CHEMICAL MFRS. ASS'N v. E.P.A.

GARWOOD, Circuit Judge:

Pursuant to the Toxic Substances Control Act (TSCA), 15 U.S.C. §§ 2601-2629, respondent, the United States Environmental Protection Agency (EPA), promulgated a final rule, 53 Fed.Reg. 28,195 (1988), 40 C.F.R. § 799.1285, requiring manufacturers and processors of the chemical cumene (isopropyl benzene) to perform certain toxicological testing of it to determine its health and environmental effects. Petitioners, the Chemical Manufacturers Association (CMA), a trade association representing the chemical industry, and five of its members engaged in cumene manufacturing or processing, bring the instant proceeding in this Court pursuant to TSCA § 19, 15 U.S.C. § 2618, to obtain judicial review of the rule and have it set aside.

Background

Statutory framework

The TSCA, enacted in 1976, P.L. 94-469, 90 Stat. 2003 et seq., provides, among other things, for the EPA to both substantively regulate the manufacturing and processing of chemicals, TSCA § 6, 15 U.S.C. § 2605, and to require health and environmental effects testing of chemicals by and at the expense of their manufacturers and processors. TSCA § 4, 15 U.S.C. § 2603.1 It is the latter authority—that to require testing—with which we are here concerned. In this connection, Congress declared, "It is the policy of the United States" that

Implementing this policy, the EPA is authorized by TSCA § 4(a), 15 U.S.C. § 2603(a), to require testing under certain circumstances.2 It is required that the EPA find there to be "insufficient data and experience upon which the effects of" the manufacturing or processing of the chemical "on health or the environment can reasonably be determined or predicted," and that testing of the chemical "with respect to such effects is necessary to develop such data." It is also required that there be a finding either that the chemical's manufacturing or processing "may present an unreasonable risk of injury to health or the environment," TSCA § 4(a)(1)(A)(i), or that, as stated in TSCA § 4(a)(1)(B)(i):

Here the final rule was based entirely on TSCA § 4(a)(1)(B). That section does not require, and the EPA did not here find, that manufacturing or processing of the chemical may present an unreasonable risk of health or environmental injury.

The testing ordered under TSCA § 4 must be "to develop data" respecting those "health and environmental effects" as to which there is insufficient "data and experience and which are relevant to a determination" that the manufacturing or processing of the chemical "does or does not present an unreasonable risk of injury to health or the environment." TSCA § 4(a).

TSCA § 4(e) provides for the establishment of a committee of representatives from specified federal agencies and federally funded institutions—now known as the Interagency Testing Committee (ITC)—to recommend to the EPA those chemicals to which the EPA "should give priority consideration for promulgation of" a testing requirement under TSCA § 4(a), and requires that the EPA, within a year from such a recommendation as to a given chemical, either initiate a rulemaking proceeding for testing under TSCA § 4(a) or publish its reasons for not doing so.3

Before requiring testing under TSCA § 4, the EPA must follow the rulemaking standards of the Administrative Procedure Act, 5 U.S.C. § 553, generally requiring publication of the proposed action and a minimum thirty-day period for participation in the rulemaking through comment, and also must allow for (and transcribe) oral, as well as written, presentation of "data, views, or argument," and make and publish "the findings described in § 4(a)(1)(A) or 4(a)(1)(B)." TSCA § 4(b)(5).

While the TSCA grants the EPA broader testing than regulatory authority,4 Congress also plainly intended the EPA to consider the economic impact of any actions taken by it under the TSCA,5 and expressly provided for judicial review of EPA testing (as well as regulatory) orders under the TSCA, with the review in each case to be under the substantial evidence rule.6

Cumene

Cumene, a colorless liquid with a sharp odor, is produced or processed at a total of some sixteen plants in the United States, nearly all concentrated in the vicinity either of Houston or Philadelphia. Cumene is commercially manufactured by the reaction of benzene and propylene under elevated temperature and pressure in the presence of a catalyst, most often solid phosphoric acid. Production was approximately 3.3 billion pounds in 1984, with annual capacity estimated at about 4.4 billion pounds. Cumene is one of the top fifty chemicals produced in the United States. About five percent of cumene produced is exported, with some ninety-nine percent of the balance being used to make two other industrial chemicals, phenol and acetone, and in this processing cumene is chemically transformed and is present only in trace quantities in the acetone and phenol products distributed in commerce. Some of the plants produce the cumene they process, while others purchase it from producers.

