Petitioners in this action seek review of a final order by the National Highway Traffic Safety Administration (NHTSA) rescinding the automatic crash protection requirements of Federal Motor Vehicle Safety Standard 208 ten months before the standard's effective date. 46 Fed.Reg. 53,419 (Oct. 29, 1981) (Notice 25). The standard would have required that large and mid-size automobiles manufactured after September 1, 1982, and all automobiles manufactured after September 1, 1983, carry passive restraints such as airbags or "passive" seatbelts. Airbags are cushions stored under the dashboard that, when triggered by a frontal collision, fill with stored or rapidly generated gas to protect the rider from collision with the car's interior. Passive seatbelts, also called "automatic" seatbelts, move into place automatically when a passenger enters a vehicle and closes the door. Petitioners State Farm Mutual Automobile Insurance Company (State Farm) and the National Association of Independent Insurers (NAII) challenge NHTSA's rescission of the standard as arbitrary, capricious, an abuse of discretion, and a violation of law as defined by section 10 of the Administrative Procedure Act, 5 U.S.C. § 706 (1976).

We agree. This case is complicated because it has far-reaching implications and involves a politically controversial safety standard, but the determining principle is simple. An administrative agency, possessing power delegated by the legislative branch of government, must comply with the legislative requirement that its decisions be reasoned and in accordance with the purposes for which power has been delegated. NHTSA's rescission of the safety standard presents a paradigm of arbitrary and capricious agency action because NHTSA drew conclusions that are unsupported by evidence in the record and then artificially narrowed the range of alternatives available to it under its legislative mandate. NHTSA thus failed to demonstrate the reasoned decisionmaking that is the essence of lawful administrative action.

The procedural history of the case before us is extremely complex. The standard that has now been rescinded was the subject of approximately 60 notices of proposed rulemaking, hearings, amendments, and the like between 1969 and 1981. There were separate adjudications before this court and in the Sixth Circuit, and successful as well as unsuccessful attempts in Congress to control the evolution of the regulation. A number of these events must be described in some detail before we turn to the issues in this case, because they help to put into perspective the course of the agency's action.

Congress enacted the National Traffic and Motor Vehicle Safety Act of 1966 (the Safety Act), 15 U.S.C. §§ 1381 et seq. (1976 & Supp. IV 1980), in response to the alarming number of deaths and personal injuries on the nation's highways. The stated purpose of the Safety Act was "to reduce traffic accidents and injuries to persons resulting from traffic accidents." 15 U.S.C. § 1381.1 The Safety Act directs the Secretary of Transportation or his delegate2 to issue motor vehicle safety standards that "shall be practicable, shall meet the need for motor vehicle safety standards that "shall be practicable, shall meet the need for motor vehicle safety, and shall be stated in objective terms." 15 U.S.C. § 1392(a). In issuing these standards, the Secretary is also directed to consider "relevant available motor vehicle safety data," whether the proposed standard "is reasonable, practicable and appropriate for the particular type of motor vehicle or items of motor vehicle equipment for which it is prescribed," and "the extent to which such standards will contribute to carrying out the purposes" of the Safety Act. 15 U.S.C. § 1392(f)(1), (3), (4).3

Under these guidelines, the Department of Transportation (DOT) issued the original Standard 208 in 1967, requiring seatbelts in all cars. 32 Fed.Reg. 2408, 2415 (Feb. 3, 1967). By July 1969, however, DOT had concluded that the level of seatbelt use was too low to reduce traffic injuries to an acceptable level. It issued a notice of proposed rulemaking to consider "the prompt development and installation of passive restraint systems," 34 Fed.Reg. 11,148 (July 2, 1969), defined as protective systems that require "no action by vehicle occupants." 36 Fed.Reg. 8296 (May 4, 1971). The agency conducted a lengthy rulemaking proceeding on passive restraint systems. It revised Standard 208 in 1970 to include passive protection requirements, 35 Fed.Reg. 16,927 (Nov. 3, 1970), and after a series of modifications and petitions for reconsideration published a final amendment to Standard 208 in 1972. 37 Fed.Reg. 3911 (Feb. 24, 1972).

The Federal Register during this period provides a strong reminder of how frequently an agency sometimes acts to focus or clarify a regulation before the regulation is promulgated. Not all of the twenty-four notices and amendments to the standard issued between 1970 and 1972 are important to the case presented to us here, but two of these modifications are significant. When the first notice, entitled "Inflatable Occupant Restraint Systems," was published in 1969, the agency's emphasis was clearly on airbags. 34 Fed.Reg. at 11,148. In 1971, however, the agency observed that "some belt-based concepts have been advanced that appear to be capable of meeting the complete passive protection options," leading it to add a new section to the standard "to deal expressly with passive belts." 36 Fed.Reg. 12,858, 12,859 (July 8, 1971).

The second modification relevant here was the subject of a separate notice issued on the same day. For the first time, perhaps because passive seatbelts had expressly been added to the proposed standard, the agency suggested that passive restraint systems contain an "emergency release" mechanism to facilitate extrication of passengers following a crash. The agency cautioned, however, that the emergency release capability could not be allowed to nullify the advantages of the passive restraint system:

In the case of passive safety belts, it would be required that the release not cause belt separation, and that the system be self-restoring after operation of the release. An example of such a system would be a lever on the belt retractor that releases the locking mechanism, allowing the belt to pay out freely. The self-restoration requirement could be fulfilled, for example, by a lever that frees the belt only while continuous pressure is exerted, or by a time-delay mechanism. In the case of an air bag system, deflation could constitute the "automatic" release.

36 Fed.Reg. 12,866 (July 8, 1971).

In its final form, the 1972 version of Standard 208 called for "complete passive protection" on vehicles manufactured after August 15, 1975. In the interim, vehicles built between August 1973 and August 1975 were to carry either passive restraints, or lap and shoulder belts coupled with an "ignition interlock" that would prevent starting the vehicle if the belts were not connected. Most car makers chose the second option, and the ignition interlock was one of the factors that caused the 1972 rule to founder in both the courts and in Congress. In Chrysler Corp. v. Dep't. of Transportation, 472 F.2d 659 (6th Cir. 1972), the Sixth Circuit held that "the Agency's decision to require passive restraints is supported by substantial evidence," but that the testing procedures required of passive belts did not satisfy the Safety Act's requirement that standards be "objective." Id. at 675. The Sixth Circuit's decision did not affect ignition interlocks, but by late 1974 the public's irritation at being unable to start a car without fastening seatbelts led Congress to reject the entire standard. The Motor Vehicle and Schoolbus Safety Amendments of 1974, Pub.L.No.93-492, § 109, 88 Stat. 1482 (codified at 15 U.S.C. § 1410b) (hereinafter cited as the 1974 Amendments), effected two main changes. First, Congress banned any federal motor vehicle safety standard requiring ignition interlocks or continuous buzzers to warn that seatbelts were not in use. Id. § 1410b(b)(1).4 Second, the 1974 Amendments sharply reduced DOT's discretion to modify Standard 208 in the future. If a modified standard could be satisfied by any system other than seatbelts only, the amended safety standard would have to be submitted to Congress where it might be vetoed by concurrent resolution of both houses. Id. § 1410b(b)(2).5

Predictably, the actions of the Sixth Circuit and Congress precipitated a new flurry of notices in the Federal Register. NHTSA proposed new warning systems to replace the prohibited continuous buzzers. See, e.g., 39 Fed.Reg. 42,692 (Dec. 6, 1974). It postponed the effective date for passive restraint systems so as to comply with the mandate of Congress and the testing procedures required by the Chrysler decision. See, e.g., 40 Fed.Reg. 16,217 (April 10, 1975). But most important, the agency had to comply with the congressional ban on ignition interlocks, and this had profound consequences for the case before us now. The 1974 Amendments were enacted on October 27, 1974. Nine months earlier, however, an agency rulemaking had reopened the question of emergency release mechanisms on passive seatbelts, 39 Fed.Reg. 3834 (Jan. 30, 1974), leading NHTSA to change its position of 1971 that these belts should not be detachable. In April 1974, NHTSA had adopted the suggestion of one automobile manufacturer that emergency release of passive belts be accomplished by a conventional latch-permitting belt separation-provided the restraint system was guarded by an ignition interlock and warning buzzer to encourage reattachment of the passive belt. 39 Fed.Reg. 14,593 (April 25, 1974). The newly enacted amendments to the Safety Act obviously made this provision untenable. Rather than return to the regulation as it was proposed in 1971, however, the agency simply eliminated the ignition interlock and buzzer requirements. 39 Fed.Reg. 38,380 (Oct. 31, 1974). Passive belts still were required to have an emergency release mechanism in the form of a latch mechanism that caused release at a single point by pushbutton action.

Despite these modifications, NHTSA retained its interest in a standard calling for mandatory passive restraints. It postponed the effective date of the requirement to August 31, 1976, but emphasized that this postponement was only for a single year. "The NHTSA intends to propose the long-term requirements for occupant crash protection ... as soon as possible." 40 Fed.Reg. 33,977 (Aug. 13, 1975). Shortly before that date, however, Secretary of Transportation William Coleman initiated a new rulemaking on the issue. 41 Fed.Reg. 24,070 (June 14, 1976). After hearing testimony and reviewing written comments, Coleman suspended the passive restraint requirement altogether. Although he found such restraints technologically and economically feasible, the Secretary based his decision on the expectation that there would be widespread public resistance to the new systems. Instead of a mandatory passive restraint standard, Coleman proposed a demonstration project involving up to 500,000 cars with passive restraints in order to smooth the way for such a standard at some later date. Department of Transportation, The Secretary's Decision Concerning Motor Vehicle Occupant Crash Protection (December 6, 1976) (Coleman Decision), Joint Appendix (J.A.) 2065.

Coleman's successor as Secretary of Transportation, Brock Adams, reopened the passive restraint rulemaking only four months later. 42 Fed.Reg. 15,935 (March 24, 1977). Following another round of written comments and a public hearing, Adams decided that the demonstration program was unnecessary. He issued a new mandatory passive restraint regulation, known here as Modified Standard 208. See 42 Fed.Reg. 34,289 (July 5, 1977). This regulation, covering eleven pages of the Code of Federal Regulations, 42 C.F.R. § 571.208 (1977), ordered a "phasing-in" of passive restraints based on vehicle size, beginning with large cars manufactured for the 1982 model year and extending to all cars manufactured for the 1984 model year and beyond.

Like its predecessor, the 1977 rule had to withstand testing in Congress and the courts. In Pacific Legal Foundation v. Dep't. of Transportation, 593 F.2d 1338 (D.C.Cir.), cert. denied, 444 U.S. 830, 100 S. Ct. 57, 62 L. Ed. 2d 38 (1979), this court upheld Modified Standard 208 as a rational, nonarbitrary regulation consistent with the agency's mandate under the Safety Act. Modified Standard 208 also fared well in Congress, which did not exercise its authority under the legislative veto provision of the 1974 Amendments. No action was taken by the full House of Representatives. The Senate committee with jurisdiction over NHTSA affirmatively endorsed the standard, S.Rep.No.481, 95th Cong., 1st Sess. (1977), and a resolution of disapproval was tabled by the Senate. 123 Cong.Rec. 33,332 (1977).

