Full opinion text
Opinion for the court filed by Chief Judge J. SKELLY WRIGHT. J. SKELLY WRIGHT, Chief Judge: Chrysler Corporation has petitioned for review of a final order of the Administrator of the Environmental Protection Agency (EPA), issued on November 20,1978, directing Chrysler to recall all 1975 vehicles equipped with 360 and 400 cubic inch displacement (CID) engines having two-barrel carburetors and catalytic converters. The Administrator determined that a substantial number of the vehicles in the recall class fail to conform to the applicable carbon monoxide emission standards when in actual use, even though they have been “properly maintained and used.” Having found this violation of Section 207(c)(1) of the Clean Air Act, 42 U.S.C. § 7541(c)(1) (Supp. II 1978), the Administrator ordered Chrysler to submit a plan for remedying the nonconformity. Jurisdiction of this court is pursuant to Section 307(b) of the Act, 42 U.S.C. § 7607(b) (Supp. II 1978). Chrysler has sold approximately 208,000 vehicles in the recall class, all equipped with catalytic converters designed to reduce carbon monoxide emissions to within federal standards. Not long after these vehicles had been sold, however, EPA discovered that many of them were failing to meet the standards. After extensive tests EPA determined that excessive emissions were primarily caused by misadjustment of the carburetor idle mixture of the vehicles. The Agency gathered evidence that the misad-justments were the inevitable result of certain defects in the design of the emission control system of the recall class, and later initiated an administrative proceeding to require Chrysler to recall the vehicles and correct this design. An initial hearing before an Administrative Law Judge (ALJ) resulted in a recall order against Chrysler, which the company appealed to the Administrator. The Administrator determined, first, that as a matter of law a manufacturer must be held responsible in a recall action for nonconformities primarily caused by design defects, provided the manufacturer foresaw or should have foreseen the consequences of the defects but failed to take available steps to obviate them, and, second, on the evidence, that Chrysler’s recall class must be recalled under this standard. Chrysler disputes both of these positions. On the legal issue Chrysler argues that since the nonconforming vehicles in the recall class were misadjusted, they were not “properly maintained” within the meaning of the Act and thus should not be recalled. On the factual issue Chrysler asserts that the Administrator’s conclusions were not supported by substantial evidence in- the record. Because we agree with the Administrator’s interpretation of Section 207(c)(1) and determine that there was substantial evidence to support his findings of a violation, we affirm. I. STATUTORY FRAMEWORK In 1970 Congress passed the Clean Air Amendments of 1970, Pub.L.No.91-604, 84 Stat. 1676, which required auto manufacturers to reduce carbon monoxide emissions by 1975 to one-tenth of former levels: to 3.4 grams per mile. See 36 Fed.Reg. 12657 (1971). The Administrator may, however, postpone implementation of this statutory standard on grounds of technological feasibility or other factors. See Section 202(b)(5) of the Act, 42 U.S.C. § 7521(b)(5) (Supp. II 1978). In 1973 the Administrator postponed implementation of the 3.4 grams per mile standard and set an interim carbon monoxide emission standard of 15 grams per mile. This 15 grams per mile standard was in effect during the 1975 model year, with which we are concerned. To comply with the Act manufacturers must design, build, and equip each new vehicle to conform to emission standards at the time of sale and to be free from defects in material or workmanship that would cause the vehicle to fall below the standards within a five-year or 50,000-mile period after sale. Section 202(a)(1), (d)(1), 42 U.S.C. § 7521(a)(1), (d)(1) (Supp. II 1978). Manufacturers must provide a warranty to purchasers to this effect. Id. § 207(a)(1), 42 U.S.C. § 7541(a)(1). If the purchaser of a vehicle maintains and operates it in accordance with the written instructions of the manufacturer, see id. § 207(c)(3), 42 U.S.C. § 7541(c)(3), and it fails to conform to emission standards during the warranty period (thus subjecting the owner to penalty or other sanction) the manufacturer must remedy the nonconformity at its own expense. Id. § 207(b), 42 U.S.C. § 7541(b). To ensure compliance with the Act EPA conducts a three-stage testing process. Under authority of Section 206(a), 42 U.S.C. § 7525(a), the Agency examines prototypes of new vehicles or new vehicle engines to determine whether they will conform to the emission standards and issues a “certificate of conformity” to vehicles passing this test. As a part of this examination the Agency inspects the written maintenance and use instructions provided to the purchasers. Until the 1977 amendments went into effect the Agency determined whether such instructions were “reasonable and necessary