Bethlehem Steel Corp. v. Gorsuch

No. 82-2884 (7th Cir. August 22, 1984)

The court rules that the Environmental Protection Agency (EPA) erred in partially approving a 1974 Indiana state implementation plan (SIP) revision with the effect of making the SIP provision significantly more stringent than the state intended. Initially the court rules that the legality of EPA's 1975 partial approval is properly before it. Petitioner timely challenged EPA's disapproval of a 1979 state-delayed compliance order based on the 1974 SIP. EPA actions on such orders, the court rules, are subject to the time limitations on review established by § 307(b)(1) of the Clean Air Act. The statute is silent on whether the requirement of prompt challenge to final agency action applies where a timely challenge is made to one action — the 1979 EPA disapproval — whose legality depends on the legality of an earlier action — the 1975 partial approval — when the legal effect of the earlier order was not made clear until EPA issued the more recent order. Under the circumstances, the court concludes, it makes more sense to analyze the issue as one of laches.It is unnecessary to consider the scope or applicability of that doctrine here, since a 1980 remand on this issue by the court implies that petitioner was justified in not challenging the 1975 EPA order at the time it was issued.

Turning to the merits, the court rules that EPA's 1975 partial approval of the Indiana SIP revision was unlawful as it applied to noncombustion sources. EPA approved only a 40 percent opacity standard, disapproving a 15-minute daily exception. The court emphasized that it is not questioning EPA's authority to approve only selected provisions in a SIP revision. It further reaffirms an earlier decision upholding the partial approval at issue in this case as it applied to combustion sources. In that context the partial approval had the effect of precluding a weakening of the state's pre-existing rule, which unquestionably applied to combustion sources. In the present case, however, the partial approval had the effect, not of reinstating the prior state rule, but of imposing a new, more stringent requirement on noncombustion sources.It was questionable whether the pre-1975 rule applied to noncombustion sources at all, and if it did the impact was relatively limited. The 1974 rule, as partially approved by EPA, applied to noncombustion sources as the state intended, but deprived them of the daily exception, which appeared necessary to make the rule economically feasible for petitioner. This type of partial approval violates the procedures specified in the Clean Air Act. In § 110(a)(3)(A) the Act requires EPA to disapprove an inadequate SIP provision and promulgate a substitute provision, actions that must be subject to specified review procedures. EPA's partial approval was, in effect, promulgation through a shortcut procedure of a substitute provision never contemplated by the state. As such, the court rules, it must be overturned, since the erroneously promulgated rule has the potential to require cessation of operations at petitioner's noncombustion sources and the error thus is not harmless.

Counsel are listed at 14 ELR 20295.

Before CUDAHY and POSNER, Circuit Judges, and WILKINS, Senior District Judge.^*

[14 ELR 20741]

POSNER, Circuit Judge.

The Bethlehem Steel Corporation petitions us to set aside an order of the Environmental Protection Agency made under the authority of the Clean Air Act, as amended, 42 U.S.C. §§ 7401 et seq. The petition and the EPA's response raise several difficult issues: the scope of a 1972 air pollution regulation issued by the State of Indiana; the EPA's power under the Clean Air Act to alter, by partially approving, state air pollution regulations; and the timeliness of Bethlehem's challenge to the partial approval.

In 1972 the State of Indiana, pursuant to section 110(a) of the Clean Air Act, as amended, 42 U.S.C. § 7410(a), submitted to the EPA a plan for implementing the EPA's air quality standards. This "state implementation plan" (SIP) included an air pollution control regulation (originally promulgated in 1968) that we shall call 1972 APC-3. Although the precise application of this regulation to noncombustion emissions from Bethlehem's Burns Harbor Works on the south shore of Lake Michigan was unclear, the state in 1974 submitted a revised plan which included a revision of 1972 APC-3 that clearly did apply to those emissions. As part of its review of the revised plan, the EPA in 1975 issued an order that (we now know) was intended as a partial approval of 1974 APC-3. 40 Fed. Reg. 50032, 50033 (Oct. 28, 1975). The state then issued an order (a "delayed compliance order" or DCO) setting a timetable for Bethlehem to comply with APC-3. The EPA disapproved this compliance order in 1979, and Bethlehem petitioned us to set aside the order of disapproval. We remanded the case, however, because the basis of the order was unclear. Bethlehem Steel Corp. v. EPA, 638 F.2d 994, 1007 (7th Cir. 1980). On remand the EPA reinstated the order, making clear that the basis of the order was that the state's compliance order had not directed Bethlehem to comply with 1974 APC-3 as it had been modified by the EPA in 1975. 47 Fed. Reg. 43377, 43378 (Oct. 1, 1982).

