Full opinion text
Opinion for the Court filed PER CURIAM. PER CURIAM : This case revisits Title IV of the Clean Air Act (“the Act”), which, inter alia, directs the Environmental Protection Agency (“EPA”) to promulgate limits on the emission of nitrogen oxides from various electric utility boilers. In Alabama Power Co. v. EPA, 40 F.3d 450 (D.C.Cir.1994), we invalidated the first set of these emission limits as exceeding EPA’s statutory authority. We are now presented with a challenge by a number of electric utilities and industry groups to the next group of nitrogen oxides emission limits promulgated under the Act: a more stringent revision of the first set of emission limits and a new set of emission limits for a second group of boilers. This time, we uphold the bulk of the challenged rule, concluding that EPA has not exceeded its authority and cognizant of the deference due to an agency dealing with largely scientific and technical matters. We vacate, however, the portion of the final rule that reclassifies certain retrofitted cell burner boilers as wall-fired boilers and remand it to EPA for reconsideration or a more adequate justification. I.Background Among the 1990 amendments to the Clean Air Act, 42 U.S.C. § 7401 et seq. (1994), was Title IV, which was designed to reduce the adverse effects of acid deposition (more commonly known as “acid rain”) from the atmosphere by limiting the allowable emissions of sulfur dioxide (SO2) and nitrogen oxides (NOx ), two principal sources of acidic compounds. See 42 U.S.C. § 7651 (1994) (congressional findings and purposes). Electric utilities such as Appalachian Power contribute to NOx emissions through the burning of coal, which releases nitric oxide (NO) that reacts with elements in the air to form nitrogen dioxide (N02) and other harmful pollutants. In 1990, electric utility emissions constituted approximately 32 percent of total annual NOx emissions. See Acid Rain Program; Nitrogen Oxides Emission Reduction Program, 61 Fed.Reg. 67,112, 67,112 (1996). In order to encourage a reduction in NOx emissions, Title IV directs EPA to set NOx emission limits for two groups of coal-fired electric utility boilers and divides those boilers into two additional groups for the purpose of setting compliance deadlines. A boiler therefore may be a “Group 1 boiler” or a “Group 2 boiler,” depending on its type, and may be a “Phase I boiler” or a “Phase II boiler,” depending on when it is subject to emissions limitations. There are both Group 1 boilers and Group 2 boilers in each of the compliance phases. One method of reducing NOx emissions is to retrofit coal-fired boilers with an emission control device. For Group 1 boilers, such a device commonly consists of what is termed “low NOx burner technology,” which, as we noted in Alabama Power, is an emission control method that limits the amount of oxygen available to react with the nitrogen in the coal and thus reduces the amount of nitrogen oxides formed. Alabama Power, 40 F.3d at 452 n. 3. The emissions from Group 2 boilers are more difficult to reduce, and thus Group 2 boilers are retrofitted with a greater variety of emission controls, including selective catalytic reduction, selective non-catalytic reduction, gas reburning, and plug-in and non-plug-in retrofits. Each control method can achieve varying levels of NOx emissions reduction. As we noted in Alabama Power, Congress intended in enacting Title IV “to tie the obligation of utilities to meet the NOx emission limit to the use of low NOx burners.” Alabama Power, 40 F.3d at 455. To this end, section 407(b)(1) requires that the Group 1 limits be set at a maximum of 0.45 pounds per million British thermal units (“Ib/mmBtu”) for tangentially fired boilers and 0.50 Ib/mmBtu for dry bottom wall-fired boilers, unless EPA finds that these rates cannot be achieved using “low NOx burner technology,” a term not explicitly defined in the statute. The limits, which were to be set by May 15, 1992, would then apply to Group 1, Phase I boilers starting on January 1, 1995. 42 U.S.C. § 7651f(b)(l). EPA was permitted to revise the Group 1 limits by January 1, 1997, to apply to Phase II boilers if it determined that “more effective low NOx burner technology [was] available.” 42 U.S.C. § 7651f(b)(2). If no such revision were undertaken by January 1, 1997, the limits established for the Group 1, Phase I boilers were to go into effect for the Group 1, Phase II boilers. EPA was also required to set by January 1, 1997, the NOx emission limits for Group 2 boilers. These limits were to be based “on the degree of reduction achievable through the retrofit application of the best system of continuous emission reduction, taking into account available technology, costs and energy and environmental impacts; and which is comparable to the costs of nitrogen oxide controls” set for the Group I, Phase I boilers. 42 U.S.C. § 7651f(b)(2). On March 22, 1994, well past its statutory deadline, EPA promulgated the Group 1, Phase I emission limits. See Acid Rain Program; Nitrogen Oxides Emission Reduction Program, 59 Fed.Reg. 13,538 (1994). The final rule defined “low NOx burner technology” to include overfire air, another emission control method, as well as low NOx burners themselves. We invalidated the rule as inconsistent with EPA’s statutory directive, holding that the term “low NOx burner technology” was an “unambiguous reference to low NOx burners” that did not permit EPA to include additional control methods. Alabama Power, 40 F.3d at 455. EPA subsequently revised the 1994 regulations on April 13, 1995, to eliminate references to overfire air and established limits of 0.45 Ib/mmBtu for tangentially fired boilers and 0.50 lb/ mmBtu for wall-fired boilers, limits identical to those provided for in the statute. See 60 Fed.Reg. at 18,763; 42 U.S.C. § 7651f(b)(l). To account for the delay in promulgation, the compliance date for Group 1, Phase I boilers was moved to January 1, 1996. 60 Fed.Reg. at 18,763. The rule at issue here, issued on December 19, 1996, promulgates the next set of emission limits under the statutory scheme: the revised NOx emission limits for Group 1, Phase II boilers as well as the NOx emission limits for Group 2 boilers. 61 Fed.Reg. at 67,112. EPA revised the Group 1 limits after determining, as required by section 407(b)(2), that boilers with low NOx burners were achieving lower emission levels than the limits promulgated in 1996 and therefore that more effective low NOx burner technology was available. (This determination was the result of a regression analysis in which EPA constructed equations capturing the reduction achieved by Group 1, Phase I boilers and applied these equations to the uncontrolled emission rates of Group 1, Phase II boilers.) In establishing the Group 2 emission limits, EPA interpreted its statutory directive to require a comparison of the cost-effectiveness of Group 2 controls and low NOx burner technology and thus promulgated emission limits for Group 2 boilers based on control technologies that were shown to be as cost-effective in reducing NOx emissions as low NOx burner technology. In addition, relying on section 407(a), which states that a boiler becomes an “affected unit” for purposes of the NOx emission limits at the same time it becomes an affected unit for purposes of the SO2 emission limits (established elsewhere in Title IV), EPA set January 1, 2000, as the date by which compliance with the new limits must