Natural Resources Defense Council, Inc. v. Herrington

Opinion for the Court filed by Circuit Judge WALD.

CONTENTS

Page

I. Background .................................................................................. 349

II.DOE’s Definition of “Significant Conservation of Energy” ....... 354

A. The Development of DOE’s Definition ....................................... 354

B. The Validity of DOE’s Definition .............................................. 357

C. DOE’s Rationale for its Definition ............................................. 362

III. DOE’s Method of Determining Savings That Would Result From Standards ............................. 368

A. Statutory Authority for Subtracting “Base-Case” Savings From “Standards?’. Savings ..... 369

B. The ORNL Model ..................................................................... 370

IV. Maximum Technological Feasibility ............................................. 376

A. The Statute .............................................................................: 376

B. DOE’s Treatment of Maximum Technologically Feasible Levels ___ 377

C. DOE’s Compliance with EPCA .................................................. 379

1. DOE’s Failure to Determine Maximum Technologically Feasible Levels ................................................................................ 379

2. DOE’s Refusal to Consider Standards Based on All Technologically Feasible Design Options .............................................. 381

a. Prototypes .................................................................... 381

b. Foreign Market Design Options ......................................• 388

c. The Five-Year Payback Period _______________________________________ 389

d. Lead Times ................................................................... 392

e. Specific Design Options _________________________________________________ 393

3. DOE’s Reliance on 1980 Data______________________________________________ 393

V. Economic Justification ................... 395

A. DOE’s Analysis of Benefits ...................................................... 395

1. DOE’s Discussion of the Nation’s Need to Save Electricity and Energy Savings .................................................................. 395

2. DOE’s Use of a 10 Percent Real Discount Rate ................... 397

*347 CONTENTS Page

3. DOE’s Calculation of Benefits from Central Air Conditioner Standards ..................................................................... 399

a. Possible Reduction of Peak Load Electrical Demand ---------- 399

b. Failure to Consider High-Efficiency Models ..................... 402

c. Flaws in the Cost Efficiency Curve ............................... 403

d. Hours of Operation ....................................................... 403

B. DOE’s Analysis of Burdens ....................................................... 404

1. The Financial Impacts Model ............................................... 404

2. The FIM Results for Central Air Conditioner Standards ........ 407

3. Forgone Investment and Reductions in Performance or Utility 409

C. DOE’s Weighing of Burdens Against Benefits .............................. 410

VI. DOE’s Refusal to Allow Cross-Examination ................................ 410

VII. DOE’s Failure to Prepare an Environmental Assessment or Environmental Impact Statement ...................................................... 414

WALD, Circuit Judge.

The petitions in these consolidated cases require us to interpret section 325 of the Energy Policy and Conservation Act (EPCA or the Act), 42 U.S.C. § 6295 , which was enacted in 1975 as part of a “comprehensive national energy policy.” S.Rep. No. 516, 95th Cong., 1st Sess. 116 (1975), U.S.Code Cong. & Admin.News 1975, p. 1762 (conference report). In its initial version, section 325 required the Federal Energy Administrator to prescribe energy efficiency improvement targets for thirteen named household appliances, called “covered products.” See EPCA § 325(a)(1)-(2), Pub.L. No. 94-163, 89 Stat. 871, 923-24 (1975). 1 If the Administrator determined that manufacturers of any of the covered products were not likely to achieve the aggregate gain in efficiency specified in the target by 1980, he was directed to begin a proceeding to prescribe a mandatory “energy efficiency standard” for that appliance. See id. § 325(a)(4)(A)-(B), 89 Stat. at 924.

In 1978, however, Congress amended EPCA to “eliminate[] the target approach and improve[ ] the procedures for establishing standards to ensure that efficiency improvements will be made expeditiously.” H.R.Rep. No. 496, Pt. 4, 95th Cong., 1st Sess. 46 (1977), U.S.Code Cong. & Admin. News 1978, pp. 7659, 8493; see National Energy Conservation Policy Act (NECPA) § 422, Pub.L. No. 95-619, 92 Stat. 3206, 3259 (1978). The amended version of section 325(a) orders the Secretary of Energy to prescribe energy efficiency standards for the thirteen covered products without first establishing industry targets. Under the Act, “[e]nergy efficiency standards for each type (or class) of covered products ... shall be designed to achieve the maximum improvement in energy efficiency which the Secretary determines is technologically feasible and economically justified.” EPCA § 325(c). If the Secretary prescribes a standard at a level lower than the maximum technologically feasible level, the Secretary must explain why that lower level was chosen. Id. § 325(i)(3). The Act also declares that if, for any type or class of covered product, 2 a standard would not *348 result in significant conservation of energy or would not be technologically feasible or economically justified, the Secretary shall not prescribe a standard. Id. § 325(b). A determination that no standard is warranted for a particular appliance, like the issuance of a mandatory standard, preempts any state-law efficiency requirements for the appliance, although the state may then apply to the Secretary for an exemption from the preemption provision. Id. §§ 325(b), 327. The Secretary was further instructed to give priority in, formulating standards to nine of the covered products. Id. § 325(g). 3

The final rules challenged here apply to eight of the nine covered products given priority under EPCA. 4 The Department of Energy (DOE) 5 determined that for seven of those eight products, a mandatory standard would not result in significant conservation of energy. 6 DOE also determined that for all eight products, a mandatory standard would not be economically justified. See 48 Fed.Reg. 39,376, 39,376 (1983); 47 Fed.Reg. 57,198, 57,198 (1982). These “no-standard standards” terminated the rulemaking for the eight products considered, and thus preempted state regulation without imposing any federal regulation at all. 7 The petitioners are the Natural Resources Defense Council (NRDC), Consumers Union of the United States, Representative , Richard Ottinger, the California State Energy Resources Conservation and Development Commission, the State of Minnesota, and the State of New York. 8 The State of Texas, the Northwest *349 Power Planning Council, the Florida Department of Community Affairs, the Oregon Department of Energy, and the Iowa Commerce Commission have intervened in support of the broad attack on the rules mounted by petitioners. Various appliance manufacturers and trade associations have intervened in support of the Secretary of Energy.

We agree with the petitioners that in important respects DOE’s determinations are unsupported by substantial evidence and are contrary to law. See EPCA § 336(b)(2). We also find that DOE violated section 336(a)(2) of EPCA by refusing to allow interested persons a meaningful opportunity to question DOE employees who participated in the rulemaking. Finally, we find that DOE did not adequately evaluate the environmental consequences of these rules, as section 102(2)(C) of the National Environmental Policy Act of 1969, 42 U.S.C. § 4332 (2)(C), requires. We therefore grant the petitions and hold the rules under review to be unlawful.

I. Background

EPCA was originally enacted in the aftermath of the oil embargo imposed against the United States in 1973 and 1974 by certain petroleum-producing countries. As the House Report commented:

The 1973-74 Arab embargo resulted in a 2.2-million-barrel-per-day reduction on imports of crude oil and petroleum products — an interruption equal to about one third of the United States’ imports and [to] about 12 percent of overall petroleum supplies. The embargo was. accompanied by a 7 percent decrease in real Gross National Product for the first quarter of 1974, and an increase of unemployment of approximately 425,000 persons for that quarter. Dependence of the United States on petroleum products has risen since September, 1973, from 33 percent to approximately 36 percent in December, 1974____ Continuing dependence on foreign petroleum, and the economic consequences of another embargo, require that the United States take steps to effectively reduce its vulnerability to future import interruptions.

H.R.Rep. No. 340, 94th Cong., 1st Sess. 20 (1975), U.S.Code Cong. & Admin.News 1975, p. 1782. This emergency vividly illustrated the serious long-term economic and national security problems that continuing dependence on foreign sources of energy would create. To meet these challenges, President Ford called for “the strongest and most far-reaching energy conservation program we have ever had.” 11 Weekly Comp.Pres.Doc. 40, 41 (Jan. 20, 1975).

Congress responded by enacting EPCA, a wide-ranging statute establishing a comprehensive national energy policy. According to the conference report, EPCA was designed to:

. (1) maximize domestic production of energy and provide for strategic storage reserves of crude oil, residual fuel oil and refined petroleum products;

(2) ... minimize the impact of disruptions in energy supplies by providing for emergency standing measures;

(3) provide for domestic crude oil prices that will encourage domestic production in a manner consistent with economic recovery; and

(4) reduce domestic energy consumption through the operation of specific voluntary and mandatory energy conservation programs.

S.Rep. No. 516, 94th Cong., 1st Sess. 116-17 (1975) U.S.Code Cong. & Admin.News 1975, p. 1957. The energy conservation programs alone included separate provisions designed to:

Establish mandatory average fuel economy performance standards for any passenger automobiles and other light duty highway vehicles;

Require energy labeling of major home appliances and certain other consumer products, and authorize energy efficiency standards for major appliances;

*350 Authorize block grants-in-aid for States to assist in the development and implementation of state-administered energy conservation programs; and

Establish a program to encourage increased efficiency of energy use by American industry;

Establish a program for energy conservation within the Federal Government; and

Promote the use of recycled oil.

Id. at 118-19 , U.S.Code Cong. & Admin. News 1975, p. 1959.

The EPCA appliance program had four elements. First, the Federal Energy Administrator was required to prescribe test procedures for measuring the energy efficiency and operating costs of thirteen named products. See EPCA § 323, Pub.L. No. 94-163, 89 Stat. 871, 919 (1975). Second, the Administrator was required to promulgate rules compelling manufacturers of covered products to label those products with information developed from the test procedures. Id. § 324, 89 Stat. at 920. Third, the Administrator was required to prescribe voluntary energy efficiency improvement targets, set at the highest efficiency that was economically and technologically feasible, for each of the thirteen covered products. Id. § 325(a)(1)(A), (a)(2), 89 Stat. at 923, 924. Targets for ten of those products were given priority. See id. § 325(a)(1)(A), 89 Stat. at -923. Targets for priority covered products were to be fixed so that the aggregate efficiency of all appliances belonging to those product types manufactured in 1980 would exceed the efficiency of the same appliances manufactured in 1972 by at least 20 percent. Id. § 325(a)(1)(B), 89 Stat. at 924. Finally, if the Administrator determined that any target for a product type was not likely to be achieved, he was directed to begin a proceeding to prescribe a mandatory energy efficiency standard. Id. § 325(a)(4)(B). If, in that proceeding, the Administrator found that for the product type or for product classes within that type (a) improvements in energy efficiency were technologically feasible and economically justified, and (b) an unregulated market, as influenced by the labeling program, would not result in the production and purchase of products at the maximum technologically feasible and economically justified efficiency, EPCA required the Administrator to prescribe a mandatory energy efficiency standard. Id. § 325(a)(4)(C). EPCA also set specific deadlines for the Administrator in carrying out the provisions of the Act. 9

In 1978, President Carter proposed a series of bills aimed at a complete overhaul of national energy policy. Among the major initiatives suggested were nationwide programs to encourage conservation and efficiency planning in transportation, buildings, appliances, and industry; cogeneration, district heating, and utility reform measures; price control and tax measures for oil and natural gas; incentives to stimulate production of Alaskan oil, oil from the outer continental shelf, shale oil, liquified natural gas, and synthetic fuels; and development programs for coal, nuclear, hydroelectric, solar, and geothermal energy. These proposals were, in various forms, enacted in five statutes: the Public Utility Regulatory Policies Act of 1978, Pub.L. No. 95-617, 92 Stat. 3117; the Energy Tax Act of 1978, Pub.L. No. 95-618, 92 Stat. 3174; the Powerplant and Industrial Fuel Use *351 Act of 1978, Pub.L. No. 95-620, 92 Stat. 3289; the Natural Gas Policy Act of 1978, Pub.L. No. 95-621, 92 Stat. 3350; and NECPA, Pub.L. No. 95-619, 92 Stat. 3206 (1978). Together, these statutes were known as the National Energy Act.

NECPA, though only part of a mammoth legislative program, itself sought to strengthen and expand energy conservation programs in a number of ways. Title II of the Act inaugurated various programs involving public utilities in efforts to conserve the residential use of energy and authorized certain grants to low-income families for energy-saving improvements in housing; Title III created energy conservation programs for schools, hospitals, arid other buildings owned by local governments and public institutions; Title IV established conservation programs for automobiles, appliances, and industrial equipment; Title V set forth energy conservation measures for the federal government, including the use of solar heating and cooling and a photovoltaic energy program; and Title VI contained miscellaneous conservation and reporting measures.

The Carter administration’s National Energy Plan, which announced the details of this legislative package, declared that:

Most [major home appliances] could achieve significant reductions in energy use with relatively small increases in cost. Current legislation relies mainly on voluntary efforts to meet industry-wide average target, and permits the establishment of mandatory standards only after long delays. New legislation is proposed to streamline the regulatory process. The present voluntary program will be replaced by mandatory minimum standards on certain major home appliances as soon as possible.

Executive Office of the President, The National Energy Plan 43 (1977). In keeping with these views, the administration proposed a radical revision of the EPCA appliance program. Specifically, the administration's proposal would have required DOE to prescribe energy efficiency standards for seven of the thirteen covered products named in EPCA. See National Energy Act: Communication from the President of the United States, H.R.Doc. No. 138, 95th Cong., 1st Sess. 30 (1977) [hereinafter cited as H.R.Doc. No. 138]. Those seven products were selected because they consumed relatively large amounts of electricity — together, about 80 percent of the energy used in American households. See S.Rep. No. 409, 95th Cong., 1st Sess. 38 (1977). The proposal permitted but did not compel DOE to prescribe standards for the six lower-consumption covered products. See H.R.Doc. No. 138 at 30. Under certain conditions, DOE could also prescribe standards for further appliances not named in EPCA or in the new legislation. 10 The proposal did not require “significant” conservation; it simply demanded that standards be technologically feasible and economically justified. See id. at 30-35 .

The House amended this proposal to require that DOE prescribe standards for all thirteen named products. The House bill retained discretionary standards only for appliances not named in the Act that DOE voluntarily designated for coverage. Having mandated standards even for the six low-consumption covered products, the House also decided DOE should not pre *352 scribe standards that would not result in “significant” energy conservation. See H.R.Rep. No. 496, Pt. 4, 95th Cong., 1st Sess. 46-47 (1977). The Senate Bill required DOE to prescribe standards for nine of the covered products and authorized discretionary standards for the remaining four covered products and for other appliances designated by DOE. See S.Rep. No. 409, 95th Cong., 1st Sess. 32, 77 (1977). Like the administration proposal, the Senate bill contained no significance provision. See S. 2057, 95th Cong., 1st Sess. § 201(a), reprinted in S.Rep. No. 409, 95th Cong., 1st Sess. 13-14 (1977).

The conference report combined the two versions. Congress accepted the House provision requiring standards for all thirteen covered products, along with the significance provision, but standards for the nine appliances singled out in the Senate bill were given priority. See H.R.Rep. No. 1751, 95th Cong., 2d Sess. 114-15 (1978). The President signed NECPA into law on November 9, 1978. See 14 Weekly Comp. Pres.Doc. 1,978 (Nov. 9, 1978).