Bethlehem came back to us, asking that we set aside the order of disapproval on the ground that the EPA-modified version of 1974 APC-3 was invalid and therefore the state could not be made to make Bethlehem comply with it. It was invalid, Bethlehem argued, because the EPA, bypassing statutory procedures for the revision of a state implementation plan, had improperly used the technique of "partial approval" to make 1974 APC-3 a stricter regulation than the state had intended either in the state's own version of 1974 APC-3 or in the previous regulation, 1972 APC-3. The EPA disagreed that it had made the state's regulation stricter, alternatively claimed the power to make it stricter by partial approval, and argued that in any event Bethlehem should have challenged the partial approval back in 1975. By a split decision we affirmed the EPA's order earlier this year. Bethlehem filed a petition for rehearing with suggestion of rehearing en banc, and though the suggestion was turned down the panel unanimously decided to reconsider the case. We vacated our decision, asked the parties to submit supplemental briefs directed to specific questions put by the panel, see Bethlehem Steel Corp. v. Gorsuch, 732 F.2d 97 (7th Cir. 1984) (en banc) (per curiam), and now proceed to decide the case anew.

The same Indiana implementation plan that contained 1972 APC-3 also contained 1972 APC-5, which placed a ceiling on the amount of particulates emitted by the coking operations at the Burns Harbor Works. Coke is made by heating coal in oxygen-free ovens which give off smoke and dust ("particulate emissions") that are vented through stacks. These emissions are noncombustion emissions, because no combustion occurs in the ovens.The ovens are heated by combustion, however, and the smoke produced by this combustion is vented through other stacks.1972 APC-5 clearly covered both combustion and noncombustion emissions.

Indiana's APC-3 (of whatever vintage) regulates opacity, a visual measure of pollution, rather than particulate level, a chemical measure. A tiny pencil of smoke, harmless to health but impossible to see through, would be 100 percent opaque. The parties describe opacity limitations as proxies for limitations on particulate levels. Cf. National Lime Ass'n v. EPA, 627 F.2d 416, 446-51 (D.C. Cir. 1980); Portland Cement Ass'n v. Ruckelshaus, 486 F.2d 375, 400-01 (D.C. Cir. 1973), after remand, Portland Cement Ass'n v. Train, 513 F.2d 506 (D.C. Cir. 1975). Particulate levels are more difficult to measure than opacity — which, literally, is measured by looking at the plumes of smoke (the people who do this are called "plume observers"). Levels of particulates in noncombustion omissions are especially difficult to measure because such emissions are highly irregular.

1972 APC-3 placed a 40 percent opacity limitation on emissions from the Burns Harbor Works. But there was a question of coverage. One sentence in the regulation stated: "No person shall operate any combustion installation so as to produce, cause, suffer or allow smoke to be emitted, the appearance, density or shade of which is darker than No. 2 of the Ringelmann Chart." (Although the Ringelmann Chart measures darkness rather than opacity, No. 2 on the Chart is a commonly used proxy for 40 percent opacity.) If this were all there was to the regulation it could have no application to Bethlehem's noncombustion installations, that is, to the ovens themselves as distinct from the boilers that heat them. But another sentence states: "The opacity of any color equivalent to the Ringelmann Chart may be used as prima-facie evidence in determining process emissions but may be refuted by approved stack emission tests or other evidence acceptable to the [Indiana air pollution control] Board." At the very most, however, this means that noncombustion emissions that flunk the 40 percent opacity test violate the regulation prima facie but that Bethlehem can rebut theprima facie case by showing that the level of particulates in its noncombustion emissions is within the ceiling fixed by APC-5. On this reading, 1972 APC-3 and 1972 APC-5 between them established two independent limitations on combustion emissions — opacity and particulate level — but only one on noncombustion emissions — particulate level; for with respect to noncombustion emissions the opacity limitation could be used only to force Bethlehem to prove that it was complying with APC-5, though admittedly the burden of proof might be difficult to carry because of the difficulty of measuring particulate levels in emissions from noncombustion sources.

