategies, in particular a method
known as propensity score matching. This method identifies the treatment effect by comparing voluntary participants in the program with suitably matched non-participants. Participants and non-participants are matched according to the similarity of their characteristics,
which are summarized in a propensity score. A practical problem with this approach is that
one needs a relatively large set of observations in order to be able to control for a possibly
large set of characteristics. Working with relatively small data sets is thus problematic.
Furthermore, we are concerned that analyzing the ARET program using propensity score
matching would violate one of the identifying assumptions of this method, namely, the
means stable unit treatment assumption. This assumption states that the impact of a program one one firm must not depend on who else is participating in the program. As we are
deliberately exploring spillover effects of the ARET program, the outcome of the treatment
does depend on who else is participating, and this will violate the identifying assumption.
See Wooldridge (2002, ch. 18) for a discussion of alternative treatment effect estimators.
20
subject to the participation decision modeled by
p∗ijkt = κ0 +κ1 ln(Lijt )+κ2 ln(Eijt )+κ3 ln(Iijkt )+κ4 ln(Nijt )+κ5 Hijt +µijkt (7)
Here, Eijt is the toxicity-weighted sum of all emissions reported to NPRI,
Iijkt = Eijkt /Lijt is the proxy of emission intensity for particular chemical substance k, Nijt is the population in the surrounding area of facility j, and Lijt is employment, our proxy for facility size. We choose the
log time-difference form for our estimating equation in order to eliminate
facility-specific and industry fixed effects from the regression. The log timedifference form also allows us to estimate the treatment effect without having to adjust for firm size, and the estimated treatment effect is easily interpreted as a percentage change (for small changes). We do not pool our
observations but estimate our model separately for each substance k. We
do not adjust Eijkt for toxicity as the regressions are chemical-specific. However, when we consider a facility’s overall emissions Eijt we aggregate using
toxicity scores. Our model tries to capture the effects of emission intensity
and facility (firm) size. Introducing emission intensity allows us to capture
a firm’s “rung on the abatement ladder.” As argued earlier, facilities with
higher substance-specific emission intensities (the laggards) may embark
on more aggressive emission reductions than facilities with low substancespecific emission intensities (the leaders). Introducing facility size allows us
to capture facility-level economies of scale in dealing with emission abatement.
There is an important caveat in our analysis. Capturing the treatment
effects chemical by chemical re-estimates the participation equation each
time. Because not all facilities produce the same chemicals, the participation equation may be estimated with different facilities from our sample.
While there is no a-priori reason to believe that the participation decision
works differently for different subgroups of our sample, it is nevetheless a
reason for concern because the participation decision is at the facility level
and integrates the benefits from participation across all chemicals. We therefore also consider an alternative approach where we sum up chemicals by
weight. We are unable to adjust for toxicity when adding up chemicals because of unavailable toxicity scores for many substances.
8
Evaluating ARET: Results
Table 9 reports the results of our treatment effect regressions separately for
each of the 28 substances that were reported to the NPRI between 1994 and
2000. For practical purposes we also eliminate substances from our analysis for which have fewer than 50 observations across facilities and years.
This leaves us with merely 17 ARET substances from the B-1, B-2, and B-3
categories, or about 15% of the total number of substances. While this cov21
ers only a small part of the ARET program, the fact that large numbers of
facilities report these substances indicates their relative importance.
Table 9 suppresses the results of the treatment participation equation
and only shows the estimation results for the equation of interest, the treatment effect (column ARET), and the estimation bias (column atanh(ρ)). In
what follows we index chemical substances with the letter k.
First, the estimation results for the emission intensities (column ln(Iijkt ))
are consistently negative and in all but three cases significant. This means
that more pollution-intensive facilities are reducing emissions faster than
less pollution-intensive facilities. For dichloromethane, a facility that is 50%
more pollution-intensive than another will reduce emissions 1.4% faster per
year. The effect is much larger for anthracene (24% faster) and asbestos
(17% faster).23 As this effect is independent of ARET participation, there
must be a benefit to emission reductions independent of ARET. This result
is consistent with firms reacting independently to the threat of regulation
as in Antweiler (2003), where the most pollution-intensive firms are on the
lowest rungs on the “abatement ladder.”
Second, we also observe that larger facilities are reducing emissions
faster than smaller facilities as captured by our facility size proxy employment in column ln(Lijt ). The effect is significant for 8 of the 17 chemicals we
analyze. This result may be indicative of facility-specific economies of scale
in abatement technology or abatement opportunities. It may also be a result of greater stakeholder or regulatory pressure on larger facilities whose
emissions are, overall, more noticeable. The effect we observe varies in magnitude. For anthracene and asbestos, facilities that are double the size of
another facility will reduce emissions 33% and 26% faster per year, respectively. Given the cumulative nature of ARET participation, this amounts to
very large differences in emission reductions.
The column ARETijt is of primary interest in our analysis; it shows
the estimate of the average treatment effect (γ ∗ ). A negative sign on the
estimate indicates that ARET facilities are reducing emissions faster than
their non-ARET counterparts, whereas a positive sign on the estimate indicates that ARET facilities are reducing emissions slower than their nonARET counterparts. For the 17 B-group substances analyzed by us, we
find that the treatment is statistically insignificant for ten of the substances.
In two cases (ethylene oxide and acetaldehyde) we find a significant positive sign, indicating that ARET was unsuccessful in the sense that participants were outperformed by their non-ARET counterparts.24 For five substances (anthracene, tetrachloroethylene, asbestos, dichloromethane, and
23
Calculated as 100%[1.5a − 1] with estimates a from table 9.
Table 3 reveals that emissions of ethylene oxide decreased for ARET facilities while
they increased for non-ARET facilities. The positive estimate of the treatment effect is thus
due to self-selection bias of the participating facilities, of which they were relatively few
for this substance. Emissions of acetaldehyde increased substantially for both ARET and
non-ARET facilities.
24
22
formaldehyde) we find that ARET accelerated emission reductions. In these
five cases the magnitude of the treatment effect is economically meaningful
and actually quite sizable: ARET participants are reducing emissions faster
than non-participants by 19% per year (anthracene), by 26% per year (tetrachloroethylene), by 1.8% per year (asbestos), by 36% per year (dichloromethane), and by 30% per year (formaldehyde).25 As ARET membership
spans about six years, ARET facilities virtually eliminate these substances.