be achieved. Finally, EPA determined that certain retrofitted cell burner boilers should be reclassified from Group 2 to Group 1, thereby subjecting them to the stricter emission limits applicable to the latter group. Appalachian Power and petitioner Arizona Public Service Company now seek review of these portions of the final rule, claiming that EPA’s actions both exceeded its statutory authority and were arbitrary and capricious. II. Discussion A. The Group 1, Phase II Emission Limits Appalachian Power’s first challenge is to EPA’s statutory authority to revise the Group 1 emission limits, which is bounded by the requirement that the agency must determine that “more effective low NOx burner technology is available.” See 42 U.S.C. § 7651f(b)(2). EPA argues that a plain reading of this language reveals that it may revise the Group 1 emission limits if performance data show that available technology can achieve lower emission limits. It contends that because Congress required the agency to evaluate the capabilities of low NOx burner technology when setting the initial Group 1 limits, Congress must have intended that EPA would reevaluate those capabilities in deciding whether to revise those limits. Appalachian Power argues, however, that “more effective” technology refers to a change in burner design and not merely to a proven increase in the effectiveness of existing burners. Because EPA has not provided evidence that “more effective” burners than those justifying the April 1995 limits exist, Appalachian Power argues, the revised rates must be vacated as contrary to the statute. Because resolution of this question turns on the interpretation of the statutory phrase “more effective low NOx burner technology,” the two-step Chevron framework governs our analysis. Under Chevron, we must first consider, having studied the statutory text and the legislative history, “whether Congress has directly spoken to the precise question at issue. If the intent of Congress is clear, that is the end of the matter; for the court, as well as the agency, must give effect to the unambiguously expressed intent of Congress.” Chevron U.S.A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 842-43, 104 S.Ct. 2778, 2781, 81 L.Ed.2d 694 (1984); see also Ohio v. United States Dep’t of the Interior, 880 F.2d 432, 441 (D.C.Cir.1989). If Congress has not directly addressed the precise question at issue, “the question for the court is whether the agency’s answer is based on a permissible construction of the statute,” Chevron, 467 U.S. at 843, 104 S.Ct. at 2781 — -that is, whether it is “reasonable in light of the Act’s text, legislative history, and purpose.” Southern Calif. Edison Co. v. FERC, 116 F.3d 507, 511 (D.C.Cir.1997). Chevron’s first step does not take us very far. There is nothing in the statutory section at issue or in the legislative history to suggest what Congress meant by “more effective low NOx burner technology” or, more specifically, whether the phrase can encompass varied applications of existing controls as well as new controls. The word “technology,” for example, is a chameleon word, capable of many meanings, but here even its context does not help us to identify a precise meaning. Elsewhere in the Act, “technology” is defined as comprising “methods, systems, and techniques,” which lends support to EPA’s interpretation, see, e.g., 42 U.S.C. § 7479(3) (1994) (defining “best available control technology” as an emission limitation achievable through “application of production processes and available methods, systems, and techniques”), but this single reference cannot suffice to determine Congress’s intent with respect to low NOx burners. The legislative history also provides few clues as to the breadth of the phrase. Although the 1990 amendments to the Act were ultimately the product of a conference committee, the language in section 407(b)(2) was derived largely from the Senate bill. During debate on the bill, several senators referred to “low NOx burner technology” as the standard by which other emission control methods would be judged, but none of these senators elaborated on what the phrase included. See, e.g., 136 Cong. Reo. 35,627 (1990) (statement of Sen. Cochran) (“[Ujtilities will be allowed to comply with nitrogen oxide reduction requirements through the application of low NOx burner technology. This technology is a reasonable, cost-effective method which has proven to be successful in achieving significant NOx reductions.”); 136 Cong. Reo. 5046 (1990) (statement of Sen. Lott) (emission rates are based on “the application of low-NOx burner technology, a much more reasonable and cost-effective method proven to successfully achieve significant NOx reductions”). Similarly, the conference report accompanying the final version of the bill notes simply that “[t]he NOx reductions from existing units mandated under section 407 are to be accomplished by use of conventional, available burner technology (‘low-NOx’ burners),” H.R. Conf. Rep. No. 101-952, at 344 (1990), a declaration that does not aid us in divining whether the phrase “more effective low NOx burner technology” implies a change in burner. Because our search for Congress’s intent has been less than fruitful, we go on to decide, under the second step of the Chevron analysis, whether EPA’s interpretation of the phrase “more effective low NOx burner technology” to encompass improved performance of existing burners is reasonable. We believe that it is. As we have noted, it is evident that Congress intended that low NOx burner technology serve as a benchmark for emission limits promulgated under Title IV, so to the extent that the rule derived from EPA’s interpretation does not require technology beyond low NOx burners, that rule would not be inconsistent with the intent of the statute. EPA’s interpretation is also consistent with the concern with achievability that motivates the section. See, e.g., 42 U.S.C. § 7651f(b)(l) (EPA may set higher emission limits than those provided by statute if statutory rates are not achievable); id. § 7651f(b)(2) (Group 2 emission limits must be based on achievable degree of reduction). In other words, it is a fair interpretation to read Congress’s directive that more effective technology be available as authorizing more stringent limits only if those limits are achievable in practice. Moreover, the fact that Congress did not simply require a determination that “more effective low NOx burners” existed suggests that “more effective low NOx burner technology” refers to something beyond the burner hardware itself. If this were not the case, we would expect to find some evidence to the contrary in the legislative history; we find none, and Appalachian Power is unable to point us toward any. We addressed a similar issue with respect to the Clean Air Act in International Harvester Co. v. Ruckelshaus, 478 F.2d 615 (D.C.Cir.1973). At issue was a provision of the Act that authorized a one-year exemption from an emission-level requirement if, inter alia, “the applicant has established that effective control technology, processes, operating methods, or other alternatives are not available.” Id. at 624 (quoting 42 U.S.C. § 1857f-l(b)(5)(D)(iii) (1970)). We rejected the petitioners’ argument that EPA’s determination of whether technology was “available” must be based solely on “technology in being as of the time of the application.” Rather, we held that EPA was justified in determining what