Probably the second quoted sentence is even less restrictive than we have suggested. This is not because it omits "noncombustion emissions"; we may assume from the [14 ELR 20742] definition of "process" elsewhere in the state implementation plan that "process emissions" include noncombustion emissions. The important thing is that the sentence does not mention any specific opacity limitation and just says that opacity "may be used" to make out a prima facie case. It seems therefore that 1972 APC-3 imposed no actual restriction on noncombustion emissions, but merely authorized the state to impose a restriction.

The revision of 1972 APC-3 that Indiana submitted to the EPA in 1974 eliminated the distinction between combustion and noncombustion emissions and, without equivocation, made the 40 percent opacity measure a mandatory limitation on both sources; gone was the prima facie case. But 1974 APC-3 allowed the opacity limitation to be exceeded for up to 15 minutes every 24 hours. In the course of reviewing the 1974 revisions of Indiana's implementation plan, the EPA touched briefly on 1974 APC-3 and concluded that "APC-3 must be disapproved to the extent that the 15-minute exemption provision . . . fails to meet the requirements of" the applicable federal air quality standards. The order does not state explicitly that the EPA is approving the rest of 1974 APC-3 but does say (after discussing the other APC's in the revised Indiana plan) that "with the exception of the above-noted disapproval actions and material returned to the State or held in abeyance pending further review, the proposed revisions meet the substantive and procedural requirements of Section 110 of the Clean Air Act . . . and are hereby approved as revisions to the Indiana Implementation Plan, effective immediately." 40 Fed. Reg. at 50033. This order, insofar as it pertains to 1974 APC-3, is the "partial approval" challenged by Bethlehem.

The EPA argues that this challenge comes too late. A proceeding to review an EPA order approving an original or revised state implementation plan (or part thereof) must be filed in the court of appeals within 60 days after the order is issued. See section 307(b)(1) of the Clean Air Act, as amended, 42 U.S.C. § 7607(b)(1). (This provision dates only from 1977, but a similar provision, though allowing only 30 days, was in force in 1975.) The EPA issued its order partially approving 1974 APC-3 in 1975, and it was not till 1979, when the agency disapproved the compliance order issued by the Indiana board, that Bethlehem sought judicial review. But the EPA's order disapproving the state's compliance order was itself reviewable under section 307(b)(1). Although such orders are not among those mentioned by name as orders reviewable under this section, they fall within the section's catch-all provision making reviewable "any other final action" by the EPA under the Clean Air Act. This provision, which the Supreme Court read broadly in Harrison v. PPG Industries, Inc., 446 U.S. 578, 587-92 (1980), appears to cover an EPA order disapproving a delayed compliance order, since the consequence of disapproval is that the polluter must either comply immediately or pay a noncompliance penalty. See 42 U.S.C. §§ 7420(a)(2)(A), (B)(iv); Currie, Air Pollution: Federal Law and Analysis § 8.07, and § 9.06 at p. 9-19 (1981). Bethlehem made timely application for review of the EPA's 1979 order, and the validity of the order depends on the legality of the EPA's version of 1974 APC-3. The EPA disapproved the state's compliance order because the state was ordering compliance with the wrong regulation — the state's version of 1974 APC-3 rather than, as Bethlehem thought, the previous regulation, but in any event not the EPA's version. This ground of disapproval is invalid if the EPA's version of 1974 APC-3 is unlawful.