It is, however, plausible that the apparent ARET effect for these substances reflects other factors that have not been controlled for in our regressions. In the case of asbestos, the NPRI program restricted the conditions
under which asbestos was to be reported (limiting reporting to only the
“friable form” of asbestos) in 1995, which would have had the effect of reducing at least some facilities’ reports even in the absence of any abatement
effort. Such facilities may have been the ones most inclined to sign on to
ARET. Tetrachloroethylene was identified as toxic as defined by the Canadian Environmental Protection Act in 1993, and as a result was the subject
of two intensive stakeholder negotiations (concerning dry cleaning and degreasing) in the mid-1990s to determine whether a voluntary or regulatory
approach should be taken. Greater reductions of this substance may thus
reflect the very explicit regulatory threat, which might have had a comparable effect even in the absence of ARET. Moreover, technological change apparently independent of environmental concerns was reducing emissions
of tetrachloroethylene from the dry cleaning sector dramatically during the
1990s.26 This may have provided an opportunity for facilities adopting new
technology to selectively sign up for the ARET program to claim credit for
emissions reductions they were going to achieve in any case. Similar to
tetrachloroethylene, dichloromethane was also identified as “CEPA-toxic”
in 1993 and a stakeholder issue table was established to advise the government on policy options. While a comparable regulatory threat thus exists in this case as well, it is noteworthy that the stakeholder group explicitly recommended that facilities releasing dichloromethane join ARET to
demonstrate the utility of a voluntary approach. However, even there, Environment Canada’s own evaluation of the ARET program speculates that
facilities used participation in ARET to avoid or delay a more stringent regulation (Environment Canada 2000).
The last column atanh(ρ) shows the estimate (and direction) of the program participation self-selection bias.27 When ρ is positive, the OLS esti25
Since ∆ ln(E) = ln(Et+1 /Et ), the estimate γ of the treatment effect can be expressed as
a percentage change 100%[exp(γ ∗ ) − 1].
26
The dry cleaning “issue table” estimated that in the absence of any regulatory intervention, emissions of tetrachloroethylene from this sector would fall from 7111 to 5549 metric
tons, or over 20 percent from 1990 to 2000.
27
For reasons of numerical stability during the maximization of the likelihood function,
Stata estimates the inverse hyperbolic tangent atanh(ρ) = ln[(1 + ρ)/(1 − ρ)]/2 rather than
ρ directly. For small values of ρ, it holds that atanhρ ≈ ρ. The two diverge as ρ → ±1.
23
mate of the treatment effect is biased upward, and when ρ is negative, the
OLS estimate of the treatment effect is biased downward. Keeping in mind
that the desirable ARET treatment effect is one that is numerically negative
(leading to emission reductions), a positive ρ indicates that OLS underestimates the usefulness of ARET, whereas a negative ρ indicates that OLS overestimates the usefulness of ARET. In the five cases where we find that ARET
has had a positive environmental impact, ρ is positive and OLS would have
underestimated the usefulness of ARET. These results confirm that correcting for the self-selection bias of a voluntary program is necessary. However,
the direction of the self-selection bias is contrary to the assumption that only
“easy achievers” joined the ARET. Apparently, ARET was joined by many
facilities for whom participation was indeed a challenge.
The scarce coverage of ARET substances by the NPRI severely restricts
our ability to empirically analyze the success of the ARET program. For the
number of substances for which we can evaluate ARET participation, ARET
cannot be heralded a comprehensive success. For only five of the seventeen
B-group substances is it plausible that ARET has had a significant positive
impact on the speed of emission reductions. And as our comments above
suggest, special circumstances may have influenced ARET participation for
some of these five chemicals.
The last row of table 9 reports the results from a weight-aggregated
treatment effect model at the facility level. As before, larger firm size and
higher prior emission intensity are associated with faster emissions reductions. The estimated treatment effect is numerically positive and significant.
According to this specification, ARET participants were slower at reducing
emissions than their non-ARET peers. This is an even more pessimistic conclusion than the results from the chemical-by-chemical analysis.
Our skepticism concerning ARET’s impact is reinforced by other reviews
of the program. The federal government commissioned a survey of participants from the pulp and paper, base metals smelting and refining, and
chemical manufacturing sectors, which collectively account for 79% of reductions from 1993 to 2000 (Gunn, Moffett and Bregha 1999). The study
concluded that ARET “does not appear to have had a widespread, direct impact on emissions.” Respondents from the pulp and paper industry, which
alone contributed 59% of ARET reductions after 1993, cited new federal and
provincial regulations as the primary motive for emissions reductions in
that sector. Respondents from the metals sector also emphasized the impact
of existing regulations. Other reasons for emissions reductions included
technological upgrades that had concurrent environmental benefits and the
threat of regulation, with the voluntary Responsible Care program being
cited by respondents from the chemicals sector.
24
9
Conclusions
A central premise of the turn toward more cooperative approaches to environmental policy is that negotiated programs will be more effective because
participating firms will have “bought in” to their objectives (Georg 1994).
Thus one might expect a voluntary challenge like ARET that was negotiated
with partners from a broad range of industries to be more effective than a
unilateral government challenge such as 33/50. However, others have questioned the price of business consent in terms of sacrifice of environmental
objectives (Hajer 1994, Harrison 1999). As Ayres and Braithwaite (1992)
note, “The very conditions that foster the evolution of cooperation are also
the conditions that promote the evolution of capture”. Consistent with this,
we find many troubling aspects of the ARET program that are difficult to
imagine as part of a traditional government statute or regulation. These include the option for participants’ to choose their own base years, and the
lack of transparency of participants’ reports, which makes it virtually impossible to systematically and comprehensively evaluate the success of this
program. It is especially troubling to us that the Canadian federal government would sponsor a program that did not allow for transparency and
verifiability, and thus accountability. It would seem that business’ consent
came at a high price indeed.
The lack of integration evident among Environment Canada’s own programs is also disconcerting. Fewer than half of the ARET priority substances were included on the NPRI list developed at roughly the same time.
If the ARET stakeholders’ identification of priority substances based on
their persistence, toxicity, and bioaccumulative capacity was correct, one
would think those substances should have been reported to NPRI. If, on
the other hand, the NPRI list was a more reasonable reflection of priorities,
then the ARET program’s focus was misplaced. It is also troubling that the
federal government has persistently trumpeted the success of ARET in the
face of limited and even at times contrary evidence. The ARET program has
strained credulity by taking credit for reductions made years before the program began. As late as 2003, the final program report concluded that “The
ARET program [...] succeeded in reducing releases into the environment by
over 72% from base year levels” (ARET 2003), even though roughly half of
those reductions were achieved before the program began and three years
after the federal government’s own program evaluation had concluded that
ARET was not a significant factor in accounting for the remaining reductions (Environment Canada, Review Branch 2000).
Our study also reveals interesting facts about which firms participated in
the ARET voluntary challenge program and why. It appears that large and
emission intensive facilities are more likely to join an environmental voluntary challenge program than facilities that are either small or relatively
clean. Most importantly, participation in the voluntary challenge increases
along with the participation of other facilities in the same industry. This
25
points to the existence of intra-industry spillovers or benefits. This in turn
is consistent with several possible hypothesis: that industry-wide participation in a voluntary challenge program serves as an inoculation against future regulation, that participation generates benefits through intra-industry
technology sharing, or that coordination has benefits via joint marketing.