Congress has plainly considered Modified Standard 208 politically controversial, however. Riders were attached to appropriation bills for 1979 and 1980 that prohibited DOT from implementing the passive restraint standard in those years. Pub.L.No.95-335, § 317, 92 Stat. 435, 450 (1978); Pub.L.No.96-131, § 317, 93 Stat. 1023, 1039 (1979). The passive restraint standard was not scheduled to be implemented until 1981, however, and these measures emphasized the need for experimentation and research. See, e.g., H.R.Conf.Rep.No.1329, 95th Cong., 2d Sess. 14 (1978) ("The conferees intend that the language of this provision permits broad research and development activities related to the provisions of occupant restraint standard No. 208"). Of far greater importance were congressional efforts in 1980 to modify rather than nullify the standard. See, e.g., H.R.Rep.No.1371, 96th Cong., 2d Sess. (1980). These efforts, discussed in detail at pp. 224-229 infra, narrowly failed to become law, and Modified Standard 208 remained in effect.

Automobile manufacturers also reacted to Modified Standard 208. On May 22, 1978, NHTSA issued yet another notice of proposed rulemaking in response to a petition from General Motors (GM) concerning emergency release mechanisms on passive belts. GM sought reconsideration of the agency's 1974 decision to require latch mechanisms on passive belts.

(V)ery little was known at the time of that decision regarding designs of passive belt systems. Further, automatic belt "defeat" rates were not a significant issue at that time since no mandate existed which would have required the use of passive restraints on all vehicles .... While this conclusion was fully justified at that time we do not believe it has the same degree of validity when applied to the mandatory incorporation of automatic restraint systems. As we noted previously maximum usage will be a major concern beginning with the effective dates of the mandate established in June 1977; thus necessitating additional consideration for designs which show promise in helping to ensure that vehicle occupants avail themselves of the inherent safety benefits of automatic belt restraints.

43 Fed.Reg. 21,912, 21,914 (May 22, 1978) (quoting GM proposal). Specifically, GM sought authority to use a "spool release" mechanism that would allow emergency exit from vehicles without requiring separation of the belt. NHTSA noted its "interest" in "the anticipated rate of installation of passive belts and in any passive belt designs that would seek to minimize the rate of disconnection." Id. at 21,913. It granted GM's proposal six months later, although in a broader form so as to permit even greater experimentation with emergency release designs. 43 Fed.Reg. 52,493 (Nov. 13, 1978). Automobile makers began gearing up to comply with the safety regulation; by October 1981, they had expended a substantial portion of $550 million capital investment necessary to produce the required equipment.6

In February 1981, approximately one month after taking office, Secretary of Transportation Andrew Lewis reopened the rulemaking yet again. 46 Fed.Reg. 12,033 (Feb. 12, 1981). He based his decision at least in part on "the fact that economic circumstances have changed since the standard was adopted in 1977" and the "difficulties of the automobile industry," citing high unemployment, sales "at a very depressed level," and losses "by even the largest of the domestic manufacturers." Id. Two months later, the agency ordered a one-year delay in the application of the standard to large cars, extending the deadline to September 1982. 46 Fed.Reg. 21,172 (April 9, 1981). This notice also observed that the "economic situation of the industry and consumers and the economy as a whole have drastically changed since the standard was adopted in 1977." Id. at 21,174. On the same day, NHTSA proposed the possible rescission of the entire standard. 46 Fed.Reg. 21,205 (April 9, 1981). Both decisions were announced by the White House Press Office on April 6, 1981, as part of a larger package of economic recovery measures. See Actions to Help the U.S. Auto Industry at A-34 (April 6, 1981), J.A. 1281, 1321. After receiving written comments and holding public hearings, NHTSA issued a final rule (Notice 25) that rescinded the passive restraint requirement, and amended Standard 208 to eliminate this requirement. 46 Fed.Reg. 53,419 (Oct. 29, 1981) (Notice 25). This court denied motions to stay the rescission in December 1981, and these petitions for review followed.

The issues raised by Notice 25 are best introduced by a general discussion of the safety problem and the regulatory task faced by the agency. The starting point for both NHTSA and the petitioners in this action is the fact, based on survey data and accident reports, that American motorists overwhelmingly fail to use their seatbelts. The usage rate of seatbelts is not only low, but falling. In 1977, DOT estimated that usage of manual seatbelts was around 20%. 42 Fed.Reg. at 34,290. Today, NHTSA estimates that nationwide usage has fallen to 11%. 46 Fed.Reg. at 53,422.7

Although individuals must decide whether to wear a seatbelt, NHTSA and Congress have recognized that seatbelt usage has serious social consequences. It has been estimated that one American dies in a traffic accident every 11 minutes. S.Rep.No.481, 95th Cong., 1st Sess. 2-3 (1977). NHTSA predicts that there will be 61,710 motor vehicle fatalities in 1984, and that "because of the growing number of small cars, passenger car occupant deaths could increase by as much as 7,000 by 1990." NHTSA Final Regulatory Impact Analysis, Rescission of Automatic Occupant Protection Requirements (October 1981) (hereinafter cited as RIA), at V-12 n.4, XI-3, J.A. 160 n.4, 263. A significant number of these deaths-and an even larger number of serious injuries-could be prevented if motorists wore seatbelts. The resulting benefits to society include reduced insurance premiums, lower medical and rehabilitative costs, and incalculable savings in terms of human tragedy. Estimating the value of these benefits is necessarily inexact, but estimates can be made. In 1977, for example, DOT predicted that passive restraints could prevent approximately 12,000 deaths and over 100,000 serious injuries annually. 42 Fed.Reg. at 34,29 8. In Notice 25, NHTSA estimated that if the usage rate of seatbelts rose from 11% to 60%, Modified Standard 208 would save 8,750 lives and 176,900 serious injuries each year. RIA at IV-71, J.A. 127. The agency estimates the long-term savings in various kinds of insurance premiums alone at $4.3 billion annually, if these fatalities and deaths could be avoided. Id. at V-14, J.A. 162.8

Investment in safety also has its costs, of course. In 1977, for example, Secretary Adams estimated that passive seatbelts would exceed the cost of manual seatbelts by $25 per car, and that installation of airbags would cost $112 per vehicle. 42 Fed.Reg. at 34,293. These costs have risen with inflation. In 1981, NHTSA estimated that the marginal cost of passive restraints would vary between $50 and $150 per vehicle. The average figure was $89, which includes $15 for added fuel costs over the life of the car. 46 Fed.Reg. at 53,423; RIA at VI-40, J.A. 202.9 The cost of airbags has also risen sharply and depends to a great extent on the volume of production. If all vehicles manufactured annually were equipped with airbags, the resulting economies of scale would hold airbag costs to between $200 and $330 per car. At lower volumes, however, such as only 10,000 units a year, the cost of airbags could climb as high as $1,200 per vehicle. RIA at VI-10, J.A. 172. NHTSA estimated the cost of Modified Standard 208 to be approximately $1 billion per year. 46 Fed.Reg. at 53,423; RIA at VI-49, J.A. 211.

The regulatory problem facing NHTSA was therefore simple in theory, although extremely difficult in application. It had to predict the savings that would result from Modified Standard 208, and compare that savings with the cost of the standard, so as to conduct "a 'commonsense' balancing of safety benefits and economic cost." United States v. General Motors Corp., 518 F.2d 420, 435 (D.C.Cir.1975); see H & H Tire Co. v. Dep't. of Transportation, 471 F.2d 350, 353-54 (7th Cir. 1972).10 NHTSA found that the costs of the passive restraints required by Modified Standard 208 could be justified only if nationwide seatbelt usage rose by 13 percentage points, from 11% to 24%. RIA at A-10, J.A. 284. NHTSA concluded that no such increase would occur, and therefore rescinded the standard.

As discussed above, Modified Standard 208 as it stood in 1981 could have been satisfied by airbags or by either of two kinds of passive seatbelts. Both the "continuous" and the "detachable" passive seatbelts provide the emergency release mechanism required by the standard. On continuous belts, this mechanism may consist of a "spool-out" device that expands the belt but does not detach it. On detachable belts, the mechanism allows separation of the belt in the same way that manual seatbelts are buckled and unbuckled.

In Notice 25, NHTSA first found it "reasonably certain" that if Modified Standard 208 were implemented, "the overwhelming majority of new cars would be equipped with automatic belts that are detachable." 46 Fed.Reg. at 53,421. Sixteen automobile manufacturers had participated in the rulemaking. Two stated that they planned to use the detachable belts; three said they expected to do so, although they did not make a definite commitment; five predicted that the automotive industry as a whole would rely on detachable belts, without discussing their own design plans at all; and the remaining six were silent on the question.11

NHTSA then analyzed the efficacy of the detachable passive seatbelt. Passive seatbelts-both detachable and continuous-have been in use for some years, and the record shows that these devices increase seatbelt use on every model.12 NHTSA observed, however, that the data could be interpreted in greatly different ways.13 It also questioned, for a number of reasons, whether the data could be used to predict overall usage rates under Modified Standard 208.14 Finally, NHTSA analyzed the operation of the detachable belts, and found them functionally equivalent to manual seatbelts already in use.

Most planned automatic belts would be like today's manual lap and shoulder belts in that they can be easily detached and left that way permanently .... Some belt designs may be detached and permanently stowed as readily as the current manual lap and shoulder belts. Once a detachable automatic belt is detached, it becomes identical to a manual belt. Contrary to assertions of some supporters of the standard, its use thereafter requires the same type of affirmative action that is the stumbling block to obtaining high usage levels of manual belts.

46 Fed.Reg. at 53,421.

NHTSA's final step was to predict the usage rate of detachable passive belts that could be expected under the standard. Although Notice 25 is somewhat unclear as to the precise finding, NHTSA clearly believed that any increase in usage would be minimal. The agency

cannot reliably predict even a 5 percentage point increase as the minimum level of expected usage increase. The adoption of a few percentage points increase as the minimum would, in the agency's judgment, be more consistent with the substantial uncertainty about the usage rate of detachable automatic belts. Based on the data available to it, NHTSA is unable to assess the probability that the actual incremental usage would fall nearer a 0 percentage point increase or nearer some higher value like a 5 or 10 percentage point increase.

46 Fed.Reg. at 53,423. NHTSA cautioned that "the agency is not able to agree with assertions that there will be absolutely no increase in belt use as a result of automatic belts," id. at 53,425, but it repeatedly emphasized that any increase would be "extremely small due to the substantial similarity of the design and methods of using detachable automatic belts and manual belts" and that "detachable automatic belts may contribute little to achieving higher belt usage rates." Id. at 53,423.

Based on these three steps, NHTSA concluded that the savings from increased seatbelt usage under Modified Standard 208, if any, would not exceed the costs of the regulation:

In view of the possibly minimal safety benefits and substantial costs of implementing the automatic restraint requirements, the agency is unable to conclude that the incremental costs of the requirements are reasonable. The requirements are, in that respect, impracticable.

46 Fed.Reg. at 53,423. It therefore concluded that the standard should either be revised or rescinded.

2. Modified Standard 208 as NHTSA considered revising it

NHTSA recognized that its analysis of detachable passive belts did not apply to other kinds of passive restraints. "(T)he question then arises whether the agency should amend the standard to require that automatic belts have a use-inducing feature"-i.e., be continuous rather than detachable-that "would increase belt usage." 46 Fed.Reg. at 53,423. The agency rejected a refinement of the standard as "impracticable," however, for reasons of cost, equity, public reception, and safety.

Of these factors, only the last two received emphasis in Notice 25.15 NHTSA's first reason for not amending Modified Standard 208 to require "use-compelling features" on passive belts drew on the negative earlier reaction to ignition interlocks. "The history of the Congressional action which removed this authority from NHTSA suggests that Congress would look with some disfavor upon any similar attempt to impose a use-compelling feature on a belt system." Id. at 53,424. Second, NHTSA concluded that revising the standard to require only continuous belts would be "counterproductive" because of "irrational" reactions by users:

Recent attitudinal research conducted by NHTSA confirms a widespread, latent and irrational fear in many members of the public that they could be trapped by the seat belt after a crash. Such apprehensions may well be contributing factors in decisions by many people not to wear a seat belt at all.... (I)t would be highly inappropriate to impose a technology which by its very nature could heighten or trigger that concern.