As the Act required, DOE proceeded to develop final test procedures for the thirteen covered products. See 10 C.F.R. § 430.22 -.23 & apps. A-0 (1985) (codifying current test procedures); 44 Fed.Reg. 49, 51 (1979) (tabulating Federal Register citations for test procedure notices). After issuing an advance notice of proposed rule-making, see id., DOE published proposed rules on June 30, 1980 that set forth minimum efficiency standards for all eight of the products considered in the rulemaking. See 45 Fed.Reg. 43,976 (1980); see also infra note 15. DOE’s general approach was to consider standards based on the most efficient combination of those design options commercially available in 1980. DOE did not consider standards based on prototype technologies or other advanced designs not yet on the market. DOE then developed several different possible standards at different efficiency levels for each product type or class, and evaluated those possible standards to decide if they were economically justified. See 45 Fed.Reg. at 43, 983-85. Following that path of analysis, DOE concluded that standards for all eight covered products would result in significant conservation and were, on balance, economically justified. See id. at 43,976 . DOE also prepared a lengthy environmental assessment of the proposed rules, although it did not prepare a full-scale environmental impact statement. See U.S. Department of Energy, Environmental Assessment: Energy Efficiency Standards for Consumer Products (1980) [hereinafter cited as 1980 Environmental Assessment], J.A. at 225.

EPCA required DOE to promulgate final rules by January 2, 1981. 11 In January of 1981, the outgoing administration informed a congressional committee that except for “fine-tuning,” work on the standards was nearly complete. 12 However, in February *353 of 1981 the incoming administration announced that it would seek repeal of the mandatory appliance program 13 and in any event intended to undertake a comprehensive reassessment of the data supporting the proposed rules. See 46 Fed.Reg. 13,-51-7.(1981).

In October of 1981 — over eight months after the statutory deadline — NRDC and Consumers Union of the United States filed suit to compel promulgation of appliance efficiency standards. See Natural Resources Defense Council, Inc. v. Edwards, Civ. No. 81-2546 (D.D.C.). On April 2, 1982, DOE issued a new notice of proposed rulemaking, which proposed a completely new and vastly higher definition of “significant” conservation of energy, entirely new views about the benefits and burdens of mandatory standards, and tentative determinations that no standard would be issued for any of the eight products involved in this rulemaking. See 47 Fed.Reg. 14,424 (1982). DOE’s notice did not state when final rules would be prescribed, and NRDC and Consumers Union accordingly pressed forward with their litigation. DOE settled the suit by agreeing to promulgate rules no later than October 29, 1982, unless prevented by good cause. See 47 Fed.Reg. 57,198, 57,200 (1982). DOE failed to meet that deadline. On December 22, 1982, DOE published final rules determining that standards for kitchen ranges and ovens and for clothes dryers would not result in significant conservation of energy and in any event would not be economically justified. See id. at 57,198 . The December rulemaking also finally adopted a slightly revised version of the new definition of significant conservation proposed in the April notice. See id. at 57,202-09 .

On August 30, 1983, DOE announced its refusal to prescribe mandatory standards for the six remaining priority products involved in this case. See 48 Fed.Reg. 39,376 (1983). DOE concluded that for seven of the eight appliances considered in the December 1982 and August 1983 rules, standards would not result in significant conservation and would not be economically justified. See id. at 39,376 . For the eighth, central air conditioners, DOE concluded that a standard would result in significant conservation but would not be economically justified. See id. at 39,402, 39 ,-406. DOE’s decisions were based only on technology available for appliances commercially sold in the United States by November of 1980. For reasons discussed below, DOE declined to assess prototypes and other advanced technologies, foreign models, or models that had appeared in the domestic market between 1980 and the issuance of final rules in 1982 and 1983. See id. at 39,379-80 ; see also infra at 76-100. These final rules ended federal consideration of standards for the eight appliances and preempted state efficiency regulation of those appliances. See supra note 7. DOE also decided that it was not required to prepare an environmental assessment or environmental impact statement for its *354 “no-standard standard” determinations. See 48 Fed.Reg. at 39,407-08; 47 Fed.Reg. 57,198, 57,216-17 (1982).

In Parts II-V, we consider petitioners’ arguments that the final rules are in violation of EPCA and unsupported by substantial evidence. Parts VI and VII discuss whether DOE allowed petitioners a reasonable opportunity to question DOE employees, and whether DOE was required to consider more thoroughly the environmental consequences of the final rules.

In reviewing DOE’s construction of EPCA, we first determine whether Congress “has directly spoken to the precise question at issue,” Chevron, U.S.A., Inc. v. Natural Resources Defense Council, Inc., — U.S. -, 104 S.Ct. 2778, 2781 , 81 L.Ed.2d 694 (1984); if so, we “must give effect to the unambiguously expressed intent of Congress,” id. 104 S.Ct. at 2781-82 (footnote omitted). “If Congress’ intent is not clear, however, the court ‘must conduct the “narrower inquiry into whether the [agency’s] construction was ‘sufficiently reasonable’ to be accepted by a reviewing court.” ’ ” Eagle-Picher Indus., Inc. v. United States Envtl. Protection Agency, 759 F.2d 922 , 927 n. 5 (D.C.Cir.1985) (quoting General Motors Corp. v. Ruckelshaus, 742 F.2d 1561, 1567 (D.C.Cir.1984) (en banc)) (further citations omitted).

Although DOE developed the rules under review through informal rulemaking, 14 EPCA expressly provides that the substantial evidence standard guides our review of factual findings. See EPCA § 336(b)(2). “When reviewing the policy judgments made by the Secretary, including those predictive and difficult judgment calls the Secretary is called upon to make, we will subject them to searching scrutiny to ensure that they are neither arbitrary nor irrational — in other words, we must determine whether ‘the decision is based on a consideration of the relevant factors and whether there has been a clear error of judgment.’ ” California v. Watt, 668 F.2d 1290 , 1302 (D.C.Cir.1981) (quoting Citizens to Preserve Overton Park, Inc. v. Volpe, 401 U.S. 402, 416 , 91 S.Ct. 814, 824 , 28 L.Ed.2d 136 (1971)); see Industrial Union Dep’t, AFL-CIO v. Hodgson, 499 F.2d 467, 472-76 (D.C.Cir.1974); cf. Weyerhaeuser Co. v. Costle, 590 F.2d 1011, 1024-28 (D.C.Cir. 1978).

II. DOE’s Definition of “Significant Conservation of Energy”

A. The Development of DOE’s Definition

EPCA instructs the Secretary not to issue standards for a type or class of covered product if he “determines, by rule, that the establishment of such standard will not result in significant conservation of energy.” ÉPCA § 325(b)(2). In the June 1980 proposed rules, DOE announced a two-part definition of significant conservation, both parts of which had to be met for a proposed standard to pass the test. First, a standard governing a product type or product classes within that type would have been required to result in national energy savings for the product type exceeding 840,000,000 kilowatt hours (2.867 X 1012 Btu) per year. Second, a standard would have been required to result in energy savings per unit over the product class involved exceeding 30 kilowatt hours (102,-390 Btu) per year. See 45 Fed.Reg. 43,976, 44,031-32 (1980); U.S. Department of Energy, Economic Analysis: Energy Efficiency Standards for Consumer Products § 6.3 (1980) [hereinafter cited as 1980 Economic Analysis Document], J.A. at 612-15.

DOE explained that this definition was drawn from figures in EPCA itself. Under EPCA as it was first enacted, the Federal Energy Administrator was required to pre *355 scribe energy efficiency improvement targets for the same thirteen products that under NECPA became subject to mandatory standards. See EPCA § 325(a)(1)(A), (a)(2), Pub.L. No. 94-163, 89 Stat. 871, 920, 924 (1975). Of those products, ten were given priority. See id. Congress demanded that the targets for those priority products be designed to achieve an overall improvement in efficiency of at least 20 percent, measured by comparing the aggregate efficiency of all models of the ten priority products manufactured in 1972 against the same figure for models manufactured in 1980. See EPCA § 325(a)(1)(B), 89 Stat. at 924. DOE also noted that section 325(a)(2) of EPCA, as amended by NECPA, authorized DOE to prescribe energy efficiency standards for appliances not named in section 322(a) if, among other conditions,

(A) the average per household energy use within the United States by products of such type (or class) exceeded 150 kilowatt hours (or its Btu equivalent) for any 12-calendar-month period ending before such determination [and]

(B) the aggregate household energy use within the United States by products of such type (or class) exceeded 4,200,-000,000 kilowatt-hours (or its Btu equivalent) for any such 12-calendar-month period

EPCA § 325(a)(2)(AMB).

DOE evidently reasoned that Congress must have thought significant conservation of energy was possible from standards governing appliances that consumed energy at the minimum levels specified in section 325(a)(2), or Congress would not have authorized DOE to consider standards for those appliances. Cf. 1980 Economic Analysis § 6.3 at 6.4, J.A. at 613. In light of former section 325(a)(1)(B), DOE settled on a 20 percent reduction in energy consumption for an appliance that just met the minimum consumption levels as a reasonable definition of “significant” savings. A 20 percent reduction in energy consumption for an appliance that consumes 4,200,000,-000 kilowatt hours nationally per year yields a savings of 840,000,000 kilowatt hours (2.867 x 1012 Btu) per year — the amount specified in the first part of the 1980 definition of significance. A 20 percent reduction in energy consumption for an appliance unit that consumes 150 kilowatt hours per year yields a savings of 30 kilowatt hours (102,390 Btu) per year — the amount specified in the second part of the definition. See 45 Fed.Reg. at 44,031-32. These levels, DOE announced, would screen out standards that produced only “marginal conservation,” id. at 44,032, and so fulfill the purposes of the significance requirement. 15

On April 2, 1982, the new administration issued proposed rules that completely repudiated the 1980 approach to defining significance. See 47 Fed.Reg. 14,424, 14,429-31 (1982). 16 DOE first announced that sav *356 ings would be assessed as significant or insignificant over the 19-year period 1987 to 2005, see id. at 14,426-27, rather than on the annual basis proposed in 1980. DOE's basic approach was to compare, for each product, three measures of the difference between projected energy consumption during that period if standards were not imposed and if they were imposed. DOE then examined those figures, which it viewed as gauging the savings attributable to standards, for “significance.” 17

DOE’s notice set forth its three measures of energy consumption and the level at which savings under each measure would be deemed significant. DOE argued that the overall goal of NECPA’s conservation programs was to reduce national dependence on imported oil. For this reason, a saving for each of the thirteen product types would be deemed significant only if that saving made “a significant contribution to reducing this Nation’s energy dependence on foreign nations.” Id. at 14,-429. According to DOE, the definition proposed in 1980 recognized as significant “savings so low as to be non-measurable in national reports,” id. at 14,429 n. 14, and was therefore not rigorous enough under DOE’s new approach. In place of the 1980 definition, DOE proposed three alternative tests for significant savings. 18 Under the first of these tests, savings were significant if they amounted to at least 10,000 barrels per day (bpd) of oil or an equivalent amount of natural gas. DOE later explained that this test was not satisfied if the combined savings of oil and gas equaled the equivalent of 10,000 bpd, but required that the full figure be met for one or the other of the two commodities separately. See 47 Fed.Reg. 57,198, 57,208 (1982).

DOE’s second test for significance was based on its finding that conservation of energy would also be “significant” if the electricity saved “would have a significant impact on the environment or significantly reduce the need for additional generating capacity.” 47 Fed.Reg. 14,424, 14,430 (1982). DOE noted that according to the environmental assessment accompanying the June 1980 proposed rules, those rules would not have had a significant effect on the environment. Id. at 14,430; see 1980 Environmental Assessment at 5-11, J.A. at 241. It concluded that “[bjecause the standards proposed in the June 30 proposal were predicted to create greater energy savings than the standards analyzed in this proposal, DOE believes that no single standard considered in today’s proposal could have a significant effect on the environment.” 47 Fed.Reg. at 14,430. DOE further decided that to have an effect on the need for new generating capacity, a national saving of electricity would have to exceed 1 percent of national usage. See id. That figure was therefore chosen as the second threshold level for significance.

Finally, DOE recognized that the first two proposed tests had “a built-in bias towards products that use a great deal of energy in the first instance.” Id. In particular, DOE noted that those tests demanded an annual savings greater than the total annual consumption of some covered products. Standards for these products could not have resulted in significant savings even if the standard reduced energy consumption by the product to zero. DOE therefore proposed a third test, purportedly based on former EPCA section 325, under *357 which savings would be significant if “the energy saved as a result of the standard, as measured by the ORNL Model, [would] be 20 percent of the energy that would be used by the product in the no-standard case.” Id.

These tests were hotly debated in the comments on the proposed rules, and DOE responded in its December 22, 1982 notice of final rules by making several changes. Specifically, DOE abandoned its earlier decision to calculate energy savings over the 19-year period from 1987 to 2005, on the ground that many of the covered products had very different useful lives. Covered products with different useful lives necessarily have different “penetration rates”— that is, the shorter the useful life of a covered product, the more quickly new appliances of that type will tend to account for a large proportion of all such appliances in use. The penetration rate of a covered product in turn affects how quickly more efficient models will influence national energy consumption. DOE therefore adopted the average life of a product, beginning in 1987, as the appropriate period for analyzing savings for that product under the first two tests for significance. See 47 Fed.Reg. 57,198, 57,203 (1982).

The third proposed test presented a special timing problem because, as several commenters informed DOE,

[S]ome products would not meet the test even if a standard required every new product sold to use no energy at all. This anomalous result occurred because so much of the energy being measured in the standards case would be energy consumed by products purchased before the standards were put into effect.

Id. at 57,208. For this reason, DOE settled on the “one-year period following the year of the retirement of the average product purchased immediately preceding the application of a standard,” id., as the year during which the third test was to be applied. For example, if standards were put into effect in 1987, the measuring year for an appliance with an average life of 18 years would be 2005. Products purchased in 1986 — the last year before standards were imposed — would on average wear out in 2004. Thus in 2005, according to DOE, relatively few pre-standard models would remain to skew the calculation. Finally, DOE noted that former section 325 of EPCA referred to a 20 percent improvement in efficiency, while DOE’s definition required a 20 percent decrease in energy use. In fact, however, a 20 percent improvement in efficiency mathematically results in only a 16.67 percent decrease in fuel use, and accordingly DOE substituted that lower percentage in its test. See id.

Although commenters voiced numerous other criticisms of DOE’s proposed definition of significance, DOE adhered to that definition with the revisions described above. Under DOE’s final definition, energy savings from a proposed standard were significant only if they met at least one of the following three tests: (1) “the standard would result in the saving of 10,000 bpd of oil or the saving of natural gas equivalent to 10,000 bpd of oil over the period of the average life of the product in question beginning with the year 1987,” id. at 57,209; (2) “the standard would result in the saving of one percent of national electricity use over the period of the average life of the product in question beginning with the year 1987,” id.; or (3) “the savings attributable to a standard for a product were equal to 16.67 percent of the energy that would be used by that product in the absence of a standard measured over the one year period following the period of the average life of the product purchased in the last year before the standard would be imposed,” id. at 57,209.

B. The Validity of DOE’s Definition

EPCA does not define “significant conservation of energy,” and DOE maintains that its interpretation of this statutory term is entitled to great deference. We agree that Congress left DOE with substantial discretion to set specific levels of significance, but no one disputes that the levels selected must be consistent with the express terms and underlying congression *358 al intention of the Act. We must, as the Supreme Court has directed, decide whether EPCA’s definition is “a reasonable choice within a gap left by Congress.” Chevron, U.S.A., Inc. v. Natural Resources Defense Council, Inc., — U.S. -, 104 S.Ct. 2778, 2793 , 81 L.Ed.2d 694 (1984). Answering that question requires us to determine whether, as DOE thought in 1980, Congress meant to exclude only “marginal” savings as insignificant, or as DOE thought in 1983, Congress licensed DOE to create so formidable an obstacle that it blocked standards for seven of the eight priority products at issue here. To conduct that inquiry, we first look to the structure and text of the Act and then examine the rationale DOE offered for its most recent position. We conclude that DOE’s definition of significant conservation is inconsistent with the congressional design reflected in NECPA.

