dissenting:
The court today upholds the National Traffic Highway and Safety Administration’s (NHTSA) rollback of its bumper performance "standard from the old requirement that front and rear bumpers withstand impacts of 5.0 mph (5.0/5.0 mph standard) to the new requirement that front and rear bumpers need only withstand impacts of 2.5 mph (2.5/2.5 mph standard). The majority finds that the agency did not act arbitrarily and capriciously in determining that the lessened standard (1) would be in accordance with the safety requirements of the National Traffic and Motor Vehicle ' Safety Act of 1966, 15 U.S.C. § 1391 et seq. (1982) (Safety Act), and (2) would represent, overall, the most cost-effective standard and thus the proper standard under the Motor Vehicle Information and Cost Savings Act of 1972, 15 U.S.C. § 1911 et seq. (1982) (Cost Savings Act).
I cannot agree with the majority’s conclusion or with most of its steps in reaching that conclusion. In my judgment, NHTSA’s actions in this rulemaking were clearly arbitrary and capricious and not in accordance with its statutory mandates. NHTSA first dealt in a wholly inadequate manner with the safety concerns which prompted promulgation of the bumper standard originally. Then, having dismissed these concerns, NHTSA proceeded to perform a cost-benefit analysis that appears, given the contortions that the agency went through to reach its final conclusion, to have been solely a formalistic exercise aimed at justifying a preordained result. Throughout these deliberations, the agency ignored clear legislative direction as to what policies were to be given precedence. The majority’s approval of these actions in the name of deference to agency expertise is, I believe, an unacceptable retreat from our judicial responsibility to carefully review agency decisionmaking and an unsupportable condonation of agency failure to act in accordance with explicit instructions from the legislative branch. Consequently, I dissent.
I. Background
As the majority notes, the bumper standard at issue here was promulgated under two separate statutes: The Safety Act and the Cost Savings Act. Thus the 5.0/5.0 mph standard that was revoked in this rule-making represented both a motor vehicle safety standard under the Safety Act and a bumper standard under the Cost Savings Act. Consequently, alteration of this standard must, as the majority recognizes, be reviewed to ensure that such alteration was in accordance with the requirements of both relevant Acts.
A. Statutory Framework
The Safety Act provides for promulgation of motor vehicle safety standards to protect the public “against unreasonable risk of accidents” and “against unreasonable risk of death or injury to persons, in the event accidents do occur.” 15 U.S.C. § 1391(1), (2). The 5.0/5.0 mph bumper standard was originally set forth as a motor vehicle safety standard under this Act in 1971. This safety standard remained in its original form until the agency decision under review here.1 Thus the alteration of *153the bumper standard from 5.0/5.0 mph to 2.5/2.5 mph was, first of all, a change in a safety standard. Consequently, the court must first review that alteration to ascertain that the agency’s decision was in accordance with the strictures of the Safety Act.
Our review under the Safety Act, however, is only the first step. For the 5.0/5.0 mph standard was also a bumper standard promulgated under the Cost Savings Act, and we must also review the alteration of the standard for its compliance with that Act. Under the Cost Savings Act, NHTSA is to formulate a bumper standard so as to obtain “the maximum feasible reduction of costs to the public and to the consumer.” 15 U.S.C. § 1912(a), (b)(1). This Act, in addition to specifying this overall criterion, also specifies particular factors for the agency to consider in determining the most cost-reducing bumper standard:
(A) the cost of implementing the standard and the benefits attainable as the result of implementation of the standards;
(B) the effect of implementation of the standard on the cost of insurance and prospective legal fees and costs;
(C) savings in terms of consumer time and inconvenience; and
(D) considerations of health and safety, including emission standards.
15 U.S.C. § 1912(b)(1). The Cost Savings Act also states explicitly that the Safety Act considerations are to be given priority: “Bumper standards under this subchapter shall not conflict with motor vehicle safety standards promulgated under * * * the * * Safety Act.” Id. § 1912(b)(2). This statutory language is supported by consistent indications throughout the legislative history of the Cost Savings Act that safety concerns were to. be overriding. See, e.g., S.Rep. No. 92-413, 92d Cong., 1st Sess. 20 (1971) (“For example, if a cost-beneficial bumper standard increased the likelihood of, or severity of, injury to occupants, then the Secretary would not promulgate that particular standard. The considerations of safety would take precedence.”); H.R.Rep. No. 92-1033, 92d Cong., 2d Sess. 23 (1972) (“Safety must remain the paramount and overriding purpose of the Department.”). Thus any alteration of the bumper standard to achieve cost reductions as mandated by the Cost Savings Act can take place only after a determination that such alteration meets all of the criteria for an alteration of the bumper standard as a motor vehicle safety standard under the Safety Act. Any review under the Cost Savings Act, therefore, must give precedence to the Safety Act review.
B. Standard of Review
As the Supreme Court noted recently in Motor Vehicle Mfrs. Ass’n v. State Farm Mutual Auto. Ins. Co., 463 U.S. 29, 103 S.Ct. 2856, 2865, 77 L.Ed.2d 443 (1983), revocation or amendment of a motor vehicle safety standard under the Safety Act is to be reviewed according to the judicial review provisions of the Administrative Procedure Act. Thus the reviewing court is to set aside the agency’s action if it is found to be “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2)(A) (1982).2 And, since the Cost Savings Act *154also provides for the applicability of the APA review provisions, see 15 U.S.C. § 1913(c), a revocation or amendment of a bumper standard under that Act is to be reviewed under the same standard.
As discussed by the Supreme Court in State Farm, the scope of review under the arbitrary and capricious standard is narrow. This does not mean, however, that an agency has unlimited discretion or that a reviewing court may abdicate its review responsibility by simply deferring to the agency without thoroughly investigating and evaluating the agency’s actions. As the Supreme Court noted, the agency is required to “examine the relevant data and articulate a satisfactory explanation for its action including a ‘rational connection between the facts found and the choice made.’ ” State Farm, supra, 103 S.Ct. at 2866-2867, quoting Burlington Truck Lines v. United States, 371 U.S. 156, 168, 83 S.Ct. 239, 245, 9 L.Ed.2d 207 (1962). For example, an agency has failed to meet this general requirement, and has acted in an arbitrary and capricious manner,
if the agency has relied on factors which Congress has not intended it to consider, entirely failed to consider an important aspect of the problem, offered an explanation for its decision that runs counter to the evidence before the agency, or is so implausible that it could not be ascribed to a difference in view or the product of agency expertise.
State Farm, supra, 103 S.Ct. at 2867.
Judicial review of agency action requires that the reviewing court “consider whether the decision was based on a consideration of the relevant factors and whether there has been a clear error of judgment.” Id., quoting Bowman Transportation, Inc. v. Arkansas-Best Freight System, Inc., 419 U.S. 281, 285, 95 S.Ct. 438, 441, 42 L.Ed.2d 447 (1974); Citizens to Preserve Overton Park, Inc. v. Volpe, 401 U.S. 402, 416, 91 S.Ct. 814, 823, 28 L.Ed.2d 136 (1971). As noted by this court in Int’l Ladies’ Garment Wkrs. Union v. Donovan, 722 F.2d 795, 815 (D.C.Cir.1983), cert. denied, — U.S. -, 105 S.Ct. 93, 83 L.Ed.2d 39 (1984), such review “is not merely perfunctory. We are to engage in a ‘searching and careful’ inquiry, the keystone of which is to ensure that the [agency] engaged in reasoned decisionmaking.” See also Telocator Network of America v. FCC, 691 F.2d 525, 537 (D.C.Cir.1982).
This formulation points to an acknowledged disparity between the depth of our review and the ultimate scope of that review: Although the ultimate scope may be narrow, the depth must be sufficient for us to be able to comprehend the agency’s handling of the evidence cited or relied upon. The purpose of this in-depth review is to educate ourselves so that we can properly perform our reviewing function: determining whether the agency’s conclusions are “rationally supported.” United States v. Allegheny-Ludlum Steel Corp., 406 U.S. 742, 749, 92 S.Ct. 1941, 1946, 32 L.Ed.2d 453 (1972); Ethyl Corp. v. EPA, 541 F.2d 1, 34-35 n. 74 (D.C.Cir.) (en banc), cert. denied, 426 U.S. 941, 96 S.Ct. 2663, 49 L.Ed.2d 394 (1976). For, although data interpretation and analysis are functions that often lie within an agency’s realm of expertise, see, e.g., Office of Communication of United Church of Christ v. FCC, 707 F.2d 1413, 1440 (D.C.Cir.1983), it is our duty to review those functions to ascertain whether the agency’s actions were complete, reasoned, and adequately explained. See id. The mere fact that an agency is operating in a field of its expertise does not excuse us from our customary review responsibilities. And, where the agency’s reasoning, although complex, is rational, clear, and complete, we must affirm. Contrarily, where the agency’s reasoning is irrational, unclear, or not supported by the data it purports to interpret, we must disapprove the agency’s action. See Ethyl Corp., supra, 541 F.2d at 36.