The Act is explicit that any action by the EPA which could be reviewed in the court of appeals under section 307(b)(1) cannot later be reviewed in civil or criminal enforcement proceedings. 42 U.S.C. § 7607(b)(2). But this does not tell us how far back a polluter who is properly in the court of appeals to challenge a reviewable order of the EPA can reach to question the validity of prior orders on which the validity of the order under review depends. As the statute is silent on this question it seems to us that the better argument for the EPA is not that Bethlehem's challenge to the 1975 order is barred by the 60-day (at the time 30-day) statute of limitations in section 307(b)(1) but that it is barred by the equitable doctrine of laches, a doctrine nowadays applied well beyond the limits of the original equity conception. Compare Smith v. Clay, 3 Brown Ch. 646, 29 Eng. Rep. 743 (1767), with United States v. Darnell, 716 F.2d 479-80 (7th Cir. 1983) (per curiam). It can be argued, however, that the doctrine of laches should have no application to a case such as this. Although a polluter can challenge an original or revised state implementation plan as soon as the EPA approves it, provided there is a reasonable likelihood that the plan will impose costs on him, see, e.g., Indiana & Michigan Elec. Co. v. EPA, 733 F.2d 489, 490-91 (7th Cir. 1984); Currie, supra, §§ 9.07-9.09, it does not follow that he must challenge it then.The polluter may be content to comply if given enough time to do so and if the means of enforcement chosen by the state are not too onerous. It is significant in this connection that "the EPA may not deny approval of a DCO because EPA prefers that the state use different means of enforcement." Bethlehem Steel Corp. v. EPA, 651 F.2d 861, 871 (3d Cir. 1981) (citing our first decision in the present case, Bethlehem Steel Corp. v. EPA, 638 F.2d 994, 1006 (7th Cir. 1980)). Maybe it was only when the EPA disapproved the compliance timetable established by Indiana that Bethlehem concluded it could not live with 1974 APC-3 as rewritten by the EPA. We would not condone waiting to challenge a regulation till a magistrate refused to quash inspection warrants obtained by the EPA to determine compliance with the regulation, as in Public Service Co. v. EPA, 682 F.2d 626, 636 (7th Cir. 1982). That kind of belated challenge is stonewalling; but perhaps waiting to see whether regulation will really pinch, before challenging it in court, is not.

The counterargument is that a company should not be allowed to wait many years after a regulation clearly applicable to it is promulgated before challenging the lawfulness of the regulation. To allow that would be to invite companies to drag their heels in complying with regulations. The well-known delays in compliance with the Clean Air Act make us reluctant to extend such an invitation. But we find it unnecessary in this case to determine the scope and applicability of the doctrine of laches in proceedings to review orders under the Clean Air Act. We also find it unnecessary to decide as an original matter whether Bethlehem knew in 1975 that the result of the EPA's order had been to put 1974 APC-3 into effect minus the 15-minute blow-off period. If it did not know this, and was reasonable in not knowing, it had no actual controversy with the EPA at that time, and could not be faulted for not suing then even if air pollution regulations must always be challenged at the earliest possible opportunity. See Bethlehem Steel Corp. v. EPA, 723 F.2d 1303, 1306 (7th Cir. 1983); Bethlehem Steel Corp. v. EPA, 536 F.2d 156 (7th Cir. 1976). Bethlehem argues that until the EPA disapproved Indiana's compliance order, it had thought that the legal effect of the agency's action in disapproving the 15-minute exemption was to disapprove 1974 APC-3 in its entirety rather than to approve it partially, and that therefore when Indiana ordered Bethlehem to comply with APC-3 it meant comply with the previous regulation, 1972 APC-3.