We think the basic decisionmaking structure of the Act suggests that DOE has made a fundamental mistake. First, the elaborate NECPA provisions governing standards and their legislative history leave no doubt that Congress expected DOE to prescribe mandatory efficiency standards. Congress apparently intended the threshold significance determination only to weed out standards promising such negligible benefits that they were not worth the trouble of further examination. We note that DOE may not issue a standard it has disqualified under the significance provision even if that standard imposes absolutely no burdens at all. Consequently, even an efficiency standard with no technological or economic drawbacks whatever — one that offers a completely painless way to energy conservation — will be discarded if it fails to achieve “significant” conservation. If, on the other hand, the standard under consideration survives this first look, DOE must still decide whether the standard would be economically justified. In reaching that decision, DOE is explicitly directed to weigh “the total projected amount of energy savings likely to result directly from the imposition of . the standard.” EPCA § 325(d)(3). Thus, a finding that a proposed standard results in significant conservation is far from a prologue to inevitable promulgation of a mandatory standard; instead, that finding simply triggers a much more thorough review in which the amount of energy a standard would save is assessed in light of any other benefits and countervailing burdens of the standard.

We think it unlikely that the Congress that enacted NECPA and its four related energy statutes intended DOE to throw away a cost-free chance to save energy unless the amount of energy saved was genuinely trivial. 19 Moreover, that general view is bolstered by the few specific figures Congress included in the sections of NECPA that created the appliance program. In particular, review of the energy consumption level at which DOE was authorized to prescribe discretionary standards for an appliance is revealing.

As we note above, section 325(a) of EPCA permits DOE to prescribe an energy efficiency standard for an appliance if its average per-household energy consumption exceeds 150 kilowatt hours, and its aggregate national household use during one year exceeds 4,200,000,000 kilowatt hours, or .014335 quadrillion Btu (Quads). Congress plainly thought that saving some part of the energy consumed by an appliance operating at those levels would be *359 significant, or it would not have authorized DOE to prescribe mandatory standards for such an appliance. Our review of DOE’s figures suggests that standards for three of the appliances DOE considered in this rulemaking would have saved more energy per year than the total annual energy consumption necessary to qualify an appliance for discretionary standards. These standards, in other words, would have saved more energy than would have been saved if a standard instantly reduced to zero the total energy consumption of an appliance operating at the minimum level for discretionary standards. Yet for two of those three appliances, DOE branded the savings from standards insignificant, and thus rejected savings that exceeded 100 percent of consumption for appliances Congress was willing to regulate.

We must briefly sketch some technical background in order to illustrate this anomalous result. The consumption figures in section 325(a) measure energy use at the site of consumption, i.e., the energy actually consumed by an appliance in the household. However, the ORNL model calculates the energy a standard would save in terms of source consumption, or the energy used as input to powerplants that in turn generate electricity used in the household. Because of inefficiencies in the generation and transmission of electricity, a given number of Quads in site savings yield a substantially larger number of Quads in source savings.

Under section 325(a), DOE may prescribe a discretionary standard for an appliance if, among other criteria, its national energy consumption exceeds .014335 Quads. According to DOE’s figures, an appliance that consumes .that amount of electricity at the site of use consumes approximately .0483 Quads annually at the source of generation. 20 The table below compares DOE’s estimate of savings from standards for the six covered products considered in the 1983 final rules, calculated over the average lifetime of an appliance purchased in 1987, with the total consumption of an electrical appliance during that same period.

Table 1

Appliance Lifetime (years)

DOE-Estimated Savings (Quads)

Consumption of Appliance Qualifying for Discretionary Standards .014335 (Furnaces) or .0483 Quads (all others) times Lifetime

Refrigerators (Level 3) 19 1.65 0.92

Freezers (Level 3) 21 0.58 1.01

Water Heaters (Level 3) 13 1.18 (electricity only) 0.63

Furnaces 23 0.17 0.33

Room Air Conditioners (Level 4) 15 0.61 0.72

*360 Appliance Lifetime (years)

DOE-Estimated Savings (Quads)

Consumption of Appliance Qualifying for Discretionary Standards .014335 (Furnaces) or .483 Quads (all others) times Lifetime

Central Air Conditioners 12 1.23 0.58 (Level 4)

Source: EPCA § 325(a)(2)(B); 48 Fed. Reg. 39,376 , 39,385, 39,390, 39,392, 39,394, 39,396, 39,398, 39,402 (1983). 21

Thus, DOE’s definition does not recognize as significant energy savings equal to the total annual consumption of an electrical appliance that in the view of Congress might appropriately be subject to a mandatory standard, even when that savings is calculated according to DOE’s own figures and over the periods DOE itself selected for analysis. 22 We think that such a high definition of significance is inconsistent with the congressional decision to authorize discretionary standards for appliances con *361 suming the amounts of energy Congress specifically designated. Congress determined that discretionary standards for such appliances might he appropriate if they would result in “substantial” conservation of energy. Congress plainly thought that substantial conservation was possible for an appliance operating at the minimum level stipulated in the statute; and it therefore could not have thought that the total energy consumption of such an appliance might reasonably be considered an insignificant amount to save. 23

In the December 1982 notice, DOE responded to a somewhat different version of the argument we have presented in the following way:

DOE does not believe that because the oil or gas and the electricity savings tests would require greater savings of energy by the eight products subject to this rule than is required to be used by some other products in order to be possibly considered for standards means that these tests are unlawful or precluded by the statute. As indicated above, the oil or gas and electricity savings tests were designed with a bias to products that use large amounts of energy.

47 Fed.Reg. 57,198, 57,204 n. 11 (1982). DOE’s comment, however, is directed only to the first two of its three tests. It does not acknowledge that, as we have shown, a standard might save an amount of energy equal to the total consumption necessary for a discretionary standard and nonetheless fail all three tests. 24

More importantly, we think that DOE’s approach is inadequately sensitive to the relationship between different passages in section 325 of EPCA. In effect, DOE points out that a fixed amount in energy savings might be a very small percentage of total use for a high-consumption appliance and a much larger percentage for a low-consumption appliance. DOE concludes that it may therefore reasonably regard energy savings as insignificant for a high-consumption appliance, even if the same amount would be significant for a low-consumption appliance.

In the abstract, we have no occasion to disagree with that idea. For example, we do not hold that the Act forbids DOE to set levels of significance for each product type as a percentage of the energy consumed by that product type, provided that the levels selected reasonably accommodate the policies of the Act. Our point is *362 simply that DOE’s definition of significance must show some awareness of the range of energy savings Congress thought worth pursuing. Congress unmistakably believed that savings for an appliance that just qualified for discretionary standards might be worth pursuing, even though the absolute amount of energy saved would be small. Perhaps DOE need not say that precisely the same amount of energy such a standard might realistically save would be significant if saved by a high-eonsumption appliance. However, DOE has gone much further than that. Obviously, the total consumption of an appliance subject to discretionary standards is much higher than the amount of energy Congress could have expected such a discretionary standard to' save. Yet DOE has rejected as insignificant savings equal to the total minimum consumption of an appliance subject to discretionary standards.

As we have discussed, a determination that a standard does not result in “significant” savings means that DOE must discard the standard at a preliminary stage of decisionmaking as unworthy of detailed study. Moreover, DOE must reject the standard even if it imposes absolutely no costs on manufacturers, consumers, or anyone else. In our view, DOE’s definition rejects savings that, judged against figures Congress included in the text of the Act, must be regarded as significant. At least in the absence of strong countervailing evidence of congressional intent, we think the anomalous and disproportionate results of DOE’s definition are a powerful sign that DOE has passed beyond the limits of its statutory discretion. We turn, then, to the specific rationales DOE offers for its three tests to determine if they satisfactorily explain the strange consequences of those tests.

C. DOE’s Rationale for Its Definition

DOE claimed that the definition of significance announced in its December 1982 notice was based on congressional intent, although it had previously made the same argument for the radically different definition of significance advanced in June of 1980. In support of its first two tests, DOE cited the general findings in NECPA that stressed the importance of reducing national dependence on oil imports. See NECPA § 102(a), Pub.L. No. 95-619, 92 Stat. 3206, 3208 (1978); 47 Fed.Reg. 57,198, 57,205 (1982); 47 Fed.Reg. 14,424, 14,429 (1982). These broadly stated findings, however, describe overall goals for the national energy program from “all sectors of our Nation’s economy,” NECPA § 102(a)(3), 92 Stat. at 3209, to be “approached progressively in a step-by-step pattern to produce savings of a significant proportion.” National Energy Act, Part I: Hearings Before the Subcommittee on Energy and Power of . the House Committee on Interstate and Foreign Commerce, 95th Cong., 1st Sess. 2 (1977) (statement of Rep. Dingell). DOE could reasonably decide that a standard must make some progress towards the goals of the Act to be considered significant, but those general goals furnish little support for the specific tests DOE has adopted. Moreover, the appliance program is, of course, merely one program in a vast plan encompassing five statutes and scores of initiatives. Yet DOE steadfastly refused to consider that fact in deciding how great a contribution each standard must make to national conservation efforts in order to be deemed significant. Commenters echoed the view of Representative Dingell, Chairman of the House Sub-Committee on Energy and Power and a member of both of the House Ad Hoc Committee on Energy and the NECPA Conference Committee, that “conservation must be approached on a nickel and dime basis” and that “the cumulative impact of a series of conservation initiatives, which in themselves might appear insignificant, could be enormous,” id., but DOE was unmoved. It merely replied that the statute “makes clear that the determination of whether a standard will save significant energy is to be made on an individual product type basis, not on the basis that standards for all products together would save significant energy.” 47 Fed.Reg. 57,198, 57,203 (1982).

*363 DOE is right to think that under section 325, standards for each product type must result in significant conservation. But DOE is plainly wrong to think that the overall conservation possible under the appliance program, and indeed under other aspects of the 1978 national energy program, has no relevance at all to what qualifies as a reasonable definition of significant conservation. DOE itself relied heavily on the ultimate end Congress declared for its program: a gradual end to massive dependence on foreign sources of energy. But DOE cannot isolate that end from the means Congress selected — namely, a massive plan involving numerous programs. No single program was expected to be decisive in achieving Congress’ ultimate goal, let alone a single standard within the appliance program. In this sense, the cumulative savings possible from the appliance program as a whole is certainly relevant to whether the conservation that standards for a particular product type might achieve should be deemed significant. By rejecting that linkage, DOE has plucked the general goals Congress set for NECPA from their place in the statutory scheme, and failed to consider important evidence of how far standards for one product type should be expected to go towards achieving those goals.

DOE also relied on 1979 decisions by two congressional committees rejecting a contingency plan for restricting the use of lighting for. advertising during energy

shortages. See 47 Fed.Reg. 14,424, 14,429 (1982). The committees found that the plan would save only 4,400 barrels of oil per day, which they regarded as inadequate to justify enactment of the program. See S.Rep. No. 98, 96th Cong., 1st Sess. 10-11 (1979); H.R.Rep. No. 122, 96th Cong., 1st Sess. 4 (1979). An unfavorable recommendation by two congressional committees on a substantially different proposal after passage of the legislation at issue is weak support indeed for DOE’s test. Congress might well have thought that a single nation-wide emergency program should produce higher savings than the savings demanded for a standard governing any one of thirteen covered products, particularly since the appliance program was itself only part of the large-scale national conservation program embodied in the 1978 energy acts.. Moreover, DOE set its first test at 10,000 barrels of oil per day, or more than double the 4,400 barrels of oil per day that the rejected program would have saved. If stronger support for DOE’s definition cannot be found in congressional intent, this isolated, unrelated, and post-enactment example of legislative inaction cannot rescue it.

In addition, more direct evidence of congressional intent casts further doubt on the relevance of these committee decisions. The report on NECPA by the House Ad Hoc Committee on Energy predicted that the Act would result in the following savings:

Table 2

Estimate of oil import savings achieved by H.R. 8444

Thousands of barrels Initiative: per day, in 1985

Transportation 360

Buildings 420

Appliance Program 50

Industry/utility conservation 170

Increased residential/commercial gas availability 120

Oil and natural gas pricing 260

Synthetic use L, 420

Coal conversion 820 to 1,150

Total Savings 2,620 to 2,950

*364 H.R.Rep. No. 543, 95th Cong., 1st Sess. 80 (1977) (footnotes omitted). The House Report commented that the NECPA conservation programs, considered as a bundle, would “provide[] a solid start” toward reducing oil imports and “changpng] the way we use energy.” Id. Of those programs, Congress expected the lowest immediate return from appliance regulation, which was expected to save 50,000 barrels per day of imported oil in 1985 — less than one-half the benefit projected from the program with the next smallest return and less than 2 percent of the total oil import savings projected from NECPA during that year. 25 Similarly, the minority members of the House Committee on Interstate and Foreign Commerce commented that “appliances consume only 2.7 percent of the energy consumed annually in the United States.” H.R.Rep. No. 496, Pt. 4, 95th Cong., 1st Sess. 291 (1977) (minority views), U.S.Code Cong. & Admin.News 1977, p. 8679. 26 Yet those members supported the program. See id. Congress, in short, had a realistic notion of the relatively small energy savings to be expected from the appliance program', but nonetheless decided to regulate appliance efficiency as part of a national assault on unnecessary energy use.

DOE’s first test for significance, in contrast, requires that standards for each of the eight covered products considered in the rulemaking save 10,000 barrels of oil a day or an equivalent amount of natural gas, even though Congress projected a 1985 saving from standards for all thirteen appliances of only 50,000 barrels a day. That congressional estimate, moreover, describes the saving Congress thought probable, and not the smallest saving that justified detailed evaluation of proposed standards. In light of these statistics, we think DOE’s reliance on the obscure fate of quite a different program in committees of a later Congress is misplaced.

DOE justified its second test for significance on the ground that “a national savings of electricity would at least have to exceed 1 percent of national usage over the period of analysis in order to have an effect on the need for new generating capacity.” 47 Fed.Reg. 14,424, 14,430 (1982). DOE followed its statement of this test with the following table:

Table 3

Appliance 1980 Electricity Usage Electric usage _in quads Percent of total electric usage

Central space heater .0.96 3.8

Room space heater 0.62 2.5

Air conditioner room 0.38 1.5

Air conditioner central 1.47 5.9

Water heater 1.35 5.4

Refrigerator 1.17 4.7

Freezer 0.44 1.8

Range 0.54 2.1

Clothes dryer 0.45 1.8

*365 Id. (citing as source ORNL Residential Energy End Use Model) (footnotes omitted).

According to this table, each of the eight covered products identified in the table accounts for small percentages of national electricity consumption. Only a very dramatic drop in electricity consumption by a product would enable even the high-consumption covered products to meet DOE’s test, and evidently some of the low-consumption appliances could not meet the test at all. 27 We view as suspect a test that demands a fixed amount of energy savings greater than the savings that low-consumption covered products could ever possibly achieve. By including low-consumption appliances in the mandatory standards program, Congress showed that it believed improved efficiency for those products could result in significant conservation. A test for significance that demands far higher savings than standards for low-consumption ■ covered products could realistically ever produce is thus out of scale with the congressional view that these products should be included in the Act.

Moreover, our dubiousness about this test is heightened by DOE’s equivocal response to comments submitted during the rulemaking. DOE summarized those comments by noting that:

[S]everal commenters attempted to translate the aggregate electricity savings for all products using electricity, if standards were adopted for all those products, into additional new capacity that could be avoided. These estimates ranged from 16.8 GWe [gigawatts] ... to 17.6 ... GWe in the year 2005. Each of these commenters also reflected these. estimated avoided capacities in terms of powerplant equivalents, which ranged from 17 1000 MW [megawatt] plants to 22 800 MW plants.