to assure the proper functioning” of the emission control system. Section 207(c)(3), 42 U.S.C. § 1857f-5a(c)(3) (1976). The second stage of the testing process takes place after vehicles come off the assembly line. At this time the Agency examines sample vehicles to ensure that they conform to the requirements of the Act. Authority to conduct this examination derives from Section 206(b), 42 U.S.C. § 7525(b) (Supp. II 1978). If the Administrator determines that some or all of the vehicles off the assembly line fail to conform to the applicable regulations, he must suspend the certificates of conformity for those vehicles until the manufacturer corrects the deficiency. The third stage of the testing process takes place while the vehicles are in actual use. EPA and related state agencies test sample vehicles to determine whether they continue to satisfy emission standards during the statutory period. If a “substantial number” of the vehicles in any class or category fail to conform to the emission standards “although properly maintained and used” by the owner, then the Administrator must notify the manufacturer and require it to submit a “plan for remedying the nonconformity” at the manufacturer’s expense. Unlike the discovery 'and cure of nonconformity of individual vehicles under the warranty provisions, the remedy at this stage is recall of the entire class of vehicles in order to correct the design, material, or workmanship defect causing the nonconformity. This provision—-Section 207(c)(1) of the Act, 42 U.S.C. § 7541(c)(1) (Supp. II 1978)—is the statutory basis for the order under review. II. THE RECALL CLASS A. Design of the Emission Control System To reduce carbon monoxide emissions and bring its vehicles into conformity with the interim federal standards Chrysler, like the other American auto manufacturers, installed catalytic converters in its 1975 model year vehicles. A catalytic converter can reduce carbon monoxide emissions by 60-80 percent by promoting a chemical reaction among the carbon monoxide, hydrocarbons, and oxygen. This reaction produces two harmless byproducts, carbon dioxide and water. An adequate supply of oxygen in the exhaust stream is essential to effective operation of the catalytic converter. Unless there is enough oxygen to oxidize all the emissions, the catalytic converter will begin to work poorly and at some point cease to operate altogether. There are two major methods of supplying enough oxygen: installation of an air pump to introduce additional oxygen into the system, and precise adjustment of the carburetor idle to ensure that sufficient oxygen is mixed with the fuel. The principal disadvantage of air pumps is that they increase the cost of the system by about $50 per car; they also decrease gasoline mileage and may in some cases inhibit proper oxidation of the emissions. On the other hand, Chrysler engineers recognized that air pumps generally decrease emissions more effectively than does the carburetor idle adjustment system. It was for just that reason that Chrysler used air pumps in its 1975 model year vehicles subject to the more stringent emission standards of California. Nevertheless, Chrysler did not use air pumps in the recall class vehicles it sold in other states, because Chrysler engineers thought them unnecessary to achieving compliance with federal standards. Instead, Chrysler decided to rely on the carburetor adjustment method to ensure an adequate supply of oxygen to the catalytic converter. The company instructed owners to take their vehicles to a mechanic for servicing at intervals of 15,000 miles or whenever they detected a malfunction. If all went well, the mechanic would adjust the carburetor precisely in accord with Chrysler’s instructions, as detailed in the service manual and summarized on a permanent label affixed to the underside of the engine hood. By means of this adjustment enough oxygen would be mixed with the fuel in the idle circuit to enable the catalytic converter to oxidize most of the carbon monoxide in the exhaust. A correct adjustment of the fuel-air mixture in the idle circuit is called a “lean” setting, meaning a low proportion of fuel to air. This setting is measured by the percentage concentration of carbon monoxide in the exhaust as detected upstream of the catalyst. Chrysler calculated that a 0.5 percent carbon monoxide setting would generally enable the catalytic converter to reduce carbon monoxide emissions to about 12 grams per mile—well within the interim standard. In order to provide a greater margin of safety, Chrysler decided to specify a 0.3 percent carbon monoxide setting. When the carburetor idle is set to Chrysler’s specifications, most of the vehicles pass the carbon monoxide emission standards; when the idle mixture is adjusted to “richer” levels, most of the vehicles fail the standards. Unfortunately, the idle adjustment method has apparently not worked as well in practice as in theory. EPA researchers have linked the poor emission control performance of the recall class in actual