Id. Finally, NHTSA suggested that this concern might not be so "irrational" after all. It suggested that "there are compelling safety reasons" why continuous belts-even those with emergency release features-should not be mandated.

In the event of accident, occupants wearing belts suffer significantly reduced risk of loss of consciousness, and are commonly able to extricate themselves with relative ease. However, the agency would be unable to find the cause of safety served by imposing any requirement which would further complicate the extrication of any occupant from his or her car, as some use-compelling features would.

Id.

Although these three reasons exhaust NHTSA's explicit discussion of its refusal to modify the passive restraint standard, it seems clear that the agency also relied on a fourth concern, albeit one stated only in the context of detachable belts. Much of Notice 25 focuses on concern about public attitudes toward government safety regulations, suggesting that a backlash against a passive restraint standard "might cause significant long run harm to the safety program." Id. at 53,424. Because detachable passive belts and manual belts may be perceived as functionally identical, "it is not unreasonable to conclude that the public may regard the automatic restraint requirements as an expensive example of ineffective regulation." Id. Adverse public reaction might lead some car owners to "cut the automatic belts out of their cars, thus depriving subsequent owners of the cars of the protection of any occupant restraint system," and could bring "a poisoning of popular sentiment toward efforts to improve occupant restraint systems in the future." Id. "A public that believes it is the victim of too much government regulation by virtue of the standard might well resist such parallel efforts (advertising campaigns and educational programs) to enhance voluntary belt usage." Id. at 53,425-26.16

Our review in this case proceeds under both the substantive sections of the Safety Act and the provision for judicial review of informal rulemaking in the Administrative Procedure Act (APA), 5 U.S.C. § 553 (1976). Section 103(b) of the Safety Act, 15 U.S.C. § 1392(b), states that the APA "shall apply to all orders establishing, amending, or revoking a Federal motor vehicle safety standard." The 1974 Amendments further specify that "(s)ection 553 of title 5 (the APA) shall apply" to occupant crash protection standards promulgated under the congressional review procedures. 15 U.S.C. 1410b(c)(2). As a result, the standard of review appears easily formulated. It is well established that the familiar "arbitrary and capricious" test applies to informal rulemaking conducted pursuant to section 553 of the APA. See, e.g., Camp v. Pitts, 411 U.S. 138, 93 S. Ct. 1241, 36 L. Ed. 2d 142 (1973); Pacific Legal Foundation v. Dep't. of Transportation, 593 F.2d at 1343.

The appropriate scope of judicial review remains the most troublesome question in this case, however, because we are called upon to review the rescission rather than the promulgation of an agency rule. The scope of review in such a situation appears to be a matter of first impression, even though judicial review of orders revoking a standard is specifically authorized by the Safety Act and the APA. The appropriate scope of our review is also a question of some complexity. At first view, rescission more resembles agency refusal to act than an agency decision to act, and the distinction has significance for the degree of judicial deference paid to the agency. As we recently noted in WWHT, Inc. v. FCC, 656 F.2d 807, 818-19 (D.C.Cir.1981), it is only in the rarest and most compelling circumstances that courts overturn an agency's " 'expert' determination not to pursue a particular program or policy at a given time." Although the WWHT court held that an agency's denial of a petition for rulemaking was subject to judicial review, the opinion emphasized that "the scope of review of such a determination must, of necessity, be very narrow." Id. at 809. In Natural Resources Defense Council v. SEC, 606 F.2d 1031 (D.C.Cir.1979), the court reached a similar conclusion in reviewing a situation in which the agency terminated a rulemaking proceeding without issuing a rule:

As is typical in informal rulemaking cases under section 4 of the APA, 5 U.S.C. § 553, many of the issues raised here are within the province of agency expertise and do not readily lend themselves to judicial oversight.... (O)ur review of the Commission's factual, and particularly its policy, determinations will perforce be a narrow one ....

Id. at 1052-53. These cases may be distinguished, of course, from the one before us. NHTSA has not denied a petition for rulemaking, or failed to issue a rule after a proceeding, but has rescinded a rule that has already been promulgated. Even so, the parallels are obvious, and dictate caution in formulating the appropriate scope of review here.

In recent years, however, courts have increasingly emphasized that the "arbitrary and capricious" standard encompasses intensive as well as deferential judicial scrutiny, depending in part on "the nature of the particular problem faced by the agency." Natural Resources Defense Council, 606 F.2d at 1050. In Citizens to Preserve Overton Park, Inc. v. Volpe, 401 U.S. 402, 415, 91 S. Ct. 814, 823, 28 L. Ed. 2d 136 (1972), the Supreme Court interpreted the arbitrary and capricious test to require a "substantial inquiry" subjecting the agency's action to "a thorough, probing, in-depth review."17 The prior adjudications under the Safety Act concerning Standard 208 also illustrate that this review may be searching as well as deferential. Each decision reviewed NHTSA's issuance of the passive restraint standard for whether it was supported by "substantial evidence" within the meaning of 5 U.S.C. § 706(2)(E), even though in each instance the agency had promulgated the standard after informal rulemaking. In Chrysler Corp. v. Dep't. of Transportation, the Sixth Circuit concluded that the substantial evidence test was required by the Safety Act, which mandates that "all of the evidence before the agency ... shall be included in the record" submitted to the reviewing court. 15 U.S.C. § 1394(a)(1) (referring to 28 U.S.C. § 2112(b)). 472 F.2d at 668.18 In Pacific Legal Foundation v. Dep't. of Transportation, the court conducted a "thorough, probing, in-depth review" of the record, but declined to follow the Sixth Circuit's invocation of the substantial evidence test because any difference was "largely semantic":

We do not follow this reasoning because we agree with the emerging consensus of the Courts of Appeals that the distinction between the arbitrary and capricious standard and substantial evidence review is largely semantic, and that "in the review of rules of general applicability made after notice and comment rulemaking, the two criteria do tend to converge." Associated Industries of New York State, Inc. v. Dep't. of Labor, 487 F.2d 342, 349-50 (2d Cir. 1973) .... (W)e agree with Judge Lumbard that "when an agency engages in substantive rulemaking, it abuses its discretion (or acts arbitrarily and capriciously) if its actions are not supported by substantial evidence." Nat'l. Nutritional Foods Ass'n v. Weinberger, 512 F.2d 688, 705 (2d Cir. 1975) (Lumbard, J., concurring in the result).

593 F.2d at 1343 n.35. See Recording Industry Ass'n v. Copyright Royalty Tribunal, 662 F.2d 1, at 7-8 (D.C.Cir. Aug. 27, 1981); Sierra Club v. Costle, 657 F.2d 298, 323 n.67 (D.C.Cir.1981); Paccar, Inc. v. NHTSA, 573 F.2d 632, 636 (9th Cir.), cert. denied, 439 U.S. 862, 99 S. Ct. 184, 58 L. Ed. 2d 172 (1978); American Public Gas Ass'n v. FPC, 567 F.2d 1016, 1029 (D.C.Cir.1977), cert. denied, 435 U.S. 907, 98 S. Ct. 1456, 55 L. Ed. 2d 499 (1978); Bunny Bear, Inc. v. Peterson, 473 F.2d 1002, 1006 (1st Cir. 1973).

In short, although all parties in this case agree that we should apply the arbitrary and capricious standard of review, they differ markedly about the intensity and rigor with which that standard should be applied. NAII contends that there is a "heavy burden" on NHTSA to explain that its rescission is "rational and supported by substantial evidence." Brief for Petitioner NAII (NAII Brief) at 19. State Farm urges that our review be "intensive and exacting," Brief for Petitioner State Farm (State Farm Brief) at 21. Although NHTSA "welcomes the most intense scrutiny" and contends that Notice 25 is "supported by the evidence in the record," Brief for Respondent NHTSA (NHTSA Brief) at 19, the agency also urges that we exercise "a high degree of deference to the agency's determination." Id. at 24 (quoting Natural Resources Defense Council, 606 F.2d at 1050).

Before we decide this question, it may be useful to ask why the same verbal standard of review should be given different scope in different contexts. Part of the answer was suggested in Pacific Legal Foundation v. Dep't. of Transportation, where the court explained that its "probing" review was required by the fact that Secretary Adams' decision to issue the 1977 rule (Modified Standard 208) had come on the heels of Secretary Coleman's decision four months earlier that such a standard was not required:

In addition, because the order under review here reversed a prior policy, the agency must provide "an opinion or analysis indicating that the standard is being changed and not ignored, and assuring that it is faithful and not indifferent to the rule of law."

593 F.2d at 1343-44 (quoting Columbia Broadcasting System, Inc. v. FCC, 454 F.2d 1018, 1026 (D.C.Cir.1971)). As Judge Leventhal observed over a decade ago, sharp changes of agency course constitute "danger signals" to which a reviewing court must be alert. Joseph v. FCC, 404 F.2d 207, 212 (D.C.Cir.1968). He elaborated in a later case:

Judicial vigilance to enforce the Rule of Law in the administrative process is particularly called upon where, as here, the area under consideration is one wherein the Commission's policies are in flux. An agency's view of what is in the public interest may change, either with or without a change in circumstances. But an agency changing its course must supply a reasoned analysis indicating that prior policies and standards are being deliberately changed, not casually ignored ....

Greater Boston Television Corp. v. FCC, 444 F.2d 841, 852 (D.C.Cir.), cert. denied, 403 U.S. 923, 91 S. Ct. 2229, 29 L. Ed. 2d 701 (1971). See RKO General, Inc. v. FCC, 670 F.2d 215, 223-24 (D.C.Cir.1981); Local 777, Democratic Union Organizing Comm. v. NLRB, 603 F.2d 862, 882 (D.C.Cir.1978) (agency must announce "principled reason" for reversal of policy).

The difficulty with these observations is that they blend judicial review of agency adjudications and licensing with review of agency rulemaking. Intuitively, at least, it seems obvious that concern for consistency in an agency's decisionmaking should be greatest when the agency departs from an adjudicative precedent or a well established licensing rule, and less grave when the agency reconsiders the value of a "quasi-legislative" rule of general applicability. Precedents by definition must be the basis for future reliance, and a sudden agency departure suggests that parties may have been treated unfairly. See, e.g., Boston Edison Co. v. FPC, 557 F.2d 845, 849 (D.C.Cir.), cert. denied sub nom. Towns of Norwood, et al. v. Boston Edison Co., 434 U.S. 956, 98 S. Ct. 482, 54 L. Ed. 2d 314 (1977) (should agency decide to reverse its course, it must give notice "and apply the changed standard only to those actions taken by parties after the new standard has been proclaimed as in effect"). In contrast, even the eleventh-hour decision not to promulgate a regulation has less impact on regulated parties. This may explain why rulemaking proceedings that terminate short of agency actions, e.g., Natural Resources Defense Council v. SEC, 606 F.2d 1031, and agency decisions not to conduct rulemaking proceedings at all, e.g., WWHT, Inc. v. FCC, 656 F.2d 807, are tested under a "very narrow" reading of the arbitrary and capricious test. The agency's refusal to act may reflect its desire to use scarce resources on more pressing problems, or its judgment that a problem is trivial or nonexistent. See Moog Industries, Inc. v. FTC, 355 U.S. 411, 413, 78 S. Ct. 377, 379, 2 L. Ed. 2d 370 (1958). Agency departure from precedent raises obvious problems, but why should courts have similar concerns about erratic agency policymaking or reversals in the course of rulemaking?