technology would be considered “available” based on predicted improvements in existing technology, “subject to the restraints of reasonableness.” Id. at 628-29. See also Portland Cement Ass’n v. Ruckelshaus, 486 F.2d 375, 391-92 (D.C.Cir. 1973) (in determining “achievable” emission limits, EPA may make reasonable projection based on existing technology). We can find no significant difference between a determination that “available” technology includes predicted improvements in existing technology and a determination that “more effective” technology includes actual improved performance in existing technology. We thus think it reasonable, as a preliminary matter, for EPA to find that “more effective low NOx burner technology” exists if improved performance for already-existing burners can be shown. Appalachian Power next argues that even if “more effective low NOx burner technology” is given the meaning we approve today, EPA has failed to show that boiler performance has improved. EPA asserts that its regression analysis shows that boilers retrofitted with low NOx burners can achieve lower emission levels than the limits deemed adequate by the 1995 rule. This improvement in performance, EPA contends, may be attributable to a number of improvements in the effectiveness of the technology surrounding low NOx burners, including the accumulated experience of boiler operators, improved operating practices, more advanced burner tuning and management software, and improved automation. Appalachian Power challenges this conclusion, arguing that the revised limits set by the rule require technology beyond the capability of low NOx burners. Because Appalachian Power’s challenge, although framed in the most general of terms, is at root a challenge to EPA’s analytical model, we must consider whether the use of that model was arbitrary and capricious. See 42 U.S.C. § 7607(d)(9) (1994) (authorizing reversal of actions under the Act found to be “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law”). Our analysis is guided by the deference traditionally given to agency expertise, particularly when dealing with a statutory scheme as unwieldy and science-driven as the Clean Air Act. As we have previously noted, so long as EPA “acted within its delegated statutory authority, considered all of the relevant factors, and demonstrated a reasonable connection between the facts on the record and its decision,” we will not interfere with its conclusion. Ethyl Corp. v. EPA, 51 F.3d 1053, 1064 (D.C.Cir.1995). Statistical analysis is perhaps the prime example of those areas of technical wilderness into which judicial expeditions are best limited to ascertaining the lay of the land. Although computer models are “a useful and often essential tool for performing the Herculean labors Congress imposed on EPA in the Clean Air Act,” Sierra Club, 657 F.2d at 332, their scientific nature does not easily lend itself to judicial review. Our consideration of EPA’s use of a regression analysis in this case must therefore comport with the deference traditionally given to an agency when reviewing a scientific analysis within its area of expertise without abdicating our duty to ensure that the application of this model was not arbitrary. As we have noted, it is only when the model bears no rational relationship to the characteristics of the data to which it is applied that we will hold that the use of the model was arbitrary and capricious. See American Iron & Steel Inst. v. EPA, 115 F.3d 979, 1005 (D.C.Cir. 1997); Chemical Mfrs. Ass’n v. EPA, 28 F.3d 1259, 1265 (D.C.Cir.1994). Therefore, while we will examine each step of EPA’s analysis to satisfy ourselves that the agency has not departed from a rational course, we will not take it upon ourselves, as nonstatistieians, to perform our own statistical analysis — a job more properly left to the agency to which it was delegated. EPA’s determination of the revised Group 1 emission rates involved four steps: (1) the construction of a database consisting of Group 1, Phase I boilers already employing low NOx burner technology; (2) the derivation of two equations (one each for tangentially fired boilers and wall-fired boilers) that captured the percent reduction in emissions from the uncontrolled emission rates achieved by the boilers in the database; (3) the application of these equations to the uncontrolled emission rates of Group 1, Phase II boilers; and (4) the setting of emission rates for Group 1, Phase II boilers based on the range of data resulting from the application of the equations. We examine each of these steps in turn. 1. Construction of the Database EPA began its determination of whether the Group 1 limits should be revised by constructing a computerized database consisting of all known boilers that had installed only low NOx burners subsequent to November 15, 1990 (the date the amendments to the Act were enacted), and for which there existed at least 52 days of data measured by continuous emission monitors (“OEMs”). This database consisted initially of 24 wall-fired boilers and 9 tangentially fired boilers. 61 Fed.Reg. at 67,121. In response to the recommendations of several commenters that various boilers be included in or excluded from this database, EPA formalized and expanded its selection criteria into Data Quality Objectives (“DQOs”) — “rigorous and precisely defined rule tables” used to select candidates for the database. Id. at 67,122. Application of the DQOs resulted in a new database consisting of 39 wail-fired boilers and 14 tangentially fired boilers, a result that EPA believed would “increase[ ] the overall representativeness of the database.” Id. at 67,123-24. EPA then considered the lowest average NOx emission rate each boiler in the database had sustained for at least 52 days when CEM data were available (the “low NOx period”). To take into account the fact that the emissions rate immediately after low NOx burner installation might not be representative of a boiler’s emissions rate at full operating capacity, EPA also analyzed emission rates for a time period beginning 30 days after resumption of operation after installation until the end of the available CEM data as well as for a time period beginning with the first hour of the low NOx period until the end of the available CEM data. Id. at 67,-125. In response to comments that suggested that the 52-day period alone was insufficient to determine actual emission rates, EPA selected for the final rule the time period beginning with the first hour of the low NOx period until the end of the available CEM data (the “post-optimization period”) as the basis for assessing low NOx burner performance. Id. at 67,126. As part of its procedural challenge to the rule, Appalachian Power argues that EPA violated the rulemaking requirements of the Act by not disclosing the DQOs until the final rule, apparently invoking the Act’s requirement that EPA’s notice of proposed rulemaking (“NPRM”) include, inter alia, “the methodology used in obtaining the data.” 42 U.S.C. § 7607(d)(3)(B) (1994). We disagree. While it is true that the DQOs did not appear in the NPRM in precisely the same form or to the same extent as they did in the final rule, it is not the case that any significant DQO appeared for the first time in the final rule. In the NPRM, EPA listed two criteria that governed selection of boilers for the database: (1) whether units had installed only low NOx burners and had installed them after the date of enactment of the 1990 amendments and (2) whether post-retrofit hourly emission rate data, measured by OEMs and sustained for at least a 52-day period, was available. 