We also find that participation is high in industries that have faced stricter
environmental regulation in the past, or which are threatened by stricter
regulation in the future. We find no support for the hypothesis that participation is driven by ‘green leaders’ among firms in an attempt to invite
regulation that solidifies competitive advantages.
Our evidence on the environmental impact of ARET is mixed at best.
For the priority substances (class A-1) we can only draw inferences from
the data reported to the NPRI during 2001-2003 after the ARET program
was completed. There we find that many of these highest priority toxins have not been virtually eliminated as targeted. In particular, several
polycyclic aromatic hydrocarbons (PAHs) are still emitted in large quantities. Moreover, discrepancies between voluntary ARET reports prior to
2001 and mandatory NPRI reports thereafter for these substances suggests
either significant increases in releases in recent years or very selective reporting by ARET participants in earlier years. Our analysis of the ARET
treatment effect of the B-group substances that were reported to the NPRI
during 1994-2000 does not support the notion that ARET was comprehensively a successful program. We find a positive impact of ARET only for
five of the seventeen substances for which there is sufficient data for analysis. Where we find a positive impact of ARET, it has been environmentally
meaningful. However, even there, regulatory developments or changes in
reporting requirements may offer plausible alternate explanations for these
reductions (and, implicitly, ARET participation).
There are several lessons to be learned from ARET. First and foremost, a
voluntary challenge program must provide transparency and accountability. Progress should not be self-reported based on arbitrary base year data,
but should be subject to precise and comprehensive reporting standards
and measured against clear targets. It is heartening that the federal government’s Policy Framework for Environmental Performance Agreements (Environment Canada 2001), adopted since the conclusion of ARET, reflects these
lessons.28 However, the question remains whether the apparently minimal
impacts of ARET justify another kick at the can. While our findings concerning intra-industry spillovers suggest some potential for governments to
pursue industry-specific programs that are narrower in scope, we find little
basis for the current enthusiasm for voluntary approaches in the record of
the ARET Challenge.
28
The Policy Framework stipulates four principles for negotiating VCPs: effectiveness
(achieving measurable environmental results); credibility (capacity to deliver targeted environmental results); transparency and accountability (verifiability of program commitments); and efficiency (programs should cost not more than alternatives for equivalent
environmental results).
26
References
Alberini, Anna and Kathleen Segerson (2002) “Assessing Voluntary Programs to
Improve Environmental Quality,” Environmental and Resource Economics, 22 (12), 157–184.
Anton, Wilma Rose Q., George Deltas, and Madhu Khanna (2004) “Incentives
for environmental self-regulation and implications for environmental performance,” Journal of Environmental Economics and Management, 48 (1), 632–654.
Antweiler, Werner (2003) “How Effective is Green Regulatory Threat?,” American
Economic Review, 93 (2), 436–441.
and Kathryn Harrison (2003) “Toxic Release Inventories and Green Consumerism: Empirical Evidence from Canada,” Canadian Journal of Economics,
36 (2), 495–520.
ARET Secretariat (1997) Environmental Leaders 2: Voluntary action on toxic substances, Environment Canada.
(1999) Environmental Leaders 3, Environment Canada.
(2003) Environmental Leaders 4, Environment Canada.
Arora, Seema and Timothy N. Cason (1995) “An Experiment in Voluntary Environmental Regulation: Participation in EPA’s 33/50 Program,” Journal of Environmental Economics and Management, 28 (3), 271–286.
and
(1996) “Why Do Firms Volunteer to Exceed Environmental Regulations? Understanding Participation in EPA’s 33/50 Program,” Land Economics,
72 (4), 413–432.
and
(1999) “Do Community Characteristics Influence Environmental
Outcomes? Evidence from the Toxics Release Inventory,” Southern Economic
Journal, 65 (4), 691–716.
Ayres, Ian and John Braithwaite (1992) Responsive Regulation: Transcending the
deregulation debate, New York: Oxford University Press.
Darier, Eric and Debora VanNijnatten (1996) “Lessons learned from ARET: A
qualitative survey of perceptions of stakeholders,” Technical Report, School
of Policy Studies, Queen’s University, Kingston, Ontario. Final Report.
and
(2001) “Voluntary or Involuntary? ARET and the Caprices of Policy
Instrument Choice,” in William Leiss, ed., In the Chamber of Risks: Understanding Risk Controversies, McGill-Queen’s University Press, chapter 9.
Dauvergne, John (1998) “Integrating Voluntary and Regulatory into Toxics Risk
Management: Industry Canada’s Perspective.” Presentation to the Policy Forum sponsored by the Eco-Research Chair Program in Environmental Policy,
Queen’s University, 9 November 1998.
27
DeCanio, Stephen J. and William E. Watkins (1998) “Investment in Energy Efficiency. Do the Characteristics of Firms Matter,” Review of Economics and Statistics, 80 (1), 95–107.
Environment Canada (2001) Policy framework for environmental performance agreements. ISBN 0662658795.
Environment Canada, Review Branch (2000) “Evaluation of the Accelerated Reduction and Elimination of Toxics Initiative (ARET).” Final Report.
Gallon, Gary (1998) “Accuracy is optional in reporting voluntary success,” Alternatives Journal, 24 (2), 12.
Gamper-Rabindran, Shanti (forthcoming) “Did the EPA’s Voluntary Industrial
Toxics Program Reduce Emissions? A GIS Analysis of Distributional Impacts
and By-Media Analysis of Substitution,” Journal of Environmental Economics
and Management.
Georg, S. (1994) “Regulating the Environment: Changing from Constraint to Gentle Coercion,” Business Strategy and the Environment, 3 (2), 43–49.
Greene, William H. (2003) Econometric Analysis, 5th ed., Upper Saddle River, NJ:
Prentice Hall.
Gunn, Cynthia, John Moffett, and Francois Bregh (1999) “Review of ARET’s Impact in Pulp and Paper, Base Metals Smelting and Refining, and Chemical
Manufacturing.” Report prepared for Environment Canada.
Hajer, M .A. (1994) “Verinnerlijking: the limits to a positive management approach,” in G. Teubner, L. Farmer, and D. Murphy, eds., Environmental Law
and Ecological Responsibility, New York: John Wiley and Sons.
Harrison, Kathryn (1999) “Talking with the Donkey: Cooperative Approaches to
Environmental Protection,” The Journal of Industrial Ecology, 2 (3), 51–72.
(2001) “Voluntarism and Environmental Governance,” in E. A. Parson, ed.,
Governing the Environment: Persistent Challenges, Uncertain Solutions, Toronto:
University of Toronto Press.