The answer to this question lies in the fact that an agency is not a legislature.19 Congress delegates rulemaking power in the anticipation that agencies will perform particular tasks. Reviewing courts are required to strike down agency actions that exceed this mandate. See, e.g., 5 U.S.C. § 706(2)(C). Even when there is no claim that the agency has exceeded its jurisdiction, as there is not in this case, sudden and profound alterations in an agency's policy constitute "danger signals" that the will of Congress is being ignored. The few cases in which agency decisions not to institute rulemaking have been overturned, for example, primarily involve plain errors of law, suggesting that the agency has been blind to the source of its delegated power. See, e.g., NAACP v. FPC, 520 F.2d 432 (D.C.Cir.1975), aff'd, 425 U.S. 662, 96 S. Ct. 1806, 48 L. Ed. 2d 284 (1976) (Commission erred in concluding that it lacked jurisdiction to promulgate regulations concerning racial discrimination by licensees); NORML v. Ingersoll, 497 F.2d 654 (D.C.Cir.1974) (bureau erred in rejecting filing of petition for rulemaking for reasons going to the merits). Cf. Los Angeles Women's Coalition v. FCC, 584 F.2d 1089 (D.C.Cir.1978) (per curiam) (remanding Commission's denial of hearing on petition to deny license for further development of factual issues). In Geller v. FCC, 610 F.2d 973, 979 (D.C.Cir.1979) (per curiam), the court reversed the agency's "plainly misguided" refusal to consider the possible effect of newly enacted copyright legislation on its earlier regulations. As we recently described the Geller rule, "an agency may be forced by a reviewing court to institute rulemaking proceedings if a significant factual predicate of a prior decision on the subject (either to promulgate or not to promulgate specific rules) has been removed." WWHT, Inc. v. FCC, 656 F.2d at 819. This articulates in specific terms the general principle that administrative agencies derive their power from the laws of Congress and have no authority to act inconsistently with their statutory mandate. The same tenet may be identified in adjudications under 5 U.S.C. § 706(1) to "compel agency action unlawfully withheld or unreasonably delayed."

In determining the scrutiny with which the arbitrary and capricious standard should be applied to NHTSA's rescission of Modified Standard 208, then, we must first consider the extent to which NHTSA's action may be inconsistent with the congressional purpose behind the Safety Act. It may seem unusual to discuss this legislative history before a precise standard of judicial review has been formulated, but in this case there is no better way to undertake such a task.

Our review of the legislative history of the 1974 Amendments to the Safety Act and the subsequent congressional reaction to Modified Standard 208 suggests that the standard has come as close as an agency-made regulation can come to being affirmatively endorsed by Congress, without Congress actually having done so. Although Congress has always considered the standard politically controversial, the regulation has received sufficient congressional approval to raise doubts that NHTSA's rescission necessarily demonstrates an effort to fulfill its statutory mandate. Three separate periods of the standard's history in Congress merit close attention. The first is late 1974, when Congress banned the ignition interlock and continuous buzzer but did not foreclose NHTSA's pursuit of a passive restraint standard. The second period is late 1977, when NHTSA submitted Modified Standard 208 to both houses of Congress in accordance with the legislative veto provisions of 15 U.S.C. § 1410b, but concurrent resolutions of disapproval were not enacted. The final period is early 1980, when Congress contemplated new amendments to the Safety Act that would have refined Modified Standard 208 but would not have abolished it.

In 1974, Standard 208 had not yet been modified by Secretary Adams. It called for the mandatory installation of passive restraints by August 15, 1976, and provided that during the preceding two-year period automobiles either should have passive restraints, or should be equipped with an ignition interlock and continuous buzzer to encourage use of manual seatbelts. As noted above, see pp. 210-211 supra, the public reaction to the interlock and continuous buzzer was swift and furious, and a chief purpose of the 1974 Amendments was to ban them. Congress could not help knowing about the pending date for mandatory passive restraints, however, and the instructive aspect of the 1974 period is that Congress encouraged NHTSA to proceed with such a standard under carefully constructed legislative conditions.

The 1973 Senate bill for NHTSA authorizations contained no provision dealing with the passive restraint standard. But see 119 Cong.Rec. 16,054 (1973) (Senator Magnuson) (noting that airbags "have proven to be both reliable and life saving," and proposing $3 million authorization "to equip General Services Administration vehicles with airbag systems"). In the House, however, opponents of the ignition interlock broadened their criticism to include passive restraints as well. Representative Wyman, who later described himself as "the prime sponsor of the interlock prohibition," 120 Cong.Rec. 35,637 (1974), proposed a broad amendment to the Safety Act that would have made it impossible for NHTSA ever to promulgate a standard requiring mandatory passive restraints. Id. at 27,822. Representative Moss proposed to amend this amendment so as not to affect the "pending regulation, which already has been published, on which comment has been received, for the passive restraint system to become effective" as planned. Id. But Representative Wyman insisted on his version, arguing that passive restraints should be available only as options, and the Wyman amendment passed overwhelmingly. Id. at 27,822-23.

When the House and Senate bills went to conference, however, the Wyman amendment was discarded. Instead, the conference version proposed:

No occupant restraint system other than a belt system could become effective until Congress was given an opportunity to consider such standard for sixty days of continuous session (except that DOT could permit a manufacturer (at his option) to comply with a standard with a nonbelt system instead of a belt system).

H.R.Conf.Rep.No.1452, 93d Cong., 2d Sess. 45 (1974). The Senate approved the report without a recorded vote, 120 Cong.Rec. 35,037 (1974), but most relevant comments were favorable. Senator Magnuson suggested that

exciting new technology on the immediate horizon ... will save more lives than all of the other (motor vehicle) standards heretofore promulgated put together.... Congress has vested the Department with all of the authority it needs to get the job done. The burden is now with the Department to move forward with a sense of urgency. The problem is urgent.

Id. at 35,036; see id. (statement of Senator Hartke) (noting high benefit-cost ratios of passive restraint systems). The House also approved the conference report without a recorded vote, id. at 35,637, with Representative Wyman also urging its adoption, id.

As the 1974 Amendments were interpreted by members of Congress, NHTSA was required to follow four main steps before a "nonbelt" safety standard could be promulgated.20 Representative Staggers, chairman of the House Committee on Interstate and Foreign Commerce with jurisdiction over the legislation, explained that the 1974 Amendments required (1) a modification in DOT's notice and comment requirements in order "to permit interested persons to present oral presentation"; (2) specific provision for "input" by members of Congress; (3) transmittal of any proposed standard to Congress; and (4) application of the legislative veto provisions of 15 U.S.C. § 1410b. See 120 Cong.Rec. 35,63 6 (1974). With these conditions, the way was clear for NHTSA to try again. Congress obviously anticipated that a test of the passive restraint standard would come when a revised standard was sent to it on a later date. NHTSA complied with these procedural requirements, and transmitted Modified Standard 208 to Congress on June 30, 1977.

Concurrent resolutions to disapprove Modified Standard 208 were introduced in each house almost immediately upon congressional receipt of the standard. See, e.g., 123 Cong.Rec. 21,760 (1977) (resolution introduced by Senator Griffin on June 30, 1977); id. at 24,168 (seven identical resolutions introduced in the House on July 20, 1977). These resolutions were then sent to the appropriate committees. In the House, the Subcommittee on Consumer Protection and Finance of the Committee on Interstate and Foreign Commerce concluded hearings on September 23.21 The subcommittee voted by voice vote to recommend to the full committee that the resolution of disapproval not pass. The full committee voted to table the resolution of disapproval on October 12. As a result, the standard was not considered by the House and no concurrent resolution of disapproval was passed.22

In the Senate, four days of hearings were held by the Consumer Subcommittee of the Committee on Commerce, Science, and Transportation. The subcommittee voted unanimously, 5-0, to disapprove the resolution of disapproval. The full committee voted by voice vote to report the resolution to the Senate, again with the recommendation that the resolution of disapproval not pass. S.Rep.No.481, 95th Cong., 1st Sess. (1977). On October 12, the full Senate voted 65-31 to table the resolution. 123 Cong.Rec. 33,332 (1977). Modified Standard 208 could have been disapproved only by concurrent resolution of both houses; neither house voted to disapprove the standard, and Modified Standard 208 thus went into effect.

Throughout the Senate review, Modified Standard 208 received not grudging acceptance but positive support. The Senate report stated that the standard "would provide major increased protection for front-seat automobile occupants" that could save "more than $3.5 billion annually." S.Rep.No.481, 95th Cong., 1st Sess. 2-3 (1977).23 On the floor, one speaker after another affirmed the need for the standard. See, e.g., 123 Cong.Rec. 33,318 (Senator Ford) ("the hearing record contains overwhelming support for the Department of Transportation's rule"); id. at 33,319 (Senator Ribicoff) (noting "well proven" efficiency of airbags); id. at 33,320 (Senator Durkin) ("the only way to have our citizens protected by passive restraints is to mandate them"); id. at 33,325 (Senator Bentsen) (the rule is "in the best interest of the public .... (P)assive restraint systems are the most effective way to improve vehicle safety"); id. at 33,329 (Senator Magnuson) ("it is time to put this matter to rest"); id. at 33,330 (Senator Baker) (supporting standard because "there is ample evidence indicating that air bags are effective in preventing injuries, that they would save thousands of lives and prevent many more serious injuries annually if installed on all cars, and that they are not hazardous"). Even critics of the standard focused more on the desirability of further testing and promotion of the systems than outright opposition to the standard. See, e.g., id. at 33,322 (Senator Goldwater) ("trying to argue against the idea of saving lives is a lot like arguing against free beer and mother love. It is a difficult thing to do"); id. at 33,325 (Senator Cannon) (urging DOT to reinstitute Coleman demonstration project); id. at 33,327 (Senator Griffin) (seeking more experimentation or reinstatement of Coleman demonstration project).

The refusal by Congress in 1977 to disapprove Modified Standard 208 did not terminate debate on the question of passive restraints. Opponents of passive restraints in the House criticized not only the standard, but the fact that the 1977 resolution of disapproval had been "bottled up" in committee, thus depriving its supporters of opportunity for floor debate. See, e.g., 124 Cong.Rec. H 5308 (daily ed. June 12, 1978) (Representative Shuster); id. at H 5313 (Representative Devine). As discussed above, see p. 12 supra, riders were attached to NHTSA appropriations bills for 1979 and 1980 that prohibited DOT from implementing the passive restraint standard in those years. Congress recognized that the standard was not scheduled to be implemented until 1981, however, and these measures did not interfere with NHTSA research and testing of passive restraints. See id. at H 5309 (Representative Conte); id. at H 5313 (Representative Staggers). Representative Dingell, a co-sponsor of the 1979 rider, observed that "(t)he amendment does not interfere with the progression toward the effective date, model year 1982, of the passive restraint standard." 125 Cong.Rec. H 8055 (daily ed. Sept. 18, 1979).24

The passive restraint standard was again examined in great detail by the 96th Congress during deliberations on the Motor Vehicle and Cost Savings Authorization Act of 1980. The Senate bill, S 1159, made no reference to the standard. See 125 Cong.Rec. S 9166 (daily ed. July 11, 1979). When the House bill, H.R. 2585, came to the floor, a number of speakers praised the standard. See, e.g., id. at H 12,282 (daily ed. Dec. 19, 1979) (Representative Staggers) (standard will "save not one, but many thousands" of lives); id. (Representative Mineta) (noting "grave concern about any amendments that would have the effect of delaying implementation of the automatic restraint program," and that "(e)ach time the standard has been reviewed, it has been reconfirmed"); id. (Representative Maguire) ("we should go forward aggressively with this program"). Other speakers found it "disheartening" that "the Congress again appears to be going along with (NHTSA) in its continuing tendency toward excessive reliance on the airbag." Id. at H 12,283 (Representative Cleveland).