*155With this standard of review in mind, I will first examine whether the agency’s rulemaking decision was a reasoned and lawful decision under the Safety Act; that is, I will ascertain what the agency’s decisionmaking processes were and determine whether they were arbitrary and capricious. Second, I will examine whether or not the agency’s decision was arbitrary and capricious under the Cost Savings Act.3 Only if the rulemaking survives both of these inquiries can it be upheld. I turn now to these inquiries.
II. Review of the Agency’s Decision for Compliance With the Safety Act
As the majority notes, the agency’s findings at issue here under the Safety Act relate to two separate safety issues: (1) the issue of bumper protection, in low-speed collisions, of vehicle safety features which, if damaged and malfunctioning, may later cause additional accidents; and (2) the issue of bumper contributions to energy management in collisions at speeds above the design speed of the bumper (crashworthiness). The majority finds that the agency adequately fulfilled its responsibilities under the Safety Act. I disagree with this conclusion for several reasons. First, I am uncertain, due to a lack of clarity on the part of the agency, as to precisely what the agency did in terms of its Safety Act obligations. I am thus dubious that the majority’s assumptions as to the agency’s actions are correct. Second, even if the majority’s assumptions are correct, I disagree with its conclusion that the agency’s actions complied with the Safety Act. In this section, I will state and discuss my concerns.
A. The Agency’s Lack of Clarity and the Majority’s Assumptions With Respect to the Agency Actions
Any investigation of an agency rulemaking decision begins with an examination of the agency’s conclusions and the findings that purportedly support those conclusions. This elementary step, however, is hindered in this case by the agency’s complete lack of clarity regarding its findings. It is clear that the agency concluded that safety concerns did not prevent its proposed rulemaking. Precisely what findings it relied upon in reaching that conclusion are unfortunately obscure. Specifically, it is unclear whether the agency found that the reduction in the bumper standard would have no reduction in safety or no significant reduction in safety.
In its final rulemaking statement, NHTSA noted that it had found that the rollback of the bumper standard from 5.0/5.0 mph to 2.5/2.5 mph would “cause no reduction in vehicle safety” with regard to either safety feature protection or crash-worthiness. 47 Fed.Reg. 21820 (1982), Joint Appendix (JA) 85 (emphasis added). Specifically, the agency found that “adoption of the 2.5-mph/2.5-mph alternative [would] not have any measurable effect on the risk that future accidents might be caused by safety components which malfunction due to damage incurred in prior low speed collisions and which are left unrepaired.” Id. at 21827, JA 92. The agency also found that “reducing the test speeds for the safety criteria [would] not measurably affect the high-speed crash energy management of cars.” Id. Later, however, in its findings relating to its denial of the petition for reconsideration of its decision, NHTSA stated simply that it had “concluded that reduction of the 5/5 bumper standard would not have any significant effect on safety.” 47 Fed.Reg. 56643 (1982), JA 4 (emphasis added). Yet this conclusion was based entirely on the same evidence as its previous statement, with only one exception: the question of crash-worthiness at speeds lower than 30 mph. There, the agency noted that it had concluded that the change in the bumper design speeds would “have a negligible effect” on crashworthiness at speeds of 10-15 mph and higher. Id.
*156Thus the findings that we are to review here were expressed in different terms at different times. The finding of no reduction in safety accompanying the final rule was transformed somehow into a finding of no significant reduction in the statement accompanying the denial of the petition for reconsideration. Consequently, precisely what the agency concluded is unclear from its official statements.4
Nor is it clear from the underlying record. In fact, my review of the record indicates manifest ambiguity not only on this point but also on the question whether the agency ever considered safety concerns under the Safety Act at all. In the Preliminary Regulatory Impact Analysis (PRIA) the four-page discussion of safety was extremely cursory and entirely speculative; no conclusions were stated and the Safety Act was not mentioned. In the preliminary notice of rulemaking the only discussion of safety was a solicitation of data regarding any reduction on meeting the motor vehicle safety standard for crash-worthiness (not the bumper standard), see 46 Fed.Reg. 48267 (1981), JA 375; the agency did not mention the Safety Act and, in fact, seemed to assume that the bumper standard was no longer a motor vehicle safety standard under that Act. See id. at 48262, JA 369.
In the Final Regulatory Impact Analysis (FRIA), responding to ■ heated comments that safety concerns were paramount, the agency did mention the Safety Act briefly, noting that it applied only to significant safety risks and stating that it had concluded that “no meaningful adverse safety effects [would] occur.” FRIA at IV-3, JA 196. In its more detailed discussion of the safety issues, however, it did not appear to be considering safety effects in light of the significance criterion. Thus it was not clear whether it had determined that the effects did not require application of the Safety Act or whether they were in compliance with its requirements.
The agency’s conclusions were also unclear in the more detailed discussion of the safety issues. The agency stated that the safety effects would be “at the upper range very minor.” FRIA at IV-15, JA 208. Specifically, it stated that there would be “no discernible effect on crash energy management.” Id. It also stated that
lower impact speed requirements with no change in uniform height requirements, will bring about any significant increase in damage to bumper standard-related safety components [sic]. Even in such cases the inoperation of such safety components is itself a very minor factor in accident causation. Such accidents show at most a conjectural relationship with injuries and fatalities[.]
Id.
Finally, the final notice of rulemaking, in addition to its stated conclusions of no reduction in safety, contained two short incomprehensible paragraphs stating in a wholly conclusory manner that the agency was not conceding that the Safety Act criteria applied but that, if they did, the rule-making here was in accordance with them.5 *157In asserting this conclusion, however, the agency in no way explained how those criteria were met in this case. Nor did it point to a relevant explanation in the underlying documents. In fact, only its assertion and its “no significant effect” conclusion support its ultimate conclusion that the change complied with the Safety' Act.
I believe this lack of clarity on the part of the agency renders impossible affirmance of its decision here, for several reasons. First, it is inadequately clear that the agency considered the Safety Act requirements at all. This is not a casual or superficial problem that can be glossed over: Safety concerns were mandated by Congress to prevail. Thus failure to properly consider the Safety Act requirements evidences fundamental inconsistency with the legislative mandate. Where such a critical issue is at question, we should hesitate to attribute actions to the agency that would validate its decision unless its path truly “ ‘may reasonably be discerned.’ ” State Farm, supra, 103 S.Ct. at 2867, quoting Bowman Transportation, Inc., supra, 419 U.S. at 286, 95 S.Ct. at 442.
Second, even assuming, as the majority does, that the agency did consider the Safety Act applicable, its interpretation of the requirements of that Act appears to me to be inadequately justified and very possibly inconsistent with the Act’s explicit mandate. As noted above, the Safety Act provides that motor vehicle safety standards are to be promulgated to protect the public from unreasonable risks of accident and of injury from accidents. The Safety Act also provides certain criteria that the agency must consider in prescribing such standards. Specifically, the agency must
(1)consider relevant available motor vehicle safety data, including the results of research, development, testing and evaluation activities conducted pursuant to this chapter;
(2) consult with the Vehicle Equipment Safety Commission, and such other State or interstate agencies (including legislative committees) as [it] deems appropriate;
(3) consider whether any such proposed standard is reasonable, practicable and appropriate for the particular type of motor vehicle or item of motor vehicle equipment for which it is prescribed; and
(4) consider the extent to which such standards will contribute to carrying out the purposes of this chapter.
15 U.S.C. § 1392(f). These criteria merely specify what factors the agency is to consider in its rulemaking process; they do not add to or subtract from the statutory mandate of the agency to regulate against unreasonable risks. The legislative history, however, does give some indication of how the agency is to treat these factors. For example, both the House and Senate Reports on this bill noted specifically that safety was to be the “overriding” or “paramount” concern. See S.Rep. No. 1301, 89th Cong., 2d Sess. 6 (1966) (“safety shall be the overriding consideration in the issuance of standards under this bill”); H.R. Rep. No. 1776, 89th Cong., 2d Sess. 16 (“[m]otor vehicle safety is the paramount purpose of this bill”).