use to certain characteristics of the carburetor idle adjustment process that make precise adjustment difficult and undesirable to owners and mechanics. To explain what EPA finds to be wrong with the design of this process, we must describe in some detail the way the idle mixture is adjusted and the relation of such adjustment to driveability of the vehicles. The adjustment process is cumbersome and time-consuming, taking the mechanic approximately 30 to 40 minutes, according to an EPA investigation. The mechanic must first obtain a Chrysler Huntsville exhaust emission analyzer or other approved infrared analyzer, and must verify that it is warmed up and calibrated according to the manufacturer’s instructions. He must check to see that the sample lines and connections of the sampling system for the analyzer are free of leaks, and must also warm up the vehicle’s engine and allow it to idle for no more than ten minutes. Then the mechanic must remove the plug from the threaded catalyst tap on the vehicle and install the sample line of the analyzer in the tap upstream of the catalyst. This will generally require the mechanic to crawl under the car or to use a hoist. He must then start up the vehicle’s engine and verify that the idle RPM and timing are within specification. At this point he must measure the idle carbon monoxide concentration. If it exceeds specified limits he must adjust the mixture screws on the carburetor to achieve a “leaner” mixture of fuel and air in the idle circuit, checking back and forth between the analyzer and the carburetor to monitor the effect of his adjustment on the idle mixture. A clockwise turn of a screw will decrease the amount of fuel discharged through the idle port and into the idle circuit; a counterclockwise turn will increase the amount of fuel, resulting in a “richer” mixture. The adjustment screws are highly sensitive and must be turned in small fractions of a rotation in order to set the adjustment with sufficient precision. After making the proper adjustment the mechanic must also balance the mixture screws on a two-barrel carburetor for lowest level of hydrocarbons or smoothest idle within the prescribed specifications. An EPA expert who interviewed 27 Chrysler dealership mechanics and also performed the maintenance procedures concluded that “the procedure is cumbersome, time consuming and most importantly will usually result in customer dissatisfaction.” Joint Appendix (JA) 1:368 (Gockel testimony). The complex and time-consuming nature of the adjustment procedure is significant because the manufacturer did not allow enough time in its reimbursement schedules for dealership mechanics to make carburetor adjustments, according to a survey of Chrysler mechanics. This meant that many mechanics had to complete the specified adjustments without full compensation for their time. Moreover, Chrysler did not reimburse mechanics for a second adjustment in the event a customer was dissatisfied with the results of the first, as frequently happened when the idle was set to so “lean” a mixture. Various alternative idle adjustment methods employed by mechanics for many years are both less time-consuming and less likely to displease the customer. But as both EPA and Chrysler experts agree, alternative procedures are likely to produce mixtures “richer” than those needed by the emission control system. Several specific factors make the required procedures more difficult to perform. First, the Administrator found that precisely calibrated and fully operating infrared analyzers are often unavailable, and that most dealers were not attaching the analyzers to the upstream tap, as specified by Chrysler. Evidence in the record shows that 41 percent of nondealership service facilities lacked such analyzers as of August 1975. Moreover, in 1975 Chrysler undertook a survey of emission testing equipment at 39 dealerships; that survey revealed that “[a] significant number of analyzers had leaks in the sampling systems large enough to cause incorrect readings.” JA V:2001. One dealer in the survey had no analyzer at all; one had a “totally inoperative” analyzer; one had an analyzer so unstable that it could not be checked; ■ one had its analyzer at the shop for repairs. Eleven other analyzers at the dealerships had major failings, such as faulty calibration or a sticking carbon monoxide meter. The Chrysler engineer analyzing the results flatly stated that “routine maintenance of the instruments is not performed.” JA V:2001 (memorandum from A. T. Weibel to W. S. Fagley). These results are corroborated by tests undertaken by California and New Jersey. Moreover, a majority of mechanics at Chrysler dealerships did not regularly use the exhaust analyzer at the catalyst probe, as required by Chrysler’s instructions. A second special problem with the emission control system adopted by Chrysler is that the carburetor adjustment screws are so sensitive that tiny rotations of the screws will grossly affect the idle fuel-air mixture. A mere V20 of a turn of the mixture screws