(2002) “Challenges in Evaluating Voluntary Programs,” in Paul Stern, ed.,
New Tools for Environmental Protection: Education, Information, and Voluntary
Programs, Washington, DC: National Academy Press.
and Werner Antweiler (2003) “Incentives for Pollution Abatement: Regulation, Regulatory Threats, and Non-Governmental Pressures,” Journal of Policy
Analysis and Management, 22 (3), 361–382.
Hartman, Raymon S. (1988) “Self-Selection Bias in the Evolution of Voluntary Energy Conservation Programs,” Review of Economics and Statistics, 70 (2), 448–
458.
28
Industry Canada (1998) The Power of Partnerships: Industry and Government Working Together for Economic Growth and a Cleaner Environment, Industry Canada,
Sustainable Technologies and Services Industries Branch. ISBN 0-662-63541-8.
Khanna, Madhu and Lisa A. Damon (1999) “EPA’s Voluntary 33/50 Program: Impact on Toxic Releases and Economic Performance of Firms,” Journal of Environmental Economics and Management, 37, 1–25.
and Wilma Rose Q. Anton (2002) “Corporate Environmental Management:
Regulatory and Market-Based Incentives,” Land Economics, 78 (4), 539–558.
, Wilma Rose H. Quimio, and Dora Bojilova (1998) “Toxics Release Information: A Policy Tool for Environmental Protection,” Journal of Environmental
Economics and Management, 36, 243–266.
King, Andrew and Michael J. Lenox (2000) “Industry Self-Regulation Without
Sanctions: The Chemical Industry’s Responsible Care Program,” Academy of
Management Journa, 43 (4), 698–716.
Lutz, Stefan, Thomas P. Lyon, and John W. Maxwell (2000) “Quality Leadership
when Regulatory Standards are Forthcoming,” Journal of Industrial Economics,
48 (3), 331–348.
Lyon, Thomas P. and John W. Maxwell (2003) “Self-Regulation, Taxation and
Public Voluntary Environmental Agreements,” Journal of Public Economics, 87,
1453–86.
and
(2004) Corporate environmentalism and public policy, New York and
Melbourne: Cambridge University Press.
Maxwell, John W., Thomas P. Lyon, and C. Steven Hackett (2000) “Self-Regulation
and Social Welfare: The Political Economy of Corporate Environmentalism,”
Journal of Law and Economics, 43 (2), 583–617.
Nagelkerke, N. J. D. (1991) “A note on a general definition of the coefficient of
determination,” Biometrika, 78 (3), 691–692.
OECD (2000) Voluntary Approaches for Environmental Policy: An Assessment, Paris:
Organization for Economic Cooperation and Development. ISBN 9264171312.
(2003a) Voluntary Approaches for Environmental Policy: Effectiveness, Efficiency
and Usage in Policy Mixes, Paris: Organization for Economic Cooperation and
Development. ISBN 9264101772.
(2003b) Voluntary Approaches: Two Canadian Cases, Paris: Organization for
Economic Cooperation and Development. ENV/EPOC/WPNEP(2002)10/FINAL. The original draft was prepared by Mary Jane Middlekoop, François
Bregha and John Moffet of the consultancy firm “Stratos´´, Ottawa, Canada.
Porter, Michael and Claas van der Linde (1995a) “Green and Competitive,” Harvard Business Review, 73 (5), 120–134.
29
and
(1995b) “Toward a New Conception of the EnvironmentCompetitiveness Relationship,” Journal of Economic Perspectives, 9 (4), 97–118.
Prakash, Aseem and Matthew Potoski (2005) “Covenant with Weak Swords: ISO
14001 and Firms’ Environmental Performance,” Journal of Policy Analysis and
Management, 24 (4), 745–769.
Rivera, Jorge, Peter de Leon, and Charles Koerber (2005) “Is Greener Whiter Yet?
The Sustainable Slopes Program after Five Years.” George Washington University School of Business Working Paper.
Segerson, Kathleen and Thomas J. Miceli (1998) “Voluntary Environmental
Agreements: Good or Bad News for Environmental Protection?,” Journal of
Environmental Economics and Management, 36 (2), 109–130.
Toffel, Michael W. and Julian D. Marshall (2004) “Journal of Industrial Ecology,” Improving Environmental Performance Assessment: A Comparative Analysis of Weighting Methods Used to Evaluate Chemical Release Inventories, 8 (1-2),
143–172.
U.S. Environmental Protection Agency Risk-Screening Environmental Indicators
User’s Manual, Technical Appendix A, Listing of All Toxicity Weights for TRI Chemicals and Chemical Categories (2002).
VanNijnatten, Deborah L. (1998) “The day the NGOS walked out,” Alternatives
Journal, 24 (2), 10–15.
Videras, Julio and Anna Alberini (2000) “The appeal of voluntary environmental
programs: which firms participate and why?,” Contemporary Economic Policy,
18 (4), 449–461.
Welch, Eric W., Allan Mazur, and Stuart Bretschneider (2000) “Voluntary Behavior by Electric Utilities: Levels of Adoption and Contribution of the Climate
Challenge Program to the Reduction of Carbon Dioxide,” Journal of Policy
Analysis and Management, 19 (3), 407–425.
Wooldridge, Jeffrey W. (2002) Econometric Analysis of Cross Section and Panel Data,
Cambridge, MA: MIT Press.
30
Tables and Figures
Table 1: Summary of ARET Substance Lists
A-1
A-2
B-1
Criteria
Reduction
Targets
Number of
Chemicals
Persistence
Bioaccumulation
Toxicity
Short-Term (by 2000)
Long-term
all
—, actually reported
covered by NPRI
—, before 2000
covered by EPA-CHHI2
—, and by NPRI
yes
yes
yes
90%
100%
30
25
18
0
6
1
1
B-2
B-3
yes
yes1
yes
50%
no
yes
yes
50%
yes
no
yes
50%
no
no
yes
50%
2
2
1
1
2
1
8
7
3
2
5
2
33
29
10
10
25
10
44
30
17
15
34
16
Note: 1 no consensus on whether these substances meet persistence and bioaccumulation and
toxicity criteria. 2 risk-screening toxicity scores (chronic human health indicator) provided by
the U.S. Environmental Protection Agency.
Table 2: Industry Composition of ARET Members
3221
2122
3251
2211
3241
3259
2111
3255
3261
3311
3256
3313
3212
3254
3262
3314
3252
3312
3329
NAICS-4 Industries
Pulp, Paper & Paperboard Mills . . . . . . . . . . . . . . . .
Metal Ore Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Basic Chemical Mfg. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Electricity Generation, Transmission & Dist. . . . .
Petroleum & Coal Products Mfg. . . . . . . . . . . . . . . .
Other Chemical Product Mfg. . . . . . . . . . . . . . . . . . .
Oil & Gas Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . .
Paint, Coating & Adhesive Mfg. . . . . . . . . . . . . . . . .
Plastic Product Mfg. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Iron & Steel Mills & Ferro-Alloy Mfg. . . . . . . . . . .