Cumene occurs in the environment from a variety of sources apart from its commercial manufacturing and processing. Cumene is a natural product that is present in many foods, automobile and truck exhaust, and as a natural component of crude oil. It is also present in a variety of consumer products and in cigarette smoke. The major source of cumene in most urban environments is probably from hydrocarbon fuel combustion, primarily by land transportation vehicles such as cars and trucks.

Test rule proceedings

In November 1984, the ITC, in its fifteenth report to the EPA, recommended that cumene be considered for required testing for health effects—short-term genotoxicity, chronic effects including oncogenicity, and teratogenicity and reproductive toxicity—and for ecological effects, namely, acute and chronic toxicity to estuarine and freshwater fish and invertebrates. 49 Fed.Reg. 46,931, 46,939 (1984). In April 1985, the EPA held public meetings, participated in by CMA, in reference to the ITC recommendation concerning cumene testing.

In late November 1985, the EPA responded to the ITC recommendation by issuing its proposed test rule on cumene. 50 Fed.Reg. 46,105 (1985). The proposed rule includes these findings:

CMA submitted written comments on the proposed rule in February 1986, and in April 1986, the EPA held a one-day public hearing on the proposed cumene test rule at which CMA made a presentation in opposition to the proposed rule.

The EPA published its final rule requiring cumene testing on July 27, 1988, to be effective September 8, 1988. 53 Fed.Reg. 28,195 (1988), 40 C.F.R. § 379.1285. The EPA considered the comments submitted by CMA in response to the proposed rule, and accepted some of them. The EPA recognized that CMA's presentation raised "some doubt" as to whether a study the EPA had relied on in connection with the proposed rule's findings concerning mutagenic and oncogenic effects was "positive or equivocal" and that other cumene test results submitted by CMA were negative in these respects, so that the EPA "has decided that no further testing in these areas is necessary at this time." 53 Fed.Reg. 28,196. Apparently for this reason, the final rule does not contain the proposed rule's finding that cumene "may present an unreasonable risk of mutagenic and oncogenic effects," or any other finding that cumene may present an unreasonable risk of injury to health or the environment. And, unlike the proposed rule, which rested in part on TSCA § 4(a)(1)(A), as well as § 4(a)(1)(B), the final rule rests entirely on § 4(a)(1)(B). In other respects, however, the EPA in the final rule rejected nearly all the criticisms leveled by CMA at the proposed rule,7 and made findings under TSCA § 4(a)(1)(B) similar to those made respecting that section in the proposed rule. The final rule requires health effects, environmental effects, and chemical fate testing.8

The principal factual criticisms leveled by CMA at the proposed rule were that its estimate of three million pounds annual release into the environment from cumene manufacturing and processing facilities was excessive, that the contribution to cumene in the environment from such releases was insignificant as compared to other sources, that occupational exposure was not significant, and that cumene in the aquatic environment was not shown to have come from cumene manufacturing or processing facilities.

As to the three million pounds annual release into the air from cumene manufacturing and processing facilities, CMA criticized the data relied on by the EPA as being based on a five-year-old study of two manufacturing facilities which did not allow for the fact that most cumene plants are now covered by the Clean Air Act and as a result "are now, or will soon be, subject to the more stringent control requirements imposed by state and federal authorities" that would reduce leaks and similar sources of fugitive emissions found in the EPA study. CMA also supplied its own study, which indicated that approximately 844,000 pounds of cumene were released annually as fugitive emissions, based on a study of eleven of the sixteen plants accounting for eighty percent of the manufactured and processed cumene. The EPA in the final rule did not address the criticism of its study, but did reject the CMA study, concluding that it did not address processing emission, particularly, as was often the case, where processing was done "at the same site [as manufacturing] using a separate physical system," and that "fugitive emissions of cumene from processing are generally estimated to be twice those from manufacturing." The EPA also noted that in extrapolating its figures, CMA "has incorrectly assumed that cumene emissions are strictly proportional to the amount of cumene manipulated, regardless of whether the cumene was manufactured or processed" and also had not considered "the age or size of the plant." 53 Fed.Reg. 28,197.