The agency, however, in its general discussion of the requirements of the Safety Act, ignored the factors the statute requires it to consider and the explicit statutory mandate regarding unreasonable risks and noted that all it would have to do would be to find that any safety differences between a 5.0 mph bumper and a 2.5 mph bumper were insignificant. See FRIA at IV-3, JA 196. It based its reliance on this single criterion, to the exclusion of any *158consideration of the explicit statutory factors, on a single sentence in the Senate Report on this bill. See id. In that Report the Senate Committee considering the bill noted that “[t]he Secretary is not expected to issue a standard covering every component and function of a motor vehicle, but only for those vehicle characteristics that have a significant bearing on safety.” S.Rep. No. 1301, supra, at 6. The agency, relying solely on this lone sentence, evidently decided that a finding of insignificance would be enough to allow it to disregard entirely the explicit statutory requirements. Thus it is not clear to me that the agency’s action, under the Safety Act, was supported by adequate findings as required by the statute or even governed by the appropriate statutory standard.
The majority defends the agency’s emphasis on significance, rejecting application of the statutory standard of unreasonableness (and apparently condoning the lack of the statutorily required findings), by noting that “[t]he principle that an ‘unreasonable risk’ provision requires even insignificant risks to be eliminated if that can be done at (presumably) insignificant cost would turn many areas of regulation into unending pursuit of the trivial.” Maj. op. at 1344 n. 5. My point is not that the agency must regulate every unreasonable risk, however small. Rather, my point is that a finding of significance properly is relevant only in the agency’s initial determination regarding whether a particular vehicle characteristic should be subject to regulation. As the Senate Committee asserted, the agency is only expected to regulate “those vehicle characteristics that have a significant bearing on safety.” S.Rep. No. 1301, supra, at 6. Here, that threshold finding was clearly made by the agency when it decided to issue a standard for bumper performance. And no party to this proceeding challenges this initial determination. Having decided that bumpers are significant enough to warrant regulation, the proper standard to apply in determining what, specific bumper standard to promulgate is the explicit statutory standard of unreasonableness. To import into this second determination an additional significance requirement, as the agency did, is to ignore specific legislative direction regarding the proper inquiry.6
Finally, the lack of clarity confuses the determination of whether the agency’s decisionmaking was rational because it is difficult to know what the decision was. Most importantly, the majority conducts its entire review under the assumption that the agency’s findings, even its findings of no reduction in safety or no measurable reduction in safety, were essentially findings of no significant reduction. This exclusive focus is problematic because it may enable approval of the agency’s decision when that decision could not stand if measured against the no reduction findings.
The majority’s treatment of the agency’s action assumes away all of these issues. In. reaching its conclusion the majority assumes that the agency’s findings were findings of no significance and that significance is the sole and controlling relevant criterion under the Safety Act, and then concludes that the agency’s findings were reasonably supported and adequate under that Act (although it does not assert that the agency considered the evidence under the Safety Act). I believe, at the outset, that our review is made impossible by the *159lack of clarity with respect to whether or how the agency confronted or resolved these issues and that remand to the agency is therefore appropriate at the outset.7 Nevertheless, because I also disagree with the majority’s conclusion that, even accepting its assumptions as to the agency’s actions, those actions were reasonable and not arbitrary and capricious, I will discuss these specific disagreements below. I do not, however, concede the propriety of the majority’s underlying assumptions.
B. Finding of No Significant Reduction in Safety
The majority approves what it takes to be NHTSA’s determinations that the reduction in bumper standards will not significantly reduce safety (1) by reducing protection of other safety features or (2) by reducing the car’s ability to withstand collisions. Even assuming the agency did make these findings, I see no rational connection between them and the record upon which they are supposedly based.
1. Effects of change in bumper standard on safety feature protection. According to the majority, the agency relied on a single study in reaching its conclusion that the proposed change in the bumper standard would have no significant effect on other safety features that might prevent accidents. Specifically, it relied on the Tri-Level Study of the Causes of Traffic Accidents prepared by the Institute for Research in Public Safety at Indiana University. See JA 2743-2809 (Executive Summary of Tri-Level Study). The agency noted that “very few accidents occur as a result of malfunctioning of those vehicle components which are subject to the safety criteria of the bumper standard.” 47 Fed. Reg. at 21827, JA 92. It stated further that it had concluded “that far fewer accidents could be attributed, and only by speculation, to a failure to repair such components after they had been damaged in the only type of collision relevant to this discussion, i.e., one which might occur at an impact speed between 2.5 and 5.0 mph.” Id.
In determining whether the agency’s projected conclusions here are rationally based on the record, a reviewing court must trace the agency’s path from initial data to final result. In this instance the agency started with the assumption, “for purposes of analysis,” FRIA at IV-11, JA 204, that a reduction in the bumper standard could result in increased damage to other safety features. The agency, in making this assumption, made no attempt to quantify the extent of this additional damage. In fact, the agency rejected totally the only data submitted on this point, a State Farm study that indicated that more damage to safety features occurred with Model Year 1973 cars (which were required to comply approximately with a 2.5 mph bumper standard) than with Model Year 1974 cars (which were required to comply with a 5.0 bumper standard). The agency dismissed these data because the MY 1973 car bumpers were not sufficiently similar to the contemplated 2.5 mph bumpers to draw any conclusions from these data because bumper heights in 1973 were not uniform (although petitioners pointed out that in 1974 only 1974 model cars had uniform bumpers) and the Fuel System requirements did not apply. See FRIA at *160IV-9 to -11, JA 202-204. The agency thus concluded that State Farm had not shown that the difference between the two systems was noticeable and consequently downgraded the likelihood and extent of such a difference.8 Starting with this unknown difference, the agency then proceeded to conclude that even if such a difference existed it would have little effect on actual safety. This conclusion was based on a series of discounts of the chance that reduced bumper protection of other safety features would result in decreased safety in fact.
This series of discounts began with the statistical data in the Tri-Lével Study. Those data indicated that, of all accidents included in that study, 0.44-0.48% were certainly caused by problems with safety features that would be protected by the bumper standard, 0.93-1.67% were probably caused by such problems, and 1.6-5.7% were possibly caused by such problems. See Executive Summary of Tri-Level Study at 21, JA 2768. First, without any explanation the agency ignored entirely the “possible” data and stated only the certain and probable ranges indicated by the study. See FRIA at. IV-12, IV-13, JA 205, 206. Next, noting that non-functioning vehicle lights and signals comprised the largest component of the causal percentages, the agency proceeded to conclude that because “the study reports that older vehicles are *161over-involved in accidents resulting from mechanical problems * * *, it would thus not be unreasonable to attribute natural wearing out of bulbs, or less commonly, high impact crashes, as the most typical causes of inoperation.” FRIA at VI-13, JA 206. The agency’s conclusion, however, does not follow from the study’s findings. The study, with respect to older vehicles, notes simply that “[t]he probability of a vehicle 8 years of age or older causing an accident (by virtue of a mechanical problem) is 2.1 times greater than for vehicles in general.” Id., quoting Tri-Level Study at 71; see also Executive Summary of TriLevel Study at 23, JA 2770. This fact indicates merely that an older vehicle is more likely than a newer vehicle to cause an accident because of a mechanical problem. This fact does not in any way indicate what percentage of the accidents that are in fact caused by mechanical problems are caused by older vehicles: Only with data regarding the mix of vehicles involved in these accidents or regarding the mix of vehicles on the road overall could the inference drawn by the agency logically be drawn. Thus, even if the agency were correct in its (wholly unsupported) assumption that mechanical defects in older vehicles are more likely to be caused by natural wearing out of parts or by high speed crashes than by failure to repair damage from low-speed crashes, it still did not show any reasonable basis for concluding that these are “the most typical causes of inoperation” of those mechanical components. FRIA at IV-13, JA 206.
The agency then further discounted the safety effects of those accidents that may result from mechanical problems by noting that “the contribution of such accidents to injury was extremely rare.” FRIA at IV-14, JA 207. This conclusion was based on its observation of the study’s conclusion that other factors were overrepresented in personal injury accidents. Although the Tri-Level Study does indicate that other factors are more likely to correspond to personal injury accidents, the percentage of accidents caused by problems with malfunctioning bumper-related safety features that resulted in some personal injury was estimated to be perhaps 20%, surely not an “insignificant” percentage. See Executive Summary of Tri-Level Study at 25, JA 2772.
Cumulating all these discounts — quantified and unquantified, proven and speculative — the agency concluded, in the majority’s view, that any safety effects of this type from changing the bumper standard to 2.5 mph would not be significant. Although the reasoning underlying this conclusion is quite opaque, it appears that the agency’s logic was along the following lines: Because, by its calculations, the accidents potentially affected by any change in the bumper standard would form only a small percentage of all personal injury accidents, those accidents were insignificant, even though the magnitude of the differences in safety protection provided by various bumper standards was unknown and might be quite large.