Soap, Cleaning Compound & Toilet Prep. Mfg. .
Alumina & Aluminum Production & Processing
Veneer, Plywood & Engin. Wood Product Mfg. .
Pharmaceutical & Medicine Mfg. . . . . . . . . . . . . . . .
Rubber Product Mfg. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Non-Ferrous (exc. Al) Production & Processing .
Resin, Synth. Rubber, & Fibre & Filament Mfg. .
Steel Product Mfg. from Purchased Steel . . . . . . .
Other Fabricated Metal Product Mfg. . . . . . . . . . . .
All Other Industries . . . . . . . . . . . . . . . . . . . . . . . . . . . .
31
Facilities
ARET
All
66
123
39
87
34
116
30
62
21
38
21
119
15
110
13
104
13
165
12
20
8
55
7
45
6
46
6
29
6
42
6
38
5
37
5
65
5
61
30 1,554
ARET Share
Faci. Empl.
53.7 % 67.3 %
44.8 % 54.7 %
29.3 % 60.7 %
48.4 % 83.4 %
55.3 % 73.9 %
17.6 % 40.8 %
13.6 % 26.7 %
12.5 % 26.1 %
7.9 %
6.4 %
60.0 % 90.3 %
14.5 % 30.1 %
15.6 % 36.2 %
13.0 %
9.4 %
20.7 % 17.3 %
14.3 % 29.5 %
15.8 % 50.0 %
13.5 % 38.9 %
7.7 % 14.1 %
8.2 % 11.0 %
32
n
11
51
90
14
10
126
5
19
44
150
17
82
35
2
87
13
20
5
310
65
4
275
15
219
195
183
2
42
111
719.0
400.1
.3000
783.6
743.4
200.8
.2450
870.2
782.3
203.1
456.6
172.5
6.014
15.02
.0050
2143.
1066.
1.000
1116.
1.204
1041.
2843.
2261.
40.04
.1410
1287.
105.6
186.1
315.5
2.000
188.8
5.114
18.77
410.9
3.120
1995
9.364
2.080
26.65
6.028
7.571
1.147
.0100
7.769
238.6
2242.
26.09
147.1
225.5
1994
9.900
1.938
24.17
7.409
9.064
.5770
.0120
17.51
.0410
2237.
50.84
168.9
310.1
4.000
140.0
1.507
18.43
.0050
2599.
1736.
1993
24.80
10.04
9.775
5.788
24.40
.1250
.0850
836.7
752.8
312.0
1349.
1.383
1996
9.100
2.977
28.83
16.71
6.395
.8860
.0100
.4900
208.2
2341.
22.83
201.3
128.2
2.350
277.4
.8860
10.39
.0050
1996.
1171.
.1430
694.3
727.3
282.7
.7800
860.7
643.0
270.0
1262.
1.985
.0080
.4640
208.3
2218.
18.52
132.0
105.8
.1960
200.5
.2300
4.950
.0310
1477.
925.6
.0100
.2570
220.9
2445.
18.00
148.0
111.4
.8160
230.1
.3270
5.190
.0110
1672.
1203.
1791.
2.898
1998
9.040
1.617
8.383
26.01
3.036
1997
8.000
3.851
19.29
18.27
4.289
1647.
1.934
6510.
708.1
311.8
.2000
1.418
689.7
.0050
.4940
114.3
2388.
21.69
125.0
95.71
1.190
867.4
.2830
8.185
.3660
1337.
947.1
1999
10.55
1.093
8.976
26.95
1.774
1803.
1.180
6126.
657.2
356.9
.1900
.7300
649.7
952.1
.1200
6.000
.4400
1133.
726.7
.3100
18.05
2219.
18.53
121.6
119.7
2000
12.70
4.850
18.22
19.71
.8000
1727.
1.628
7234.
676.5
367.0
.1900
92.56
752.9
1095.
.1020
8.686
.3740
1047.
609.8
.0080
.3110
30.29
1777.
19.21
100.8
96.07
2001
12.94
9.751
13.07
14.17
.5890
1920.
2.263
4938.
751.0
289.4
.3210
.9320
746.2
939.6
.0940
7.908
.2620
854.0
521.9
.0040
.3380
64.45
1086.
15.35
43.29
83.74
2002
11.13
10.18
7.193
7.949
.7910
-36.5
-3.5
+102
-19.0
-6.8
+18.7
†
-3.7
-23.8
-14.7
+198
-88.3
-57.2
†
-59.7
-60.7
+6.1
-96.7
+91.3
-85.3
-100
-46.9
-78.6
-50.2
-27.5
-66.2
-98.0
-62.3
ARET
share1
0.0
55.2
23.9
100.0
63.9
9.3
100.0
5.4
41.7
35.5
95.0
40.3
97.0
0.0
55.1
40.6
96.0
100.0
86.3
87.9
0.0
36.5
87.5
66.6
21.9
23.7
94.5
1.3
3.1
NPRI
-11.0
+51.8
-46.9
+91.4
-91.7
-100
-46.9
-97.8
-49.2
-7.9
-63.6
-44.5
-59.6
-85.7
+395
-95.6
-50.8
†
-54.4
-55.0
-100
+81.2
+22.7
∆%2
ARET
†
-3.1
+3.6
+26.6
-9.0
+8.4
-100
+223
-98.5
-48.4
+4.9
+59.7
+127
+255
-85.7
†
-99.7
†
ARET
-11.0
+124
-28.4
†
-99.8
-100
Note: n is the number of reporting facilities. 1 share is the percentage of emissions by ARET-listed facilities relative to all NPRI reporters. 2 percentage
changes are based on the average of the years 1994-1996 (denominator) and 1999-2002 (numerator). Changes for ARET include facilities that were ARET
members for at least 4 years. All other facilities are grouped together as non-ARET (ARET). † indicates changes over +1000%. 3 mixed isomers. All emissions
are reported in metric tons.