As to the comparison of cumene plant emissions to those from other sources, the EPA conceded that taking the nation as a whole, land vehicle emissions contributed about five times as much cumene to the environment as did cumene manufacturing and processing facilities. However, the EPA concluded that "in communities close to" these facilities "it appears that these facilities emit approximately 3.6 times the amount of cumene emitted by land vehicles exhaust and, hence, are the dominant source of atmospheric cumene." Id. The EPA also noted that fifteen to sixteen million people live within a fifty-kilometer radius of all cumene manufacturing and processing facilities, and that ninety-seven percent of the cumene and sixty-six percent of phenol capacity "are concentrated in areas with a population of about 7 million people." Facilities within the latter areas "are predicted to emit some 2.58 million pounds of cumene per year into the atmosphere .... By comparison, automobiles in these areas are predicted to emit only 0.47 million pounds per year"—so that cumene emissions from these facilities would be about five and a half times as much as from vehicle exhaust. Further, "the half-life of cumene in the atmosphere"—conceded to be "on the order of one or two days"—"is long enough to allow some transport" so "the vast majority of atmospheric cumene in these areas must come from cumene manufacturing and processing facilities." Id. The EPA also noted that its "worst case" modeling studies— based on hypothetical atmospheric inversions which "trapped" cumene emitted from plants for twenty-four hours — showed concentrations up to 59.9 parts per billion (ppb) within a one-kilometer radius from plants and up to 3.1 ppb at a five-kilometer radius. Id. The EPA noted, but did not expressly respond to, CMA contentions that the "`worst case' figures may correspond to occasional peak emissions, but do not reflect long-term average emission levels." The EPA did observe that in areas without cumene manufacturing or processing facilities, monitoring indicated "cumene concentrations at or below 2 ppb in the air," while for places with such facilities, the monitoring indicated "much higher concentrations of cumene in the air" with "[s]ome of the highest monitored cumene concentrations (6 and 11 ppb)" being "near the Shell Oil Company manufacturing complex in Deer Park, TX." Id.9

The EPA also rejected CMA's contentions based on a survey of 393 routinely and 346 intermittently exposed cumene facility employees, involving 1,487 samples, of which all but thirty-five were below one part per million (ppm), six were at 4.01 to 30 ppm, four at 3.01 to four ppm, and twenty-five at 1.01 to two ppm. The EPA found this survey flawed because of acknowledged water vapor interference and noted, "the entire set of data in the survey may be suspect." The final rule goes on to state: "Nevertheless, the information provided by the survey is of concern to the Agency because of the potential for chronic adverse health effects to workers from exposure levels reported." 53 Fed.Reg. 28,198.10 Having initially recited CMA's contention "that worker exposure to cumene at manufacturing and processing facilities is neither `substantial' nor `significant' under section 4(a)(1)(B)," the final rule concludes the worker exposure discussion by stating, "the Agency believes that occupational exposure to cumene, when considered along with the potential for general population exposure to cumene, meets the exposure criteria needed to make a section 4(a)(1)(B) finding under TSCA (i.e., the chemical is produced in substantial quantities and there is potential for substantial human exposure)." Id.

The final rule also observes that "EPA does not believe ... that the current health effects data base for cumene is adequate to allay the concern that cumene may present a threat of chronic adverse health effects at levels presently in the environment," and that "[t]he available acute and subchronic data are not sufficient to reasonably predict the dose-response curve for chronic human exposure." Id.