This conclusion is irrational for two reasons. First, the “insignificant” percentage was very probably not as small as its erroneous assumptions indicated. All of the unsupported assumptions or conclusions made by the agency in reaching its result had the effect of understating the percentages at issue. Thus it is not clear that the correct percentage would be insignificant even according to whatever (undisclosed) criteria of insignificance the agency was using.
Second, as petitioners point out, the assumption that a safety standard is necessarily insignificant if it affects only a small percentage of accidents runs directly counter to the statutory mandate of the Safety Act. In that Act Congress dealt specifically with vehicular features that might impair or improve highway safety. That vehicular factors are a relatively small part of all safety risks does not mean that any vehicular factor alone is prima facie insignificant (most vehicular accident causes, by definition, will affect only a small percentage of even vehicular causes generally, let alone all accident causes). Thus the fact that a particular risk is merely a small *162component of a larger risk does not mean that the smaller risk itself is insignificant. This conclusion gains support from this court’s decisions under the Safety Act regarding defects. See United States v. General Motors Corp., 518 F.2d 420, 438 n. 84 (D.C.Cir.1975) (“We use the term ‘significant’ to indicate that there must be a nonde minimus [sic] number of failures. * * The number of failures need not be and normally will not be a substantial percentage of the total number of components produced.” (emphasis added)). Thus the agency’s conclusion that the fact that any potential safety effects would occur in only a small percentage of cases meant that those safety effects were insignificant is not, without more, consistent with its statutory mandate. Nor did the agency make any attempt to explain why this particular small percentage was insignificant. That is, it formulated no criteria for defining what an insignificant safety risk might be. It simply concluded that this particular small percentage was insignificant. This is the essence of unreasoned and unsupported decisionmaking. Cf. Industrial Union Dep’t, AFL-CIO v. American Petroleum Institute, 448 U.S. 607, 655, 100 S.Ct. 2844, 2870, 65 L.Ed.2d 1010 (1980) (plurality opinion) (noting that the agency may ascertain what risk levels are significant but requiring that it “determine, in the first instance, what it considers to be a ‘significant’ risk”).9
The majority’s review of the agency’s action here is completely unconvincing. First, the majority accepts in its entirety the asserted assumption that any risk that is a small percentage of total risk is insignificant. Second, the majority approves each and every assertion made by the agency whether or not that assertion is supported in the record. The majority notes with acceptance each discounting of the percentage figure of accidents potentially caused by bumper-related safety features. Then, noting that these “massive” discounts would “obviously” apply, maj. op. at 1345, the majority concludes that it was reasonable for the agency to conclude that the difference between the 5.0 and the 2.5 mph bumper systems was not measurable and not a significant safety consideration. The majority makes no attempt to evaluate in any way the intermediate steps taken by the agency or the reasonableness of those steps. It simply notes those steps and accepts them. I cannot agree that this superficial summary comprises the searching and careful inquiry to which we are bound. Nor can I agree with the majority’s conclusion that the agency’s action on this point was reasonable.
2. Effects of change in bumper standard on vehicle crashworthiness. NHTSA also, according to the majority, concluded that the change in the bumper standard would have no significant effect on the crashworthiness of motor vehicles in crashes at speeds higher than the design speeds. The agency based this conclusion on estimates submitted by car manufacturers and by State Farm. The manufacturers submitted comments to the effect that a bumper system of either 5.0 or 2.5 mph design would contribute only about 5% of the total energy management in 30-mph *163crash tests. State Farm submitted evidence to the effect that a 2.5 mph bumper would be only 25% as effective as a 5.0 mph bumper in managing crash energy. See FRIA at IV-4 to -5, JA 197-198.
Several manufacturers also submitted comments to the effect that “in 30 mph perpendicular barrier crashes * * * in' certain cases it may be possible to have a better optimized and more controllable occupant protection system for high speed protection if one didn’t have to take into account the requirements of low speed protection.” Volvo of America Corporation, Docket Comments at 3, JA 1655 (quoted in FRIA at IV-5 to -6, JA 198-199). See also General Motors Corporation, Docket Comments at A-5, JA 1773; Ford Motor Company, Docket Comments at 24, JA 1835; Volkswagen of America, Inc., Docket Comments at II, JA 1722.
Finally, one car manufacturer (Volvo) indicated that a 5.0 mph bumper might in fact be more “aggressive” than a 2.5 mph bumper and that it thus might actually be more dangerous, especially in side impacts. See FRIA at IV-4 to -5, JA 198-199. Another manufacturer (General Motors), however, noted that there would be no difference in aggressiveness between the two types of bumpers. Id. at IV-5, JA 199.
From these comments the agency concluded that “[t]he difference in the energy management capability of 5.0 mph bumpers and 2.5-mph bumpers is negligible at crash speeds such as those [30 mph] specified in the safety standards regulating the crash-worthiness of new cars.” 47 Fed.Reg. at 21827, JA 92.
Although not disputing this particular conclusion, the Insurance Institute for Highway Safety, in its comments submitted in support of the petition for reconsideration of the agency decision, noted that “the agency did not address * * * the contribution of 5 mph bumpers to energy management in crashes at speeds of less than 30 mph, which produce the vast majority of occupant injuries. * * * At these speeds, 5 mph bumpers with energy absorbers reduce the violence of the impact forces * * * far better than at 30 mph.” Comments of IIHS at 10, JA 22.
The agency, in its statement accompanying the denial of the petition for reconsideration, stated that
the agency fully considered that issue as it related to all speeds, notwithstanding the fact that certain of the illustrations cited by the agency involved 30 mph crashes. As IIHS pointed out, it is true that severe injuries can and do occur in crashes at speeds as low as 10-15 mph. However, at those speeds as well as. at higher speeds, 5 mph bumpers contribute insignificantly to changes in the management of the crash energies to which vehicle occupants are subjected.
47 Fed.Reg. at 56643, JA 4. The agency went on to explain this conclusion by noting that 5.0 mph bumpers could only have significant effects on crashworthiness at these lower speeds if they had the effect of slowing the decrease in velocity of a car upon crashing. And it noted that it had determined that 5.0 mph bumpers would not have this effect because only differences in total energy absorption capability can have this effect and these differences are small percentages of the total energy at crashes at 10-15 mph and above. The agency also commented that the potentially more aggressive nature of the 5 mph bumpers could actually adversely affect crashworthiness.
For all but the last of these apparently factual statements, the agency produced no evidentiary support. It produced no support for its statement that it had considered crashes at lower speeds in its final rulemaking (or in the preliminary rulemaking assessment). And, according to the administrative record, the only facts considered in the rulemaking concerned the 30-mph crashes. The agency also produced no support for its conclusion that 5 mph bumpers have only a minimal effect on total energy management at speeds above 10-15 mph. The comment of the agency with respect to bumper aggressiveness was the only, statement supported by the record. That support, however, was of du*164bious validity. The only car manufacturer that noted that 5 mph bumpers might increase the aggressiveness of the car in collisions was Volvo, a manufacturer whose estimate of primary weight effects the agency excluded in another part of the rule-making because, as the majority notes, it had “very small sales in the United States.” Maj. op. at 1353. The agency, however, neglected to mention a comment to precisely the opposite effect by General Motors, see FRIA at IV-6, JA 199, clearly not an insignificant contributor to the entire American car market. Thus the agency’s finding of “no significant effect” at all speeds above 10 mph also has no basis in the record and is therefore arbitrary and capricious.
The majority dismisses any objections to the agency’s conclusions with the comment that “[petitioners do not advance any data showing that there is better crash energy management with a 5.0 mph bumper system.” Maj. op. at 1350. With all due respect, this is just not the point. The point is not whether petitioners can mount a conclusive evidentiary showing that their (common sense) view — that stronger bumpers will provide greater protection in automobile collisions — is correct. The question is whether the agency’s ultimate conclusions and subsidiary findings to the contrary are adequately reasoned from the available evidence.
The majority chalks up the agency’s conclusions to its expertise in generalizing from field studies and defers to those conclusions on that basis. See maj. op. at 1350 n. 10. Although a reviewing court, as discussed above, should defer to an agency in such matters, such deference should not equal total capitulation. The agency here started with manufacturer estimates for 30 mph crashworthiness. It then, in considering the petition for reconsideration and not before, concluded that it could extrapolate these estimates to crashworthiness at all injury-producing lower speeds as well. It based these extrapolations on factual assertions that are supported in the record by only one piece of evidence, and a dubious piece of evidence at that. I do not believe that deference extends this far.