47 Fed.Reg. 57,198, 57,207 (1982) (citations and footnote omitted). DOE first complained that the commenters had impermissibly aggregated total savings from standards covering all covered products. When these aggregations were broken down into the savings attributable to standards for each product type, DOE maintained that:

Under even [the] methodology [of commenter American Council for an Energy-Efficient Economy (ACEEE),] the largest reduction in need for new, incremental capacity attributable to one standard (electric water heaters) would be only 36 percent of the aggregate figure presented [for total electricity saved as a result of standards]. Under DOE’s base case projection, the largest reduction would be only 17 percent of the aggregate figure presented, or approximately 3 GWe or three 1000 MW powerplant equivalents nationwide. As discussed earlier, Section 325(b) requires DOE to make a determination of significant, savings on an individual product type basis, not on the basis of the aggregate savings if standards were imposed on all products.

Id. 28 As we have explained, we agree that savings for standards must be evaluated separately by product type, but DOE may not therefore close its eyes to the cumulative effect of imposing standards. We thus cannot agree that DOE’s first reason supports its rejection of the comments submitted.

DOE also argued that:

*366 [I]t is not at all clear that even a 17 to 18 GWe theoretical reduction of additional new capacity is significant in terms of actually affecting potential new, incremental capacity. The 17 to 18 GWe capacity savings projected by the commenters would be less than 2 percent of the approximate forecast capacity for the year 2005, and these savings would for the most part be spread over large parts, if not all, of the Nation. The reuslt of this is that in some utility services areas, the level at which decisions are made as to the need for new, incremental capaci-' ty, these marginal savings could be so small compared to other factors, e.g., demographic changes, that they would have no impact on actual construction decisions. Thus, one cannot reliably assume that whatever number of gigawatts one may mathematically back out as a result of reduced kilowatt hours attributable to a standard represents the actual gigawatts of capacity construction foregone.

Id. (footnote omitted). Part of this passage relies on the demographic uncertainty that is inevitably involved in projecting electricity use long into the future. However, that uncertainty results mostly from the lengthy period for analysis DOE itself chose, and characterizes many of DOE’s own predictions.

The quoted passage also argues that the savings in question would be spread out over the entire nation, and consequently those savings might not affect the need for additional generating capacity in any area. Petitioners maintain that the market in electricity is essentially a national one, and that incremental savings achieved across the nation really do diminish the need for new generating facilities. See Oral Comments of Angelo F. Orazio at 9 (May 20, 1982), J.A. at 3116. DOE did not details its reasons for rejecting that argument, and we therefore have some difficulty in assessing the reasonableness of the opposing : views in this subsidiary controversy. However, we are not now considering whether DOE’s general doubts about the effect of savings on the need for new generating facilities would, if DOE’s basic approach to significance were sound, support the particular quantities of energy DOE specified in its test. As we have discussed, DOE’s definition of significant savings produces incongruous results in light of other EPCA provisions, and we are now searching for an agency explanation that might prop up this faltering definition. DOE’s argument does not persuade us that DOE has, after all, acted within the discretion Congress allowed it.

Finally, DOE argued that its third test is grounded in former section 325 of EPCA. That section required the Federal Energy Administrator to set target figures for appliance efficiency improvement so that the ten priority appliances named in the Act would register an aggregate efficiency improvement of 20 percent between 1972 and 1980. Thus, unlike DOE’s third test, section 325 established a target, not a threshold figure below which regulation would be abandoned as not worth the trouble. That target required an aggregate improvement for all ten appliances of 20 percent, thus recognizing that standards for some covered products might fall short of that level. In addition, the 20 percent target was based on a comparison of 1972 efficiency to 1980 efficiency and so is quite unlike DOE’s comparison of its standards case with a projected no-standards base case. Given these major differences, we are unconvinced by DOE’s argument.

We think DOE’s argument is also at odds with other signs of legislative intent. The Senate Report on NECPA, for example, set forth the following table:

*367 Table 4

Estimated Improvements in Average Appliance Efficiency Over 1972 Levels (Percent)

Expected Expected Improvements Improvements From From EPCA Appliance President’s Program_Initiative

Refrigerators 27 47

Freezers 17 30

Clothes Dryers 5 ele[c]., 8 elec.,

10 gas 18 gas

Water Heaters 10 elec., 18 elec.,

14 gas 25 gas

13 oil 23 oil

Room Air Conditioners 15 54

Clothes Washers 15 20

Furnaces 5 23

S.Rep. No. 409, 95th Cong., 1st Sess. 147 (1977). The first column of the table indicates the improvement in efficiency the administration expected between 1972 and 1985 under EPCA as originally enacted. Id. The second column indicates the improvement in efficiency the administration expected during the same period under its proposal for mandatory standards, which in somewhat modified form became NECPA.

These projections were submitted to Congress by the Deputy Administrator of the Federal Energy Administration. He commented that under the administration proposal,

[mjinimum efficiency standards shall be prescribed as soon as possible for refrigerators/refrigerator-freezers, freezers, water heaters, room air-conditioners, kitchen ranges and ovens, central air-conditioners, and furnaces. These products account for about 80 percent of the energy used in American households. Standards for these products would be required because significant improvements in the energy efficiency of each is [sic] possible and a substantial national savings would result

Id. at 132 (statement of David J. Bardin) (emphasis added).

Because the table shows energy efficiency improvements as a comparison between 1972 statistics and 1985 projections, the figures it states are not directly comparable to those DOE used in calculating its 16.67 percent test. Nonetheless, the range in the table is instructive. For example, the table predicts that with standards, clothes washers would be 20 percent more efficient in 1985 than in 1972; without standards, the difference would be 15 percent. For electric clothes dryers, the figures are 8 and 5 percent; for gas clothes dryers, 18 and 20 percent. As these numbers show, Congress knew that standards for some covered products would produce quite modest incremental gains in efficiency and consequently in energy conserved. Congress was also authoritatively informed by the Federal Energy Administration — the predecessor agency to DOE in administering the appliance program — that the savings revealed by the table for each of the covered products were significant. On that understanding, Congress enacted the program.

We emphasize once again that we do not view the table we have reprinted above or any of the other legislative history we have discussed as absolutely binding DOE to any particular definition of significance. “[A]n agency to which Congress has delegated policy-making responsibilities may, within the limits of that delegation, properly rely on the incumbent administra *368 tion’s views' of wise policy to inform its judgments.” Chevron, U.S.A., Inc. v. Natural Resources Defense Council, Inc., — U.S.-, 104 S.Ct. 2778, 2793 , 81 L.Ed.2d 694 (1984). Nonetheless, when an agency does not reasonably accommodate the policies of a statute or reaches a decision that is “not one that Congress would have sanctioned,” id. 104 S.Ct. at 2783 (quoting United States v. Shimer, 367 U.S. 374, 383 , 81 S.Ct. 1554, 1560 , 6 L.Ed.2d 908 (1961)), a reviewing court must intervene to enforce the policy decisions made by Congress. “[A] new administration may not choose not to enforce laws of which it does not approve, or to ignore statutory standards in carrying out its regulatory functions.” Motor Vehicle Mfrs. Ass’n v. State Farm Mutual Auto. Ins. Co., 463 U.S. 29 , 103 S.Ct. 2856 , 2875 n. *, 77 L.Ed.2d 443 (1983) (Rehnquist, J., concurring in part and dissenting in part).

We have acknowledged that Congress granted DOE considerable discretion to define significant conservation. This case does not require us to define the precise limits of that discretion: all we must decide is whether DOE has reasonably enforced the decisions Congress made when it enacted NECPA. For the reasons stated above, we believe DOE has failed to do so, and we accordingly hold that its definition of significance is inconsistent with the Act.

III. DOE’s Method of Determining Savings That Would Result From Standards

Petitioners launch two attacks on DOE’s method of determining when “the establishment of [a] standard will not result in significant conservation of energy.” EPCA § 325(b)(2) (emphasis added). First, they claim that DOE should not have calculated savings by comparing a standards case, which projected future energy consumption by particular appliances if standards were imposed, with a base case, which projected the same statistic in the absence of standards. Instead, petitioners assert that DOE should have “simply assessed] how much energy would be conserved if existing appliances were improved to comply with minimum efficiency standards.” Br. of Petitioners NRDC et al. at 54. As we understand this claim, petitioners argue that DOE should have estimated energy consumption by a product type or class over some future period of years if appliance efficiency remained at current levels, and should then, have subtracted from that figure projected energy consumption over the same period if standards were imposed. Petitioners do not really avoid the need for some base case to use in determining the amount of energy that standards would conserve; in effect, they dispute only DOE’s assumption that appliance efficiency will improve even in the absence of standards.

Second, petitioners challenge DOE’s use of an econometric computer model called the Oak Ridge National Laboratory (ORNL) Residential Energy End-Use Model to project future energy consumption under both the base and standards cases. Among the ORNL model’s assumptions is that as the cost of energy to consumers increases, consumers will become much more likely to purchase efficient appliances. Because the data fed into the ORNL model predicted sharp increases in the cost of energy, the assumed relationship between energy cost and consumer behavior caused the model to predict that even in the absence of standards, appliances would become much more efficient in the immediate future. That prediction substantially lowered projected energy consumption under the base cases for all the product types or classes considered in the rulemaking, and thus also lowered the figures for energy savings that would result from standards. Overall, the ORNL model predicted that for the period 1987 to 2005, appliance efficiency standards would save 32.8 Quads, as compared to the energy that appliances at 1978 levels of efficiency would consume. See 47 Fed.Reg. 14,424, 14,427-28 (1982). But the model also predicted that even in the absence of standards, improved appliance efficiency caused by market forces such as increased energy prices would save 27.6 Quads. See *369 id. Thus, the model indicated that 84 percent of the energy that more efficient appliances would save under standards compared to present efficiency levels would have been saved anyway; only the remaining 16 percent of energy savings would result from standards.

We examine these two challenges in turn.

A. Statutory Authority for Subtracting “Base-Case” Savings From “Standards” Savings

The essence of DOE’s position is straightforward: energy savings do not “result” from standards if the same savings would have been achieved without standards. We think it clear that nothing in the ordinary meaning of the statutory language renders DOE’s interpretation unreasonable. Petitionérs, however, rely on the legislative history of section 325 for their contrary interpretation. They point out that under EPCA as originally passed, Congress relied mainly on a manufacturer’s voluntary compliance with target efficiencies, coupled with a labeling program to encourage informed consumer decisions, in order to increase appliance efficiency. Mandatory standards were authorized only if, after actual experience with these programs, DOE found that they were “not likely to be sufficient” to achieve the target efficiencies. NECPA, in contrast, called for the immediate issuance of standards in place of the old program, which employed standards only after milder government intervention had failed. According to petitioners, this shift shows that Congress affirmatively rejected further experimentation with the chance that in the absence of mandatory standards, appliance efficiency would nonetheless improve. 29 On this theory, DOE was no longer to consider whether market forces and labeling would achieve the desired improvements; instead, DOE was commanded to require those improvements at once, whether or not they might occur anyway.

However, the legislative history does not bear out petitioners’ account of Congress’ intent. The House Report explains succinctly what Congress hoped NECPA would accomplish:

First, the cornerstone of the national energy plan is conservation, and the proposal permits [the Federal Energy Administration] immediately to begin the process of establishing energy efficiency standards. If the present program were to continue without change, FEA would be permitted to set standards for a type of appliance only after a target had been established, energy labels had been in place for 18 months, data had been collected on industry performance, and there was evidence that targets were not being achieved. At the May 11,» 1977, hearings, the FEA contended that this process “could delay [for] some 2 to 3 years the imposition of standards that will guarantee the attainment of technologically feasible and economically justifiable energy improvement targets.”

The second purpose of utilizing minimum standards, as opposed to industry targets, is to place each manufacturer on an equal footing; all would be required to achieve the same energy efficiency. Under the target approach, there would be little incentive by a manufacturer to exceed a target, and to do so might place a given manufacturer at a competitive disadvantage.

The third purpose of the proposal is to eliminate uncertainties among manufacturers under the present target-setting *370 approach. The uncertainty in the present system arises from the fact that targets are industry-wide and a particular manufacturer must decide on what its contribution to the target should be. If the manufacturer chooses to exceed the target, he must consider the long-term effect on his company. He has no certainty over whether his appliances will ensure that the target is met, or whether standards will eventually be established.

H.R.Rep. No. 496, Pt. 4, 95th Cong., 1st Sess. 45 (1977), U.S.Code Cong. & Admin. News 1978, p. 8492. This passage certainly suggests congressional impatience with the delays that the wait-and-see approach of EPCA as originally enacted had caused. Congress plainly did not want DOE to delay the issuance of standards while it observed the effects of the target and labeling programs. However, neither NECPA nor its legislative history suggests that DOE may not project future gains in appliance efficiency and act on those projections in framing its method for deciding what savings would result from standards. 30

We conclude, therefore, that DOE’s use of projected gains in appliance efficiency to calculate energy savings resulting from standards was not unlawful. We must therefore consider whether the computer model that generated those projected gains — and the factual assumptions driving that model — were reasonable tools to carry out DOE’s general approach to figuring energy savings.

B. The ORNL Model

As we have recently reaffirmed, “[a]n agency may utilize a predictive model so long as it explains the assumptions and methodology it used in preparing the model. If the model is challenged, the agency must provide a full analytical defense.” Eagle-Picher Indus., Inc. v. United States Envtl. Protection Agency, 759 F.2d 905 , 921 (D.C.Cir.1985) (footnotes omitted). However, we will defer to an agency's judgment to use a particular model if the agency examines the relevant data and articulates a reasoned basis for its decision. See id. at 921-22; Small Refiner Lead Phase-Down Task Force v. Environmental Protection Agency, 705 F.2d 506 , 535 (D.C.Cir.1983); Sierra Club v. Costle, 657 F.2d 298, 332-33 (D.C.Cir.1981).

The ORNL model was originally “designed to produce an analytical tool which would assist decision-makers in their evaluation of the impact of conservation strategies and policies.” 1982 Economic Analysis Document at 396 (app. E), J.A. at 1333. The model was not developed specifically for this rulemaking: government-supported scientists began work on the model in 1975, before passage of NECPA, and ■ an initial version of the model was completed in 1976. See id. Numerous changes in the model have been made since then, including changes made as a result of comments received during this rulemaking. See 48 Fed.Reg. 39,376, 39,380 (1983); see also McMahon, Residential End Use Demand Modeling: Improvements to the ORNL Model in Beyond the Energy Crisis 297 (1981).

DOE used the ORNL model to compare “the energy used without a standard during [the periods selected by DOE] to the energy used with a particular standard during the same period,” 48 Fed.Reg. at 39,-380, and thus to compute “the amount of energy saved during that period which is attributable to the standard,” id. Petitioners attack the model’s market penetration algorithm. This algorithm quantified DOE’s assumption that as fuel prices go *371 up, market distortion — that is, the tendency of consumers to make economically irrational decisions to buy inefficient appliances — will go down. See generally E. Hirst & J. Carney, The ORNL Engineering-Economic Model of Residential Energy Use 37-44 (1978). DOE expressed the overall costs of an appliance to a consumer as the “life cycle cost” of the appliance, which included “total purchase price, energy use over the life of the product, and the maintenance costs over the service life of the product, discounted to their present values.” 45 Fed.Reg. 43,976, 43,984 (1980). As DOE explained,

The Model recognizes that the existing market-place is not perfect; that is, all manufacturers do not make and all consumers do not purchase the type of equipment that would always be a consumer’s most energy efficient choice from a life cycle cost standpoint alone. The Model calculates the difference between the optimal and the projected average life cycle costs of equipment in a base year from actual prices and sales information of equipment in that year and projected energy prices. The difference between the optimal and projected life cycle costs is defined as the market distortion. In years after the base year, the Model projects the level of market distortion based on the initial level as modified by the market penetration algorithm.