In the final rule, the EPA does not dispute CMA's comments that the EPA ground and drinking water surveys showed cumene was rarely detected in water and then generally in trace amounts, with no positive findings in areas shown to be in the vicinity of cumene manufacturing or processing facilities, but observes that, although "no conclusions can be made concerning the presence or absence of cumene ... in the waters near these [cumene] facilities," nevertheless the "EPA does know that a number of cumene bearing waste streams are generated from industrial processes and that cumene is discharged to the aquatic environment," and finds CMA's contentions "to be less than convincing because monitoring data from waters near cumene manufacturing and processing facilities are not available for evaluation." Id. The final rule further notes that "detection of cumene in surface water also suggests that cumene has a long enough half-life to build up detectable concentrations in surface water systems." Id. at 28,199.11

The crucial ultimate findings of the final rule are stated as follows:

Discussion

The ultimate contested finding here is that "cumene ... enters the environment in substantial quantities, with the potential for resulting substantial human exposure to cumene, from its manufacture, processing, use, and disposal."

CMA attacks this finding from both factual and legal perspectives.

Factual challenges

CMA urges that the finding that an estimated three million pounds per year fugitive cumene emissions enter the atmosphere from cumene manufacturing and processing facilities is exaggerated, preferring its own estimate of approximately 844,000 pounds. It likewise challenges the finding that in the areas where most of the cumene facilities are located and in the vicinity of which about 13.5 million people live, the cumene fugitive emissions from such facilities are predicted to be the greater source of exposure to the general population. We reject these challenges and conclude that these findings are supported by substantial evidence on the rulemaking record taken as a whole. TSCA § 19(c)(1)(B). The EPA's expert studies support the three-million-pound findings, and CMA did not so compellingly demonstrate that the EPA studies were fatally flawed as to require their rejection in this respect. On the other hand, the EPA identified sufficient defects in the CMA study to justify a determination not to rely on it. Cf. Chemical Manufacturers Association v. U.S.E.P.A., 859 F.2d 977, 979 (D.C.Cir. 1988). Of course, we deal with a subject matter—estimated fugitive plant emissions—that does not lend itself to precision or certainty, and we do not claim, and nor did the EPA, that either has been achieved. But then, neither is required. Section 4(a)(1)(B)(i), clause (I), allows a "may reasonably be anticipated" finding, while clause (II) uses the phrase "may be." These contrast to the more certain "is or will be" of the opening clause of paragraph (B)(i) respecting quantity of production. Further, the key word "substantial" in section 4(a)(1)(B)(i) suggests that rough approximation suffices. There is substantial evidence in the record taken as a whole to justify the conclusion that the three million pounds figure is a reasonable ball-park estimate which is closer to the mark than CMA's 844,000 pounds. Similarly, we conclude that the record adequately supports the finding that, in the areas near cumene facilities, emissions therefrom predictably would be a more significant source of human exposure than the other major known source, land vehicle emissions. This conclusion could reasonably be reached from the plant emission estimates and the essentially unchallenged vehicle emission data, which indicated that in the areas where most cumene facilities were located cumene emission therefrom were about 3.6 times those of vehicles and, in portions of these areas where about seven million people lived, were about five and a half times those of vehicles. The EPA's finding in this respect also received some support from monitoring data, the one- to two-day atmospheric half-life of cumene and the "worst case" studies. Even if only two million pounds of cumene a year were released into the atmosphere from cumene facilities, and even if such emissions were the source of only twice as much cumene exposure as from vehicles to seven million residents in the general areas of the facilities, it does not appear to us that this necessarily would or should have made a difference in the ultimate test rule, nor do we understand CMA to contend otherwise.

CMA also challenges the underlying finding that "releases of cumene to the aquatic environment are expected as a result of cumene-bearing wastewater discharged from cumene manufacturing and processing facilities." We sustain this challenge. The EPA has pointed to no substantial record evidence tending to support this assertion, neither any analysis of any waste-water discharge (or, indeed, any water) from (or in) any such facilities, nor any evidence that any cumene has ever been detected in any water in the vicinity of any such facilities. Cumene was detected in ground water in only a few of the many samples taken and then in only low ppb concentrations, but none of the positive samples were in the vicinity of or otherwise linked to any cumene facilities. While 204 of 4,000 samples of industrial waste-water streams were identified which contained cumene (though 97.9 percent of these cumene-bearing streams were recycled or reused and thus did not enter the environment), none of these were from cumene manufacturing or processing facilities. Moreover, no evidence was cited indicating that cumene manufacturing or processing facilities were similar in relevant aspects to the facilities shown to have cumene in their waste-water streams (or in their discharged waste-water streams). Thus, there is simply no substantial evidence to support the EPA's above finding in this respect.13 We do not conclude, however, that this deficiency necessarily invalidates the EPA's findings under section 4(a)(1)(B) either as to entering the environment in substantial quantities or as to potential resulting substantial human exposure. The record as a whole appears to indicate that the EPA on both counts primarily focused on the emissions or entry into the atmosphere (and workplace exposure) as opposed to release or entry into the aquatic environment, and no attempt was made to even generally categorize the magnitude or extent of the latter. It seems likely that the ultimate section 4(a)(1)(B)(i) findings would have been made in the absence of the subsidiary finding as to release into the aquatic environment.14 The EPA should, however, clarify this on remand.