In sum, therefore, to the extent that it considered the Safety Act at all, the agency first substituted the questionable criterion of “significance” for the explicit statutory criteria of the Act. Then, it mistakenly interpreted the meaning of the term “significant.” Finally, the agency formulated a chain of calculations, some of which were erroneous, to show that, under that mistaken interpretation of a questionable standard, the safety effects of changing the standard would be acceptable under the statute. This comprised its entire treatment of safety — the factor which legislative direction clearly indicated should be given conclusive weight. Considering these layers of error on the part of the agency, I would hold, under the Safety Act, that the agency has acted arbitrarily and capriciously and not in accordance with law in determining that there would be no unacceptable safety effects from the change in the bumper standard.
III. Review of the Agency’s Decision Under the Cost Savings Act
Although, as detailed above, I would find that the agency’s inadequate and irrational consideration of the safety ramifications of its rule under the Safety Act preclude affirmance, I also disagree with the majority’s conclusion that the agency’s decision was sufficiently reasonable to withstand judicial scrutiny under the Cost Savings Act. My disagreement with the majority’s wholesale acceptance of the agency’s cost-benefit analysis under that Act ranges from the basic approach to the specific choices. I will detail each area in which my review has resulted in my concluding that the agency acted unreasonably: the treatment of health and safety costs, the calculation of benefits from changing the bumper standard, the calculation of the costs of changing the standard, and the agency comparison of the costs and benefits of the alternatives it considered.
*165A. Treatment of Health and Safety Costs
My first, and perhaps most major, area of concern is with the treatment of health and safety costs under the Cost Savings Act. Both the agency’s and the majority’s safety analyses go only to their conclusion that the Safety Act does not prevent the proposed change in the standard. Thus the agency prefaced the only discussion of safety in its Final Regulatory Impact Analysis with a discussion of how the Safety Act required only the maintenance of “standards on significant problems, not all problems.” FRIA at IY-3, JA 196. It then went on to find that the Safety Act did not preclude its chosen action. The majority also notes, in reviewing the agency’s findings with respect to safety, that “[t]he Safety Act’s mandate is not * * * categorical. Not all risks of accident or injury are to be eliminated, but only those that are ‘unreasonable[.J ” Maj. op. at 1343.
Although the Safety Act mandates a reasonableness analysis when the agency seeks to revoke or amend the bumper standard as a motor vehicle safety standard, such an analysis does not conclude the agency’s responsibility in this case. For the Cost Savings Act explicitly requires that the agency, in its determination of what bumper standard will reduce costs to the maximum feasible extent, take into account “considerations of health and safety.” 15 U.S.C. § 1912(b)(1)(D).10 This requirement is entirely separate from the provision that bumper standards promulgated under the Cost Savings Act cannot conflict with motor vehicle safety standards under the Safety Act. See 15 U.S.C. § 1912(b)(2). Taking into account the settled rule of statutory interpretation that a statute is to be read, where possible, to give effect to every provision of that statute, I conclude that 15 U.S.C. § 1912(b)(1) requires that, under the Cost Savings Act, NHTSA not simply determine whether a safety risk is unreasonable but also incorporate any safety effects from a change in the bumper standard into the cost-benefit analysis mandated by the Cost Savings Act itself.11
It is possible that the agency originally intended its safety findings of no effect on safety to obviate any need for consideration of the costs of any reduction in safety under its Cost Savings Act cost-benefit analysis. Under the agency’s (and the majority’s) current interpretation of its safety findings, however, it has not fulfilled the Cost Savings Act requirements. The agency (and the majority) should not be able to have it both ways. If the agency found no safety effects, those findings should be reviewed as such under the Safety Act analysis. If it found no significant effects, the lack of inclusion of the costs of those effects in its cost-benefit analysis constitutes the “fail[ure] to consider an important aspect of the problem” noted by the Supreme Court to be a primary indicium of arbitrariness and caprice. See State Farm, supra, 103 S.Ct. at 2867. It is also not in accordance with the law. See 5 U.S.C. § 706(2)(A) (1982). Thus the agency decision should be overturned on this basis alone.
*166B. Calculation of Benefits of Changing Bumper Standard
In its analysis NHTSA stressed the benefits associated with degradation of the bumper standard — benefits that it estimated would consist mainly of cost savings to the industry (which savings the agency assumed would be passed through in their entirety to consumers due to the competitive nature of the automobile manufacturing industry). In particular, the savings would stem from the reduced weight of the bumper systems themselves (“primary weight” savings) and the reduced weight of the vehicle components that support the bumper systems (“secondary weight” savings). The cost savings estimated to result from the change in the bumper standard would, the agency found, come about in two ways: Some would occur directly as cost savings due to lesser materials requirements and some would occur indirectly as fuel savings. The majority traces through petitioners’ arguments that these, estimates were unacceptably biased and that certain factors were inadequately considered and resolved by the agency. With respect to most of these issues, I believe those decisions are either reasonably supported by the record and within the agency’s field of expertise or sufficiently insignificant not to warrant reversal alone. I will just say that, with respect to most of these cost savings issues, the agency’s conclusions were based on specific evidence before it and the decisions that it made with respect to that evidence were within its area of expertise, even though it consistently chose to credit those data that tended to support the ultimate conclusion it finally reached.
With respect to one issue raised by petitioners, however, I cannot accept the majority’s approval of the agency’s decision. Specifically, petitioners note that secondary weight reductions, according to industry statements, would not occur immediately and that those weight reductions should, therefore, have been discounted to their present value in the benefit calculations. In the face of this comment the agency acknowledged that it was true that production of the 2.5 mph bumper vehicles would not begin immediately. See 47 Fed.Reg. at 56647, JA 8 (“Production requirements and market pressures will constrain any move immediately to 2.5 mph bumpers only.”). But, having acknowledged this, the agency totally ignored the criticism of its discounting methodology.
The majority takes the agency’s bald statement acknowledging that the 2.5 mph bumpers would not be produced immediately and interprets it as a statement that the criticism of the methodology with respect to the secondary weight savings is invalid because “the same could be said of primary weight reductions.” Maj. op. at 1354. The majority then proceeds to refine this creative interpretation as follows: The agency’s admittedly unreal assumption for the purposes of its cost-benefit analysis does distort the magnitude of the costs and benefits that the new standard would produce but it does not distort the relevant calculation, i.e., the “relative net benefit of one standard over the other.” Id. (emphasis in original).
I cannot agree with either step of the majority’s reinterpretation of the agency statement. First, I do not agree that an admission that the 2.5 mph bumpers would not begin to be produced immediately can fairly be construed by this reviewing court to carry with it the relatively complete explanation attributed to the agency by the majority. All the agency said was that the IIHS comment was factually accurate, at least in part. Second, the explanation extracted by the majority from the agency’s unelaborated statement is flawed in and of itself. As I understand the majority’s explanation, it means that even though the costs and benefits of the 2.5/2.5 mph standard would change if discounted from, say, five years in the future to the present, the 5.0/5.0 standard’s or the 5.0/2.5 standard’s costs and benefits would change as well. Thus the discounted present values of the future costs and benefits of all the standards will bear the same relation to each other as do those future costs and benefits *167without discounting.12 This explanation, although perhaps superficially attractive, makes two assumptions that are nowhere supported in the record — and one of which is, in fact, contradicted by a statement of the agency itself. Specifically, this explanation assumes that the calculated costs and benefits of each standard will accrue at the same time. It also assumes that implementation of whatever standard was chosen (except the 5.0/5.0 standard) would occur at the same point in the future.
The first assumption is explicitly contradicted by a statement that the agency made in the same place as its generalized statement that the majority is “interpreting”: The agency, in its denial of the petition for reconsideration, notes that, although
the agency does not expect many vehicles to be produced with 2.5 mph bumpers that do not have some corresponding changes in secondary weighty t]o the extent that any such vehicles are produced in the short term, the associated costs are normal transition costs associated ' with any change in requirements. The possibility that the rulemaking may have such short-term costs is outweighed by the significant benefits that owners of future cars will experience as a result of the new standard.