The last time this case was here, the court found, consistently with Bethlehem's argument, "that even if we were to agree that the Administrator [of the EPA] indeed has the authority to approve revisions partially, the record utterly fails to support that he in fact did so in this case." 638 F.2d at 1007. This is not quite so surprising a finding as it may seem to be in light of the parties' and our description of this case as a case about partial [14 ELR 20743] approval. The words "partial approval" do not appear in the EPA's 1975 order. The order states that 1974 APC-3 is "disapproved to the extent . . ." but does not state explicitly, through in retrospect can be read to have implied (and was so read in Public Service Co. v. EPA, supra, 682 F.2d at 635-36), that the rest of 1974 APC-3 — that is, the extension of the 40 percent opacity limitation to noncombustion sources, but with no 15-minute blow-off period — was approved and put into effect. If Bethlehem's 1979 challenge to the 1975 order was untimely because Bethlehem should have realized sooner than it did that the EPA had indeed put all but the exception into effect, this court in 1980 would not have remanded; it would have dismissed the review proceeding as untimely. Evidently, the court believed — on grounds that seem at least plausible but that in any event we are disinclined to reexamine four years later — that Bethlehem was justified in not challenging the EPA's order altering 1974 APC-3 when the order was first issued. The merits of Bethlehem's challenge to the 1975 order are therefore properly before us.

The Clean Air Act authorizes the EPA to approve a state implementation plan "or any portion thereof." 42 U.S.C. § 7410(a)(2). Although the section that authorizes the EPA to approve state-proposed revisions in the plan (42 U.S.C. § 7410(a)(3)(A)) does not contain the "any portion thereof" language, we held in Public Service Co. v. EPA, supra, 682 F.2d at 631-34 — and reaffirm today — that the EPA has the power to partially approve revised as well as original plans submitted by the state. State implementation plans are revised frequently; and original or revised, they are long documents containing many different regulations. Indiana's current plan is 67 pages long. See Atlantic Environmental Associates, Inc., Regulations and Non-Regulatory Revisions to State Implementation Plan: Indiana, EPA 450/2-81-034, July 1981. It is only common sense that the EPA should be allowed to approve some regulations and disapprove others, rather than have to approve or disapprove the whole package.

Public Service Co. upheld the EPA's approval, minus the 15-minute exception, of the very 1974 APC-3 that is before us in this case. 682 F.2d at 636. But only combustion installations (coal-fired electric generating plants) were involved in that case. As to them the only effect of the state's proposed revision was to create a 15-minute safety valve, because the previous regulation had subjected these installations to the same mandatory 40 percent opacity limitation — just with no safety valve. The EPA must approve a proposed revision in the state's plan if the revision complies with the Clean Air Act, see 42 U.S.C. § 7410(a)(3)(A); so it must also be authorized to disapprove a revision that does not comply. That is what it did, insofar as combustion installations were concerned, when it disapproved the 15-minute provision in 1974 APC-3; it disapproved the state's attempt to relax its regulation of those installations. As to them the EPA's "partial approval" of 1974 APC-3 was actually an explicit disapproval; but whether it is called, it was snugly within the EPA's power.

But with regard to noncombustion emissions, 1974 APC-3, at least when stripped of its 15-minute blow-off provision, stiffened the preexisting regulation. It made the 40 percent opacity limitation mandatory for such sources for the first time. Before then as we have seen the limitation only put the source to its proof of compliance with APC-5, and maybe didn't even do that. Now the source would have to comply with the limitation even if its emissions did not exceed the particulate levels specified in APC-5. Bethlehem contends that its coke ovens cannot comply with a 40 percent opacity limitation 24 hours a day. If this is right they would have to be shut down, even though the history of APC-3 suggests that it was intended to be just a back-up for APC-5. It was to avert a shut down (Bethlehem contends) that the state included the 15-minute blow-off period in the proposed 1974 APC-3.

With the merits of these contentions (on which see generally United Elec. Co. v. EPA, 427 U.S. 246, 256-69 (1976), and the lime and cement association cases cited earlier) we have no concern. The only thing that matters to our review is that 1974 APC-3, as modified by the EPA to delete the 15-minute provision, made Indiana's regulation of noncombustion sources more stringent than it had ever been. How much more stringent is uncertain. Maybe Bethlehem would not have been able to prove compliance with APC-5 if by exceeding the limits in APC-3 it had triggered the duty to prove compliance. This is speculation, though; what is clear is that by revising 1974 APC-3 as it did, the EPA converted 40 percent opacity from (at most) a trigger to a ceiling, thus making the state's control of noncombustion emissions more stringent on its face than the state had ever intended it to be. Moreover, it is not clear that 1972 APC-3 imposed even a prima facie opacity limitation on noncombustion emissions; there may have been no limitation. In these circumstances we think it was incumbent on the EPA to show that the apparent increase in stringency brought about by the replacement of 1972 APC-3 by the EPA's edited version of 1974 APC-3 was not real. It has had ample opportunity to make this showing but has not done so. For purposes of this decision we must assume therefore that the EPA's action brought about a real, and not merely theoretical — a significant and not merely inconsequential — increase in the strictness of Indiana's restrictions on the opacity of noncombustion emissions.