On the floor, however, Representative Stockman proposed an amendment to the Safety Act that would have denied NHTSA funds to enforce or administer an occupant restraint system

unless such standard or regulation also permits the purchaser of a passenger car to select any occupant restraint system which, if installed in the passenger car purchased by such purchaser, would comply with the requirements of Federal Motor Vehicle Safety Standard Number 208 (49 Code of Federal Regulations 571.208), relating to the installation of active seat belt systems, as in effect at the end of June 29, 1977.

Id. at H 12,285. Representative Stockman explained that the amendment "preserves the right of the consumer to choose either an active restraint system, which is the lap-shoulder belt that we have on cars today, or a passive restraint system, which in practice means the airbag for large cars, and the automatic seatbelt for small cars." Id.

Representative Scheuer, chairman of the reporting committee, did not oppose the Stockman amendment "because it is only a symbolic amendment with no real or direct legal impact on the Department of Transportation's passive restraint standard." Id. He explained:

The standard is scheduled to go into effect beginning in 1982. This amendment, however, is an amendment to a 1-year bill authorizing the appropriations of funds for carrying out the National Traffic and Motor Vehicle Safety Act for fiscal year 1980 only. Further, the amendment continues to allow for the installation of airbags or passive seatbelts, while providing for an additional option-active seatbelts. This is an option which would have existed regardless of whether this 1-year amendment was adopted.

Id. Representative Stockman responded, however, that "this is a proposal we have not voted on previously. This is a compromise solution that is designed to mandate the introduction of these things into the market, mandate the offering of passive restraints on every car that is sold by any manufacturer in the U.S. market, but give the consumer the choice of which system he will actually choose to have on his car." Id. The Stockman amendment passed overwhelmingly, 320-73. It should be noted that most of the members who had earlier praised Modified Standard 208-including Representatives Mineta, Scheuer, and Staggers-were among those voting with the majority. Id. at 12,287.

Like the Wyman amendment six years earlier, the Stockman amendment was also discarded when the House and Senate bills went to conference. The Conference Report proposed several important revisions to Modified Standard 208, however, in the form of a proposed amendment to the Safety Act. First, the amendment would have accelerated the date by which the standard applied to small cars. H.R.Conf.Rep.No.1371, 96th Cong., 2d Sess. 15 (1980). The legislative history suggests two reasons for this change. "With the anticipated increase in smaller, more fuel efficient (and more dangerous) cars on the Nation's highways, it is in the interest of public safety that passive occupant restraint protection should be required in these cars as soon as possible." Id. Moreover, the standard as promulgated would not have applied to small cars until two years after the effective date for compliance on large cars, and it was felt that this "would place domestic manufacturers at a great disadvantage against foreign competition." 126 Cong.Rec. S 13,499 (daily ed. Sept. 25, 1980) (Senator Cannon).

Second, the conference substitute recognized the trend by automobile manufacturers toward phasing out various large car models in favor of smaller ones. It therefore exempted "certain smaller manufacturers from having to install automatic occupant restraint systems in mid-sized cars that will (no longer) be produced after December 31, 1982," although these manufacturers would still be required to comply with the applicable requirements for models produced the following year. H.R.Conf.Rep.No.1371, 96th Cong., 2d Sess. 15 (1980). Third, the conference substitute would have required that

beginning with model year 1983, each seatbelt assembly installed in a passenger car must be detachable by the user in a manner which does not impair the subsequent reattachment and performance of the assembly. The conferees intend that the passive belt can be detached at any point, including one adjacent to the inboard anchor.

Id. at 17.

The final and perhaps most significant revision proposed by the conference report concerned airbags. The five automobile manufacturers with the largest sales-GM, Ford, Toyota, Nissan, and Volkswagen-would have been required to "tool and offer for sale" either as an option or as standard equipment airbags on at least one car line in any three of the four model years between September 1981 and September 1985. The conference report provided careful definitions of "tooling up" and "car line,"25 and added:

It is clearly the expectation of the conferees that consumers have a meaningful choice in the market between vehicles equipped with automatic safety belts and with airbags. It is the hope of the conferees that the public will be informed of such choices and that automatic safety belts and airbags be made available at reasonable cost.

Id. at 16.

As Representative Maguire observed during the House consideration of the conference report, the requirement that larger car companies offer airbags "is a major change in policy." 126 Cong.Rec. H 10,196 (daily ed. Oct. 1, 1980). Prior to this time, proponents of the passive restraint standard had always emphasized that

(t)he standard is a performance standard which does not require that any specific technology be utilized by automobile manufacturers.... Thus, the standard does not mandate that air cushion restraint systems (the air bag) be used. Passive belts may be used, or any other technology which meets the standard.

S.Rep.No.481, 95th Cong., 1st Sess. 1 (1977); see id. at 17-18 (explaining reasons for not requiring airbags as a mandatory option). Now, for the first time, a committee of Congress "has gone on record to assure consumers freedom of choice in the selection of restraint systems by requiring that some airbags be made available to consumers who want to buy them." 126 Cong.Rec. H 10,196 (daily ed. Oct. 1, 1980) (Representative Maguire); see id. at H 10,194 (Representative Scheuer).

This "major change in policy" raised certain procedural problems, however, because Congress frowns on conference reports that depart significantly from the bills passed in the separate houses. The Senate adopted the conference report, without a recorded vote. 126 Cong.Rec. S 13,506 (daily ed. Sept. 25, 1980). As had been the case in 1977, most comments on the standard were favorable. See, e.g., id. at S 13,502-03 (Senator Warner) (supporting "the concept for automatic occupant protection" because vehicle fatalities and injuries have reached "epidemic proportions," costing society billions of dollars annually). In the House, however, points of order were raised against the conference report because of clauses 3 and 4 of House Rule XXVII, rules of scope and germaneness designed to prevent conferees from writing new legislation in conference.

In such a situation, the conference report may still be considered in two ways. The more common is to obtain a rule from the House Committee on Rules waiving the points of order. Alternatively, the conference report can be considered under a suspension of the rules, although this requires a two-thirds vote before the report is passed. The sponsors of the 1980 conference report chose the second course, a highly unusual move suggesting their confidence that the report would be accepted. See 126 Cong.Rec. H 10,194 (daily ed. Oct. 1, 1980) (Representative Scheuer) ("The conferees carefully constructed a compromise which is simultaneously supported by (DOT and NHTSA) and which GM and Ford have both said they can live with"). The reaction in the House was mixed, however. Compare id. at H 10,204 (Representative Mineta) ("The agreement reached by the conferees is a strong endorsement of the automatic crash protection standard 208") with id. at H 10,198 (Representative Frenzel) ("this is the first time that the Congress will have ever mandated airbags by law and, because of that, in my judgment, this conference report ought to be rejected out of hand"). A majority of the House voted in favor of the report, 209-192, but this was less than the two-thirds vote required under the suspension of the rules.

The sponsors of the conference report then sought and obtained a rule that would waive points of order. H.R.Rep.No.1500, 96th Cong., 2d Sess. (1980) (10-4 vote of House Committee on Rules in favor of such a rule); see 126 Cong.Rec. H 11,912 (daily ed. Dec. 4, 1980) (Representative Bolling). This rule was strongly opposed on the floor, however, and was narrowly defeated, 165-168. Id. at H 11,918; see id. at H 11,912 (Representative Quillen) ("If we do not draw that line now, we will be inviting future conferees to write completely new legislation in conference in total disregard of the wishes of the House"). As before, however, Congress recognized that the battle over amendments to Modified Standard 208 did not prevent the standard from taking effect. Representative Dingell observed:

Mr. Speaker, this is not a partisan question.... The question is not even whether airbags or passive restraints are going in, because under existing law passive restraints, as opposed to airbags, which include airbags but which also include other devices such as passive belts, are still required under regulation and some passive belts and others are and new ones will shortly be on the market.

Id. at H 11,913. Minutes later, Representative Dingell repeated that "if we reject this proposal, Airbags and Motor Vehicle Safety Standard 208, which is a performance standard, will still be in place and will not be affected by the action of the House in rejecting the action of a group of runaway conferees who have openly flaunted and disregarded the will of the House." Id. at H 11,916.

This latter effort to waive points of order against the conference report was made on December 4, 1980, the day before the end of session. On December 5, Representatives Dingell and Broyhill urged passage of "a compromise bill," H.R. 8379, which Representative Dingell explained "is again an attempt to see to it that the consumer has an opportunity for choice in the purchase of passive restraints without mandating air bags in passenger automobiles." 126 Cong.Rec. H 12,119 (daily ed. Dec. 5, 1980). The compromise bill also proposed to reverse the effective dates of Modified Standard 208, requiring that small cars comply before large cars, but would have retained the requirement that all cars have passive restraints by 1984. A large number of the representatives who had voted in favor of the Conference Report now voted against H.R. 8379, however. Representative Leland commented:

To reverse decisions that have been made in behalf of the people of America and their safety, in my estimation, is a matter of madness and in the vernacular of the community from whence I come, this bill is jive.

Id. at H 12,123. See id. (Representative Eckhardt); id. (Representative Conyers). A majority voted in favor of the compromise bill, 205-126. But because it too had been considered under a suspension of the rules, H.R. 8379 also was rejected, and never even came up for consideration in the Senate.

In summary, each of the three periods during which Congress closely considered the passive restraint standard represents a different type of legislative action. In 1974, Congress banned the ignition interlock but did not foreclose NHTSA's pursuit of a passive restraint standard. In 1977, Congress allowed the standard to take effect when neither of the concurrent resolutions needed for disapproval was passed. In 1980, a majority of each house indicated support for the concept of mandatory passive restraints, and a majority of each house supported the unprecedented attempt to require some installation of airbags. None of these acts carries the weight of positive law, of course. The 1977 failure to disapprove the standard cannot be read as a congressional enactment-even though it is easier to obtain a two-house veto than passage of a bill-and the 1980 efforts failed to become law. Reading this legislative history as a whole, however, suggests a congressional commitment to the concept of automatic crash protection devices for vehicle occupants that we may not take lightly.

Based on the legislative reaction to the passive restraint standard discussed above, we conclude that rescission of the standard must be subject to "thorough probing, in-depth review" lest the congressional will be ignored. On the facts of this case, our review must be as "searching and careful" as the judicial review in Pacific Legal Foundation v. Dep't. of Transportation, where the issue was the promulgation rather than the rescission of Modified Standard 208.