A2
B1
B1
B2
B2
B2
B2
B2
B2
B2
B2
B2
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
ARET Substance (by List)
p-Dichlorobenzene
Anthracene
Bis(2-ethylhexyl) phthalate
1,2-Dichloroethane
1,4-Dioxane
Asbestos (friable form)
Benzyl chloride
Carbon tetrachloride
Chloroform
Dichloromethane
Ethylene oxide
Tetrachloroethylene
1,3-Butadiene
2,4-Dinitrotoluene
Acetaldehyde
Acrylamide
Acrylonitrile
Aniline (and its salts)
Benzene
Chlorine dioxide
Epichlorohydrin
Formaldehyde
Hydrazine (and its salts)
Hydrogen sulphide
Methyl isobutyl ketone
Phenol (and its salts)
Quinoline (and its salts)
Toluenediisocyanate3
Trichloroethylene
NPRI-Reported Emissions by Year [tonnes]
Table 3: NPRI-Recorded Emissions of ARET-Listed Substances
Table 4: Average Annual Emissions of ARET A-1 Substances as reported to
the NPRI During 2001–2003
Emissions
Facilities
[tonnes/year]
ARET A-1 Substance
ARET All [%] ARET
All [%]
Phenanthrene
61 125 48.8 62.24 159.1 39.1
Pyrene
62 121 51.2 62.91 78.64 80.0
Fluoranthene
62 127 48.8 48.01 73.39 65.4
Benzo(b)fluoranthene
60 111 54.1 30.07 34.16 88.0
Benzo(e)pyrene
55
95 57.9 26.97 29.15 92.5
Benzo(a)anthracene
58 118 49.2 20.04 22.55 88.9
Benzo(a)pyrene
60 119 50.4 16.76 18.19 92.1
Benzo(g,h,i)perylene
60 106 56.6 12.87 13.91 92.5
Benzo(k)fluoranthene
58 102 56.9 11.10 12.26 90.5
Indeno(1,2,3-CD)pyrene
59 108 54.6 10.77 11.82 91.1
Benzo(a)phenanthrene
54
88 61.4 3.835 10.15 37.8
Benzo(j)fluoranthene
50
71 70.4 7.064 7.500 94.2
Dibenzo(a,h)anthracene
58 102 56.9 4.743 5.091 93.2
Dibenzo(a,i)pyrene
38
60 63.3 1.911 1.944 98.3
Perylene
47
80 58.8 .5391 .9585 56.2
Hexachlorobenzene
92 391 23.5 .0107 .0369 29.1
Dibenz(a,j)acridine
37
54 68.5 .0087 .0100 87.5
7H-Dibenzo(c,g)carbazole
36
51 70.6 .0047 .0055 85.0
p,p’-Methylenebis(2-chloroaniline)
0
1
0.0
Note: The table is sorted in descending order of average annual emissions. This table records 19
ARET A-1 substances reported to the NPRI. Eleven other ARET A-1 substances are not reported
to the NPRI. All emissions are reported in metric tonnes.
33
Table 5: Emissions Reported by ARET-participating Facilities to ARET and
to the NPRI, by ARET Group and Chemical Substance
CAS Substance
B1
B1
B2
B2
B2
B2
B2
B2
B2
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
120-12-7
117-81-7
56-23-5
123-91-1
127-18-4
67-66-3
75-21-8
100-44-7
107-06-2
79-06-1
91-22-5
79-01-6
10049-04-4
50-00-0
71-43-2
7783-06-4
107-13-1
106-99-0
302-01-2
62-53-3
108-10-1
108-95-2
75-07-0
26471-62-5
Anthracene . . . . . . . . . . . . . . .
Bis(2-ethylhexyl) phthalate
Carbon tetrachloride . . . . . .
1,4-Dioxane . . . . . . . . . . . . . . .
Tetrachloroethylene . . . . . . .
Chloroform . . . . . . . . . . . . . . .
Ethylene oxide . . . . . . . . . . . .
Benzyl chloride . . . . . . . . . . .
1,2-Dichloroethane . . . . . . . .
Acrylamide . . . . . . . . . . . . . . .
Quinoline (and its salts) . . .
Trichloroethylene . . . . . . . . .
Chlorine dioxide . . . . . . . . . .
Formaldehyde . . . . . . . . . . . .
Benzene . . . . . . . . . . . . . . . . . .
Hydrogen sulphide . . . . . . .
Acrylonitrile . . . . . . . . . . . . . .
1,3-Butadiene . . . . . . . . . . . . .
Hydrazine (and its salts) . .
Aniline (and its salts) . . . . .
Methyl isobutyl ketone . . .
Phenol (and its salts) . . . . . .
Acetaldehyde . . . . . . . . . . . . .
Toluenediisocyanate (m.i.)
Base
Year
25.10
20.60
81.40
34.30
266.0
746.0
145.0
.0200
70.30
1.730
1.550
126.0
2,007
1,815
3,918
16,474
24.50
367.0
5.920
1.100
282.0
329.0
437.0
.0060
ARET Self-Reported
1993 2000
1993
2000
∆[%] ∆[%]
22.30 7.870
–11
–69
1.480 11.10
–93
–46
39.50 .5500
–51
–99
24.30 .4300
–29
–99
178.0 3.610
–33
–99
425.0 65.00
–43
–91
104.0 16.80
–28
–88
.0200 .0050
0
–75
20.10 22.80
–71
–68
1.730 .1500
0
–91
1.290 .1900
–17
–88
45.00 22.30
–64
–82
1,627 408.0
–19
–80
1,511 412.0
–17
–77
2,880 996.0
–26
–75
11,263 4,369
–32
–73
22.40 7.180
–9
–71
321.0 119.9
–13
–67
5.920 2.040
0
–66
.0061 .4330
–99
–61
263.0 131.0
–7
–54
264.0 168.0
–20
–49
279.0 261.0
–36
–40
.0060 .0040
0
–33
95-00
Avg.
14.03
11.43
2.223
3.830
39.74
156.1
19.77
.0080
20.17
.3533
.6233
24.50
588.3
509.3
1,550
6,121
9.558
131.4
2.638
.1442
149.8
200.2
247.2
.0047
NPRI
95-00 Diff.
Avg.
[%]
2.268 +519
3.074 +272
.2582 +761
3.698
+4
38.28
+4
70.59 +121
19.20
+3
.0072
+12
18.94
+7
.2178
+62
.1950 +220
21.37
+15
837.9
–30
593.1
–14
1,292
+20
4,331
+41
8.068
+18
126.1
+4
1.754
+50
.1430
+1
146.5
+2
76.51 +162
281.9
–12
.1092
–96
Note: All emissions are reported in metric tonnes. Comparison is based on average 1995-2000 emissions of ARET
facilities. A plus sign in the last column indicates that self-reported ARET emissions exceeded the NPRI-reported
emissions, and a minus sign indicates that self-reported ARET emissions were less than what ARET facilities reported to the NPRI. Columns ∆[%] show percentage changes from the base year to the reference year.
34
Table 6: Self-Reported ARET Emissions of Non-NPRI A1/A2 Chemicals
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A1
A2
CAS Substance
Unit
118-74-1
29082-74-4
1746-01-6
51207-31-9
87-86-5
194-59-2
206-44-0
193-39-5
192-97-2
218-01-9
191-24-2
207-08-9
56-55-3
205-82-3
189-55-9
198-55-0
50-32-8
129-00-0
53-70-3
85-01-8
205-99-2
224-42-0
NA - 03A
tonne
kg
kg
kg
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
Hexachlorobenzene . . . . . . . . . . . . . . . . . . . . .
Octachlorostyrene . . . . . . . . . . . . . . . . . . . . . . .
2,3,7,8-Tetrachlorodibenzo-p-dioxin . . . . .