47 Fed.Reg. at 56647, JA 8. Thus the agency admits that the cost savings associated with the secondary weight saving of a 2.5 mph standard might not accrue for some time after the benefit losses associated with that standard have begun. To airily dismiss this discrepancy as a “normal transition cost” and to assume, entirely without basis, that such costs will in the long run be outweighed by the eventual cost savings ignores basic principles of cost-benefit analysis.13
The majority’s second assumption, although not directly contradicted in the record, is not supported by the record either. Nowhere does the record note that the cost savings associated with the 2.5/2.5 standard will accrue at the same time as those associated with, for example, the 5.0/2.5 standard. And common sense would indicate that the changes associated with the 5.0/2.5 standard, requiring fewer changes from the status quo, might actually occur sooner. Particularly where these two standards were close in terms of the per-vehicle savings calculated by the agency, the assumption of no time difference in implementation is one that must be at least considered by the agency.14
In sum, the agency’s patent failure to consider even the question raised by petitioners is yet another indication of its arbitrariness and caprice. The majority’s attempt to avoid facing this fact by supplying an explanation that is nowhere apparent in the agency’s statements and that may itself be wholly invalid is not a defensible exercise of judicial review. As the Supreme Court noted in State Farm, supra, 103 S.Ct. at 2867 (citing SEC v. Chenery Corp., 332 U.S. 194, 196, 67 S.Ct. 1575, 1577, 91 L.Ed. 1995 (1947)), “We may not supply a reasoned basis for the agency’s action that the agency itself has not given.” The reasons for this general rule are *168clearly apparent in the instant situation. Although the agency may have intended to explain its action with the (potentially flawed) analysis offered by the majority, it is equally as likely that the agency either simply ignored the point raised by the comment or resolved it in an entirely different, but unfortunately undisclosed, fashion. Thus, although I conclude that the agency’s treatment of the cost savings associated with movement away from the 5.0/5.0 mph standard was largely reasonable, I cannot agree with the majority that its treatment of the potential time effects of any delay in secondary weight effects are adequate, reasonable, or, without further explanation by the agency, affirmable by this court.
C. Calculation of Costs of Changing Bumper Standard
The costs calculated by the agency to be associated with the proposed movement away from the 5.0/5.0 mph bumper standard are those increased repair and associated costs that will result from increased damage to motor vehicles in low-speed crashes. Petitioners’ main disagreement with the agency cost estimates stems, as the majority notes, from a disagreement with the particular intermediate values used by the agency, not with the methodol-' ogy used by the agency.
The primary focus of petitioners’ argument is the curves relied upon by the agency to determine the effectiveness of a 2.5 mph bumper relative to a 5.0 mph bumper (effectiveness curves). In drawing these theoretical curves the agency made certain assumptions about all the bumpers it was comparing. It assumed:
(1) No bumper system of any design speed that is in the range considered (up to 10 mph BEV) is 100 percent effective in any incremental speed range.
(2) All bumper systems reach 0 percent effectiveness at replacement speed. Bumpers are not fully effective in all crash configurations up to design speed because, for example, of override/under-ride problems. Hence full effectiveness cannot be assumed. (3) At design speed, aluminum and steel bumpers are 60 percent effective and soft facia correspondingly higher. * * * (4) Bumper Effectiveness curves are of * * * approximate [S-shape].
Calculations and Supporting Material for the Preliminary Analysis of the Bumper Standard at B-l, JA 1218. Petitioners now contend that these assumptions are unsupported and invalid.
The majority dismisses these points for reasons which are entirely unrelated to their potential merit. The majority notes that these assumptions, and the resulting effectiveness curves, first appeared in the 1979 Assessments of the Bumper Standard prepared by the agency (both Preliminary and Final). The majority then concludes that the fact that petitioners first raised this point on this appeal means that it cannot be critical to the agency’s rulemaking:
It is impossible to think so, since if they were there would have been strenuous objection much before this. Both the assumption of 60 percent effectiveness at design speed and the shape of the effectiveness curves have been a part of the agency’s methodology since the January 1979 Preliminary Assessment and were carried forward (with minor modifications in light of IIHS comments) to the June 1979 Final Assessment, which the insurance industry found not only acceptable, but congenial.
Maj. op. at 1360.
The majority’s assumption that inaction by petitioners can be taken as conclusive of the unimportance of the agency’s actions here is utterly unsupported and, in addition, not indicative of the type of searching judicial review we are obligated to perform. For one thing, there might have been other reasons for petitioners not to object to the figures in the 1979 Assessments. For example, those Assessments supported retention of the 5.0/5.0 standard, a result which petitioners have supported all along. See, e.g., Final Assessment at 88, JA 1069. And those Assessments specifically stated that *169the disputed effectiveness curves were conservative assumptions: There were other data, data that would have indicated a greater relative effectiveness on the part of the 5.0 bumpers, that the agency did not use. See, e.g., Final Assessment at 31, F-18, F-19, JA 1012, 1167, 1168. Perhaps petitioners’ failure to object can be attributed to the fact that even with these conservative assumptions the 5.0 standard was found to be the optimal one. Further, with respect to the use of these estimates in the current rulemaking, the main alternatives presented in the PRIA were these estimates or the use of actual production data, data which the agency eventually rejected (they would have made the 5.0 standard look more favorable) as being based on inapplicably limited data. Perhaps petitioners focused on the differences between the alternatives offered in the PRIA rather than on the weaknesses of each alternative. These factors lend support to the hypothesis that there may have been alternative reasons for petitioners’ failure to object before this appeal and undercut the majority’s superficial presumption of insubstantiality.15
More importantly, whatever the reasons for the delay in making the arguments (and assuming that they were not, in fact, raised in their precise form below16), that delay is irrelevant. We review the agency’s action, not petitioners’. And I consider petitioners’ argument on this point, whenever raised, to be compelling evidence of the hasty and arbitrary path followed by the agency in reaching its foregone conclusion. That is, I believe this issue to be sufficiently critical to require more explanation than that given by the agency and endorsed by the majority (“engineering judgment”). The curves in question are the sole and entire basis for all of the calculations regarding the costs of the increased damage that could be expected from the rollback of the standard — the major component of all the costs of the rollback. Given this, I do not see how the majority can dismiss them as narrow points.
If one looks beyond the majority’s facile conclusion that the absence of prior complaints aimed at the precise points raised here conclusively demonstrates their uncritical nature to the underlying merits of petitioners’ claim, one sees more evidence of unsupported and unreasoned decision-making on the part of the agency. For example, the first assumption contested by petitioners is that of when the bumpers reach 0% effectiveness.17 After the 1979 Preliminary Assessment IIHS commented that the assumed replacement speeds were not valid. See Comments of IIHS at 3-5, JA 2240-2242. In the 1979 Final Assessment the agency did not portray revised effectiveness curves, but it did change some of the effectiveness numbers (shown in tabular form). One change was that the relative effectiveness at low speeds of 2.5 bumpers was increased, although 0% effectiveness, according to the IIHS comments, should have occurred at a lower speed. The majority notes that the fact that other *170changes occurred as well indicates that “[t]he logical inference is that the agency did in fact rethink and redraw its curves following IIHS’s suggestion.” Maj. op. at 53. It may indeed be a logical inference that the agency redrew the curves, but the important point for our review here is not that it redrew the curves but how it redrew them. And nowhere in the 1979 Final Assessment are the redrawn curves shown (although values different from the original ones are used); nor is the agency’s methodology for redrawing them described or explained. Yet the relative effectiveness values tabulated in the Final Assessment were lifted without change and used in the current rulemaking.
The second assumption at issue is the selection of the 60% effectiveness number at design speeds. Although the reasoning behind this assumption is quite opaque, the statement by the agency to the effect that “[bjumpers are not fully effective in all crash configurations up to design speed because, for example, of override/under-ride problems,” Calculations and Supporting Material at B-l, JA 1218, indicates that at least one concern leading to the assumption was that bumpers of unequal heights would lock and create additional damage. Yet, as the agency has pointed out in the numerous contexts in which it has rejected 1973 real world data (the only such data) with respect to several of the variables estimated by the agency, such data cannot be applied to the proposed 2.5 bumpers, which will all be of uniform height. Where the agency rejects some data based on one rationale and bases other estimates on that same rationale, those decisions appear to me to be unreasoned, at least without some explanation of this inconsistency on the part of the agency. Further, even without that inconsistency, the question of what other factors prompted the agency to arrive at this figure remains unanswered.
Finally, the question as to the shape of the effectiveness curves remains. Although the general shape of the curves appears to be invoked by several of the parties here involved, it seems that an agency should be required to at least explain the basis for that shape. The majority notes that the agency explained that these were based on the agency’s engineering judgment and that deference to its result is therefore appropriate. I agree wholeheartedly that deference to an agency’s engineering expertise is appropriate, but I cannot agree that, where the agency merely refers to its expertise and does not in any way explain the derivation of the curves at issue, it has adequately shown that deference is warranted. Agencies cannot just cite their expertise and magically unveil a number. They must state the underlying assumptions, uncertainties, and policy decisions with respect to those uncertainties. In consequence, I would find the agency’s action in formulating these curves to be inadequately justified. Consequently, reliance on those assumptions, without more, is arbitrary and capricious.
D. Agency Comparison of Costs and Benefits of Alternatives
Having formulated estimates of the costs and benefits of the alternatives before it, the agency, rather than arriving at one figure for each variable, formulated a range of values for each. From these ranges, it then extracted a high and low range for the costs and the benefits of each alternative. Using these ranges, the agency then compared the cost-effectiveness of each alternative. To accomplish this, it evaluated the ranges it had before it in four separate ways. See maj. op. at 1364-65. With three of these, the 5.0/2.5 alternative came out to be the most cost-effective. With the fourth, the 2.5/2.5 alternative prevailed. The agency then determined that this fourth method was the most accurate and proceeded to declare that the 2.5/2.5 alternative had been shown to be the best of those considered.