CMA also levels what are, or depend on, essentially legal challenges to the final rule.

Statutory construction challenges

CMA asserts that a finding of "substantial" under either of clauses (I) or (II) of section 4(a)(1)(B)(i) requires a showing that the quantity or level of exposure is such that if the chemical were "highly toxic" it could "realistically be expected" to cause health or environmental harm. CMA urges that this follows from the undoubted facts that what ultimately concerned Congress in enacting the TSCA was avoidance of injury to human health or the environment, rather than simple scientific curiosity or other concerns, and that under section 4(a) testing may be ordered only to develop data needed to determine that manufacturing or processing (or distribution, use, or disposal) of the chemical "does or does not present an unreasonable risk of injury to health or the environment." This position has at least a surface plausibility, and we are not persuaded by so much of the EPA's reply thereto as suggests that CMA's position amounts to no more than importing the "unreasonable risk of injury" requirements of section 6(a) or section 4(a)(1)(A)(i) into section 4(a)(1)(B)(i). CMA's referenced position would not require the EPA to find or have evidence indicating that the chemical even might be or is suspected of being toxic, but rather would only require the EPA to assume hypothetically that the chemical were "highly toxic" and to then assess whether the quantities involved were "substantial" in relation to that hypothetical level of toxicity.

Ultimately, however, we are not persuaded that the EPA is obliged to adopt CMA's referenced construction of "substantial." The TSCA contains no definition of "substantial," nor does its legislative history. And the legislative history does not in any other respect lend any specific support to CMA's position. In these circumstances, Congress is deemed to have implicitly delegated to the EPA the power to define or interpret "substantial," and we will sustain the agency's interpretation as long as it is rational and consistent with the statutory scheme and legislative history. See Chemical Manufacturers Association, 859 F.2d at 984. The statutory scheme does not compel CMA's construction, because section 4(a)(1)(B)(i) contains no reference to "risk" or "injury," or "toxicity," or the like—hypothetical or otherwise —while its alternative, section 4(a)(1)(A)(i), does. Moreover, CMA has not persuaded us that there is, or that Congress believed there to be or assumed, any generally accepted concept of what might be the very highest level of toxicity that any of the chemicals whose toxic properties were unknown might be expected to have under various conditions. In a sense, general uncertainty about chemicals and their effects is a major theme of the statute, e.g., sections 2(b)(1), 4(a)(1)(B)(ii), and the legislative history. See, e.g., H.R.Rep. No. 1341, 94th Cong., 2nd Sess. (1976) at 3 ("It is often many years after exposure to a harmful chemical before the effects of its harm become visible.").