2,3,7,8-Tetrachlorodibenzofuran . . . . . . . . .
Pentachlorophenol (PCP) . . . . . . . . . . . . . . . .
7H-Dibenzo(c,g)carbazole . . . . . . . . . . . . . . .
Fluoranthene . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Indeno(1,2,3-CD)pyrene . . . . . . . . . . . . . . . . .
Benzo(e)pyrene . . . . . . . . . . . . . . . . . . . . . . . . .
Benzo(a)phenanthrene . . . . . . . . . . . . . . . . . .
Benzo(g,h,i)perylene . . . . . . . . . . . . . . . . . . . .
Benzo(k)fluoranthene . . . . . . . . . . . . . . . . . . .
Benzo(a)anthracene . . . . . . . . . . . . . . . . . . . . .
Benzo(j)fluoranthene . . . . . . . . . . . . . . . . . . . .
Dibenzo(a,i)pyrene . . . . . . . . . . . . . . . . . . . . . .
Perylene . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Benzo(a)pyrene . . . . . . . . . . . . . . . . . . . . . . . . .
Pyrene . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Dibenzo(a,h)anthracene . . . . . . . . . . . . . . . . .
Phenanthrene . . . . . . . . . . . . . . . . . . . . . . . . . . .
Benzo(b)fluoranthene . . . . . . . . . . . . . . . . . . .
Dibenz(a,j)acridine . . . . . . . . . . . . . . . . . . . . . .
Cadmium (inorganic/respirable/soluble)
Base
Year
.0691
.7000
.1460
.8958
.5700
.0441
66.00
6.430
4.490
25.50
3.770
8.850
16.40
.3000
4.440
1.190
12.90
22.80
10.30
25.20
4.470
.0441
126.0
ARET Self-Reported
1993
2000
1993
∆[%]
.0674
0
–2
0
0
–100
.0032 54E-5
–98
.0249 .0042
–97
.0531 .0040
–91
.0436 .0013
–1
64.70 9.860
–2
6.190 1.270
–4
4.470 .9300
–0
24.50 5.730
–4
3.750 .9200
–1
8.450 2.190
–5
15.70 4.310
–4
.2600 .0951
–13
3.820 1.410
–14
1.030 .3800
–13
12.30 4.140
–5
21.70 7.780
–5
9.650 3.720
–6
23.20 10.80
–8
3.860 2.110
–14
.0436 .0413
–1
90.20 30.20
–28
2000
∆[%]
–100
–100
–100
–100
–99
–97
–85
–80
–79
–78
–76
–75
–74
–68
–68
–68
–68
–66
–64
–57
–53
–6
–76
Note: The table is sorted in descending order of reported emission reductions from the base year to year 2000.
Columns ∆[%] show percentage changes from the base year to the reference year.
35
Table 7: Self-Reported ARET Emissions of Non-NPRI B1/B2/B3 Chemicals
B1
B1
B1
B1
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B2
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
CAS Substance
Unit
88-06-2
57-97-6
28804-88-8
77-47-4
534-52-1
111-44-4
58-90-2
75-27-4
30777-19-6
238-84-6
7440-41-7
74-87-3
226-36-8
124-48-1
NA - 07
NA - 10A
NA - 14A
NA - 04A
NA - 05A
NA - 08A
NA - 11A
NA - 15A
NA - 02A
NA - 13A
NA - 06A
576-26-1
606-20-2
86-30-6
621-64-7
121-14-2
62-75-9
106-93-4
120-83-2
64-17-5
60-35-5
593-60-2
107-02-8
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
kg
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
tonne
2,4,6-Trichlorophenol . . . . . . . . . . . . . . . . . . . .
7,12-Dimethylbenz(a)anthracene . . . . . . . .
Dimethylnaphthalene . . . . . . . . . . . . . . . . . . .
Hexachlorocyclopentadiene . . . . . . . . . . . . .
4,6-Dinitro-o-cresol (and its salts) . . . . . . . .
Bis(2-chloroethyl)ether . . . . . . . . . . . . . . . . . .
2,3,4,6-Tetrachlorophenol . . . . . . . . . . . . . . . .
Dichlorobromomethane . . . . . . . . . . . . . . . . .
Benzo(b)fluorene . . . . . . . . . . . . . . . . . . . . . . . .
Benzo(a)fluorene . . . . . . . . . . . . . . . . . . . . . . . .
Beryllium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Methylene chloride (chloromethane) . . . .
Dibenz(a,h)acridine . . . . . . . . . . . . . . . . . . . . .
Chlorodibromomethane . . . . . . . . . . . . . . . . .
Cyanides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mercury (elemental, inorganic) . . . . . . . . . .
Zinc (inorganic/respirable/soluble) . . . . .
Chromium (Cr6+) . . . . . . . . . . . . . . . . . . . . . . .
Cobalt (inorganic, soluble) . . . . . . . . . . . . . .
Lead (all forms except alkyl) . . . . . . . . . . . .
Nickel (inorganic/respirable/soluble) . . .
Uranium (inorganic/respirable/soluble)
Arsenic (inorganic) . . . . . . . . . . . . . . . . . . . . . .
Silver (soluble inorganic salts) . . . . . . . . . . .
Copper (inorganic salts) . . . . . . . . . . . . . . . . .
1,6-Dimethylphenol . . . . . . . . . . . . . . . . . . . . .
2,6-Dinitrotoluene . . . . . . . . . . . . . . . . . . . . . . .
N-Nitrosodiphenylamine . . . . . . . . . . . . . . .
n-Nitrosodi-n-propylamine . . . . . . . . . . . . .
2,4-Dinitrotoluene . . . . . . . . . . . . . . . . . . . . . . .
n-Nitrosodimethylamine . . . . . . . . . . . . . . . .
1,2-Dibromoethane (Ethylene dibromide)
2,4-Dichlorophenol (and its salts) . . . . . . . .
Ethanol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Acetamide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Vinyl bromide . . . . . . . . . . . . . . . . . . . . . . . . . .
Acrolein . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Base
Year
4.980
.1000
22.20
.0060
.2900
.0160
.4300
.1200
3.100
4.210
.1300
1,093
.0334
.3000
242.0
29.60
3,321
26.90
13.10
1,896
1,424
.1400
342.0
3.310
1,105
.0640
.1200
.0084
.0095
1.503
.0441
1.370
1.390
570.0
.0010
.0010
13.70
ARET Self-Reported
1993
2000
1993
∆[%]
.9700 .0141
–81
.0890 .0328
–11
19.30 7.500
–13
.0060 .0060
0
.2800
0
–3
0
0
–100
.1400 .0040
–67
.0770 .0240
–36
3.070 .6800
–1
3.980 1.670
–5
.0937 .0750
–28
1,064 634.0
–3
.0329 .0211
–1
.3000 .3000
0
157.0 15.20
–35
11.20 3.340
–62
1,613 716.0
–51
24.10 6.180
–10
11.30 3.340
–14
1,134 502.0
–40
579.0 431.0
–59
.1400 .0667
0
119.0 169.0
–65
2.310 1.750
–30
787.0 626.0
–29
0
0
–100
0
0
–100
0
0
–100
0
0
–100
.0010 .0010
–100
.0061 14E-5
–86
1.330 .0404
–3
1.000 .1500
–28
301.0 110.0
–47
.0010 .0010
0
.0010 .0010
0
9.060 25.40
–34
2000
∆[%]
–100
–67
–66
0
–100
–100
–99
–80
–78
–60
–42
–42
–37
0
–94
–89
–78
–77
–75
–74
–70
–52
–51
–47
–43
–100
–100
–100
–100
–100
–100
–97
–89
–81
0
0
85
Note: The table is sorted in descending order of reported emission reductions from the base year to year 2000.