61 Fed.Reg. at 1443-44. The DQOs in the final rule include these criteria as well as the following: (1) the database would be limited to Group 1 boilers; (2) boilers for which low NOx burner design, installation, and/or operations were known to be seriously flawed would be excluded; (3) boilers would have to have a pre-retrofit uncontrolled emission rate based on quality-assured data; (4) New Source Performance Standards boilers or boilers using Powder River Basin coal would be excluded because they could more easily meet low NOx emission limits than other boilers. 61 Fed.Reg. at 67,122 (Table 3). With the exception of criterion (1), which merely defines the database, the DQOs that appeared explicitly for the first time in the final rule were all intended to exclude faulty or overly optimistic data. Appalachian Power does not and, we think, could not contend that it would have challenged these quality-control DQOs had they been presented in full in the proposed rule. There is therefore no basis for us to conclude, as we must to invalidate the rule for procedural errors, that the addition of these DQOs in the final rule was “so serious and related to matters of such central relevance to the rule that there is a substantial likelihood that the rule would have been significantly changed if such errors had not been made.” 42 U.S.C. § 7607(d)(8) (1994). By including all the database criteria in the NPRM that could possibly have been subject to adverse comment, EPA has complied with its statutory directive to include “the methodology used in obtaining the data.” Moreover, we can find no apparent defects in the database itself. In constructing the database for the final rule, EPA applied the DQOs not only to those boilers used in the proposed rule analysis but also to those boilers that commenters requested that EPA consider as well as to additional boilers identified by EPA as using low NOx burner technology. This resulted in the addition of 20 boilers to the database (which ultimately contained a total of 39 wall-fired boilers and 14 tangentially fired boilers). 61 Fed.Reg. at 67,123-24. In this respect, EPA has identified all likely candidates for the boiler database as well as been responsive to comments. Appalachian Power’s assertion that the emission rates reflect boilers employing beyond-burner technology is not supported by the record. 2. Construction of the Equations Using the database, EPA constructed two linear regression equations, one for wall-fired boilers and one for tangentially fired boilers, that captured the percent reduction in emissions with low NOx burner technology as a function of the uncontrolled emission rate. As EPA noted in the preamble to the final rule, the use of a regression model rather than a simple extrapolation from the low NOx burner database would enable EPA better to predict the effect of installing low NOx burner technology on Phase II boilers. See 61 Fed.Reg. at 67,130. It is commonly understood that the more variables that are included in a regression analysis, the more likely it is that the model describes accurately the phenomenon it is being used to explain. As the Supreme Court has noted in the employment discrimination context, “the omission of variables from a regression analysis may render the analysis less probative than it otherwise might be,” but it does not render the analysis completely devoid of value. Bazemore v. Friday, 478 U.S. 385, 400, 106 S.Ct. 3000, 3009, 92 L.Ed.2d 315 (1986). Nevertheless, a number of commenters, Appalachian Power among them, argued that EPA’s analysis failed to take into account several operational factors associated with low NOx burners, including normal aging and wear of equipment, increased particulate emissions, auxiliary equipment design, and furnace configuration, all of which arguably could have an effect on the level of NOx emissions. EPA responded to this concern by using the post-optimization period rather than the 52-day period for analysis, which it believed would “reasonably account for variation in operating conditions among Group 1 boilers.” EPA Response to Comments at 63. “The claim that there are various problems due to aging of equipment that have not yet been encountered,” the agency continued, “is speculative and unsupported.” Id. While EPA’s response could have been more extensive, it does not suggest that the agency’s use of the regression models was arbitrary and capricious. As we have previously noted, the party challenging the use of such a model “cannot undermine a regression analysis simply by pointing to variables not taken into account that might conceivably have pulled the analysis’s sting.” Koger v. Reno, 98 F.3d 631, 637 (D.C.Cir. 1996) (dicta). See also Segar v. Smith, 738 F.2d 1249, 1277 (D.C.Cir.1984) (noting that where there is no reason to conclude that the omitted variable correlates with the dependent variable, the omission will not affect the validity of the analysis). Rather, that party must identify clearly major variables the omission of which renders the analysis suspect. This conclusion, derived from employment discrimination cases, holds equally true in this context, even more so because of the deference due to an agency’s scientific analysis. Neither the eommenters before EPA nor Appalachian Power here before us has offered any data to support the assertion that additional factors not accounted for in EPA’s regression analysis would have a significant effect on NOx emissions. The regression equations were constructed based on the data available or reasonably predictable at the time of the final rule; to require EPA to take into account variables for which no data existed would be to require that it engage in precisely the type of arbitrary rulemaking the Act forbids. 3. Application of the Equations to the Uncontrolled Emission Rates of Phase II Boilers The next step of EPA’s analysis was to calculate, through the application of the regression equations developed to the uncontrolled rates of the Phase II boilers, the NOx emission rate each Phase II boiler could be expected to achieve through a low NOx burner retrofit. Appalachian Power’s primary challenge to this step of the analysis is that the results generated by the regression equation are faulty because they do not include the uncertainty inherent in the calculation — in other words, the true reduction in NOx emissions associated with a particular retrofit might be somewhat greater than or less than the amount yielded by the regression equation. As a result, Appalachian Power contends, because EPA based its revised emission limits on what represents the midpoint between the uncertainty boundaries, the predicted emission levels are based on levels achievable by only 50 percent of the Phase II boilers. Although Appalachian Power’s assertion that the results are subject to some uncertainty is correct, we do not believe its complaint constitutes a telling critique of EPA’s analysis. In any regression analysis, the line described by the regression equation represents the best possible fit to the data; some points will necessarily be plotted above the line and some will fall below the line (except in the rare circumstance in which the line is a perfect fit to the data). While each data point will be associated with some residual (the difference between actual and fitted values), so long as this residual is within acceptable statistical limits, the fact that some data points necessarily fall below the line does not . render the regression analysis invalid. As we have noted in similar circumstances, “[t]hat the model does not fit every application perfectly is no criticism; a model is meant to simplify reality in order to make it tractable.” Chemical Mfrs., 28 F.3d at 1264. To invalidate a model simply because it does not perfectly fit every data point “would be to defeat the purpose of using a model.” Id. at 1265. Appalachian Power does not suggest in its argument before us that the uncertainty surrounding the data points is statistically unacceptable, only that it exists. We would not deem that sufficient to label EPA’s model arbitrary and capricious. This is not, certainly, like the case in Sierra Club, in which we rejected a 92 percent sulfur removal rate that was based solely on evidence that “only one commercial scale plant and one small pilot unit can almost but not quite meet the standard.” 