Nor is it necessary to adopt CMA's position in this respect in order to prevent the obviously improper result of having the requirement for testing under section 4(a)(1)(B) wholly divorced from Congress' concerns with health and environmental injury. A finding under section 4(a)(1)(B)(i) does not alone justify a testing order. It must also be found under section 4(a)(1)(B)(ii) that "there are insufficient data and experience upon which the effects of the manufacture [or processing, etc.] ... of such substance ... on health or the environment can reasonably be ... predicted." Similarly, section 4(a)(1)(B)(iii) imposes the further requirement that the testing be "necessary to develop such data" (emphasis added) — i.e., necessary to render the experience and data sufficient as a basis on which the health and environmental effects of the manufacturing (or processing, etc.) can reasonably be predicted. In other words, if existing data and/or experience allows the EPA to reasonably predict no health or environmental injury from the manufacturing (or processing, etc.) of the chemical, then testing may not be ordered under section 4(a)(1)(B). If the EPA properly concludes that the existing data and experience do not suffice as a basis for it to reasonably predict that there will be no health or environmental injury from the manufacturing (or processing, etc.) of the chemical, then affirmative evidence and findings of risk of such injury at hypothetical toxicity levels under section 4(a)(1)(B)(i) are not necessary to provide a nexus between requiring testing under section 4(a)(1)(B) and congressional concern for health and the environment. As noted, CMA has not challenged the EPA's findings under section 4(a)(1)(B) paragraph (ii) or paragraph (iii).

CMA further urges that in order to make a proper finding under section 4(a)(1)(B)(i)(I) of "enter the environment in substantial quantities," there must be a determination respecting more than mere "entry" of such quantities into the environment. CMA urges that there must also be a determination concerning some degree of persistence in the environment of those substantial quantities. To a large extent, this argument is merely a variation of that discussed above, and for generally the same reasons we also hold that the EPA is not obliged to adopt this construction. Again, "enter" is not defined in the statute or the legislative history, and we are unable to say that in a dictionary sense it unambiguously includes a "persist" element. CMA does point to legislative history indicating that Congress may have contemplated that "exposure" was as relevant under clause (I) of paragraph (B)(i) as under clause (II), although clause (II) mentions "exposure" while clause (I) does not and instead uses the term "enter."15 However, this legislative history does not conclusively establish CMA's contention. The statute does use different words, "enter" and "exposure." Moreover, it is by no means clear that even if "enter" as used in clause (I) embraces a concept of "exposure," that it necessarily also embraces a durational or persistence requirement. It is not unreasonable to conclude that Congress assumed that if a substance did "enter the environment in substantial quantities" then that could be equated to, or referred to as, "substantial environmental exposure" without more.16

With respect to clause (II) of section 4(a)(1)(B)(i), CMA challenges the finding that there is "the potential for ... substantial human exposure to cumene" from its manufacturing and processing, on the grounds that there is no showing that the level of such exposure, to which the general population in the vicinity of the cumene facilities is subjected, is other than very low. CMA also urges in this connection that the number of cumene workers exposed is low and their exposure, while clearly higher than that of the general population, is not so high as to be potentially dangerous. To some extent this is a factual challenge, which we have at least partially rejected. And to some extent it also partakes of the previously noted and rejected legal argument that "substantial" must be measured against the risk posed by some hypothetically assumed level of toxicity. However, CMA's argument does appear to contain still another element, namely, the more generalized assertion that whether "human exposure" is "substantial" must depend on something more than simply the number of persons exposed no matter how brief the exposure or small the quantity to which exposed or what the circumstances are in which the exposure occurs. In response, the EPA's brief in this Court states that the EPA "has interpreted `significant' to refer to the quantity of a chemical to which living organisms or the environment are or may be exposed, and `substantial' to refer to the number of people exposed." This response of the EPA is troubling in at least two respects. In the first place, clause (II) does not refer to "the environment," while in clause (I), which does, the term "significant" is not used. How then can "significant," which appears only in clause (II), refer to "the environment," which appears only in clause (I)? In the second place, we find nothing in the final (or the proposed) rule here indicating that the EPA did in fact apply these definitions in this case; nor are we cited to any prior EPA interpretations where these definitions have been adopted. We do not now pass on whether the EPA might properly interpret clause (II) so that "significant" refers only to the quantity of the chemical to which humans are exposed and "substantial" refers only to the number of persons exposed to the chemical,17 because it is by no means clear that the EPA has so interpreted clause (II) (or that its decision in this case is based on such an interpretation).18 As we stated in United States v. Garner, 767 F.2d 104, 116-17 (5th Cir. 1985), "`[i]t is well established that an agency's action must be upheld, if at all, on the basis articulated by the agency itself.' ... Post hoc explanations—especially those offered by appellate counsel — are simply an inadequate basis for the exercise of substantive review of an administrative decision."