above Chrysler’s specified adjustment range will cause a substantial increase in idle carbon monoxide. A full rotation of the screw will produce an idle carbon monoxide level about 23 times Chrysler’s specifications. Chrysler attempts to pass this problem off, saying that it merely causes the adjustment procedure to “take slightly longer.” Brief of petitioner at 65. But the sensitivity of the adjustment screws is more significant than Chrysler is willing to admit. In order for the emission control system to work, the idle mixture must be set precisely, within very narrow limits. The hypersensitivity of the screws makes misad-justments far more likely, especially if mechanics are inexperienced or pressed for time. Chrysler attempted to alleviate the adjustment problem by installing plastic limiter caps on the idle adjustment screws. Ideally, such caps would prevent the mechanic from turning the screws to a mixture “richer” than specification. But the limiter caps are easily removed or damaged ; indeed, the mechanic must remove the caps in order to make certain repairs. Moreover, Chrysler’s limiter caps were such that even when in place they permitted adjustment of the idle mixture to levels many times “richer” than Chrysler’s specifications. Most important, the Chrysler procedure often disrupts the vehicles’ smooth operation. Generally speaking, a vehicle drives more smoothly and has a smoother idle with a relatively “rich” fuel-to-air mixture. “Lean” settings such as those specified by Chrysler often generate “driveability” problems and hence customer dissatisfaction. The recall class vehicles are adjustable to any setting between 0 and 8 percent carbon monoxide, as measured at the tailpipe, with “best idle” occurring between 5 and 8 percent. In contrast, the setting required for effective operation of the catalytic converter is 0.3 percent idle carbon monoxide, as measured upstream of the catalyst. As a Chrysler engineer admitted, the “lean” idle setting specified by the Chrysler “certainly affects idle quality.” JA 11:875 (Harris testimony). He stated that “rich adjustment towards the area of best idle will produce better idle by definition than a leaner setting.” JA 11:877. In particular, he noted that at leaner settings it becomes more important to balance the carburetor precisely, so as to deliver the same fuel-air mixture to all of the cylinders. At a “lean” mixture it becomes “more likely that one cylinder or two cylinders might not be receiving the proper fuel distribution.” JA 11:875. Thus the idle is more likely to be rough and uneven. A large majority of dealership mechanics surveyed by EPA stated that it is not possible to achieve “acceptable engine smoothness and drivea-bility” when the idle mixture is adjusted to Chrysler’s specifications. The mechanics also said that customers complain of rough idle, hesitation, poor performance, or surge when the idle mixture is properly set. A significant majority admitted that they do not use the specified Chrysler procedure for adjusting the idle mixture. More than two-thirds of the mechanics said that they “often” or “always” remove the carburetor limiter caps in order to make a good idle adjustment, thereby indicating that they must set the idle mixture far “richer” than operation of the emission control system would permit. As the Administrator said, “Given customer dissatisfaction with the idle quality and the fact that the best idle quality is achieved at an idle setting far richer than Chrysler’s specifications, mechanics are motivated to misadjust the vehicles.” JA 111:1219 (Decision and Order). In comparison, most Ford vehicles in the 1975 model year were equipped with air pumps, which obviate the need for precise carburetor adjustments to extremely “lean” idle mixtures. Since air pumps provide sufficient oxygen to the catalytic converter, the idle can be set at a “richer” mixture, as driveability seems to require, without disabling the emission control system. Other undesirable features of the Chrysler system were also absent from most of the systems made by Chrysler’s competitors. A Chrysler expert admitted that most of. the idle mixture screws used by other manufacturers were “substantially” less sensitive than that used by Chrysler. The limiter caps used by other manufacturers—unlike those used by Chrysler—would not permit the adjustment screws to be adjusted beyond the manufacturer’s specifications. Moreover, Chrysler was the only domestic auto manufacturer to install a system requiring use of an exhaust analyzer at the upstream tap. B. Performance and Testing of the Recall Class EPA approved the prototypes of recall class vehicles submitted for initial examination and issued them a certificate of con-' formity. However, at the second-stage of testing Chrysler and EPA received preliminary indications that the emission control system of the recall class might not perform as well as expected. Early in 1975 Chrysler submitted data to EPA from tests of the vehicles as they came off the assembly line. These tests, conducted