657 F.2d at 363. In this case, 23 of 39 wall-fired boilers and 9 of 14 tangentially fired boilers in Group 1, Phase I can currently meet the revised limits. 61 Fed.Reg. at 67,123-24 (Tables 4 and 5). Because the statute requires only a determination that more effective low NOx burner technology is “available” for a class of boilers, the fact that, as Appalachian Power claims, some individual boilers cannot currently meet the revised limits does not serve to invalidate the rule. 4. Determination of the Limitation Finally, EPA used the rates resulting from the regression equations to determine “reasonably achievable emission limitation[s]” for Phase II boilers. 61 Fed.Reg. at 67,130. Appalachian Power asserts, however, that the predicted controlled emission rates of many boilers are so close to the emissions limits that any error in the prediction would render these boilers in violation of the limits. In addition, it claims that many utilities typically attempt to overcontrol emissions so that any fine tuning of the boiler will not bring the boiler over the emission limit. If the rule deems this “overcontrolled” emissions level achievable, Appalachian Power claims, utilities will be penalized for anticipating control difficulties. For both these reasons, Appalachian Power argues that EPA should have included a compliance margin in the NOx emission limits. Again, we find this challenge insufficient to vacate the rule. The first reason that this is so is a statutory one: The Act permits EPA to revise the emissions limitations upon a finding that “more effective low NOx burner technology is available.” 42 U.S.C. § 7651f(b)(2). The fact that these boilers can achieve lower emission levels with low NOx burner technology — even if they depend on a cushion between those levels and the emissions limits — demonstrates that the statutory requirement has been satisfied. Moreover, as EPA has noted in the preamble to the final rule, boiler owners who fear that tuning may send them over the allowable limits may use the alternative emission limitations (“AEL”) and averaging options provided in the Act to ensure that their total NOx emissions from all affected units comply with EPA regulations. 61 Fed.Reg. at 67,-136. Finally, we note, as EPA has pointed out, that the limit applies to a unit’s average annual emission rate rather than to a monthly or a daily emission rate. This means that a boiler may overemit on some days and underemit on others and still be deemed in compliance with its emission limit. Given these various options, there is no reason that EPA’s failure to build a compliance margin into the. limits themselves should render them arbitrary and capricious. We therefore reject Appalachian Power’s challenge on this front, as with its other substantive challenges to the Group 1, Phase II limits. Undaunted, Appalachian Power and intervenor National Mining Association (“NMA”) mount an additional procedural challenge to the Group 1, Phase II limits: namely, that EPA justified its revision of the rates based on public health and welfare concerns not authorized in the statute. See 61 Fed.Reg. at 67,160-61. We do not find this argument persuasive. The statute provides that EPA “may revise” (emphasis added) the limitations if it determines that more effective technology is available. We have noted that when a statute uses the permissive “may” rather than the mandatory “shall,” “this choice of language suggests that Congress intends to confer some discretion on the agency, and that courts should accordingly show deference to the agency’s determination. However, such language does not mean the matter is committed exclusively to agency discretion.” Dickson v. Secretary of Defense, 68 F.3d 1396, 1401 (D.C.Cir.1995) (emphases in original). Here, it is clear that EPA’s discretion is not boundless because section 407(b)(2) requires that EPA determine that more effective low NOx burner technology is available before it is permitted to revise the Group 1 emission limits. There is no indication, however, that Congress intended to further limit EPA’s discretion to revise the Group 1 limits once such a determination has been made. Our conclusion is supported by the fact that section 407(b)(2) provides that EPA “shall” establish the Group 2 emission limits by January 1, 1997, but “may” revise the Group 1 limits by the same date. See 42 U.S.C. § 7651f(b)(2). The use of both words in the same statutory subsection — as well as the section’s further reference to the applicability of “the revised emission limitations, if any” (emphasis added) — implies that Congress intended that EPA’s discretion to revise the Group 1 limits be broader than its discretion to set the Group 2 limits. Ethyl Corp., relied on by Appalachian Power, is not to the contrary. In Ethyl Corp., we considered a statute that provided that EPA, “‘upon application of any manufacturer of any fuel or fuel additive, may waive’ ” a prohibition against introducing certain fuel additives into commerce upon a finding that “the fuel additive does not cause or contribute to a failure of vehicles to meet emission standards.” 51 F.3d at 1055, 1058 (quoting 42 U.S.C. § 7545f(4) (1988 & Supp. V 1993)). The statute also provided that if EPA did not act either to grant or tp deny a waiver application within 180 days, the waiver would be treated as granted. See id. at 1059 (quoting 42 U.S.C. § 7545(f)(4)) (“If the Administrator has not acted to grant or deny an application under this paragraph within one hundred and eighty days of receipt of such application, the waiver authorized by this paragraph shall be treated as granted.”). EPA found that Ethyl had established the factual finding but denied its waiver application for public health reasons. We held that in the context of the entire statute, the statutory language stating that EPA “may waive” the prohibition after making the requisite factual determination referred to EPA’s discretion “to either act affirmatively, granting or denying a waiver, or not to act, and instead, to let the 180-day limit run.” Id. As a result, we held that once the factual standard was met, EPA’s discretion extended only to the decision as to whether the waiver would occur through EPA’s action or EPA’s inaction. Here, by contrast, there is no statutory language that provides a more limited definition of the phrase “may