Lack of an administrative standard

Finally, CMA contends that the EPA has not articulated any understandable basis—either in the form of a general definition of or a set of criteria respecting the statutory term "substantial" or in its analysis of the specific evidence respecting cumene—for its ultimate determinations that the quantities of cumene which enter the environment from the facilities in question are "substantial" and that the potentially resulting human exposure to cumene is "substantial." We conclude that this argument is well taken, and that a remand to the EPA is necessary so that we may understand the basis on which its clauses (I) and (II) "substantial" findings were made.19

Neither the proposed rule nor the final rule sheds any very helpful light on how the EPA defines "substantial" for purposes of clauses (I) and (II) of section 4(a)(1)(B)(i) or what criteria it considers relevant to determining what is or is not "substantial" for such purposes. The proposed rule states that:

While the final rule does not repeat this language, neither does it set forth any alternate formulation or expressly modify this aspect of the proposed rule.

Both the proposed rule, id., and the final rule, 53 Fed.Reg. 28,196, expressly reference the EPA's first and second proposed test rules at, respectively, 45 Fed.Reg. 48,510 (July 18, 1980) and 46 Fed.Reg. 30,300 (June 5, 1981). The first proposed test rule is mainly concerned with section 4(a)(1)(A), although it does contain some references to section 4 testing generally and to section 4(a)(1)(B).20 The second proposed test rule, promulgated June 5, 1981, is more directly concerned with the applicable standards under section 4(a)(1)(B). The second proposed test rule provides in this connection:

What can be gleaned from all this is indeed limited. We can ascertain that the EPA's focus in clauses (I) and (II) of section 4(a)(1)(B)(i) is on quantity, and, at least avowedly, only on quantity. It also seems clear in this connection that for purposes of clause (I), the quantity being considered is that of the chemical which is released to (or enters) the environment. As to clause (II), however, the quantity focused on is unclear. Is it the quantity of the chemical to which some humans are exposed or of the humans exposed to it, or some combination of these? As previously noted, on this appeal, the EPA urges that in clause (II) the relevant quantity is that of the chemical to which any number of persons are exposed for purposes of a "significant" finding, and is the number of persons exposed to any of the chemical for purposes of a "substantial" finding. But, as we have seen, this construction is not referenced in the proposed or final rule here or in earlier EPA pronouncements; nor is it unambiguously required by the statute or legislative history. See notes 17 & 18, supra, and accompanying text. Such a post hoc explanation does not suffice to fill a void in articulated standards. Garner. Moreover, even if we know what the quantity is of—as we do in clause (I), where it is quantity of the chemical released into the environment—we have no way of ascertaining on what basis the EPA determines whether or not a given quantity is "substantial." All that we are told about that is the statement in the June 5, 1981 second proposed test rule that the EPA viewed "strict numerical definitions of substantial" as "neither feasible nor desirable at this stage," and intended "to make judgments on these factors on a case-by-case basis, at least until some additional experience is gained." In the many years since that statement, no further standards have been publicly articulated.

Under the statutory wording, it is clear that mere entry into the environment of some quantities of the chemical does not suffice for clause (I), rather the quantities entering must be "substantial"; likewise, for clause (II), just any human exposure to the chemical does not suffice, for that human exposure must be "substantial" (or "significant," but here there is no "significant" finding). A focus on quantity, without more, is hence of little help in understanding what is meant by "substantial."