in accordance with the “Federal Test Procedure,” see 40 C.F.R. § 85.075-9 (1976), involved a simulated seven-mile driving cycle performed on a dynamometer. Examination of the exhaust emitted by the vehicles during the simulation revealed that approximately 45 percent of the recall class vehicles coming off the assembly line did not conform to the carbon monoxide emission standard. Chrysler attributed this high failure rate to the newness of the vehicles, which the company predicted would conform after being driven sufficiently to be broken in. In addition, EPA received reports from four states which conducted “short” tests indicating that a large percentage of 1975 Chrysler vehicles in actual use, including those in the recall class, greatly exceeded the idle carbon monoxide concentration level specified by Chrysler, and thus were highly likely to be in violation of the federal standards. Chrysler attributed these results to test conditions in the state programs and to carburetor misadjustments in the field. On the basis of these early reports EPA initiated an investigation of the recall class. The EPA investigators used data from a variety of testing programs conducted by Chrysler, four states, and EPA itself. In addition to the Federal Test Procedure results discussed above, Chrysler submitted the results of “short” tests conducted at the assembly line, at its new car preparation center in Chicago, and at five representative dealerships in Chicago. In each instance Chrysler discovered that many of the vehicles in the recall class exceeded Chrysler’s specified idle carbon monoxide adjustment. Significantly, many of the vehicles registered a carbon monoxide concentration in the exhaust of 1.0 percent or higher—a strong indication that the vehicles were in violation of the 15 grams per mile federal standard. The results of these tests may be summarized in tabular form: Chrysler “Short” Test Results Number and Type of Vehicle Percentage Misadjusted Percentage Over 1.0% CO Off Assembly Line: 67 360 CID vehicles 46% 27% 78 400 CID vehicles 27% 17% At New Car Preparation: 48 360 CID vehicles 27% 38 400 CID vehicles 18% 18% At Dealers: 14 360 CID vehicles 29% 29% 15 400 CID vehicles 33% 27% These results tend to indicate that many vehicles in the recall class were in violation of federal emission standards even before passing into the hands of purchasers. In the Washington, D. C. area EPA conducted a “short” test of recently purchased recall class vehicles. The test revealed mis-adjustments even more prevalent than those indicated by the Chrysler test results. Of 18 vehicles tested, 10 had received no maintenance from their owners that might affect the idle mixture, while 8 had received such maintenance. Ninety percent of the vehicles in the former category were found to be misadjusted; the average idle carbon monoxide concentration of the vehicles, as measured at the tailpipe, was 3.2 percent. Half of the vehicles that had received relevant maintenance were found to be misadjusted; the average carbon monoxide concentration was 1.8 percent. “Short” tests performed by state personnel and EPA contractors in Oregon, Ohio, Illinois, and New Jersey revealed that more than 64 percent of the tested 360 CID vehicles in the recall class, and more than 50 percent of the 400 CID vehicles, registered over 1.0 percent carbon monoxide concentration, as measured at the tailpipe. EPA conducted two major tests of the recall class, on which the ALJ and the Administrator primarily relied in reaching their decisions: the Emission Factor Program and the Olson Program. The Emission Factor Program tested randomly selected vehicles by various manufacturers in actual use in four different cities, without regard to their histories of maintenance or use. It was conducted by independent testing laboratories under contract to EPA, with some participation by EPA personnel. The investigators tested vehicles under the “Federal” procedure and other procedures, including the “short” test. Eighty-five percent of the recall class vehicles registered carbon monoxide emissions in excess of the 15 grams per mile standard, under the “Federal” procedure. The average emissions of all recall class vehicles tested was over 48 grams per mile. Under the “short” test 11 percent of the recall class vehicles registered excessive carbon monoxide concentration. Unlike the Emission Factor Program, the Olson Program was specifically designed to test whether vehicles in the recall class, although properly maintained and used, were in violation of the carbon monoxide emission standards. EPA contracted with Olson Laboratories of Livonia, Michigan to conduct the test. Olson personnel obtained a randomly-ordered list of owners of all recall class vehicles in Wayne County, Michigan, and contacted these owners by letter and telephone in an attempt to persuade them to participate. On the basis of questionnaires and inspections EPA personnel eliminated all vehicles that failed certain criteria, including some maintenance requirements. Vehicles