Mere legislative silence in the face of agency action is a hazardous basis from which to infer congressional approval of the agency's interpretation of the statutes involved. "However, a consistent administrative interpretation of a statute, shown clearly to have been brought to the attention of Congress and not changed by it, is almost conclusive evidence that the interpretation has congressional approval." Kay v. FCC, 443 F.2d 638, 646-47 (D.C.Cir.1970); see Udall v. Tallman, 380 U.S. 1, 17-18, 85 S. Ct. 792, 801-802, 13 L. Ed. 2d 616 (1965). The Safety Act directs that the Secretary "shall" issue appropriate motor vehicle safety standards; ordinarily this term "is the language of command." Escoe v. Zerbst, 295 U.S. 490, 493, 55 S. Ct. 818, 819, 79 L. Ed. 1566 (1935) (Cardozo, J.). Despite the legislative battles over whether to specify an airbag requirement or when the standard should take effect, each time Congress reviewed the passive restraint standard it was essentially confirmed. "In the matter before us there is not merely silence, proposals languishing without any Congressional action, but positive action by Congress rejecting the limiting amendments." National Automatic Laundry & Cleaning Council v. Shultz, 443 F.2d 689, 706 (D.C.Cir.1971). Although there may be situations in which an agency may repeal a regulation on no basis at all, such is not the case here. NHTSA is not writing on a clean slate; it cannot suggest that the congressional actions and failures to act described above have no bearing on the agency's freedom to regulate on this question. It follows that NHTSA has the burden of explaining why it has changed course, and of showing that rescission of Modified Standard 208 was reasonable.26

This does not mean that NHTSA may not revoke the standard. If the agency clearly articulates a reasonable basis for that action, we must defer to the policy judgments and expertise of the agency. By no means may we substitute the court's judgment, or fail to "guard against the danger of sliding unconsciously from the narrow confines of law into the more spacious domain of policy." Phelps Dodge Corp. v. NLRB, 313 U.S. 177, 194, 61 S. Ct. 845, 852, 85 L. Ed. 1271 (1941). The line between substitution of judgment and judging whether action has been rational or arbitrary is sometimes fine, and to some critics imperceptible. We fully recognize, however, the side of that line from which the court derives its lawful authority. We also recognize that the course of administrative action is frequently and inevitably characterized by shifts and turns, as agencies continually reassess what is in the public interest. But courts are "not obliged to stand aside and rubber-stamp their affirmance of administrative decisions that they deem inconsistent with a statutory mandate or that frustrate the congressional policy underlying a statute." NLRB v. Brown, 380 U.S. 278, 291, 85 S. Ct. 980, 988, 13 L. Ed. 2d 839 (1965). Judicial scrutiny of agency action-including the rescission of a rule-depends on the extent to which the agency has deviated from congressional expectations. An agency is seldom locked on course, but it must have increasingly clear and convincing reasons the more it departs from the path marked by Congress.

We therefore review NHTSA's rescission of Modified Standard 208 for whether the agency has engaged in reasoned decisionmaking, making actual judgments concerning the significance of the evidence in the record and supporting its decision with "reasoned analysis." City of Charlottesville v. FERC, 661 F.2d 945, 951 n.35 (D.C.Cir.1981). We must ascertain the facts on which NHTSA relied, determine whether those facts have some basis in the record, and judge whether a reasonable decisionmaker could respond to those facts as the agency did. Recording Industry Ass'n v. Copyright Royalty Tribunal, at 7. The court must also assure itself that rescission is "based on consideration of the relevant factors," FCC v. National Citizens Comm. for Broadcasting, 436 U.S. 775, 802-03, 98 S. Ct. 2096, 2115-2116, 56 L. Ed. 2d 697 (1978); see Home Box Office, Inc. v. FCC, 567 F.2d 9, 36 (D.C.Cir.), cert. denied, 434 U.S. 829, 98 S. Ct. 111, 54 L. Ed. 2d 89 (1977), and determine that NHTSA made "a considered evaluation of the presently available alternatives." Pillai v. CAB, 485 F.2d 1018, 1029 (D.C.Cir.1973). With this statement of the appropriate scope of our review, we turn to the merits of NHTSA's action.

The rescission of Modified Standard 208 on the grounds stated by NHTSA was arbitrary and illogical for two general reasons. The agency has offered no evidence that seatbelt usage will fail to increase as was expected when the standard was first promulgated, and has therefore made no showing that the standard is unjustified as written. More important, NHTSA has failed to consider or analyze obvious alternatives to rescission, and has thus artificially foreclosed attempts to further the purpose of the Safety Act. See Pillai v. CAB, 485 F.2d 1018, 1027 (D.C.Cir.1973). When the agency so narrows its options that it fails to heed the goals that Congress has asked it to meet, the agency violates its basic legislative mandate. Simply put, NHTSA's discussion of ways in which Modified Standard 208 could have been amended was wholly inadequate, and rescission was therefore arbitrary and unlawful.

NHTSA made three findings in deciding that the standard as written could not be justified, and should either be amended or rescinded. First, most automobile manufacturers planned to use detachable passive belts. Second, once detached, a detachable belt "becomes identical to a manual belt." Finally, NHTSA "cannot reliably predict even a 5 percentage point increase as the minimum level of expected usage increase." See 46 Fed.Reg. 53,421-23.

Although the petitioners challenge each step of this reasoning, we conclude that NHTSA's first finding was not unreasonable. The prediction of how industries plan to comply with an agency regulation is exactly the sort of factual question on which the agency's expert "administrative feel" deserves the greatest deference. It was not unreasonable for NHTSA to conclude, from the comments that appear in the record, that most manufacturers did indeed plan to install detachable passive seatbelts in order to comply with the passive restraint standard.27

NHTSA's second finding is also reasonable, although it is important to note the limitations of this finding. The observation that a detachable belt, once detached, is functionally equivalent to a manual belt, does not dictate any conclusion about the usage rate of detachable belts. Any suggestion that this finding alone leads to a prediction that usage rates under the two systems would be identical is inconsistent with NHTSA's third finding, which acknowledged that even detachable passive belts would make some difference in seatbelt usage.28 The rationality of NHTSA's conclusion about the standard as written depends entirely on how well its third step-predicting usage rates with detachable belts-holds up.

In taking this third step, NHTSA turned the question on its head and thus totally misdirected its analysis. There may well be "substantial uncertainty" about the seatbelt usage rates that can be predicted if detachable belts are widely used. But the question is not whether evidence shows that usage rates will increase by the necessary amount, but whether there is evidence showing they will not. NHTSA has some burden, in other words, to show that a regulation once considered to prevent deaths and injuries efficiently can no longer be expected to do so. This evidence may take the form of statistical data or logical argument, but it must exist in some form. To state this point in the reverse, it would be unreasonable for an agency to promulgate a regulation simply because of "substantial uncertainty" that the status quo was any better. There must be some reason to support any reasoned decision.

There is not one iota of evidence to support NHTSA's conclusion that Modified Standard 208 as written will fail to increase nationwide seatbelt use by 13 percentage points or more. No logical reason has been suggested why even detachable belts-except under a certain condition whose frequency is not known-cannot lead to safety benefits that exceed their relatively small marginal costs.29 Statistics concerning usage rates of the passive seatbelts now in service suggest exactly the opposite, and NHTSA's conjectural efforts to distinguish these systems miss the point.30 If, as NHTSA contends, the passive belts now in use cannot be the basis for predicting usage rates under Modified Standard 208, then only a well justified refusal to seek more evidence could render rescission non-arbitrary. National Ass'n of Demolition Contractors v. Costle, 565 F.2d 748, 751-52 (D.C.Cir.1977). NHTSA could have conducted surveys or experimented with detachable belt prototypes, or it could have explained logically why usage rates with detachable belts would increase less than 13 percentage points. The agency did not do so; it rested its decision only on "substantial uncertainty" about these rates. Its decision to rescind the standard thus was arbitrary because the decision was supported by no record evidence whatsoever.

We do not mean this portion of our opinion to seem simplistic. Implicitly, the perpetuation of a regulation also involves a decision that its continuation is worthwhile, and reasonable decisionmaking requires that this too be supported by reasons. If NHTSA was uncertain that Modified Standard 208 would lead to the required usage increase, it could reasonably have decided to suspend or amend the regulation rather than plunge ahead in the blind faith that the standard would succeed. Indeed, this was originally how the agency framed the questions that led to the most recent rulemaking proceedings.31 But it is one thing to decide that evidence fails to show the wisdom of continuing with a regulatory program, and quite another to decide that evidence shows the regulatory program should be abandoned altogether. There is no evidence at all that detachable passive belts will fail to increase seatbelt usage by 13 or even 50 percentage points, and NHTSA thus went one step further than reason can support.

NHTSA explicitly considered a series of alternative amendments to Modified Standard 208 before deciding to rescind the standard. Its notice of proposed rulemaking

proposes a wide range of possible changes to the automatic restraint requirements .... The Department desires to ensure that it is taking the most effective and reasonable approach to addressing the serious safety problem posed by the low rate of safety belt use in all cars and by the steadily decreasing average size of new cars sold in this country.

46 Fed.Reg. at 21,205-06. The agency considered whether to reverse the sequence of compliance so that small cars would be required to comply before large cars, and whether to amend the standard so as to require simultaneous implementation on all car sizes. NHTSA also considered amending the standard in order to exempt the front center seating position, while retaining the requirements for the driver and front right passenger seats. In reviewing these possibilities, NHTSA's Regulatory Impact Analysis (RIA) was impressively thorough and careful.32

In striking contrast, neither NHTSA's decision nor its RIA devoted even a modicum of reasoned analysis to several far more obvious possible amendments to Modified Standard 208. The decision briefly discussed-and summarily rejected-amending the standard so as to eliminate compliance by detachable belts. The RIA, which was otherwise replete with economic and safety estimates and other data, did not once attempt to analyze the consequences of such an amendment. Even more striking is the agency's failure to consider amending the passive restraint standard so as to allow compliance with airbags only. In both the decision and the RIA, airbags are mentioned only in passing and only in the context of predicting that automobile manufacturers would not install them on a widespread basis. Absolutely no effort was made to compare the costs of airbags against their potential benefits, an omission that is particularly notable because none of the problems that NHTSA identified with passive belts appears to apply to airbags.

In proceeding as it did, NHTSA allowed itself to become captive to the ways in which it predicted automobile manufacturers would comply with Modified Standard 208. Lip service aside, there is no indication of the agency's awareness that it could act positively to develop a passive restraint standard that would advance the purposes of the Safety Act. Courts frequently observe that regulatory agencies do not function "as an umpire blandly calling balls and strikes for adversaries appearing before" them. Scenic Hudson Preservation Conference v. FPC, 354 F.2d 608, 620 (2d Cir. 1965), cert. denied, 384 U.S. 941, 86 S. Ct. 1462, 16 L. Ed. 2d 540 (1966). If NHTSA did not believe the standard as written would fulfill the standard's goals, its foremost obligation was to consider whether an amended standard could. The agency's reasons for not prohibiting compliance with detachable belts are hollow, and its analysis of requiring exclusive compliance with airbags is nonexistent. NHTSA was not required to follow either avenue, of course, but it may not reject these possibilities without reasoned discussion. In stating that "the central issue in this proceeding has become whether (detachable) automatic belts would induce higher belt usage rates than are occurring with manual belts," 46 Fed.Reg. at 53,425, NHTSA lost sight of its statutory obligation to devise the best passenger safety regulation it could. By artificially narrowing the options available-or ignoring those options completely-the agency acted in a totally arbitrary fashion.

As in its assessment of Modified Standard 208 as written, NHTSA's discussion of whether to require only continuous passive belts starts by asking the wrong question. The issue is not whether exclusive reliance on continuous belts would be superior to detachable belts, but whether detachable belts comply with the passive restraint standard at all. NHTSA boldly declared that "(o)nce a detachable automatic belt is detached, it becomes identical to a manual belt.... (I)ts use thereafter requires the same type of affirmative action that is the stumbling block to obtaining high usage levels of manual belts." 46 Fed.Reg. at 53,421. If so, the obvious question is whether detachable belts conform to the requirements of the passive restraint standard, which requires compliance "by means that require no action by vehicle occupants." 49 C.F.R. § 571.208, S4.1.1.1. Whether a detachable belt is in fact a passive restraint may depend on the usage rates of these belts-a matter that NHTSA has not yet established. But if detachable belts do not comply with the passive restraint regulation, NHTSA should have concentrated its analysis on safety devices that do.