We recognize that "substantial" is an inherently imprecise word. We are also aware that in this context no definition or group of criteria can be established which will function like a mathematical formula, so that for every given set of facts a specific, predictable answer will always be forthcoming. Room must be left for the exercise of judgment. As we have indicated, the EPA has considerable latitude in defining and interpreting "substantial" as it is used in clauses (I) and (II) of section 4(a)(1)(B)(i). See Chemical Manufacturers Association, 859 F.2d at 984. Moreover, case-by-case statutory interpretation by an administrative agency is not impermissible. See, e.g., INS v. Cardoza Fonseca, 480 U.S. 421, 107 S.Ct. 1207, 1221-22, 94 L.Ed.2d 434 (1987); SEC v. Chenery Corp., 332 U.S. 194, 67 S.Ct. 1575, 1580, 91 L.Ed. 1995 (1947). However, "an agency must cogently explain why it has exercised its discretion in a given manner" and "must offer a `rational connection between the facts found and the choice made.'" Motor Vehicle Manufacturers Ass'n v. State Farm Mutual, 463 U.S. 29, 103 S.Ct. 2856, 2869, 2871, 77 L.Ed.2d 443 (1983). And, "an agency's action must be upheld, if at all, on the basis articulated by the agency itself." Id. 103 S.Ct. at 2870. As we stated in Texas Power & Light v. Federal Communications Commission, 784 F.2d 1265, 1269 (5th Cir.1986), "[t]he agency must articulate its findings and the reasons for its policy choices, so that the court may ascertain whether it engaged in balanced, informed decision making." (Footnote omitted.) As observed in a frequently cited passage, "when the Secretary is obliged to make policy judgments where no factual certainties exist or where facts alone do not provide the answer, he should so state and go on to identify the considerations he found persuasive." Industrial Union Department, AFL-CIO v. Hodgson, 499 F.2d 467, 476 (D.C.Cir.1974).21 This requirement has been described as "a necessary minimum" upon which courts reviewing agency actions must "insist." National Association of Regulatory Utility Commissioners v. F.C.C., 737 F.2d 1095, 1140 (D.C.Cir. 1984), cert. denied, 469 U.S. 1227, 105 S.Ct. 1224, 84 L.Ed.2d 364 (1985).

Here, we are unable to conclude from the final rule itself, or from the administrative record, or prior EPA decisions, on what basis or in light of what criteria the EPA concluded either that the quantities of cumene found to enter the environment from the facilities in question were "substantial" or that the human exposure potentially resulting therefrom was "substantial." As to the former, no standards or criteria whatever have been articulated, either for application generally or to this particular case. As to the latter, appellate counsel has offered a rationale—the number of people exposed at any level to any cumene from the facilities in question—but it is not one the agency itself articulated, and in some respects appears inconsistent with the approach taken by the agency. See note 18, supra. TSCA section 19 casts upon us the duty of determining whether on the record as a whole the final rule is supported by substantial evidence. This requires us to determine whether adequate evidence supports the findings that "substantial" quantities of cumene enter the environment from the facilities and whether the potentially resulting human exposure to cumene is "substantial." We cannot do this except as against one or more standards or sets of criteria for determining what constitutes the referenced chemical quantities or human exposure as being "substantial." For us to initially develop such standards or criteria is necessarily to violate both the rule that such decisions are to be made in the first instance by the administrative agency and also the principle that agency action must be sustained, if at all, on the basis articulated by the agency itself.

Conclusion

We accordingly remand the case to the EPA to articulate the standards or criteria on the basis of which it found the quantities of cumene entering the environment from the facilities in question to be "substantial" and the human exposure potentially resulting to be "substantial."22 We note in this connection that merely because we have held that the EPA is not necessarily obliged to adopt or take into account a specific criterion (such as, for example only, persistence after entry) does not imply that the EPA may not properly elect to consider such criterion. Because the first tier tests are now doubtless completed, and it does not clearly appear that there are no conceivably appropriate standards or criteria under which the EPA could properly require cumene testing, we decline to now stay the EPA's final rule. Cf. Chemical Manufacturers Association v. U.S.E.P.A., 870 F.2d 177, 266 (5th Cir.), on rehearing, 885 F.2d 253 (5th Cir.1989).

CMA has also petitioned for remand to the EPA for consideration of more recent studies, which it contends were not previously reasonably available, concerning the levels of cumene in the ambient air in the vicinity of cumene facilities near Houston.23 The extent to which this information may be material may significantly depend on the criteria articulated or developed by the EPA on remand. We direct that the EPA on remand afford CMA an opportunity to present such studies (and any others the EPA deems appropriate) unless they would not be material to any of the EPA's criteria relied on for the testing. Cf. Shell Chemical Co. v. E.P.A., 826 F.2d 295 (5th Cir. 1987).

The case is therefore remanded to the EPA for further proceedings not inconsistent herewith.

REMANDED.