Under this fourth alternative the agency arrived at four possible net benefit (or net cost) figures for each alternative by combining the high and low range cost values for each alternative with the high and low range benefit values for each alternative. Thus, it had a high cost/high benefit, a *171high cost/low benefit, a low cost/low benefit, and a low cost/high benefit net value estimate for each alternative. In the fourth calculation of what these values meant, the calculation that the agency eventually relied upon, the agency excluded entirely the high cost/low benefit number and simply averaged the other three. It explained this move as being a result of its conclusion that it was “virtually impossible that the factual elements of that combination of assumptions could occur in reality, in large part because of inherent contradictions in economic or behavioral results that would be associated with such alignment.” 47 Fed.Reg. at 21825, JA 90. See also 47 Fed.Reg. at 56650, JA 11. As the majority points out, this assertion on the part of the agency “bears every evidence of having been inserted as a make-weight by someone who had not the slightest idea what he was talking about.” Maj. op. at 1366. The majority concludes, however, that the agency formulated another reason for excluding this fourth number: that “although each of these extreme assumptions may separately have some degree of probability, the probability that they will be accurate in combination is virtually zero.” Id. at 1365. As with the majority’s discussion of the discounting problem of secondary weights, however, this has two problems: First, the agency nowhere expressed this justification, and second, the justification itself is incoherent as applied to what the agency actually did.
With respect to the' first point, the agency’s conclusions with respect to the inherent improbability of the confluence of these two unlikely sets of assumptions are, wherever expressed, justified solely, or at least largely, on the basis of contradictory underlying assumptions. See 47 Fed.Reg. at 21825, JA 90; 47 Fed.Reg. at 56650, JA 11; FRIA XI-15 to -17, JA 345-347. The majority’s interpolation of an additional independent ground is simply not based on the agency’s own statements.
Second, the majority’s interpretation not only does not support the action taken by the agency, it further draws that action into question. For if there is no causal connection between the low cost/high benefit values, there is not greater probability that either will occur with or without the other. Thus the agency, if it had decided that these values were too inherently improbable to be included at all in the analysis, would have had to exclude or at least discount the low cost/low benefit and high cost/high benefit numbers. Yet it did not do so. This indicates either that the agency was indeed relying on the “makeweight” argument rejected by the majority or that the agency was incorrectly following the majority’s substitute rationale. Further, although the evidence before the agency may have supported some discounting of certain values based on their inherent probability, the rational way of dealing with this would have been to formulate probability estimates for all four values. No justification for the total exclusion of one value was given. Nor was a justification for discounting that value and not the others. Thus under neither view of the agency’s rationale does the agency’s action appear to have been reasoned decisionmaking. In sum, the majority’s explanation is no more coherent than that it rejected, and the selection of the 2.5/2.5 alternative through the methodology used was arbitrary and capricious.
Having gone through these contortions just to reach the result that the 2.5/2.5 standard was the superior one, the agency reached back to justify that result even with an analysis based on the straight averaging of all four value combinations. It noted that even using all four value combinations the 5.0/2.5 standard was only marginally superior to the 2.5/2.5 standard. (This marginal difference was $1 net additional benefits per car; the agency did not bother to translate this difference into a total (and perhaps not so insignificant) yearly cost of approximately $11,000,000.18) The agency then concluded *172that, where the two standards were essentially equivalent, it would choose the 2.5/2.5 standard anyway. The reasons for this choice it enumerated as follows: First, the 2.5/2.5 standard “imposes the least direct, immediate costs on the consumer, i.e., the least increase in the cost of a new car. To illustrate this point, if the unregulated bumper is considered the baseline,” more immediate costs would be required to comply with the bumper requirements. 47 Fed.Reg. at 21826, JA 91. The problem with this justification is that the unregulated bumper is not the baseline. The 5.0/5.0 standard is the baseline. Agencies are to start from the current rule in evaluating changes, not from some theoretical or historical unregulated state. See State Farm, supra, 103 S.Ct. at 2866. The majority notes that the agency is entitled to discount more speculative costs in favor of more certain costs. Even assuming that these costs are more certain (an assumption which does not appear to be based on anything in the record), the place to incorporate such a discount is in the cost-benefit calculation, not as a post hoc justification for choosing the alternative the agency wanted in the first place.
Second, the agency notes that the lower standard would permit innovation, which “could result in more effective bumpers at lower cost to the public than would otherwise be available.” 47 Fed.Reg. at 21827, JA 92. The agency also notes that the same standard for front and rear will promote commonality (common components) and thus reduce prices of bumpers. Assuming that such costs and/or cost savings should be considered and that they occur in actuality, the proper way to consider them is to estimate them and to incorporate them into the cost-benefit calculation. Trotting out auxiliary hypothetical and unquantified costs to rationalize one choice or another after having supposedly performed the cost analysis mandated by the statute is manifestly inconsistent with reasoned decision-making. Thus I conclude that this aspect of the agency’s analysis was arbitrary and capricious and should not be upheld by this court.
IV. Conclusion
In sum, I have looked carefully at the steps taken by the agency in promulgating this rule and have come to the conclusion that this court should properly reject it as an arbitrary and capricious exercise of agency power. The agency, in this case, stepped beyond the bounds of legitimate policy choice and moved into the field of unreasoned ratification of preordained conclusions. The majority, unfortunately, has exercised such deference to the agency’s statements, however opaque, unsupported, or unreasoned they might be, that it has abandoned entirely its mandate of rigorous judicial review. So, too, has it failed to ensure that the agency acted in consonance with its legislative mandate. This I cannot accept, and I therefore respectfully dissent.
. The majority notes that, since its promulgation, there has been considerable discussion over the validity of this standard and implies that there was inadequate support for its promulgation initially. See majority opinion (maj. op.) at 1338-42. I note first that agency discussion of a standard, especially where it has - not previously revoked that standard, in no way diminishes the standard of review of an amendment or revocation of that standard. See Motor Vehicle Mfrs Ass’n v. State Farm Mutual Auto. Ins. Co.. 463 U.S. 29, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983) (hereinafter cited as State Farm to 103 S.Ct. only) (applying arbitrary and capricious standard of review even where agency had previously suspended and then reinstated requirement). I note second that the hypothesized invalidity of a standard initially has no bearing on its revocation or amendment. It may make the agency’s job easier in showing that revocation or amendment is rational, but it does not alter the standard of review: "If Congress established a presumption from which judicial review should start, that presumption * * is not against safety regulation, but against changes in current policy that are not justified *153by the rulemaking record.” Id. at 2866 (emphasis in original).
. The Supreme Court noted in State Farm that Congress required compilation of a rulemaking record for the reviewing court, see 15 U.S.C. § 1394, State Farm, supra note 1, 103 S.Ct. at 2867, and intended that agency findings under the Safety Act be supported by "substantial evidence on the record as a whole.” See id. at 2867, citing S.Rep. No. 1301, 89th Cong., 2d Sess. 8 (1966); H.R.Rep. No. 1776, 89th Cong., 2d Sess. 21 (1966). In its review of the agency action at issue in State Farm, however, the Court focused on an arbitrary and capricious standard.
This focus accords with our decision in Pacific Legal Foundation v. Dep’t of Transportation, 593 F.2d 1338, 1343 n. 35 (D.C.Cir.), cert. denied, 444 U.S. 830, 100 S.Ct. 57, 62 L.Ed.2d 38 (1979), in which we noted that
the distinction between the arbitrary and capricious standard and substantial evidence review is largely semantic * * *. Since our review in this case, under Overton Park, involves a complete examination of the record, we agree with Judge Lumbard that "when an agency engages in substantive rulemaking, it *154abuses 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) [cert. denied, 423 U.S. 827, 96 S.Ct. 44, 46 L.Ed.2d 44 (1975) ].
. Although my review would indicate reversal on Safety Act grounds alone, I will review the Cost Savings Act analysis because the majority reaches that issue and because I differ from the majority there as well.
. The majority attempts to evade the normal and plain meaning of the terms used by the agency by concluding that the agency’s summary statement that the rule would have no effect on safety should for some reason be ignored. See maj. op. at 1344-45 n. 5. The majority further attempts to reconcile the agency’s various qualified statements ("no measurable or discernible effect”) by asserting that they all meant one thing only — that the rule would have no significant safety effects. Unlike the majority, I believe that the variety of terms used by the agency simply corroborates my concern regarding the lack of clarity as to the agency’s conclusion. Further, I do not believe that the disparate terms used by the agency can, by intricate dictionary explication, be reasonably made to be synonymous with "significant.”