revise.” There is not, as in Ethyl Corp., a provision that once the requisite factual finding is made, the Group 1 limits will be revised even if EPA chooses not to act affirmatively to do so. Section 407(b)(2) states simply that EPA “may revise” the Group 1 emission limits downward if it determines that more effective low NOx burner technology is available, limited only by the provision that the new emission limits, “if any,” will not apply to boilers already regulated in Phase I. Given the lack of alternative interpretations, we cannot conclude that the word “may” has the same effect as it did in Ethyl Corp., that is, that it gives EPA discretion only to choose the avenue by which revised emission limits will be promulgated. Rather, we believe that “may revise” refers to EPA’s discretion to revise the Group 1 limits at all even after the requisite finding on more effective low NOx burner technology has been made. For EPA to justify this decision in part by referring to the environmental concerns of the Act, see 61 Fed.Reg. at 67,119-20; EPA Response to Comments at 281, was not arbitrary and capricious — indeed, it would have been arbitrary for EPA to offer no justification, and Appalachian Power has not suggested other considerations that would have been more consonant with the statutory framework. In any event, it is clear, contrary to Appalachian Power’s argument, that EPA has sufficiently justified its decision to revise the Group 1 emission limits apart from environmental concerns. Appalachian Power is correct that the burden is on EPA to justify the change from the 1995 emission limits, which the agency agreed were “aggressive,” see 60 Fed.Reg. at 18,758-59, “[b]ut that justification need not consist of affirmative demonstration that the status quo is wrong; it may also consist of demonstration, on the basis of careful study, that there is no cause to believe that the status quo is right, so that the existing rule has no rational basis to support it.” Center for Auto Safety v. Peck, 751 F.2d 1336, 1349 (D.C.Cir.1985). As EPA has noted, the 1995 emission limits were based on data from periods only until 1992, while the current limits incorporated additional data from as recently as 1996. See 61 Fed.Reg. at 67,120; EPA Response to Comments at 27-28. This increase in available data, and the more stringent limits that analysis of that data generated, were sufficient for EPA to conclude that “there is no cause to believe that the status quo [ie., the 1995 emission limits] is right.” Because we find nothing irrational in that determination, we uphold the revised Group 1 NOx emission limits. B. The Group 2 Emission Limits By contrast to the boilers in Group 1, the boilers in Group 2 are not necessarily amenable to retrofitting with low NOx burners. For this reason, Congress did not require EPA to base its Group 2 emission limits on that single control technology. Instead, in section 407(b)(2), Congress instructed EPA to base the Group 2 emission rates on the degree of reduction achievable through the retrofit application of the best system of continuous emission reduction, taking into account available technology, costs and energy and environmental impacts; and which is comparable to the costs of nitrogen oxides controls set [for Group 1 boilers]. 42 U.S.C. § 7651f(b)(2). Appalachian Power challenges EPA’s interpretation of this statutory language, an interpretation that informed the methodology the agency used to set the Group 2 emission limits. In addition, even assuming the validity of that interpretation, Appalachian Power challenges the reasonableness of the methodology EPA employed. We reject both challenges. 1. Statutory Interpretation of Section 107(b)(2) EPA believes that the statutory provision is ambiguous, both in its specific words and in their grammatical arrangement. Although ambiguous, EPA concludes that the best reading of the key statutory phrase, “comparable to the costs of,” directs it to conduct a comparison of the cost-effectiveness of those control technologies available for Group 2 boilers with the cost-effectiveness of the NOx controls required for Group 1 boilers (ie., low NOx burner technology). And it reasons that the provision as a whole instructs it to base the Group 2 limits on the degree of emissions reduction achievable by those Group 2 technologies that compare favorably in cost-effectiveness with low NOx burner technology. For these purposes, EPA measures cost-effectiveness in dollars per ton of NOx removed ($/ton-removed). Appalachian Power, by contrast, argues that the language of the statutory provision is unambiguous. It contends that the language requires a comparison of the costs of producing electrical output using control technologies that can be used in Group 2 boilers with the costs of producing electrical output using low NOx burner technology in Group 1 boilers. Those costs, it argues, should be measured in dollars per kilowatt ($/kw) and dollars per kilowatt hour ($/kwh). In Appalachian Power’s view, only those Group 2 technologies that compare favorably to low NOx burner technology by this measure may be considered in setting the Group 2 limits. Once again, Chevron’s first step does not take us very far. The statute does not define the phrase “comparable to the costs of.” As discussed below, we agree with EPA that the words in the phrase are ambiguous, see 61 Fed.Reg. at 67,138, and that the provision as a whole is grammatically awkward, see id. at 67,139. And although the legislative history does not definitively address the meaning of the phrase, we also agree with EPA that that history is supportive of the agency’s interpretation. To begin, both sides agree that the meaning of the word “cost” is the “price paid for a thing.” App. Pwr. Br. at 33; EPA Br. at 34. Moreover, both agree that, depending upon the context, that “thing” could be either the amount of pollution removed ($/ton-removed) or the amount of electricity produced ($/kw or $/kwh). Indeed, the word “costs” is used in two places in section 407(b)(2), and perhaps the best evidence of the essential ambiguity of the word is that each side adopts the other’s definition for one of the two uses. Appalachian Power argues that l/ton-removed is the appropriate way to define “costs” when they are “tak[en] into account” in determining “the best system of continuous emission reduction.” App. Pwr. Br. at 32. It insists, however, that only $/kw and $/kwh will do for the key phrase, “comparable to the costs of nitrogen oxides controls.” Id. at 33; App. Pwr. Reply Br. at 8. EPA would use the two definitions in precisely the opposite places. See EPA Br. at 35. Turning to the grammatical structure of the statutory provision, Appalachian Power argues that its definition for the key phrase is “plainly” required because of the context in which the phrase is used. It contends that the antecedent of the word “which,” in the phrase “which is comparable to the costs of nitrogen oxides controls,” is the phrase “the best system of continuous emission reduction.” On this reading, it argues that EPA’s job is to base emission limits on “the best system of continuous emission reduction ... which is comparable to the costs of nitrogen oxide controls.” When comparing the costs of one system of controls to those of another, it argues, the appropriate comparison is the cost of producing electrical power. Even if we were to adopt this view of the provision’s grammar, it would be