The EPA can, as we have said, disapprove an attempt by the state to weaken its existing plan, as was the situation in the Public Service case with respect to combustion sources. And it can approve some proposed regulations in state's original or revised implementation plan but not others — APC-5 but not APC-3, for example. But can it, by the device of what it calls partial approval, make a particular regulation in the state's plan substantially tougher? This is a question of first impression, left open in Indiana & Michigan Elec. Co. v. EPA, supra, 733 F.2d at 492. It is a pure question of law. The issue is not whether the agency acted reasonably or unreasonably in acting as it did, but whether it followed the procedures that the statute requires it to follow when it wants to make a state implementation plan tougher.

The key to resolving this issue is to understand that the Act creates a procedure by which the EPA can disapprove a proposed regulation and promulgate its own, more stringent regulation in its stead. See 42 U.S.C. § 7410(c)(1)(B); Kennecott Copper Corp. v. Train, 526 F.2d 1149, 1151 (9th Cir. 1975). Using this procedure the EPA could have rejected Indiana's 1974 APC-3 and promulgated its own 1974 APC-3 minus the 15-minute blow-off period. But it did not follow this course. And this is not just a matter of the EPA's having been maladroit in failing to describe what it did as the promulgation of a revised state implementation plan rather than as the partial approval of the state's proposed plan. If the EPA had followed the revision route it would have had to promulgate a proposed regulation first and give the state a chance to submit a substitute regulation. See 42 U.S.C. § 7410(c)(1). More might have been required. See Buckeye Power, Inc. v. EPA, 481 F.2d 162, 170-71 (6th Cir. 1973). The EPA might have had to give interested persons an opportunity to submit oral as well as written comments. See Currie, supra, § 4.11 at p. 4-25. And since the revision raised issues that may not have been adequately canvassed in the state hearing that preceded the proposal of 1974 APC-3, another state hearing might also have been required, notwithstanding Indiana & Michigan Elec. Co. v. EPA, 509 F.2d 839, 846-47 (7th Cir. 1975), and Appalachian Power Co. v. EPA, 477 F.2d 495, 502-03 (4th Cir. 1973).It is true that the State of Indiana could, and within a few months of the EPA's partial approval did, propose to revise 1974 APC-3 (as rewritten by the EPA) [14 ELR 20744] to restore a blow-off period for noncombustion emissions. But after almost a decade the EPA has not yet acted finally on any of the state's numerous proposals. The procedural difference between the authorized route and the shortcut actually taken has turned out to be profound. And, of course, the stakes are very great; a sufficiently strict pollution control regulation could shut down the Burns Harbor Works.

The shortcut was unauthorized. Congress did not mean on the one had to create a procedure for the EPA's revising state regulations to make them stricter, and on the other to allow the EPA to skip the procedure merely by naming what it was doing "partial approval." The words themselves are against this interpretation. To partially approve is to give the proposer part of what he wanted. In partially approving the state's proposed 1974 APC-3 the EPA did not give the state half a loaf; it converted the proposal into something completely unpalatable to the state, as shown by the state's later conduct. If you ask a judge for permission not to stand while addressing the court, and the judge gives you permission to stand, what he has done is not aptly described as partial approval of your request.