. These paragraphs read:
Given the hybrid nature of the part 581 Standard, this rulemaking action was initiated under the concurrent authority of the [Cost Savings] Act and the Safety Act. Without deciding whether the criteria established for safety standards under section 103 [of the Safety Act] must necessarily be applied in all cases under the [Cost Savings] Act where any safety relationship can be asserted, the agency has concluded, based on the discussion in this notice and the FRIA, that its actions in this proceeding are in all respects in accordance *157with the applicable criteria of the Safety Act itself.
By the same token, this action does not conflict with safety standards promulgated under the Safety Act. To the extent that bumper standards may be considered to be safety standards, the 5.0-mph safety criteria of Part 581 have been determined to be unsupported, even under the Safety Act criteria, and are amended by this notice: Reducing the test speed does not make compliance with any safety standard more difficult. The changes made by this rulemaking action do not necessitate any change in efforts to comply with the existing safety standards. * * *
47 Fed.Reg. 21830 (1982), JA 95.
. This point is elucidated by the Supreme Court's analogous treatment of the Occupational Safety and Health Act of 1970. With respect to that Act, the Court held that the agency could regulate workplaces as unsafe only if it first found significant risks. See Industrial Union Dep't, AFL-CIO v. American Petroleum Institute, 448 U.S. 607, 655, 100 S.Ct. 2844, 2870, 65 L.Ed.2d 1010 (1980) (plurality opinion); id. at 664-665, 100 S.Ct. at 2875-2876 (Powell, J„ concurring in the judgment). This holding is analogous to the idea here that not every vehicle characteristic should be regulated under the Safety Act.
Later, the Court also interpreted the Occupational Safety and Health Act’s standard for setting safe levels for regulated substances. See American Textile Manufacturers Institute v. Donovan, 452 U.S. 490, 101 S.Ct. 2478, 69 L.Ed.2d 185 (1981). In that case the Court did not interject into its discussion an additional or substitute requirement of significance. I believe that the agency in this case should also have confined itself to the statutory standard.
. It is true that petitioners do not raise these concerns specifically. (They do raise the question of what findings the record must support in the sense that their opening brief focuses on the no reduction findings and asserts that safety should be placed first and their reply brief focuses on the no significant reduction findings asserted by the agency in its brief.) I do not believe this should interfere with our consideration of these issues, however. First, to the extent that the agency’s lack of clarity prevents meaningful judicial review we have an independent obligation to focus on it. Second, the agency’s lack of clarity may actually have caused petitioners to interpret statements in one way and focus their appeal on that interpretation. (For example, the agency’s statements can readily be understood to say that the Safety Act does not even apply because no reduction in safety would occur, rather than that the Safety Act is complied with by virtue of that finding.) Where the agency’s own actions may have misled those appealing its decision, it should not benefit from the ambiguous nature of those actions.
. I do not assert here that the agency’s conclusion that the data were inapplicable was wrong. As the majority notes, such decisions normally fall within the scope of the agency’s expertise. Rather, my point is that this conclusion was reached without supporting data and in contradiction to common sense. (My common sense indicates that increased protection of safety features would effect increased safety.) The majority notes that NHTSA merely concluded that State Farm’s data "falls short of demonstrating that [the difference] is noticeable.” Maj. op. at 1346, citing FRIA at IV-9, JA 202. As I note below, see text at p. 1349 infra, State Farm does not have to prove that the agency's conclusions are false. The agency itself must show that those conclusions are reasonable. That the proffered data were not conclusive does not prove their reverse.
The majority, in its eagerness to discredit the State Farm data, performs a little analysis of its own. See maj. op. at 1347 n. 7. There, it purports to show why deference to the agency is appropriate by noting that an alternative to State Farm’s interpretation of those data shows uncertain improvement for one safety component for Model Years 1974-1978 over Model Year 1973. Unfortunately, the majority fails to address how the relative percentages in its table translate into absolute numbers. (It might be that overall safety increased even where safety in one or two categories decreased.) The majority also fails to note that the MY 1974-1978 data submitted by State Farm for two other safety features show consistent and marked improvement in the average over MY 1973.
Number of Safety Items Damaged Per 100 Estimates Written
Component 2.5 mph rear bumper standard MY 1973 5.0 mph rear bumper standard MY’s 1974-1978 % improvement of 5.0 mph over 2[.5] mph
Trunk Lid
Subcompact 14.51 5.91 59%
Compact 11.55 7.27 37%
Intermediate 11.64 8.42 28%
Full Size 10.97 7.55 31%
Gas Tank
Subcompact .73 .35 52%
Compact .82 .35 57%
Intermediate .70 .23 67%
Full Size .54 .18 67%
. One way to put the agency’s conclusions in perspective is to make some tentative calculations as to the magnitude of safety effects potentially at issue here. As petitioners note, even where the percentages are small the absolute numbers might not be "insignificant." In this context, applying the percentages hypothesized above, we know that in 1980 50,000 people died and 2,850,000 were injured in 17,900,000 automobile accidents in this country. See reply brief for petitioners State Farm Mutual Automobile Insurance Company and Allstate Insurance Companies at 10. Assuming that 1% of these accidents were caused by potentially bumper-related safety feature problems, approximately 179,000 accidents were caused by such problems. Assuming 80% of these were property damage only and that 20% were thus personal injury accidents, perhaps 35,000 of these accidents involved personal injury. Even assuming that this number might be discounted further due to the fact that not all of these safety feature malfunctions resulted from crashes that would have caused negligible damage with 5.0 mph bumpers but significant injury with 2.5 mph bumpers, to totally discount this figure because of those factors — factors on which no data were cited — seems to me to define arbitrary and capricious behavior on the part of the agency.
. Even without this specific mention of health and safety costs, I believe that consideration of such costs would be mandated under § 1912(b)(1)(A), the general requirement that the costs and benefits of implementing a standard are to be taken into account.
. The majority asserts that insignificant safety effects need not be factored into the Cost Savings Act analysis. See maj. op. at 1355 n. 15. Even assuming, with the agency and the majority, that safety effects were supportably found to be insignificant and thus unimportant under the Safety Act, I do not see that that finding automatically proves their insignificance for Cost Savings Act purposes. (As noted previously, the lack of any guidance as to what the agency considers significant makes this review additionally difficult.) For example, it seems not inconceivable to me that a safety problem could be considered not sufficiently significant to regulate under the Safety Act while the costs of that problem were at the same time significant enough to factor into a Cost Savings Act cost-benefit analysis — especially where, as here, two alternative standards under the latter Act were close under that analysis. See text at p. 1357 infra.
. Taking the alternatives set forth in note 12 supra, if the costs and benefits of B accrue at the same time, 5 years in the future, its net benefit present value will remain $46.57. But if A’s costs accrue immediately (say $50) and its benefits accrue in 5 years (say $150), although its non-time-weighted net benefit will remain the same, the present value of its net benefit will be only $43.14, less than the present value of the net benefit of B.
.Again taking the alternatives set forth in note 12 supra, if B can be implemented in 5 years and A cannot be implemented until 10 years, B’s net benefit present value will again be $46.57, but A’s net benefit present value will decrease to $38.55, even assuming its costs and benefits accrue at the same time.
.The majority's emphasis on petitioners’ failure to raise their asserted objections during the 1979 proceedings would effectively undercut the general rule that when an agency reproposes a rule, even a rule that is already in force, all aspects of its decision to promulgate that rule after reproposing it are subject to comment and, subsequently, judicial review. Failure of a party to have challenged a specific aspect of the rule originally in no way precludes this. See, e.g., State of Montana v. Clark, 749 F.2d 740, 744 (D.C.Cir.1984). There are many reasons, as noted in text, why parties may choose not to address a specific aspect of a complex rulemaking. To conclude, in a later proceeding, that this choice proves the unimportance of that aspect unsupportably limits those parties’ right to participate fully in rulemaking proceedings.
. Petitioners repeatedly raised the question of the validity of the effectiveness curves by noting that what real world data there are contradict those theoretical conclusions. It is unclear to me that one must raise the precise point below that one raises on appeal; I would assume that raising a basic issue would suffice in many circumstances. Further, as the majority notes, IIHS did object to the assumption of 0% effectiveness at replacement speed when the curves were first drawn. As noted infra, the majority's dismissal of that point is as unconvincing as its conclusion that the curves were not critical to the agency’s analysis.
. Assumption (1) is not contested.
. The $11,000,000 figure comes from multiplying the $1 per car difference by the 11 million estimate of cars per year. See PRIA at VI-25, JA 502.