hard to conclude that Appalachian Power’s definition is “plain.” The syntax for which Appalachian Power argues simply does not resolve the question of “costs for what?” Although one certainly could compare the “costs” of two systems by comparing their costs for producing electrical output, one could also reasonably compare their costs for removing tons of pollution. Moreover, we cannot agree that Appalachian Power’s view of the provision’s grammar is the only reasonable one. As EPA notes, to read the phrase “best system” as the antecedent of the word “which” would require deletion of both the semicolon and the word “and” that separate the two parts of the statutory provision. See 61 Fed.Reg. at 67,139. In EPA’s view, the better reading is that the antecedent of “which” is the phrase “the degree of reduction achievable.” On this reading, EPA urges, Congress contemplated that the agency would base Group 2 limits on the degree of reduction (1) which can be reached through the best system of NOx reduction, taking into account available technology, costs and energy and environmental impacts!;] and (2) which is “comparable to the costs” of [low NOx burner] technology. EPA Br. at 6 (emphases added). The only way to determine the degree of NOx reduction that can be achieved in Group 2 boilers at costs comparable to the costs of low NOx burner technology, EPA argues, is to consider the relative costs of controls per ton of NOx removed. EPA contends that when one focuses on “degree of reduction” as the antecedent, rather than on “best system,” cost-effectiveness becomes the natural measurement to apply. EPA’s grammatical construction is plausible. But it is no more plausible than that of Appalachian Power because it, too, requires editing of the congressional text. This time, rather than make a deletion, we would need to make an addition. As noted in italics above, we would need to add the word “which,” so that it appears twice rather than only once, in order to create a parallel construction that makes “the degree of reduction” the antecedent of both of the numbered phrases. The need to make that addition, however, only highlights its absence in the actual text and confirms the essential ambiguity of section 407(b)(2)’s key phrase. Finding nothing dispositive in the statute’s language or grammar, we look next to the legislative history for guidance. Appalachian ■ Power argues that EPA’s construction is inconsistent with the purpose of section 407(b)(2), which, Appalachian Power contends, was to ensure that the dollar “cost of controls” to owners of Group 2 boilers would not exceed the dollar “cost of controls” imposed on owners of Group 1 boilers — ie., not exceed the cost of low NOx burners. We find little support in the legislative history, however, for Appalachian Power’s view of the section’s purpose. To the contrary, although we cannot say that the legislative history is dispositive, it does contain considerable support for EPA’s view that cost-effectiveness is an appropriate measure of comparison, even if it is not the only appropriate measure. Appalachian Power correctly notes that the language of section 407(b)(2) that directs EPA to set Group 2 emission limits, including the key phrase “comparable to the costs of,” comes from the Senate bill. Compare 42 U.S.C. § 7651f(b)(2), with Clean Air Act Amendments of 1990, S. 1630, 101st Cong. § 407(b)(2) (1990) (Senate bill), reprinted in Committee on EnVt & Pub. WORKS, U.S. Senate, A Legislative History of the Clean Air Act Amendments of 1990, at 4641 (1993) [hereinafter “Legislative History”]. By contrast, the House version of the bill prevented EPA from regulating certain Group 2 boilers unless EPA, inter alia, found methods “available for reducing emissions from such boilers that are as cost effective as the application of low nitrogen oxides burner technology in the ease of [Group 1] boilers.” Clean Air Act Amendments of 1990, S. 1630, 101st Cong. § 506(a)(c) (1990) (House bill) (emphasis added), reprinted in Legislative History, at 2277. Appalachian Power contends that the conference committee’s adoption of the Senate version — which did not use the phrase “cost effective” found in the House version — was tantamount to a rejection of the concept of cost-effectiveness. We disagree. As a general matter, courts often have noted the difficulty of determining the significance of Congress’s unexplained modification of language in earlier drafts of legislation, and have found that such modification does not necessarily indicate Congress’s rejection of the substance of the earlier language. See, e.g., Seatrain Shipbuilding Corp. v. Shell Oil Co., 444 U.S. 572, 594-95, 100 S.Ct. 800, 812-13, 63 L.Ed.2d 36 (1980); Edison Elec. Inst. v. EPA, 2 F.3d 438, 451 (D.C.Cir.1993). The two phrases at issue here — “comparable to the costs of’ and “cost effective” — are not incompatible. It is possible that Congress regarded the two as synonymous, and that the conference committee simply adopted the Senate’s formulation over that of the House. Elements of the legislative history of the enacted version support this reading. Perhaps most persuasive is Congress’s direction, in the conference report on the final bill, that EPA should base emission limits for Group 2 boilers on “methods that are available for reducing emissions from such boilers that are as cost effective as the application of low nitrogen oxide burner technology to [Group 1] boilers.” H.R. Conf. Rep. No. 101-952, at 344 (emphasis added). Moreover, the conference report incorporates a section of the Senate report on an earlier Senate bill. That report equated the phrase “cost-effectiveness” — as measured by $/ton-removed — with the phrase “comparable to the cost of’: Also favoring the cost-effectiveness of [section 407] is the development of new, lower-expense technologies_ [The] decreasing cost for selective catalytic reduction (SCR) may lower the expense of initial NOx reductions even further. For example, SCR has long been viewed as prohibitively expensive, but recent dramatic declines in cost have brought the per-ton-removed pnce of this technology down to as low as $600.... This is comparable to the cost of conventional control methods like low NOx burners.... S.Rep. No. 101-228, at 332-33 (1989) (emphases added). The Senate report also noted that with the NOx emission limits, the Senate “intended to compel utilities to do no more than make the most cost-effective reductions.” Id. at 332 (emphasis added). Appalachian Power’s general point, that Congress was concerned that the “costs” for Group 2 boilers be comparable to the “costs” for Group 1 boilers, is plainly correct. But there is no support for Appalachian Power’s contention that Congress intended $/kwh to define the word in the second part of section 407(b)(2), while expecting $/ton-removed to define it in the first. Congress simply did not make the fine distinctions that the parties make here between different methods of measuring “costs.” Indeed, when introducing the amendment that led directly to section 407(b)(2), various Senators referred interchangeably to the terms “cost-effectiveness,” “low cost,” and “not unreasonably expensive.” See, e.g., 136 Cong. Reo. 5045 (1990) (statement of Sen. Chafee); 136 Cong. Rec. 5045-46 (1990) (statement of Sen. Baucus); 136 Cong. Rec. 5046 (1990) (statement of Sen. Lott). And during the floor debates on the conference report, Senator Burdick — conferee and Chair of the Senate Committee on Environment and Public Works — again equated “cost” and “cost-effective,” stating that the Group 2 limits were to be set only if the costs of such reductions are as cost effective as reductions from installation of low NOx burners on other types of boilers.... This provision is carefully worded to make cost considerations the determinative factor in consideration of NOx reductions from [Group 2] boilers. 