Time after time, members of Congress as well as NHTSA have emphasized that a safety restraint requiring affirmative action is not a passive restraint. For the last decade, the agency has explained that a passive restraint requires "no action other than would be required if the protective system were not present in the vehicle." 36 Fed.Reg. at 8296 (May 4, 1971). "The essence of a passive restraint is that it provides at least the minimum level of protection without relying on occupant action to deploy the restraint." 39 Fed.Reg. at 14,594 (April 25, 1974). "Automatic restraints are systems that require no action, such as buckling a seat belt, by vehicle occupants to be effective." 46 Fed.Reg. at 12,033 (Feb. 12, 1981). See 123 Cong.Rec. 33,321 (1977) (Senator Griffin) (a passive restraint "is something that does not require any activity on the part of the occupant of the car"); S.Rep.No.481, 95th Cong., 1st Sess. 1 (1977) ("Passive restraints are defined in the rule to mean any front seat occupant restraint which protects the occupant ... without the occupant having to activate it"). It is difficult to imagine an agency decision that could be more irrational than rescinding a regulation because of problems identified in industry actions that do not comply with that regulation. If detachable belts function as poorly as NHTSA says they would, then NHTSA's discussion of these belts is worse than irrelevant because detachable belts are not passive restraints at all.

Modified Standard 208 is a performance standard, which means that compliance may be in the form of any technology "which meets the standard." S.Rep.No.481, supra, at 1; see 123 Cong.Rec. 33,318 (1977) (Senator Ford) ("any technology which meets the standard can be utilized"). It is almost embarrassingly obvious that technology that does not meet the performance standard may not be used. NHTSA's discussion of detachable belts versus continuous belts ignores this elementary point. Its decision not to amend the standard so as to allow only continuous passive belts is therefore fundamentally flawed by NHTSA's failure to consider that the standard might only permit use of such belts in the first place. Essentially, the agency seems to conclude that because some technology will not meet the passive restraint standard, it need not mandate compliance by technology that will. The absurdity of this Orwellian reasoning is obvious.

The explicit reasons stated by NHTSA for refusing to mandate compliance only with continuous passive belts also seem capricious. The first reason stated was the congressional prohibition on interlocks, and the fear that "Congress would look with some disfavor upon any similar attempt to impose a use-compelling feature on a belt system." 46 Fed.Reg. at 53,424. NHTSA's discussion of the interlock is absolutely irrelevant, of course, because the 1974 Amendments prevent it from mandating such devices. Of more serious concern is NHTSA's apparent belief that interlocks and other "use-compelling features" are equivalent. There is absolutely no evidence to support such a conclusion; indeed, every indication in the record points the other way.33 The distinction lies in the use that is "compelled": interlocks require affirmative action, but continuous passive belts do not. The very concept of a passive restraint standard presupposes "use-inducing" safety equipment. NHTSA's argument about congressional disfavor toward continuous belts, based only on the fact that such belts would work better than detachable belts, is thus belied by the legislative history discussed above.

Had Congress desired to repeal the passive restraint standard in 1974, it was free to do so. Instead, Congress legislated with great specificity concerning the aspects of the standard that it wanted to change-requiring, for example, that a warning buzzer not exceed eight seconds in duration-and retained the explicit right to review other aspects of the standard at a later date. The possibility of a later legislative veto is far different from a legislative prohibition, however, and NHTSA assumes too much when it tries to read into this congressional action a hostility to an effective passive restraint standard. NHTSA errs when it presumes that it knows how to speak more clearly than Congress. The 1974 Amendments do not provide the slightest support for NHTSA's refusal to consider a passive restraint standard that could be satisfied with continuous but not detachable belts.

The last two explicit reasons given by NHTSA for refusing to mandate compliance only with continuous belts are somewhat inconsistent: the public's "widespread, latent and irrational fear" about being trapped by seatbelts, and the concern that a "use-compelling feature" could not be required consistent with "the cause of safety." Id. Considerable doubt exists about the strength of the first finding.34 More important, NHTSA's second finding-clearly more fundamental to any agency charged with mandating safety standards-is squarely contradicted by the explicit requirement of an "emergency release mechanism" in the passive restraint standard. See 42 C.F.R. § 571.208, S.4.5.3.3(a). NHTSA has not suggested, nor is there any evidence for such a proposition, that the emergency release mechanisms on continuous belts are any more susceptible to jamming or locking in the event of a collision than the latch mechanisms on detachable or manual belts. Indeed, for all that appears on this record, the emergency release on a continuous belt may be more dependable than that on a detachable belt. In any event, NHTSA also notes that if an accident occurs, "occupants wearing belts suffer significantly reduced risk of loss of consciousness, and are commonly able to extricate themselves with relative ease." 46 Fed.Reg. at 53,424 (emphasis added). There is no evidence to support the suggestion that "further complicat(ing) the extrication" of occupants after collisions, if that occurs at all, will not be more than offset by the advantages of increased belt use.

Instead, there is every indication here that continuous belts-which would not require or be guarded by interlocks-would increase belt usage more than enough to meet the agency's justifiable concerns about the standard's cost.35 Such belts could never "become identical to a manual belt," id. at 53,421, and clearly "would increase belt usage." Id. at 53,423. NHTSA has not analyzed the magnitude of this increase, or whether a standard permitting continuous belts only would make Modified Standard 208 the cost-effective regulation that Congress expected it to be. If continuous passive belts worked effectively, of course, this would dispatch NHTSA's implicit concern about public backlash against "an expensive example of ineffective regulation." Id. at 53,424 (emphasis added). Until NHTSA at least analyzes the consequences of a continuous-belt-only standard, there is no way to know whether this form of the passive restraint standard is justified. At the same time, until NHTSA considers this obvious alternative, its rescission of the entire standard must be considered arbitrary.

If NHTSA's consideration of continuous belts was minimal, its analysis of airbags was nonexistent. Time and again, the agency failed or flatly refused to evaluate the cost-effectiveness of these devices. The only mention of airbags in the decision focused on Modified Standard 208 as written:

Instead of installing air bags in approximately 60 percent of new cars (as had been anticipated in 1977), the manufacturers apparently planned to install them in less than 1 percent of new cars. Thus, automatic belts would have been the predominant means of compliance, and installed in approximately 99% of new cars. Thus, the assumed life-saving potential of air bags would not have been realized.

46 Fed.Reg. at 53,421. It is obvious that airbags would be installed in 100% of new cars if NHTSA were to decide that no passive belts advanced the purpose of the safety standard and that airbags were the only way of complying with Modified Standard 208. NHTSA apparently did not consider this possibility. See RIA at IV-10, J.A. 66 (because airbags will be installed on a small scale, "the fatality reduction potential of air bags will not be considered in the main body of the analysis"); id. at IV-56, J.A. 112 ("Although benefits would be greater if (large cars were equipped with airbags), it does not now appear that this will happen"); id. at VI-19, J.A. 181 ("Air cushion costs are not dealt with in this analysis as the domestic manufacturers have decided to use automatic safety belts rather than air bags to comply with the standard"); id. at XI-5, J.A. 265 ("The issue of air bags, constant and controversial as it is, is not in fact squarely before the Agency at this time").

NHTSA's failure to consider this obvious alternative is particularly striking in view of the history of the passive restraint standard. "To some, FMVSS 208 was always intended to be an air bag standard." RIA at XI-5, J.A. 265. The first notice of proposed rulemaking leading to the 1972 version of Standard 208 referred to "inflatable occupant restraint systems," 34 Fed.Reg. at 11,148 (1969), and passive belts were added thereafter. See pp. 209-210 supra. In Chrysler Corp. v. Dep't. of Transportation, 472 F.2d at 664, the court suggested that the standard might be called "more descriptively, the Airbag Standard." Congress too has tended to consider airbags the central aspect of the passive restraint standard. See, e.g., 120 Cong.Rec. 35,637 (1974) (Representative Dingell) (observing "the desire of the Congress to review and, if appropriate, disapprove a standard imposing air bags on our constituents"); 124 Cong.Rec. H 5309 (daily ed. June 12, 1978) (Representative Conte) ("The airbag does not go into effect until 1982"); H.R.Rep.No.145, 96th Cong., 1st Sess. 13-15 (1979) (supplemental views) (discussing only airbags). In 1977, when Senator Griffin added Senator Helms as a sponsor of S.Res. 31 to disapprove Modified Standard 208, the resolution was described as a measure to ban "the use of air bags in automobiles." 123 Cong.Rec. 24,879 (1977). It is true that airbags also have been the most controversial aspect of the passive restraint standard. See pp. 226-229 supra (Senate but not House approved conference report that would have required automobile manufacturers to offer airbags on at least one car line). But this is all the more reason why NHTSA should at least have considered and evaluated the benefits and costs of a standard relying only on airbags before it could rationally rescind Modified Standard 208.

This is virtually self-evident. First, as NHTSA acknowledged in the RIA, the agency

has no basis at this time for changing its earlier conclusions in 1976 and 1977 that basic air bag technology is sound and has been sufficiently demonstrated to be effective in those vehicles in current use, such that it could be considered an acceptable method of compliance with the mandates of the Act.

Id. at XI-4, J.A. 264. More important, airbags seem to have none of the problems that NHTSA identified in passive seatbelts. They are less obtrusive and therefore less likely to be disconnected because of "irrational fears" about being trapped. See id. at IX-18, J.A. 233. They cannot be analogized to manual seatbelts, and thus their rate of "use" is likely to be higher than that of passive belts. See id. at IV-9, 89-90, J.A. 65, 145-46. Moreover, even when used, neither seatbelts nor airbags are completely effective in preventing deaths and injuries under all circumstances, but the "effectiveness" of airbags may be relatively higher. See id. at IV-9, J.A. 65.36 Finally, whereas NHTSA emphasized its uncertainty that discounts in insurance premiums would "materialize on a general basis" under a passive belt regime, see 46 Fed.Reg. at 53,423, it observed that a number of insurance companies already offer a 30 percent discount on policies for vehicles equipped with airbags. RIA at V-1, J.A. 149.37 Each of these points suggests that the benefits of an airbags-only standard would be significantly greater than the minimal benefits now predicted from Modified Standard 208 as written. Yet NHTSA made no attempt whatsoever to evaluate or even mention such a standard.

The agency did, of course, review the cost of airbags. As it observed, the unit cost of these devices is remarkably sensitive to economies of scale, ranging from as high as $1,200 when only 10,000 units are produced to as low as $200 when airbags are installed on a fleet-wide basis. Id. at VI-10, J.A. 172. Were the agency to analyze an airbags-only standard, the lower cost figures would obviously be the appropriate basis of its analysis.38 We may therefore not decide on this record that NHTSA's failure to consider airbags alone was irrelevant; indeed, airbags may be significantly cost-effective.39 That question is for NHTSA to decide, but it has not yet done so. The only reason that can even be inferred for NHTSA's silence on this point is that such a standard risks congressional disfavor under the legislative veto provisions of 15 U.S.C. § 1410b. The fact that Congress might veto an airbags-only standard is not a reason for ignoring such a standard; an agency cannot refuse to proceed with reasoned decisionmaking simply because Congress has explicitly stated an intention to review its conclusions at a later time.

Other than promising to undertake a campaign to encourage voluntary seatbelt use, the kind of effort that has had dubious effectiveness in the past,40 NHTSA considered no alternatives to the passive restraint standard.41 Such alternatives abound, from speed controls that might prevent vehicles from obtaining freeway speeds unless seatbelts were fastened to continuous or flashing lights on dashboards. See 39 Fed.Reg. 38,380 (Oct. 31, 1974) (recognizing that 1974 Amendment "leaves considerable regulatory discretion concerning warning systems" to replace continuous buzzers).