48 Fed.Reg. 39,376, 39,380 (1983) (citation and footnote omitted). Thus, market distortion during the base year is the difference between the life cycle cost the average consumer will in fact ultimately pay, and the life cycle cost an economically rational consumer — that is, one seeking to minimize life cycle cost — would have paid. These figures are calculated from historical data. The market penetration algorithm in the ORNL model then uses the base-year market distortion to project market distortion in subsequent years.

In general terms, the algorithm has the effect of diminishing the distortion between the optimal and the average life costs of equipment being purchased as a function of rising energy prices. In other words, as the price of fuel for an appliance rises, the market in that appliance will become more “rational,” i.e., the efficiency of equipment being purchased will be closer to the optimal efficiency (from a life cycle cost perspective) than it was when the fuel price was lower. Under the Model’s algorithm, however, distortion always remains. While higher prices reduce distortion over time, they can never eliminate it altogether.

Id.

The controversial assumptions underlying the market penetration algorithm go well beyond the proposition that consumers will respond to higher energy prices by buying more efficient appliances. DOE explained this as follows:

The ORNL Model assumes that as energy prices increase, more efficient equipment will be produced and purchased. This is consistent with almost all analyses of the subject. However, the model goes beyond many other studies in making the assumption that the consumer criteria for selecting energy efficiency will change as energy prices increase. In effect, the model assumes that longer paybacks on investments in improved efficiency of equipment will be acceptable as energy prices rise.

1982 Economic Analysis Document § 5.3.-2.6 at 144 (emphasis in original), J.A. at 1081.

Until the August 1983 rulemaking, DOE also described this second assumption through the concept of an “implicit discount rate.” See infra at 1390 (noting technical difficulties with DOE’s use of implicit discount rates). An implicit discount rate describes “all of the various factors influencing the market response to higher energy prices.” 1982 Economic Analysis Document § 5.3.2.6 at 144, J.A. at 1081.

A high implicit discount rate implies that future benefits are given relatively little value by the consumer [in making purchase decisions] (or, alternatively, that *372 the availability of efficient products in the market is limited). A low implicit discount rate indicates a relatively high value placed on future benefits and a high availability of efficient products. The reason for the relationship between the implicit discount rate and the energy efficiency of new products is that an initial investment in greater efficiency is returned in the future. Thus, the greater the value placed on future benefits, the greater the investment in efficiency improvements, and the lower the implicit discount rate.

Id. “An implicit discount rate of 100 percent means that [the average consumer demands] a full return on the investment ... within the first year [after] purchase.” 1980 Economic Analysis Document § 4.2.4 at 4-9, J.A. at 481.

Even assuming that the implicit discount rate remained the same as energy prices increased, consumers would nonetheless purchase more efficient appliances. Suppose, for example, that the implicit discount rate for an appliance is 100 percent, and consumers thus demand a full return on efficiency-improving technology in one year. As fuel prices increase, the cost of operating an appliance during the first year of ownership also increases. The average consumer will therefore be willing to pay more — up to the increased cost of operation during the first year — for a more efficient appliance, even at the same implicit discount rate. The algorithm assumes, however, that the average consumer will spend not only amounts equal to the increased cost of operation over one year on more efficient appliances, but that the consumer will also be willing to accept a longer payback period. The implicit discount rate will consequently decline.

The market penetration algorithm, then, can be seen as reflecting three increasingly specific propositions about the consequences of higher fuel prices on the market. First, the algorithm assumes that consumers will purchase more efficient appliances; second, it assumes that consumers will accept longer payback periods, and market distortion will therefore diminish; and third, it quantifies the expected change in market distortion in a specific mathematical formula.

Some of the petitioners challenged the first of these propositions during the rule-making. See Comment No. 2121 at 24 (June 18, 1982) (NRDC), J.A. at 2758. In reply, DOE pointed to technical literature suggesting a general correlation between the price of energy and appliance efficiency, the results of consumer surveys, and anecdotal evidence from manufacturers and trade associations. See 48 Fed.Reg. 39,376, 39,380-81 (1983).

DOE also noted that between 1972 and 1981, in a time of rapidly increasing energy prices, appliance efficiency improved significantly. Petitioners suggested during the rulemaking that a combination of state-mandated efficiency standards, the threat of federal standards, and “increases in efficiency as a by-product of other design objectives,” id. at 39,381, accounted for efficiency improvements during that period. DOE replied at some length in its August 1983 notice. In sum, DOE concluded (1) that commenters had not identified any efficiency-improving design changes that appliance manufacturers introduced as a “byproduct” of other design objectives; (2) that while federal and state regulatory efforts may have contributed to more efficient appliances, market pressures might well have caused similar efficiency improvements without actual or threatened regulation; and (3) that the pattern and timing of improvements in appliance efficiency in the past were generally consistent with DOE’s belief that market pressures, particularly increases in energy prices, were principally responsible for those improvements. See id. at 39,381-82 (1983). Rather than reciting all the evidence underlying this dispute, we merely note our conviction that the evidence was “fragmentary,” Comment No. 2121 at 10 (June 18, 1982) (NRDC), J.A. at 2744, conflicting, and ultimately susceptible of different interpretations. DOE’s notices acknowledged those problems, reviewed the record in *373 light of major criticisms voiced by commenters, and offered a reasoned defense of the agency’s conclusions. We therefore uphold DOE’s conclusion that rising energy prices tend to encourage production of more efficient appliances.

We pass, then, to objections focused on DOE’s theory that market distortion decreases as energy prices rise. DOE commented that it found only one empirical study directly relevant to the effect of energy prices on market distortion. DOE initially summarized the findings of the study as follows:

The only available information regarding changes in implicit discount rates as a function of energy price is an analysis of investment in thermal integrity for new houses from 1973 to 1979____ This analysis, based on survey information from 300,000 new houses and computer analysis of the energy use of the houses, shows almost no change in implicit discount rates for new houses with electric resistance heating and a relatively small (15 to 20 percent) reduction in implicit discount rates for gas-heated houses during the time period. As the 1973 data preceded the oil embargo and rapidly rising real energy prices in the residential sector during the middle 1970s, the apparent small reduction in implicit discount rates during this period suggests that the decisionmaking process on thermal integrity in new houses changes only slowly. Higher energy prices, induce higher investment in energy efficiency improvements in the shells of new houses, but they appear not to change the decision criteria.

1982 Economic Analysis Document § 5.3.-2.6 at 145, J.A. at 1082 (citation omitted). The last sentence of this passage directly challenges the assumptions of the market penetration algorithm: DOE interpreted the study to suggest that higher energy prices “appear not to change the decision criteria” of consumers, while the algorithm assumes “that the consumer criteria for selecting energy efficiency will change as energy prices rise.” Id. at 144 (emphasis in original), J.A. at 1081.

DOE’s interpretation of the thermal integrity study changed as the rulemaking progressed. In the April 1982 notice, DOE decided that the study “involved circumstances sufficiently dissimilar to render its results inconclusive in the appliance context.” 47 Fed.Reg. 14,424, 14,428 n. 13 (1982) (citation omitted). By the August 1983 notice, DOE had yet a different view:

The study indicated that, while investment in thermal integrity in electrically heated homes showed little change in implicit discount rates over the period 1973-1979, the implicit discount rates for those investments in gas-heated homes declined 15 to 20 percent. Inasmuch as the real price of gas increased almost five times as rapidly during the period as that of electricity, and the initial implicit discount rates for electricity were only a fraction of the initial implicit discount rates for gas, such a differential in changes in implicit discount rates between gas and electricity is explainable. While DOE tentatively characterized this reduction in implicit discount rates as relatively small in the April 1982 proposal, upon further analysis DOE believes such a characterization is inapt. First, [tjhermal integrity improvements are almost entirely made by “third party” builders of new housing, and therefore market imperfections in this type of investment would tend to be greater than for products directly purchased by the ultimate consumer. Second, there is no labeling program for such improvements. Third, such reductions in discount rates are not inconsistent with those projected for some appliances by the ORNL model over the same period of time. More importantly, upon reconsideration of the tentative conclusion to be derived from this study [contained in the passage quoted above from the 1982 Economic Analysis Document], DOE has determined that the results of the study support the concept of the reduction of market distortion as a result of rising energy prices.

*374 48 Fed.Reg. 39,376, 39,382 n. 25 (1983). DOE thus finally settled on the view that the study, “while certainly not conclusive, does suggest that higher fuel prices may lead to less market distortion.” Id. at 39,-382 (citation and footnote omitted).

DOE’s analysis of a technical engineering study is entitled to great deference from judges, who are hardly equipped to match the expertise of DOE’s scientists. Although we are concerned by DOE’s shifting position as the rulemaking proceeded and the market penetration algorithm came under increasing attack, we are unable to say that DOE’s use of the thermal integrity study was arbitrary or irrational. Whether the 15 to 20 percent decline in the implicit discount rate for gas-heated homes was significant and what weight should be given to reasons why that discount rate might be particularly resistant to changes were questions addressed to the informed judgment of DOE scientists and policymakers. We cannot hold that DOE acted unreasonably in concluding that the study furnished some support, although concededly not unequivocal or decisive support, for the factual assumptions embedded in the market penetration algorithm.

DOE also believed that high energy prices would tend to counteract some major causes of market failure in the appliance market. For example, appliances are often purchased by landlords, builders of new homes, and other “third-party purchasers” who will not pay the fuel bills. Such purchasers have an incentive to minimize the initial purchase price by buying initially inexpensive but inefficient appliances with high life cycle costs. DOE believed that as energy prices went up, consumers would be more likely to demand energy-efficient .housing units. These market pressures might consequently work to correct an important source of market distortion. 31 Similarly, DOE acknowledged evidence that at least for some appliances, its labeling program had been only modestly effective. 32 In DOE’s view, higher energy prices would spur more consumers to consult the labels and thereby diminish market distortion. 33

DOE sought to confirm these general views and to verify the algorithm’s specific formula for declining market distortion by testing the algorithm against past experience. As the designers of the ORNL model conceded, the algorithm’s formula was initially “ad hoc; it lacks a theoretical basis and empirical validation.” E. Hirst & J. Carney, The ORNL Engineering-Economic Model of Residential Energy Use 71 (1978). But cf. id. at 44 (although further efforts were required to validate algorithm, it yields “results that are intuitively plausible”). However, DOE’s tests provided *375 inconclusive for at least two reasons. First, DOE discovered that it could not obtain the market information required by the algorithm for any year before . 1978. DOE attempted to estimate the necessary figures, but it was concerned that incorrect estimates might bias its tests. Second, DOE’s efforts to verify the algorithm uncovered basic difficulties in DOE’s procedures:

[T]he scientists who developed the test designed it to measure changes in implicit discount rates, rather than changes in market distortion. In the course of making initial test runs, however, DOE discovered that under certain circumstances relatively minor changes in market distortion could result in drastic changes to the implicit discount rate. The clear implication of this was that implicit discount rates were not a good shorthand for market distortion, and that the test of the algorithm, if run in this manner, would be virtually useless because major deviations in the projected implicit discount rates from “actual” rates would not establish that the algorithm was inaccurate. Finally, in the course of the initial test runs and a general review of the ORNL Model, it was determined that a problem existed in the ORNL Model’s life cycle cost-efficiency curve, which has the effect of tending to exaggerate the market distortion in future years. That is, separate from the disputed formula that reduces market distortion as a function of rising energy prices, the algorithm has a bias toward a larger than appropriate market distortion.

48 Fed.Reg. 39,376, 39,383 (1983) (footnotes omitted). DOE’s scientists believed that lengthy study would be required to solve these problems. See id. at 39,383 & n. 30. Because of “the time that would be required for the analyses, the doubtful usefulness of the conclusions that would be reached, and the limited effect of the market penetration algorithm” on the results of the rulemaking, id. at 39,383 (footnote omitted), 34 DOE decided not to delay the rulemaking while its scientists studied the algorithm further. However, DOE noted that it continuously reviews the ORNL model and market penetration algorithm for purposes unconnected with this rule-making. DOE declared that if it discovered flaws in the model that might have affected the final rules under review, it would immediately reconsider the final rules. See id. at 39,384. No such reconsideration has been announced.

DOE’s efforts to verify the algorithm empirically thus proved disappointing. Moreover, the remaining support for any general correlation between market distortion and fuel prices, let alone the specific relationship described in the algorithm, was largely conjectural and anecdotal. Commenters pointed out these shortcomings at length, but they did not show that any other method of predicting future market distortion — including the assumption that it would remain stable — would produce demonstrably more accurate predictions. In short, DOE attempted to solve the difficult predictive problem before it through use of an independently created econometric model which had received favorable notice in technical literature. See id. DOE responsibly addressed alleged defects in the model by changing the model or explaining why the defects were both extremely difficult to fix and of relatively minor moment to the rulemaking. Our comments about review of a computer model in a previous case apply here as well:

*376 The safety valves in the use of such sophisticated methodology are the requirement of public exposure of the assumptions and data incorporated into the analysis and the acceptance and consideration of public comment, the admission of uncertainties where they exist, and the insistence that ultimate responsibility for the policy decision remains with the agency rather than the computer. With these precautions the tools of econometric computer analysis can intelligently broaden rather than constrain the policymaker’s options and avoid the “artificial narrowing of options that [can be] arbitrary and capricious.”

Sierra Club v. Costle, 657 F.2d 298, 384-35 (D.C.Cir.1981) (footnote omitted).

In this case, the uncertainties underlying modeling assumptions were perhaps greater, and the factual basis for the assumptions correspondingly scanter, than in many others. But DOE’s interpretation of the statute required it to hazard some predictions of market distortion, and at the time the final rules were issued the empirical foundation for any such predictions would evidently have been insecure. Under the principles of Sierra Club , we therefore decline to reject DOE’s use of the ORNL model to predict energy savings, although DOE must change its unlawful definition of “significant” savings. Our decision does not, of course, mean that in the future DOE may continue to rely on the market penetration algorithm and the ORNL model if further study in light of more complete information shows the model’s predictions to be unreliable. 35 Cf. National Lime Ass’n v. Environmental Protection Agency, 627 F.2d 416, 454 (D.C.Cir. 1980). “Recognizing that policymaking in a complex society must account for uncertainty, ... does not imply that it is sufficient for an agency to merely recite the terms ‘substantial uncertainty’ as a justification for its actions.” Motor Vehicle Mfrs. Ass’n v. State Farm Mutual Life Ins. Co., 463 U.S. 29 , 103 S.Ct. 2856, 2871 , 77 L.Ed.2d 443 (1983); cf. 1 Davis, Administrative Law Treatise §§ 6:13, :16 (2d ed. 1978). DOE may resolve even substantial factual uncertainties in the exercise of its informed expert judgment; but it may not tolerate needless uncertainties in its central assumptions when the evidence fairly allows investigation and solution of those uncertainties.

IV. Maximum Technological Feasibility

A. The Statute

EPCA requires that DOE “shall determine the maximum improvement in energy efficiency that is technologically feasible for each type (or class) of covered products.” EPCA § 325(0(3). DOE’s standards “shall be designed to achieve the maximum improvement in energy efficiency which the Secretary determines is technologically feasible and economically justified.” EPCA § 325(c). If DOE’s standard is not set at the maximum technologically feasible level, DOE “shall state in the proposed rule the reasons therefor.” Id. § 325(i)(3). Finally, if the establishment of a standard is not technologically feasible or economically justified, the Secretary is directed not to issue a standard. See id. § 325(c).