The distinction between partial approval and revising by deleting is a familiar one in law. A governor who has line-veto power may not extend the scope of a bill submitted to him by using his veto power to eliminate a limitation in the bill. See, e.g., Washington Ass'n of Apartment Ass'ns, Inc. v. Evans, 88 Wash. 2d 563, 564 P.2d 788 (1977). "The power of partial veto . . . is not the power to enact or create new legislation by selective deletions." State ex rel. Sego v. Kirkpatrick, 86 N.M. 359, 365, 524 P.2d 975, 981 (1974); see Thirteenth Guam Legislature v. Bordallo, 430 F. Supp. 405, 410 (D. Guam 1977). No more may a court invalidate a part of a statute on constitutional grounds and leave the rest in effect if the result is to create a law that the legislature would not have enacted. See, e.g., Norman's on the Waterfront, Inc. v. Wheatley, 444 F.2d 1011, 1019 (3d Cir. 1971); Comment, Extension Versus Invalidation of Underinclusive Statutes: A Remedial Alternative, 12 Colum. J.L. & Soc. Probs. 115 (1975). And a court that refuses to enforce a provision in a contract because it is against public policy will not enforce the remainder of the contract if the result will be to give the promisee a substantially better deal than he had bargained for. See Farnsworth, Contracts § 5.8 at p. 361 (1982); Restatement (Second) of Contracts § 185, comment b (1981).

No more can the EPA, in the guise of partial approval, remove words of limitation; it must follow the procedures that the Act prescribes for making state regulation stricter. True, the EPA's power of partial approval is in any event constrained by the substantive criteria in 42 U.S.C. §§ 7410(a)(2)(A)-(K); and admittedly that weakens our analogies. But the broad deference that reviewing courts must give the EPA's application of those criteria, see Chevron, U.S.A., Inc. v. National Resources Defense Council, Inc., 104 S. Ct. 2778, 2793 (1984); Connecticut Fund for Environment, Inc. v. EPA, 696 F.2d 169, 173 (2d Cir. 1982) (the EPA "has considerable discretion in deciding whether to approve a SIP or a SIP revision"), makes it particularly important that the agency follow the correct statutory procedures. Of course as with any other procedural default it is possible that if the EPA had followed the right procedural route it would have arrived at the same substantive destination. But that is not a good reason to excuse its refusal to follow required procedures — especially when it has suggested no reason for that refusal. And although this circuit recognizes a doctrine of harmless error in reviewing administrative decisions, see Illinois v. ICC, 722 F.2d 1341, 1348-49 (7th Cir. 1983), this is not a harmless-error case; the outcome under proper procedures is not foreordained.

Moreover, there is more at stake here than a procedural bobble, however grave. As is evident from the central role that Indiana's implementation plan has played in this protracted proceeding, the Clean Air Act creates a partnership between the states and the federal government. The state proposes, though the EPA disposes. The federal government through the EPA determines the ends — the standards of air quality — but Congress has given the states the initiative and a broad responsibility regarding the means to achieve those ends through state implementation plans and timetables for compliance. See Train v. Natural Resources Defense Council, Inc., 421 U.S. 60, 64-65 (1975); Bethlehem Steel Corp. v. EPA, supra, 723 F.2d at 1308-09. The Clean Air Act is an experiment in federalism, and the EPA may not run roughshod over the procedural prerogatives that the Act has reserved to the states, cf. Sierra Club v. Indiana-Kentucky Elec. Corp., 716 F.2d 1145, 1153-54 (7th Cir. 1983); Detroit Edison Co. v. EPA, 496 F.2d 244 (6th Cir. 1974), especially when, as in this case, the agency is overriding state policy.

But do not exaggerate the scope of our holding. We have reaffirmed that the EPA has the power, when confronted with a group of separate, independent regulations, to approve some and disapprove others, even though the group is contained in a revised rather than an original state implementation plan. We have made clear that even when dealing with a single regulation, the EPA can approve part of it and disapprove the rest, provided the effect is just to prevent the state from weakening its previous regulation. And even if the effect is to strengthen it, the EPA has not exceeded its authority if it can show that the increase in the stringency of regulation is apparent rather than real, or if real is minor. Needless to add, we express no view whatever on whether the EPA would be violating any substantive statutory criteria, or behaving arbitrarily, if, following correct procedures, it imposed a 24-hour 40 percent opacity limitation on noncombustion emissions from the Burns Harbor Works.

REVERSED.

* Hon. Philip C. Wilkins, of the Eastern District of California, sitting by designation.