Although the agency clearly did not have an obligation to consider every such alternative in the course of rescinding Modified Standard 208,42 the omission of one scheme stands out in particular. As discussed above, see pp. 211-212 supra, Secretary Coleman also decided that the passive restraint standard should be suspended, but he proposed a demonstration project involving up to 500,000 cars with passive restraints in order to "increase significantly the chances that passive restraints will ultimately prove acceptable (because) I am convinced that their potential safety benefits warrant this action." Coleman Decision, J.A. 2074. When Secretary Adams countermanded this decision and issued the mandatory standard in 1977, the debate in Congress focused on the distinction between these two approaches. See, e.g., 123 Cong.Rec. 33,325 (1977) (Senator Cannon); id. at 33,327 (Senator Griffin); S.Rep.No.481, 95th Cong., 1st Sess. 31 (1977) (Minority Views) ("the precise issue" is "not whether or not the development and use of air bags should be encouraged. (They should, and they would have been-much sooner, in fact-under former Secretary Coleman's voluntary program)"); H.R.Rep.No.145, 96th Cong., 1st Sess. 14 (1979) (Supplemental Views) ("We must agree that the wiser and more sensible approach would be to reinstate the Coleman decision"). It is difficult to understand why NHTSA did not reinstate the Coleman decision because that approach is also not discussed in Notice 25. Here too, the agency's rescission of the safety standard without discussing or perhaps even considering obvious alternatives was arbitrary and capricious.

The foregoing analysis illustrates several ways in which NHTSA artificially narrowed its analysis of Modified Standard 208. Although that regulation is a performance standard, the agency offered no evidence that certain technology would fail to perform. More important, it then rescinded the standard without analyzing it in terms of technology that clearly would comply.

By any measure of reasoned decisionmaking, NHTSA's action was arbitrary and capricious. Despite carefully issuing notice on an array of possible changes to Modified Standard 208, NHTSA proceeded as though the only question were whether to implement or rescind the safety standard exactly as that standard was first written five years ago. "It is easy enough for an administrator to ban everything." United States v. Nova Scotia Food Products Corp., 568 F.2d 240, 253 (2d Cir. 1977) (FDA regulation concerning smoked whitefish overturned because agency failed to consider formulating rule "with specific parameters that applied to all species of fish"). But "an artificial narrowing of the scope of the regulatory problem is itself arbitrary and capricious and is ground for reversal." Home Box Office, Inc. v. FCC, 567 F.2d 9, 36 (D.C.Cir.), cert. denied, 434 U.S. 829, 98 S. Ct. 111, 54 L. Ed. 2d 89 (1977) (overturning regulation of cable television because record evidence failed to suggest existence of problem at which regulation was aimed). See, e.g., Sabin v. Butz, 515 F.2d 1061, 1069 (10th Cir. 1975) (summary judgment in favor of government on whether regulation was arbitrary was precluded by allegation that agency had failed to consider anticompetitive factors); Pillai v. CAB, 485 F.2d at 1027-30 (vacating order extending multilateral air carrier rates because agency considered only alternative to be open rates, and thus ignored bilateral negotiations and agency's suspension powers). These cases set aside regulations promulgated without consideration of obvious alternatives, but we see no reason why a different analysis should be applied here to the rescission of a regulation already promulgated. Either form of agency action may at times be irrational.

On balance, it is difficult to find anything positive to say about NHTSA's decisionmaking in this case. After conducting an elaborate quantitative analysis in a 280-page Regulatory Impact Analysis, NHTSA essentially rejected much of this analysis as "hypothetical":For purposes of this analysis only, the Agency has again adopted this approach in order to place hypothetical quantitative bounds on benefits and costs. The Agency, for reasons spelled out in detail in the Final Rule, does not accept these hypothetical bounds, and believes that it is not possible to predict with any accuracy what the usage rates of automatic belts would be.

RIA at XI-13, J.A. 273. See id. at IV-51, J.A. 107 (emphasizing that extrapolation from existing data "must be understood to be hypothetical"). In other words, the analysis to support Notice 25 has not yet been undertaken. Moreover, several of the explanations stated in Notice 25 fly directly in the face of the agency's own analysis. In stating that the standard would raise equity problems, for example, because current users of manual belts would "subsidize" the nonusers who "will generate all of the benefits that result directly and solely from this regulation," 46 Fed.Reg. at 53,425, NHTSA ignored the fact that high insurance savings could be realized even by regular users of manual belts. See RIA at V-1, J.A. 149.43 It is difficult to avoid the conclusion that NHTSA's analysis in Notice 25 has been distorted by solicitude for the economically depressed automobile industry-which is not the agency's mandate-at the expense of consideration of traffic safety, which is.44

We do not hold, of course, that an agency charged with promulgating safety standards may never rescind those regulations once promulgated. But rescission must be supported by rational explanations, after a reasoned and good-faith effort to consider alternative means of advancing the agency's purpose. If Congress chooses, it has the authority to override Modified Standard 208 or any other regulation. See, e.g., 15 U.S.C. § 1410 (exemption from motor vehicle safety standards for manufacturers of 10,000 cars or less for reasons of "substantial economic hardship"). But NHTSA may not confuse its role with that of Congress. Based on the record and the statements in Notice 25, we must conclude that NHTSA has acted capriciously, wearing blinders that prevented it from reasoned evaluation of methods to fulfill the purposes of the Safety Act.

There are obvious problems associated with reversing NHTSA's action, however. Within hours of the issuance of Notice 25-and well before the notice's effective date-the automobile manufacturers began to dismantle their passive restraint programs, and now contend that it would require a minimum of twelve months "following the beginning of the first model year following announcement of the requirement" to restore these programs. MVMA Brief at 70 n.85.45 Moreover, because of NHTSA's muddy reasoning in Notice 25, manufacturers would justifiably feel uncertain about the technology that satisfies the performance requirements of Modified Standard 208. See id. at 69 n.84. NHTSA has yet to make the case that detachable belts would not comply, although it could make that finding or even require compliance by airbags only. We must therefore decline the suggestions of petitioners that the standard take full effect by September 1983, State Farm Brief at 21, or by September 1982, NAII Brief at 64.

At the same time, we should not simply remand this petition to the agency so that recalcitrance might succeed where rational decisionmaking might not. The implementation of a passive restraint standard has already been delayed without acceptable reasons, perhaps unconscionably so. Our decision does not foreclose rescission, but requires simply that the agency analyze obvious technological alternatives for compliance before doing so. We therefore remand to NHTSA to recommend to us 30 days from today a feasible schedule for completing analysis of these questions. If technological means are found that conform to the performance requirements of the standard such that its safety benefits are economically justified, NHTSA is further instructed to prepare a practical schedule for the implementation of a passive restraint standard.46 Our order does not, of course, foreclose a reasoned decision by NHTSA that the standard should be suspended or delayed for appropriate reasons, or impede any action that Congress might take to resolve these issues.

NHTSA is correct in expressing concern about negative public reaction to "an expensive example of ineffective regulation." 46 Fed.Reg. at 53,424. There is no basis on the record before us, however, for concluding that Modified Standard 208 is such a regulation. More important, it is erroneous to believe that "ineffective regulation" occurs only when government acts affirmatively. By rescinding the passive restraint standard without legal justification, NHTSA's arbitrary action presents a paradigm of ineffective regulation. Notice 25 has wasted administrative and judicial resources, and has possibly delayed without justification a safety standard that may be, "from an economic point of view, as important as any environmental, health, or safety rule now on the books."47 At present rates, 1 in every 60 children born today is expected to die in an automobile accident, and 2 out of every 3 will suffer injuries in a crash. See 126 Cong.Rec. S 13,503 (daily ed. Sept. 25, 1980) (Senator Warner). These figures may be low, for NHTSA concedes that the problem at which the passive restraint standard was aimed has become more urgent in recent years. See pp. 213-214 supra. But the agency concludes Notice 25 by observing:

After 12 years of rulemaking, NHTSA has not yet succeeded in its original intent, the widespread offering of automatic crash protection that will produce substantial benefits.

46 Fed.Reg. at 53,426. NHTSA may yet conduct the reasoned decisionmaking that can support the rescission of the passive restraint standard, but it may not reject twelve years of preparation for such a standard until it does so. The agency's action here thus represents "an expensive example of ineffective regulation" of the worst kind.CONCLUSION

NHTSA began this rulemaking because a number of factors had changed since Modified Standard 208 was approved by Congress in 1977. These changed factors-higher gasoline prices, smaller cars, an ailing automobile industry, and the methods of compliance being pursued by that industry-may fully justify reassessing, modifying, and even deferring the effective date of the regulation, decisions that are not before us here. There has been no showing, however, that these changes justify rescinding the standard outright. The explanations put forward in Notice 25 are arbitrary in their failure to address obviously relevant considerations. The rescission of Modified Standard 208 is reversed.

We recognize that sensitive issues are raised in this case. An administrative agency may well require less basis for its decision not to pursue a particular policy at a particular time than when it decides to act affirmatively. But agencies may not ignore the mandate they have received from Congress, and reverse course without a reasoned or rational foundation for doing so. In this regard, we emphasize that courts, administrative agencies, and Congress are partners, not adversaries. Courts do not substitute judgment for that of the agency, but ensure that agencies exercise their judgment only in accordance with the will of Congress.

NHTSA's rescission of Modified Standard 208 cannot be supported on this record. The agency has 30 days in which to submit a schedule for resolving the questions raised in this opinion, leading either to the rescission or suspension of the standard or to a judicially approved schedule for the effective implementation of that standard or an amended standard. If NHTSA finds nonarbitrary reasons for rescinding the standard, of course, its action will be affirmed. Absent such reasons, or intervening action by Congress, NHTSA may not arbitrarily veer from the course that Congress has set.

It is so ordered.

Although I agree with the conclusion in part III of the majority opinion, that "(t)he rescission of Modified Standard 208 on the grounds stated by NHTSA was arbitrary and illogical," maj. op. at 230, I rely principally on the reasons indicated in part III-A. Certain of the observations made in part III-B of the majority opinion appear to be irrefutable, but I do not accept them as a basis for decision in this case. With this one exception, I fully concur in the excellent opinion of Judge Mikva.

By way of emphasis, I think it important to highlight a theme that is implicit throughout the majority opinion. This case involves highly controversial issues that are seen to be critically important to various and competing interest groups. The role of the court is not to write or rewrite Federal Motor Vehicle Safety Standards, or to support a particular interest group, or to weigh alternative solutions to economic problems, or to respond to the political climate of the day. The role of the court-plain and simple-is to enforce the mandate of Congress as expressed in the National Traffic and Motor Vehicle Safety Act and, thus, to determine whether the action of NHTSA reflects "reasoned decisionmaking that is the essence of lawful administrative action." Maj. op. at 209. This judicial role does not change when the court, as here, is confronted with a hotly contested and highly controversial case. The court, thus, does not become a rubberstamp for lawless agency action merely because the questions posed are difficult or sensitive.

As Judge Mikva has so ably indicated in his exhaustive majority opinion, the decision of NHTSA is long on words and short on reasoned decisionmaking. Indeed, the agency decision appears to be nothing more than a determined effort to achieve a particular result without regard to the facts at hand. Plainly, this is not reasoned decisionmaking, and it cannot be sanctioned by this court.

We express no judgment on the proper outcome of this case and, therefore, we do not seek to substitute our judgment for that of the agency. Rather, as is our responsibility, we reverse here to require the agency to act in accordance with the will of Congress and pursuant to reasoned decisionmaking.