The statute thus establishes a clear decisionmaking procedure. DOE must first identify, for all product types or classes, *377 the maximum improvement in energy efficiency that is technologically feasible. If a standard at that level would be economically justified, DOE must set the standard there. If a standard requiring the maximum technologically feasible level would not be economically justified, DOE . must set the standard at the highest level that is both technologically feasible and economically justified. See id. § 325(b)(2), (c). 36 In that event, EPCA requires DOE to explain specifically why a standard achieving the maximum technologically feasible improvement in efficiency was rejected. See id. § 325(i)(3).

The legislative history is clear on the meaning of maximum technological feasibility. The NECPA Senate report declared that “ ‘[feasible’ is defined in the conference report accompanying [EPCA as originally enacted] to mean ‘capable of being carried out.’ The. committee agrees with this definition.” S.Rep. No. 409, 95th Cong., 1st Sess. 63 (1978). Thus, the phrase “technologically feasible” in NEC-PA carries over the meaning of the phrase as it was used in the original EPCA requirement that labeling rules be technologically feasible. The EPCA conference report quoted by the NECPA Senate report declared that “[t]he term feasibility is used in the strict sense, namely, ‘capable of being carried out.’” S.Rep. No. 516, 94th Cong., 1st Sess. 172 (1975), U.S.Code Cong. & Admin.News 1975, p. 2014. Similarly, the House report on EPCA noted that a “determination of ... technological infeasibility is intended to provide the narrowest possible grounds for refusing to prescribe a labeling rule.” H.R.Rep. No. 340, 94th Cong., 1st Sess. 97 (1975), U.S.Code Cong. & Admin.News 1975, p. 1859.

B. DOE’s Treatment of Maximum Technologically Feasible Levels

In the June 1980 proposed rules, DOE set forth the following general approach to maximum technological feasibility:

DOE believes that for 1986, product designs in prototype form [by] January 1980 represent the maximum technologically feasible products. Although there will be advances in development beyond some of these levels over the next 6 years, DOE believes that these developments could not be demonstrated as technologically feasible in that the long lead times would not allow such products to be introduced into the market by 1986.

Even though manufacturers have prototypes in the January 1980 time frame, it is not practical to expect a complete product line transformation considering product development cycle of engineering, prototype fabrication, laboratory testing, field testing, production engineering and production facility modification. As a result it is not realistic to expect that in 1986 the manufacturer would be able to produce only those products whose energy efficiency is equal to or greater than the maximum technologically feasible. Consequently, DOE has determined that it is not economically justifiable to propose energy efficiency standards at the maximum technological limit.

45 Fed.Reg. 43,976, 43,984 (1980). For these reasons, DOE announced that the “design options considered [as a basis for standards] were limited to those based on ‘available’ technology, defined as those technologies presently implemented in units available in the marketplace.” Id.

These quotations reflect two crucial findings for this rulemaking. First, DOE determined that designs in prototype form by January of 1980 could be brought to market by 1986, when final standards were expected to take effect. DOE accordingly based its proposed determinations of maximum technologically feasible improvements on prototypes, as well as on design options actually used on appliances that were commercially available by January of 1980.

*378 Second, DOE determined that manufacturers could not reasonably be expected to use any design option available only as a prototype in 1980 on all appliances within a product type or class. A standard would, of course, be mandatory for all products of the type or class to which the standard applied. DOE therefore concluded from its second finding that in formulating standards it should consider only technologies included in appliances actually sold in 1980. In effect, DOE declared that all design options available only as prototypes in 1980 were economically unjustified as the basis for 1986 standards. That general rule applied regardless of the individual characteristics of the prototype design, the time needed to include any particular prototype design in a specific line of products, or the product type or class involved.

The June 1980 notice proposed specific maximum technologically feasible efficiency levels for each product type or class based on both prototypes and commercially available design options. These levels were expressed as measurements of the maximum efficiency attainable by the various products, as gauged by DOE-approved test procedures. See, e.g., 45 Fed.Reg. 43,-976, 43,986 (1980) (Table 4.1-3); id. at 43,-989 (Table 4.2-3). DOE also identified efficiency levels for consideration as standards, based only on efficiencies attainable by commercial available design options. Finally, DOE examined those possible standards for economic justification, and proposed standards for all product types that, in the opinion of DOE, were technologically feasible and economically justified, and produced significant conservation of energy. See supra note 15.

In April of 1982, DOE published a new notice of proposed rulemaking. See 47 Fed.Reg. 14,424 (1982). In it, DOE followed the basic approach of the June 1980 proposed rulemaking: the 1982 notice proposed efficiency levels for consideration as standards based on design options that were commercially available in 1980, and then considered whether standards at those levels would be economically justified and conserve significant energy. However, DOE’s 1982 reexamination of the technical information supporting the 1980 rule produced lower levels as candidates for standards. 37 DOE also radically revised its analysis of economic justification and significant conservation and determined that no standard for any covered product saved significant energy and was economically justified.

The June 1980 notice had at least proposed a maximum technologically feasible 1 level, although it immediately rejected standards at that level. The April 1982 proposed rulemaking, however, did not even go that far: it simply failed to acknowledge or comment on the statutory requirement that DOE determine maximum technologically feasible levels. The final rules announced in December of 1982 and August of 1983 also neglected to adopt, discuss, or reject DOE’s 1980 determinations of maximum technologically feasible levels or to consider standards based on those levels.

In the August 1983 rulemaking, however, DOE did try to defend its refusal to consider standards based on all technologically feasible design options against objections by NRDC, Consumers Union, and others. First, DOE asserted that the economic information necessary to analyze prototype models and models available only in foreign markets as the basis for standards was simply unavailable. See 48 Fed.Reg. 35,376, 39,379 (1983). DOE also noted that EPCA does not permit DOE to issue a standard for a product type or class unless an approved test procedure for that product type or class existed. DOE conjectured that existing test procedures might be inadequate to evaluate prototype models, although it did not specifically determine that any or all of the prototypes described *379 by commenters could not be tested using those procedures. See id. Finally, DOE also argued that it did not have enough time to analyze technology that became commercially available after November 14, 1980 in developing rules issued during 1982 and 1983. See id.

DOE also used other broad rules governing its consideration of design options as a basis for standards. For example, in 1980 DOE decided not to consider design options that were likely to require a significant increase in the manufacturer’s engineering staff. DOE also preferred energy-saving design options that manufacturers could introduce on their appliances as quickly as possible over other options that might arguably have resulted in greater energy conservation. See 45 Fed.Reg. 43,976, 43,-983-84 (1980).

In addition, in 1982 DOE overturned one of its 1980 decisions in a way that narrowed the design options considered as a basis for standards still further. DOE initially intended to evaluate all options with a “payback period” shorter than the average life of the appliance. See id. at 43,983. The payback period for an energy-saving design option is the length of time it takes an average consumer to recover the higher purchase price the consumer paid to obtain the design option through the lower cost of energy used to operate the appliance. However, in 1982 DOE decided it would generally confine its analysis to design options with a payback period of less than 5 years. See 1982 Economic Analysis Document § 2.2.1 at 8, J.A. at 945. All options with a payback period exceeding 5 years were discarded as economically unjustified, even though DOE did not specifically weigh the benefits and burdens of those design options under the statutory criteria for economic justification.

C. DOE’s Compliance with EPCA

Petitioners attack DOE’s treatment of maximum technological feasibility on three related grounds. First, they argue that DOE never adopted any determinations of maximum technologically feasible levels, as EPCA required it to do. Second, they argue that DOE’s decision to consider only certain commercially available design options as the basis for standards violated DOE’s statutory duty to evaluate standards at the maximum technologically feasible levels. Third, they argue that DOE’s decision to prescribe rules in 1982 and 1983 founded on information about appliances sold in 1980 unreasonably rejected more current information. We examine these challenges in turn.

1. DOE’s Failure to Determine Maximum Technologically Feasible Levels

As we have observed, EPCA requires DOE to determine maximum technologically feasible levels for all product types and classes. See EPCA § 325(i)(3). If a standard is not prescribed at that level, DOE must specifically explain the reasons for its decision. See id. Under the statute, only two basic reasons are acceptable: DOE must find either that a standard at the maximum technologically feasible level would not be “economically justified,” under the detailed definition of that term in EPCA, or that such a standard would not conserve significant energy. See id. § 325(c), (i)(3).

DOE’s June 1980 proposed rule-making did set forth maximum technologically feasible levels for comment. The 1980 notice did not satisfy the statutory requirement that DOE determine those levels, however, since DOE could only make such a determination after providing interested outsiders with the opportunity to comment. As it turned out, the proposed levels simply never resurfaced during the rulemaking, and consequently DOE never made the required determinations. DOE contends that it did ultimately adopt the maximum technologically feasible levels proposed in the June 1980 notice, see DOE Br. at 37-38, but DOE’s citations do not support its claim. DOE first directs us to its April 1982 notice of proposed rulemaking, which under the heading “Technological Feasibility” notes that “[f]or all prod *380 ucts and classes of products, DOE believes the efficiency levels analyzed are technologically feasible.” 47 Fed.Reg. 14,424, 14,426 (1982). This passage simply notes that the standard levels analyzed — levels based on analysis excluding technologically feasible prototype designs — were technologically feasible. See supra note 39. The passage does hot determine that standards higher than those analyzed would not also be technologically feasible, and it certainly cannot be read as endorsing the June 1980 determinations.

Next DOE points to a technical support document prepared by one of its contractors. See U.S. Department of Energy, Consumer Products Efficiency Standards: Engineering Analysis Document A-2, El-1 to E9-5 (1982) [hereinafter cited as 1982 Engineering Analysis Document], J.A. at 1568, 1804-42. Normally, we would expect to find determinations that Congress expressly ordered DOE to make in the Federal Register, not in a technical support document such as the one DOE cites. In any event, the cited passages in the document do not in fact approve the June 1980 determinations. The very page cited by DOE disclaims any intention to comment on the maximum technologically feasible levels:

The previous [draft of the] Engineering Analysis contained discussions of the maximum technological feasible efficiency level[s]____ Discussion regarding standards levels have [sic] been removed at DOE’s request. All of these discussions have been omitted from this T.S.D. [technical support document] as they were inconsistent with the intent of this document which is to present the engineering basis to the cost efficiency data.

Id. at A-2, J.A. at 1568 (emphasis added).

DOE also attempts to explain its failure to determine the maximum technologically feasible levels in a different way. DOE announced in its June 1980 rulemaking that in light of the time required for “product development cycle of engineering, prototype fabrication, laboratory testing, field testing, production engineering and production facility modification,” 45 Fed.Reg. 43,-976, 43,984 (1980), manufacturers could not reasonably be expected to incorporate prototype designs into entire product lines within the 5-year phase-in period for standards. In effect, DOE declared that a standard based on prototypes would not be economically justified because such a standard vvould require manufacturers to do what could not reasonably be done in the time available. It thus would be pointless to determine maximum technologically feasible levels based on prototypes and analyze standards at those levels, since DOE already knew that no such standard could pass the statutory criterion of economic justification. On this theory, DOE says it reasonably confined its analysis to efficiency levels that might actually be prescribed as standards.

Moreover, DOE has suggested that the further analysis petitioners demand would be not only superfluous, but impossible. DOE claimed that it needed a detailed breakdown of manufacturing costs for a design option to determine whether a standard based on that option would be economically justified. According to DOE, this information was simply unavailable for prototypes, as well as for design options included on appliances sold only in foreign markets. Thus, even assuming manufacturers could change their product lines within 5 years to meet standards based on prototypes, DOE could not determine whether such standards met the statutory criteria. DOE met this problem by resolving the uncertainty against standards; it elected not to prescribe standards based on prototypes because in its view it could not ascertain whether such standards would be economically justified.

There are serious objections to DOE’s arguments. They do not establish that DOE complied with the statute, which on its face requires actual determinations of maximum technologically feasible levels and specific explanations if standards are not set at those levels. Instead, DOE seeks to show that its transgression did not affect the final rules, because DOE could *381 never have prescribed standards at maximum technologically feasible levels anyway. However, an agency may not ignore the decisionmaking procedure Congress specifically mandated because the agency thinks it can design a better procedure. Nor is it clear that DOE’s adoption or modification of the June 1980 proposed determinations would have had no value. If DOE had followed the statute, it would have been able to calculate how much energy standards set at the maximum technologically feasible levels as finally determined would have saved. That information, in turn, would show the potential energy conservation DOE sacrificed when it refused to estimate whether standards based on prototypes would be economically justified. If the conservation sacrificed were large, DOE might have been moved to change its procedures; at the very least, DOE and commenters would have seen what potential gains in efficiency and what potential energy savings were lost by DOE’s approach. 38

But even if one ignores these defects in DOE’s defense of its rulemaking, its arguments remain vulnerable on two scores. First, the argument conclusively assumes that manufacturers cannot incorporate any prototypes for any product type or class into all appliances of that type or class within the 5-year period. Second, it assumes that DOE could not reasonably decide whether standards based on prototypes were economically justified; and that consequently DOE could refuse to consider any standard based on prototypes at all. If these central assumptions fall, then DOE’s argument collapses.

Deciding whether these assumptions are accurate requires us to pass beyond the formal attack on DOE’s failure to determine maximum technologically feasible levels and to confront directly the petitioners’ second challenge on the issue of. technological feasibility. Accordingly, we turn to that challenge.

2. DOE’s Refusal to Consider Standards Based on All Technologically Feasible Design Options

Petitioners claim that DOE refused to analyze standards based on all technologically feasible design options without any adequate determination that standards based on those options would not be economically justified. In particular, petitioners note that DOE did not assess standards based on certain categories of design options, although those categories qualify as technologically feasible under DOE’s own definition of the term. Those categories are (a) prototypes, (b) design options included only on appliances not sold in the United States, (c) design options with a payback period of 5 years or more, (d) certain design options that would require a significant increase in the engineering staff of appliance manufacturers or that had long lead times, and (e) certain design options specifically identified by commenters that DOE rejected as a basis for standards on grounds petitioners deem inadequate. We address each of these categories separately-

a. Prototypes. As we have stated, DOE’s rationale for rejecting prototypes depended on two factual assumptions. The first of these is that no design option available only as a prototype in 1980 could reasonably be included on all appliances sold in 1986 within any product type or class. EPCA requires that in determining whether a standard is economically justified, DOE consider, among other factors, “the economic impact of the standard on the manufacturers.” EPCA § 325(d)(1). If no standard could have been based on prototypes without requiring manufacturers to accomplish the impossible, we agree that DOE could reasonably deem all such standards economically un *382 justified without trudging through the remaining statutory factors. 39

In our view, DOE’s finding that manufacturers could not meet any standard based on prototype technologies for any product type or class is not supported by substantial evidence. Congress broadly defined technological feasibility to mean improvements in efficiency “capable of being carried out,” see supra at 1391-92 — an inclusive definition motivated by Congress’ concern over “the rather low efficiencies of current consumer products.” S.Rep. No. 409, 95th Cong., 1st Sess. 36 (1977). DOE’s 1980 decision that prototypes were technologically feasible thus rests on solid ground. 40 But DOE’s exclusion of all prototypes from consideration for standards took away with one hand what the other had given. That decision guaranteed that standards could do no more than require all products to be as efficient as the most efficient design options already in the market made possible, even though enactment of the appliance program rested on the view that the market had failed to promote efficient appliances.

In these circumstances, DOE cannot prevail merely by asserting that many or even most prototypes could not workably have been included within 5 years in all products of a particular type or class. Even assuming that this is true, a minority of prototypes would remain as technologically feasible and, for all DOE knew, economically justified design options that were not considered in the agency’s deliberations. EPCA does not permit DOE to ignore design options meeting those two criteria. DOE must therefore show that substantial evidence supports its implicit determination that all or virtually all prototypes would have taken more than 5 years to introduce throughout a product type or class.