Columns ∆[%] show percentage changes from the base year to the reference year.
36
37
Higt [%]
0.045c
(39.0)
no
no
14,006
344
2,569
0.391
8,031.4
0.595c
(24.4)
(F)
−0.038a −0.003
(2.38)
(.199)
c
0.068
0.064c
(13.7)
(12.2)
c
c
c
c
0.044
0.040
0.040
0.045
0.040c
(38.8)
(31.5)
(31.2)
(36.5)
(31.2)
no
NAICS2 NAICS2 NAICS2 NAICS2
yes
no
yes
yes
yes
14,006
14,006
14,006
14,006
14,006
344
344
344
344
344
2,569
2,569
2,569
2,569
2,569
0.391
0.350
0.352
0.276
0.352
7,978.5 7,682.8 7,507.7 8,171.1 7,507.7
(E)
Note: Absolute values of z-statistics are given in parentheses; the square of each z-statistic is the Wald-χ2 statistic that is used
for hypothesis testing. Statistical significance at the 95%, 99%, and 99.9% levels of confidence is indicated by the superscripts a ,
b
, and c , respectively.
Industry Fixed Effects
Year Fixed Effects
Observations
ARET Members
Non-ARET Members
Nagelkerke Pseudo R2
Akaike Information Criterion
ARET Intra-Industry Share
Emission Intensity
Surrounding Population
Total Adj. Emissions
Employees
Intercept
Table 8: Logit Regressions of ARET Participation
(A)
(B)
(C)
(D)
c
−5.469
(27.3)
ln(Lijt )
0.427c
0.426c
0.535c
0.531c
(18.9)
(18.9)
(21.6)
(21.4)
c
c
c
ln(Eijt )
0.066
0.066
0.061
0.064c
(13.2)
(13.3)
(11.8)
(12.2)
c
c
ln(Nijt )
−0.049 −0.050 −0.004 −0.003
(3.38)
(3.43)
(.234)
(.199)
ln(Eijt /Lijt )
38
atanh(ρ)
0.738b (2.60)
0.051 (.260)
0.283a (2.17)
−0.182 (.450)
0.472a (2.35)
−1.965c (3.39)
0.571b (2.61)
0.093 (.200)
−0.885c (4.19)
−0.185 (1.33)
0.217 (1.00)
0.390b (2.98)
0.374a (2.11)
0.292 (1.02)
0.239 (1.29)
0.182 (1.24)
−0.133 (.360)
−0.102a (1.98)
Note: The estimates of the constant terms and the estimates of the treatment participation equation are not shown. Absolute values
of t-statistics are given in parentheses. Statistical significance at the 95%, 99%, and 99.9% levels of confidence is indicated by the
superscripts a , b , and c , respectively. (m.i.: mixed isomers)
B1
B1
B2
B2
B2
B2
B2
B3
B3
B3
B3
B3
B3
B3
B3
B3
B3
Table 9: Treatment Effect Regressions for ARET Substances
Substance
N
ln(Iijkt )
ln(Lijt )
ARETijt
c
c
Anthracene
103 −0.671 (7.69) −0.581 (7.10) −0.209b (2.65)
Bis(2-ethylhexyl) phthalate
231 −0.075a (2.40) −0.080a (2.51) −0.059 (.720)
Asbestos (friable form)
338 −0.445c (10.9) −0.443c (10.8) −0.018a (2.33)
Chloroform
59 −0.070b (2.87) −0.041 (1.01) −0.060 (.380)
Dichloromethane
443 −0.034b (2.75) −0.010 (.440) −0.443b (2.65)
Ethylene oxide
75 −0.405c (4.97) −0.461c (4.98)
0.865c (4.83)
Tetrachloroethylene
260 −0.074c (4.18) −0.034 (1.59) −0.301b (3.13)
1,3-Butadiene
100 −0.100 (1.83) −0.006 (.140) −0.066 (.210)
Acetaldehyde
145 −0.036 (1.02) −0.085 (1.75)
0.482c (3.69)
Benzene
815 −0.181c (9.93) −0.098c (4.09)
0.165 (1.39)
Chlorine dioxide
350 −0.211c (5.56) −0.033 (.460) −0.124 (.460)
Formaldehyde
713 −0.071c (4.45) −0.022 (.760) −0.356b (2.67)
Hydrogen sulphide
286 −0.062b (2.65) −0.014 (.540) −0.293 (1.69)
Methyl isobutyl ketone
521 −0.088c (5.28) −0.037b (2.57) −0.209 (1.03)
Phenol (and its salts)
466 −0.180c (7.47) −0.154c (5.23) −0.218 (1.36)
Toluenediisocyanate (m.i.)
203 −0.513c (8.29) −0.511c (8.27) −0.001 (.090)
Trichloroethylene
303 −0.028 (1.35) −0.034 (1.96)
0.066 (.270)
c
c
All Chemicals (weight sum) 3,215 −0.101 (12.8) −0.066 (6.56)
0.127a (2.30)
Figure 1: Size Distribution of ARET and non-ARET Facilities
28
Percentage Share of Facilities
26
24
Others
22
ARET
20
18
16
14
12
10
8
6
4
2
0
1−9
10−19
20−49
50−99 100−199 200−499 500−999 1k−2k
2k−5k
5k−10k
10k+
Facility Size Group (Number of Employees)
Figure 2: Distribution of Intra-Industry ARET Participation
11
10
Number of Industries
9
8
7
6
5
4
3
2
1
0
0−
10
10−
20
20−
30
30−
40
40−
50
50−
60
60−
70
70−
80
80−
90
Percentage Share of Intra−Industry ARET Participation
90−
100
Note: The histogram shows the distribution of the number of industries with different levels of
intra-industry ARET participation as measured by the measure Hijt described in the text. Only
industries with at least one ARET participant are shown; industries with no ARET participants
are excluded from the histogram. Industries are defined at the 4-digit NAICS level. To adjust for
different facility sizes, participation shares are weighted by facility employment.
39