136 Cong. Rec. 36,029 (1990) (statement of Sen. Burdick). In sum, we draw the same conclusion regarding the phrase “comparable to the costs of’ in the 1990 amendments as the Supreme Court drew regarding the term “stationary source” in the 1977 amendments to the Act: neither the statutory language nor the legislative history is dispositive of the meaning of the term. See Chevron, 467 U.S. at 861-62, 104 S.Ct. at 2790-91. Moving then to Chevron’s second step, we must consider whether EPA’s deeision to interpret the statute as contemplating a comparison of cost-effectiveness is reasonable. In light of the above discussion, there is little left to say. Given the ambiguous syntax and the multiple meanings that both parties concede may be assigned to the word “costs,” we cannot conclude that EPA’s decision to use $/ton-removed as the measurement of costs is unreasonable. Moreover, as our review of the legislative history suggests, although EPA’s interpretation may not be required by that history, it surely is consistent with and supported by it. We thus conclude that EPA’s construction of section 407(b)(2) is a permissible one. 2. Challenges to EPA’s Methodology for Determining Emission Limits Having concluded that it would base Group 2 emission limits on the capabilities of those Group 2 control technologies comparable in cost-effectiveness to low NOx burner technology, EPA developed a test for making such comparisons. EPA found that it could not rely on a comparison of median or mean costs alone, because the $/ton-removed cost for a given control technology, including low NOx burners, varied widely from boiler to boiler even within the same category of boiler, and had different cost ranges for different categories of boilers. See 61 Fed.Reg. at 67.138, 67,143 (Table 12). For this reason, EPA determined that it needed a more comprehensive statistical approach. See id. at 67.138. First, EPA excluded the cost-effectiveness figures for boilers in the top and bottom 10 percent of cost-effectiveness, so that neither the lowest nor the highest cost projects would skew the comparison. See id. at 67,-143 (Table 13); see also EPA Response to Comments at 91-92 (Joint Appendix (“J.A.”) 216-17). Next, EPA determined that the cost-effectiveness of using a given type of Group 2 control technology for a specific category of Group 2 boiler (a technology/category combination) was comparable to the cost-effectiveness of using low NOx burners in Group 1 boilers, if the median l/ton-removed cost of that Group 2 technology/category combination: (1) did not exceed by more than one-third the overall median $/ton-removed cost for low NOx burners in Group 1 boilers, and (2) did not exceed the individual medians for both of the categories of Group 1 boilers. See id. at 67,138, 67,-143. Finally, EPA required that the 90th percentile of the $/ton-removed cost range for Group 2 technology/category combinations not exceed the 90th percentile of the $/ton-removed cost range for low NOx burners in Group 1 boilers. See id. at 67,138. Using this cost-comparison test and further calculations, EPA selected appropriate control technologies and an emission limit for each of the four statutory categories of Group 2 boilers, see 42 U.S.C. § 7651f(b)(2)(A)-(D) (wet bottom wall-fired boilers; cyclones; units applying cell burner technology; and “all other types of utility boilers”). It concluded that it could not set emission limits at all for two types of boilers in the catch-all fourth category because no control technology met the comparability test. See id. at 67,114. And it concluded that one kind of control technology was not cost-effective for two types of boilers, and so could not be used in setting emission limits for those boilers. See id. at 67,143. Appalachian Power does not propose an alternative to the methodology EPA employed for setting the Group 2 emission limits. Instead, it and intervenor NMA charge that various elements of EPA’s methodology are arbitrary and capricious, are unsupported by the record, or were used without following the Act’s procedural requirements, and that we therefore must overturn the emission limits generated by EPA’s methodology. Although we have considered and find all of petitioners’ and intervenor’s myriad arguments in this area lacking in merit, we discuss below only the more important of them. a. Significance of Cost as a Factor in Selection of Controls. “Even assuming that Congress required EPA to compare the cost-of-tons reduced,” Appalachian Power argues, EPA’s comparison “is unlawful because it does not make cost a significant, much less a determinative factor.” App. Pwr. Br. at 37. Appalachian Power contends that this is the result of EPA’s choice of methodology, because when one uses a fraction that divides costs by tons-removed, the fraction is “driven” by the denominator. Id. at 37-38. The proof that this is so assertedly is in the results that EPA’s comparison test produces: the “costs” of the controls EPA has selected for Group 2, Appalachian Power claims, are “three to seven times higher than the costs of [low-NOx] burners.” Id. at 37-38 & n. 119. On its face, this is a difficult argument to understand. The key is to recognize that the argument actually mischaracterizes itself: it does not assume, as it claims, that the relevant “costs” are $/ton-removed, but rather assumes they are $/kwh. When Appalachian Power says the “costs” of the Group 2 controls are three to seven times the costs of Group 1 low NOx burners, it can say so only by measuring those costs by $/kwh — which is what it does. See id. at 38 n. 119. Hence, in this argument Appalachian Power does not assume the validity of the measurement EPA has chosen, but simply relitigates, in different terms, the same argument we have rejected above. b. Weight Given to Smaller Boilers. Appalachian Power contends that EPA manipulated its methodology to give unfair weight in Group 1 to smaller, underutilized boilers that are not cost-effective to retrofit with any controls, while giving more weight in Group 2 to larger, higher-utilized, and therefore more cost-effective boilers. This unfair comparison was made, Appalachian Power asserts, in order to ensure that high $/ton-removed Group 2 technologies would still be comparable to Group 1 controls. We see no evidence of this manipulation. Instead, as we have noted, EPA made a number of statistical adjustments, and in particular excluded figures for boilers in the top and bottom 10th percentiles, precisely to ensure that neither the lowest nor the highest cost projects skewed the comparison. See EPA Response to Comments at 91-92 (J.A. 216-17). The agency’s decisions not to impose limits on two types of Group 2 boilers because those boilers did not pass its cost-comparability test, and to exclude from consideration one kind of Group 2 control technology because it was too costly for two types of Group 2 boilers, are further evidence that EPA did not intentionally manipulate its methodology in order to ensure that expensive Gro