*383 We do not think that DOE has succeeded in that task. DOE’s announcement of its position in the June 1980 notice contains no discussion of any evidence or citation to DOE studies. DOE’s 1980 Economic Analysis Document recites essentially the same assumption about the lead time for- prototypes as DOE stated in its notice, but the document discloses no serious effort to confirm that assumption through empirical study. 41 Moreover, DOE’s assumption is in tension with the congressional finding that appliances already available in an unregulated market were simply not efficient enough to satisfy national policy goals. Congress designed the 5-year phase-in period precisely so that manufacturers could introduce new technologies required by standards. 42 In light of the potentially large number of technologically feasible design options that this single assumption excluded from consideration, we think that DOE was required to make a more thorough effort to confirm its abstract theory.

Numerous commenters in the rulemaking challenged DOE’s exclusion of prototypes from consideration and in some instances offered specific prototypes for evaluation. In responding, DOE made little effort to argue that manufacturers could not modify particular product lines to incorporate a specific prototype within 5 years. Instead, DOE typically argued either that it lacked economic information about the prototype or that the use of the prototype on all appliances within a product type or class might compromise product performance or consumer acceptance. See, e.g., 48 Fed.Reg. 39,376, 39,395 (1983) (rejecting prototype gas water heaters); 47 Fed.Reg. 57,198, 57,211 (1982) (rejecting bi-radiant ovens). Without something more than DOE’s general and unelaborated concerns, we cannot uphold the crucial assumption that absolutely no prototypes could be made standard items for any product type or class within 5 years, when DOE did not show that this all-encompassing view was true even for the specific prototypes suggested to it.

Finally, it is not at all intuitively clear that DOE’s assumption is likely to prove true. DOE was willing to consider standards based on design options included on a single appliance produced by a single company for sale in the United States before November 14, 1980. By including such design options in its deliberations, DOE acknowledged that other manufacturers might be able to comply with a standard based on design options that were not included on any appliances produced by those manufacturers in 1980. In other words, DOE thought that 5 years would sometimes be long enough for a manufacturer to redesign an entire product line and retool its manufacturing processes to incorporate a design option previously used only by one of its competitors. Nonetheless, DOE conclusively presumed that 5 years was not long enough to redesign and retool in order to incorporate any design option available only as a technologically feasible prototype, without regard to the characteristics of any particular prototype. We understand, of course, that the chance to observe a competitor’s experience with a de *384 sign option could easily shorten a manufacturer’s design and development time. Nonetheless, the exact length of that design and development time will logically depend on the particular prototype at issue. In the absence of careful analytic study of lead times for design changes in particular products, we cannot uphold DOE’s blanket announcement that manufacturers would not have enough lead time to comply with any standard for any covered product based on any prototype at all.

We must therefore examine the alternative factual assumption on which DOE founded its exclusion of prototypes. This assumption was that for all prototypes, DOE could not obtain the information about manufacturing costs and consumer acceptance it needed to decide whether a standard based on that prototype design was economically justified.

We return once again to the statute for guidance. Although EPCA generally requires DOE to prescribe the highest standard that is technologically feasible and economically justified, the Act also states that DOE may not prescribe a standard covered product “if a test procedure has not been prescribed ... with respect to that type (or class) of products.” EPCA § 325(b)(1). Section 323 of EPCA requires DOE to prescribe test procedures for all covered products, which must “be reasonably designed to produce test results which reflect energy efficiency, energy use, or estimated annual operating cost of a covered product during a representative average use cycle.” EPCA § 323(b)(1). Similarly, the legislative history for EPCA as originally passed notes that DOE could decline to prescribe a labeling rule if “no test procedure which is an adequate approximation of energy use or relative energy efficiency can be devised.” H.R.Rep. No. 340, 94th Cong., 1st Sess. 97 (1975), U.S.Code Cong. & Admin.News 1975, p. 1859. EPCA thus expressly recognizes that DOE may delay the issuance of a standard based on a particular design option if DOE is unable to measure the efficiency of an appliance incorporating that design option.

However, the text of the Act includes no parallel provisions allowing DOE to reject standards because it cannot precisely quantify costs relevant to DOE’s assessment of economic justification. Instead, EPCA lists the following general considerations as relevant to DOE’s assessment of a proposed standard:

(A) whether the standard to be prescribed is economically justified (taking into account those factors which the Secretary must consider under subsection (d) of this section),

(B) whether the standard will achieve the maximum improvement in energy efficiency which is technologically feasible,

(C) if the standard will not achieve such improvement, whether the reasons for not achieving, such improvement are adequate, and

(D) whether such rule should prescribe a level of energy efficiency which is higher or lower than that which would otherwise apply in the case of any group of products within the type (or class) to be subject to such standard.

EPCA § 325(i)(3). EPCA further requires DOE to weigh the following factors in de- ' ciding whether a standard is economically justified:

(1) the economic impact of the standard on the manufacturers and on the consumers of the products subject to such standard,

(2) the savings in operating costs throughout the estimated average life of the covered products in the type (or class), compared to any increase in the price of, or in the initial charges for, or maintenance expenses of, the covered products which are likely to result from the imposition of the standard,

(3) the total projected amount of energy savings likely to result directly from the imposition of the standard,

(4) any lessening of the utility or the performance of the covered products likely to result from the imposition of the standard,

(5) the impact of any lessening of competition determined in writing by the At *385 torney General that is likely to result from the imposition of the standard,

(6) the need of the Nation to conserve energy, and

(7) any other factors the Secretary considers relevant.

' EPCA § 325(d). DOE’s claim is that it could not weigh these statutory factors for any prototype technologies.

The structure of the Act strongly suggests a congressional expectation that DOE ordinarily could and would estimate the economic justification for all technologically feasible design options. EPCA does not merely authorize DOE to decide, if the agency so chooses, whether technologically feasible standards would be economically justified: it instructs DOE to reach a decision on that question as the next step in the statutory decisionmaking procedure. DOE was required to design its methodology to fit the Act, rather than lopping off statutory provisions that did not fit the economic models it chose. Accordingly, we examine with particular thoroughness DOE’s unorthodox excuse that it could not reasonably follow the text of the Act.

The NECPA legislative history shows that Congress was aware of the role projections and estimates were likely to play in the development of standards. For example, the conference report commented that:

The conferees recognize that some of the factors [the Secretary must weigh in assessing economic justification] cannot be quantified, such as “the need of the Nation to conserve energy.” However, where quantification is possible, it is expected that the Secretary will perform such quantification of individual factors to the greatest extent practicable. The conferees do not intend that quantification of a factor inherently enhances its weight as against nonquantifiable factors. Because the conference substitute requires the Secretary to weigh the factors, and because the increase in initial charges and any change in maintenance costs and energy savings will be quantified and compared, the House criterion [proposing a more specific test for economic justification] has been deleted.

S.Rep. No. 1294, 95th Cong., 2d Sess. 116 (1978) (emphasis added). Congress thus mandated quantitative analysis of the factors relevant to economic justification “to the greatest extent practicable.” Congress did not, however, inflexibly require that DOE’s forecasts be based on precisely and minutely verifiable details. Even more importantly, Congress flatly announced that DOE must first consider both the quantifiable and nonquantifiable statutory factors and must weigh those benefits and burdens against each other. Congress did not suggest that if DOE thinks it cannot quantify the factors precisely enough, it may refuse to make reasonable estimates of those factors, and on that ground refuse to perform the statutorily prescribed weighing.

All of this does not mean that if the statute requires DOE to consider projections that cannot reasonably be made, DOE must pretend that it has accomplished the unachievable. But it does strongly suggest that DOE’s effort to follow these congressional directions was too perfunctory. DOE has told us that it needs manufacturer’s assembly line costs for a design option to analyze the economic impact of a standard based on that design option. But we know nothing of any serious effort to determine whether these costs could be estimated for the full-scale working prototype models DOE refused to consider.

Moreover, DOE’s explanation of its method for estimating the costs of commercially available technologies leaves it far from clear why that technique could not reasonably be adapted to any prototype models. DOE’s basic approach was first to identify a “baseline” unit for each product type or class, which it defined as “the unit with an efficiency, output, and size approximately equal to that of the most commonly produced unit for the baseline year.” 1982 Engineering Analysis at C-3, J.A. at *386 1575. 43 DOE then determined what design options were eligible for consideration as a basis for standards. Next, “using engineering analysis and computer simulations as appropriate, the efficiency improvements which would result from implementing various design options were estimated.” Id. at C-4, J.A. at 1576; cf. 48 Fed. Reg. 39,376 , 39,397-98 (1983) (discussing computer simulation predictions of efficiency levels likely to be achieved by particular design changes). Finally, DOE applied a combination of manufacturing information and general rules to arrive at a cost breakdown:

Four basic cost elements (investment, materials, purchased components, and labor) were considered. Investment included capital equipment, plant, tooling, and an allowance for costs related to inventory, spare parts, line engineering, service training, and reissuing product literature. Per unit costs were estimated by amortizing these investment costs using the yearly production volume and depreciation periods of 30 years for plant, 10 years for capital equipment, 3 years for tooling and miscellaneous costs. Materials costs/unit were estimated by multiplying the estimated weight changes in materials identified in the design option analysis by the materials cost/pound. Costs/unit of purchased components] were based on quotes from suppliers. Although some larger manufacturers might make components such as compressors or vent dampers, these items were costed as purchased parts unless the majority of manufacturers made them in-house. Labor costs were obtained by estimating the change in labor in minutes for each design option and using a labor rate of $21.85/hour (direct cost plus burden). These four cost elements were summed to estimate the added factory cost.

1982 Engineering Analysis Document at C-4 to C-5, J.A. at 1576-77.

Several aspects of this discussion are relevant to the problem under examination. First, DOE does not report that it attempted to run its cost analysis on any particular prototype models and discovered that the effort required too much speculation. DOE excluded prototypes from its cost analysis as a threshold, preliminary step; it did not describe any actual experience in applying the analysis to prototypes. One might have expected that the difficulty of estimating manufacturing costs for a prototype could vary considerably depending on the design in question. For example, a prototype design utilizing a new combination of components already on the market could obviously be analyzed more easily than one requiring the manufacture of previously unproduced parts. This sort of specific analysis, however, is not reflected in the record.

Second, DOE evidently did not have precise, empirically derived manufacturing costs for the appliances it did consider as the basis for standards. DOE’s method required it to identify energy-saving design options available on separately manufactured appliances within the same product type or class. DOE then hypothesized a baseline appliance to which the design options selected for consideration had been added, and calculated the benefits and burdens of the resulting composite “model.” That composite, however, might well include a combination of design options that did not appear on any commercially available appliance; in fact, the composite model might not exist in reality at all. The record suggests that DOE relied in considerable part on computer simulations to predict the benefits and burdens of these imaginary appliances.

Essentially, DOE figured the manufacturing costs for its composite appliances by looking at manufacturing costs for different products that included the selected design options, estimating what part of the *387 manufacturing costs were attributable to the design options under study, and then including those estimates as part of the data used to generate total manufacturing costs for the hypothesized composite. If DOE could accomplish that task, it is unclear why DOE could not estimate the manufacturing costs for any full-scale, working prototypes at all. If, for example, a prototype consisted mainly of parts that were already manufactured for industrial use, DOE might have been able to estimate the assembly-line costs of the prototype using an additive approach similar to the one it adopted to estimate the cost of commercially available design options.

We acknowledged that with each step away from real-life information about benefits and costs, DOE’s estimates would necessarily become less reliable. But on this record, we cannot tell whether cost estimates for full-scale, working prototypes would require more or less speculation than cost estimates for hypothesized appliances combining design options that are commercially available only on separate appliances produced by different manufacturers. In our view, the record does not disclose a fair investigation of whether DOE’s methods, or reasonably practical methods of comparable reliability, could be adapted to measure the manufacturing costs of prototypes.

Third, DOE did not make individual cost determinations for important cost items in the manufacture of commercially available design options; instead, DOE applied general rules that might have been adapted to prototypes. For example, DOE used a standard labor rate, rather than investigating the particular cost of the labor required to effect specific manufacturing tasks. See supra at 1401. Thus, it is not clear exactly how far DOE carried its investigation of the costs particular manufacturers experienced in incorporating commercially available technologies into their products.

For these reasons, we conclude that nothing on the face of DOE’s method for estimating manufacturing costs explains why those costs could not reasonably be estimated for prototypes. We do not, of course, presume to decide that whether DOE’s method or reasonable alternatives to it in fact would have produced satisfactory results for some or all prototypes. All we decide is that the record does not show that DOE adequately investigated these questions and explained its results. Nor can we extrapolate from DOE’s description of its analytic method any reason why that method is entirely ill-suited to the assessment of prototypes. DOE has thus failed to provide any explicit or reasonably inferred rationale for its assumption that it could not, as the statute required, evaluate all technologically feasible design options to determine which of those options could be used as the basis for economically justified standards. And, as we have stated, we are not here considering an incidental weakness in factfinding on the periphery of this rulemaking; we are considering a crucial determination at odds with the text and purposes of the statute. Viewing DOE’s sweeping assumption in that light, we hold that it is not supported by substantial evidence.

DOE’s last line of defense for its refusal to consider prototypes centers on the agency’s speculation that existing test procedures might be inadequate to measure the efficiency of prototypes. But DOE never actually determined that the test procedures were inadequate for all or any of the prototypes suggested to it. DOE cannot simply guess that test procedures might be inadequate for some prototypes and for that reason refuses to evaluate all prototypes. The terms of the Act allow DOE to reject technologically feasible design options for consideration only if test procedures are in fact inadequate; and if test procedures are inadequate, DOE must attempt to devise new procedures. Because DOE never found that prototype design options were as a class untestable, we must reject this branch of DOE’s rationale as *388 inadequately supported by DOE’s own conclusions from the record. 44

In summary, DOE’s assumption that manufacturers could not within 5 years change all appliances within a product type or class to meet a standard based on prototype technologies is not, in our view, supported by substantial evidence. We also find that DOE has not identified substantial evidence supporting its view that no prototype could be analyzed to determine whether a standard based on it would be economically justified. Finally, we find that inadequate test procedures provide a basis for rejecting consideration of a design option only if, at a minimum, DOE actually finds that the test procedure cannot adequately measure the characteristics of the design option in question. DOE has not made this finding for any prototype, and thus DOE cannot on this record exclude all prototypes by reason of conjectured defects in the test procedures.

Two holdings follow from this reasoning. First, DOE’s theory that its failure to determine maximum technologically feasible levels was in effect harmless error cannot be sustained, because DOE has not even produced substantial evidence for the two factual assumptions necessary to make a harmless error theory plausible. On remand, DOE must therefore finally determine maximum technologically feasible levels in accordance with the statute. Second, DOE may not rely on the very general arguments it offered in the rulemaking under review in support of its refusal to consider technologically feasible prototypes as a class in formulating standards.

For these reasons, we cannot accept the arguments DOE advanced for excluding prototypes as a class. We do not, of course, hold that any specific prototype could workably be the foundation for a standard. Nor do we hold that EPCA prevents DOE from determining that particular classes of prototypes cannot qualify as technologically feasible. If DOE does ex-elude such classes of prototypes from consideration, however, that exclusion must rest on a stronger factual foundation than DOE supplied in this rulemaking. EPCA assigns DOE the job of gathering information relevant to these questions and making decisions. We hold only that DOE must do that job. DOE may not take a shortcut by discarding the entire class of prototypes in advance, despite its finding that prototypes qualify as “technologically feasible,” through very abstract, unelaborated, and unsupported assumptions about prototypes.

b. Foreign Market Design Options. DOE’s 1980 rulemaking stated that it would consider .standards based on all technologies that were commercially available in 1980. However, DOE ultimately refused to analyze design options available only in foreign markets as a basis for standards.

[DOE] did not have detailed data for foreign designs hot incorporated in domestic appliances and, therefore, could not analyze properly the impacts of standards based on foreign designs. Although some commenters criticized DOE for not analyzing standards using these technologies, DOE received little information on these designs. An additional problem with assessing foreign products was evident from the comments — disagreement and uncertainty as to the facts concerning the product. For example, NRDC and Whirlpool Corporation disagreed fundamentally over the capabilities of a particular Japanese refrigerator.

48 Fed.Reg. 39,376, 39,379-80 (1983).

DOE apparently wanted more specific information concerning both the manufacturing costs and performance characteristics of foreign-market appliances. What we have already said for prototypes also applies to foreign-market design options: the statute clearly instructs DOE to consider standards based on technologically feasible *389 design options, and DOE has not shown that it could not reasonably follow that requirement for foreign-market options. Concededly, the added assembly-line costs to domestic manufacturers that standards based on a particular design option would impose can be estimated most accurately if DOE can examine the actual cost of production for the design option to at least one domestic manufacturer. However, Congress knew that EPCA required DOE to make projections and estimates, and indeed DOE’s final rules rely very heavily on highly controversial predictions about changing patterns of consumer behavior between 1980 and 2005. We do not think that the difficulties DOE anticipated in estimating production costs for foreign-market options justify rejecting those options as a class from consideration.

DOE also noted that commenters disagreed over the performance characteristics of foreign-market design options. However, it is unclear that these disagreements differed in principle from similar controversies over the performance and efficiency of domestically available design options. Congress required DOE to develop test procedures for covered products so that, when necessary, DOE could make factual findings for itself about the performance of appliances under conditions simulating actual use. DOE does not argue that it could not obtain models of foreign-market appliances or that these appliances were unsuited to DOE’s test procedures and methods of analysis.

Again, we do not hold that DOE must consider any particular foreign-market design options. All we hold is that DOE has not identified adequate support in the record for its class-wide exclusion of these design options. DOE must therefore either offer a much more thorough explanation based on record evidence why it cannot consider foreign-market design options, or make more particularized determinations about its ability to test and analyze specific foreign-market options.

c. The Five-Year Payback Period. In 1980, DOE decided that it would evaluate all design options with a payback period shorter than the average life of the appliance. See 45 Fed.Reg. 43,976, 43,983 (1980). 45 Thus, a design option with a payback period shorter than the average life of an appliance is cost-effective for most consumers, since the appliance will probably last long enough for the consumer to recover the cost of the design option.

DOE repudiated that decision in 1982 and announced that it would generally evaluate only design options with a payback period shorter than 5 years. See 1982 Economic Analysis § 2.2.1 at 8, J.A. at 945. DOE informs us that for clothes dryers and kitchen ranges and ovens, DOE evaluated some design options with 9 year payback periods. See DOE Br. at 43, n.*. The table below lists the average lifetime of the product types involved in this rulemaking, as determined by DOE in 1982. As the table shows, all of the products have an average life of substantially more than 5 or even 9 years. DOE’s 1982 rule thus creates a serious threat that DOE will summarily reject design options with payback periods short enough so that the option is easily cost-effective for the average consumer, even though those payback periods may be longer than the maximum figures DOE set.

*390 Table 5

Appliance Lifetimes

Lifetime in Years [1982] _Product Type_Analysis

Refrigerator-Freezers 19

Freezers 21

Water Heaters 13

Furnaces 23 *

Clothes Dryers 18.5

Ranges and Ovens 18

Central Air Conditioners 12

Room Air Conditioners 15

1982 Economic Analysis § 2.3.2 at 34 (citation omitted) (table reprinted only in part), J.A. at 971.

Petitioners cite the text of the statute, its legislative history, and the purposes of the Act in support of their attack on the 5-year limitation. Specifically, petitioners note that before deciding whether a proposed standard is economically justified, DOE is statutorily required to weigh, among other factors,

the savings in operating costs throughout the estimated average life of the covered products, in the type (or class), compared to any increase in the price of, or in the initial charges for, or maintenance expenses of, the covered products which are likely to result from the imposition of the standard.

EPCA § 325(d)(2) (emphasis added). This provision explicitly requires DOE to compare the lifetime benefits of technologically feasible design options against their lifetime burdens in deciding whether standards based on particular options would be economically justified.

DOE’s cutoff, however, conclusively assumes that design options with a payback period of five years or more (nine years or more for the two product types mentioned above) could not possibly be economically justified, regardless of their continued benefits over longer periods. DOE did not individually consider whether such design options might be economically justified according to the statutory criteria; the whole thrust of DOE’s general rule was to exclude these design options as a class. Nor did DOE even determine whether that class of design options in the aggregate, or some substantial part of it, might include economically justified design options. In other words, the statutory criteria for economic justification were not even explicitly considered in DOE’s rejection of the class as a whole, let alone in any consideration of particular class members.

This disregard for the statutory decision-making procedure is especially worrisome in light of the NECPA legislative history, which shows that Congress was concerned over the tendency of consumers to reject efficiency-improving appliances with long payback periods. As Representative Stock-man explained, Congress viewed this consumer behavior as a kind of market failure:

There is some reason to doubt that market forces alone will bring about the needed shift to more efficient appliances. Numerous witnesses appearing before the subcommittee testified that the average consumer looks for a payback from higher purchase prices within 3 years. In the case of an appliance with a useful life of 10 years, this short payback horizon severely limits the amount of higher purchase price the consumer will accept in choosing a more efficient product over a cheaper, less efficient product.

H.R.Rep. No. 496, Pt. 4, 95th Cong., 1st Sess. 360 (1977), U.S.Code Cong. & Admin. *391 News 1978, p. 8744 (additional views of Rep. Stockman). Since the average life of most covered products involved in this rule-making is much longer than 10 years, the potential market failure caused by demanding a very short payback period is even greater than Representative Stockman suggested.

Moreover, DOE’s general rule is apparently inconsistent with the explanation DOE offered for departing from it. DOE did, in fact, evaluate an insulation design option for clothes dryers, despite the long payback period for the option. In the December 1982 rulemaking, DOE noted that:

Two ... commenters, Webster City Products (Webster City) and Whirlpool Corporation (Whirlpool), suggested that the extended payback period of this design option should disqualify it from consideration. Webster City remarked that its laboratory test investigation showed a 15 to 20 year payback period, while Whirlpool used cost figures from DOE’s Engineering Analysis to compute a 22.3 year payback period.

DOE agrees that the payback for this design option is not attractive, but this fact, given the purpose for which the design options in the Engineering Analysis are used, does not compel its elimination. The design options are used to assess the relationship between increases in the efficiency of a product and increases in the cost of that product. This relationship is then used in other analyses to project the effects that would be attributable to standards set at different levels. Design options with a range of payback periods may be useful in the derivation of the relationship between increases in efficiency and increases in cost. Of course, a standard that would require use of such a design option may not be economically justified.

47 Fed.Reg. 57,198, 57,209 (1982). This passage summarizes the fundamental problem with DOE’s 5-year and 9-year rules: those rules drop design options out of the analysis at a preliminary stage, before their benefits and burdens are fully assessed. As we have stressed, EPCA allows DOE to reject consideration of technologically feasible design options as a basis for standards only if standards based on those design options could not be economically justified. Yet in the passage quoted above, DOE itself effectively rejected the idea that all design options with long payback periods are, by virtue of that fact alone, economically unjustified. DOE sensibly decided that it must look at the cost-benefit relationship of those design options before it discarded them. By refusing even to analyze most design options with longer payback periods, DOE deprived itself of exactly the information it needed to decide whether standards based on such design options would be economically justified.

In short, DOE has justified making an exception to its general rule on grounds that entirely discredit the rule itself. DOE’s reply to Webster City and Whirlpool does not purport to explain why the particular design option in question should be considered, even though other design options with similarly long payback periods were not. Instead, DOE explained why, in general, design options with long payback periods should be considered in the engineering analysis. Having set forth these views, DOE may not rely on the inconsistent general proposition that no design option with a payback period longer than DOE’s maximum figures could be economically justified.

Finally, our difficulty in understanding DOE’s position is heightened by the lack of rationale in DOE’s announcement of its 5-year rule. The rule is stated in the assumptions section of the 1982 Economic Analysis, and is thus a predicate for rather than a conclusion from DOE’s technical analysis. Moreover, the only explanation for selection of 5 years rather than some other period as the cutoff is that payback periods shorter than 5 years are “considered to be acceptable to consumers.” 1982 Economic Analysis Document § 2.2.1 at 8, J.A. at 945. But as we have discussed, the fact that consumers demand short payback periods was itself a major cause of the market failure that Congress *392 hoped to correct. DOE cannot logically reject design options because consumers would not in the absence of standards buy appliances including those design options, when the entire point of a mandatory program was to change consumer behavior.

DOE’s refusal to consider design options with payback periods longer than 5 or 9 years, like its exclusion of prototypes, runs counter to the statutory text and legislative history of NECPA on a point with major consequences for the result of this rule-making. Yet instead of trying to reconcile its anomalous practice with the statute, DOE first offered an explanation inconsistent with the purposes of the statute and later supported an exception to the rule with arguments that undermine the entire rule. Accordingly, we hold that this restriction on the design options DOE was willing to analyze is unsupported by substantial evidence.

d. Lead Times. Next, petitioners complain that DOE excluded design options “that might entail a ‘significant increase in [the manufacturers’] engineering staff,’ and selected only those with ‘the shortest lead time.’ ” NRDC Br. at 40 (footnotes omitted). DOE described its method of minimizing the cost of increased engineering staff in the following terms:

Lead times available for implementing design changes required to meet the efficiency levels analyzed are long enough so that no significant increase in engineering staff will be required.

1982 Engineering Analysis Document at C — 1, J.A. at 1573. We read this sentence as describing a general approach to fixing appropriate lead times — i.e., periods it would take a manufacturer to introduce particular design options on all appliances covered by a standard. DOE could have assumed that standards would force manufacturers into expensive, “crash” development and retooling programs. This approach would obviously achieve energy savings more quickly, but it would also impose much greater costs on manufacturers. Or alternatively, DOE could assume, as it did, that lead times should be calculated based on more moderately paced programs to achieve compliance with standards. Under the latter view, the benefits of standards would be slower in coming, but the burdens on manufacturers would be much lighter. DOE plainly needed some method of estimating lead times, and we cannot say that the method on which it settled was unreasonable.

Petitioners also attack DOE’s method of selecting the combination of design options that would best achieve particular levels of efficiency. The basic problem DOE had to solve was that different combinations of design options, with varying lead times and costs to manufacturers, could lead to comparable gains in efficiency. Thus, in assessing the benefits and burdens of standards at a specific level of efficiency, DOE needed some theory to identify what combination of design options that achieved that level of efficiency should be used as the basis for analysis. DOE settled on the following principles:

1) Improvements must have a reasonable initial cost to the consumer.

2) Only the lowest lead time and most cost effective improvements were considered.

Other paths (series of design improvements resulting from application of other philosophies) to higher efficiency can be envisioned. For instance, the lowest cost to the consumer to achieve each level could have been assessed. This could lead to a rearrangement of the grouping of design improvements to achieve each level. Minimizing lead time ... was felt, however, to be the most effective and likely scenario for introducing energy savings at the earliest date.

1982 Engineering Analysis Document at C-3, J.A. at 1575; see also 45 Fed.Reg. 43,976, 43,983-84 (1980).

This passage does not describe a path of analysis that arbitrarily excluded design options from consideration. Rather, we see the passage as describing a reasonable method of ordering and structuring agency deliberations, and we accordingly uphold that method against challenge.

*393 e. Specific Design Options. During the rulemaking, petitioners and other commenters identified numerous energy-saving design options that, in the view of the commenters, should have been considered as the basis for standards. DOE rejected many of these suggestions for a variety of different reasons, and petitioners now challenge the legality of these decisions.

Under our view of the case, we need not resolve these disputes. As we have explained, we believe that DOE fundamentally misunderstood the statutory requirement that it first determine maximum technologically feasible improvements in efficiency, and then decide whether standards at that level or at some lower level would be economically justified. As we discuss in Part IV.C.3 below, DOE must approach that task anew. Because DOE must conduct a new rulemaking based on current technology, we need not pass on whether DOE’s determinations that particular design options, as described in the information then before the agency, were or were not reasonable. The rulemaking on remand will be based on very different information, and consequently our view of DOE’s decisions on specific design options could not provide useful guidance for the agency.

3. DOE’s Reliance on 1980 Data

Much of the information underlying DOE’s final rules was gathered in 1980 and never significantly updated, although the final rules were not issued until 1982 and 1983. Petitioners challenge DOE’s reliance on data of this vintage as unreasonable; DOE replies that any effort to collect and analyze more recent information would have delayed the rulemaking even longer. See 48 Fed.Reg. 39,376, 39,379 (1983). However, we need not resolve this dispute either. We have already decided that (1) DOE must determine maximum technologically feasible levels and (2) DOE must reexamine its refusal to consider prototype and foreign-made design options as the basis for standards, and must look carefully and thoroughly at the specific characteristics of those design options in determining whether they could reasonably be evaluated under the statutory criteria. Both of these inquiries will require detailed review of the characteristics of technologically feasible design options that are not “commercially available” within the meaning DOE assigned to that term in this rulemaking— that 'is, design options that were not available on commercially manufactured appliances sold in the United States in 1980.

Whether or not DOE acted reasonably in issuing rules in 1982 and 1983 based on 1980 information, we think it would be patently unreasonable for DOE to begin further proceedings in the last half of 1985 based on data half a decade old. As DOE itself acknowledges, the efficiency of many products involved in this rulemaking has changed dramatically since 1980. It would be wholly futile for us to require DOE to conform its decisionmaking procedures to the statute, but permit it to trudge through the correct procedure based on information that is now incontestably antique. Such a pointless exercise would make a mockery of the clear statutory emphasis on a realistically administered appliance program based on current technology.

Moreover, we believe that the text and legislative history of NECPA show that Congress would not have approved reliance on such outdated data. As we have discussed; the slow pace of agency proceedings under EPCA as originally enacted was a major reason for the enactment of the NECPA amendments. Congress wanted to avoid the delays involved in monitoring industry compliance with voluntary targets, and in part for that reason it eliminated the target approach. See H.R.Rep. No. 496, Pt. 4, 95th Cong., 2d Sess. 43-45 (1977). More importantly, Congress established a strict schedule in NECPA itself that specifically keyed the dates by which final rules were due to the dates on which DOE published an advance notice of proposed rule-making and a notice of proposed rulemaking. See supra note 11. Those notices, in turn, determine the period for notice and comment and thus the period during which *394 major parts of the record for the rulemaking would be created.

Congress, then, passed provisions intended to set a specific outer limit on the time that could elapse between the closing of the record and the promulgation of final rules. Thus, when Congress defined “technologically feasible” as “capable of being carried out,” see supra at 1392, it believed that only a relatively short period would pass between DOE’s evaluation of available technology and its promulgation of final rules. As Congress knew that the state of technology changes rapidly, it had good reasons for this insistence on prompt action. We believe Congress thought that “technologically feasible” meant “technologically feasible based on information that is reasonably current at the time the final rules are validly adopted,” not “technologically feasible based on information that was reasonably current at the time DOE first attempts to promulgate fi

This text is long and has been trimmed here. Open the source document for the complete record.