not falling within these exclusionary criteria were tested by Olson, EPA, and Chrysler personnel. If these investigators found any disablement of the emission control system or other tampering, they eliminated the tampered vehicle from the test. The distinctive feature of the Olson Program was that the investigators tested vehicles in the condition in which they were received, then performed extensive maintenance, including adjustments of the idle mixture in accordance with Chrysler’s specifications, and finally tested them a second time. In this way EPA could determine the effect of maintenance and idle adjustment on carbon monoxide emissions. Before receiving maintenance 90 percent of the vehicles tested in the Olson Program exceeded carbon monoxide emission standards. After maintenance only 25 percent of the tested vehicles failed to satisfy the standards. The results may be summarized in tabular form; Olson Testing Program Number and Type of Vehicle Percentage in Violation Average CO Emissions As Received: 10 360 CID vehicles 100% 69.02 grams/mile 10 400 CID vehicles 80% 33.61 grams/mile After Maintenance: 10 360 CID vehicles 20% 11.35 grams/mile 10 400 CID vehicles 30% 15.45 grams/mile The Administrator has called the Olson test results “[t]he most decisive and objective evidence establishing the high number of misadjustments * * *.” JA 111:1225. Of the 46 vehicles tested in the Emission Factor and Olson Programs, 37 had logged fewer than 15,000 miles. Since the 15,000-mile point is the first occasion when owners are instructed to readjust the idle mixture, vehicles logging fewer than 15,000 miles might be viewed as “properly maintained.” Moreover, all nine of the vehicles logging more than 15,000 miles had been taken to service facilities for a tune-up, in accordance with Chrysler instructions—seven of them to Chrysler dealerships. A large percentage of the tested vehicles—65 percent of the Olson vehicles and 38 percent of the Emission Factor vehicles—had received unscheduled maintenance relating to the carburetor or tuning because of owner complaints concerning such driveability problems as hesitation, rough idle, stalls, hard starting, pinging, and poor fuel economy. Evidence suggests that the misadjust-ments occurred to some degree on the assembly line, and that they tended to become statistically more severe as mileage accumulated. A compilation of Chrysler, state, and EPA test results, correlated according to vehicle mileage, may be summarized in tabular form: Test Summary Average Idle CO Percentage Percentage Misadjusted Over 1% CO 65% 65% 73% 68% No. of Mileage Vehicles 0-1,000 17 1,000-5,000 37 to W to CO Mileage No. of Vehicles Average Idle CO Percentage Misadjusted Percentage Over 1% CO 5,000-10,000 25 2.9% 72% 72% 10,000-15,000 20 3.1% 75% 75% Over 15,000 10 4.3% 90% 90% Evidence in the record indicates that these results are representative of cities all across the nation. EPA performed a statistical analysis of the results of the Emission Factor and Olson Programs. This analysis revealed, with 95 percent confidence, that no fewer than 79 percent of all 360 CID vehicles in the recall class, and no fewer than 62 percent of all 400 CID vehicles in the recall class, violate the carbon monoxide emission standards when in actual use. Using only the Olson Program results, EPA was able to project with 95 percent confidence that no fewer than 79 percent of the 360 CID, and no fewer than 49 percent of the 400 CID, vehicles in the recall class violate the standards in actual use. III. THE ORDER UNDER REVIEW A. The Notice of Nonconformity Relying on the results of state tests, the Emission Factor Program, and the Olson Program, the EPA Administrator, then Russell E. Train, issued a letter on December 8,1976 instructing Chrysler to submit a plan for remedying the nonconformity of the recall class. The Administrator said that the test results showed “that a substantial number of vehicles in [the recall class] are exceeding the 1975 Federal carbon monoxide standard in actual use,” and that “carburetor idle CO misadjustment is the primary cause of the nonconformity of these vehicles.” JA 111:912. He noted that misadjustment of the vehicles might be taken as proof that they had not been “properly maintained” under the statute, but he stated that Chrysler must nevertheless be considered liable to a recall action. He reasoned: Chrysler is responsible for these misad-justments because Chrysler as an automobile manufacturer should have foreseen that its carburetor design and adjustment procedures would cause widespread misadjustments and because of the agency relationship which exists between Chrysler and its authorized dealerships. JA 111:912-913. Chrysler took issue with the Administrator’s determination and demanded a public hearing before an AU, as provided in Section 207(c)(1) of the Act, 42 U.S.C. § 7541(cXl) (Supp. II 1978), and EPA regulations, 40 C.F.R. § 85.1807 (1979). B. The Initial Decision The case was assigned to ALJ Edward B. Finch. Both parties undertook extensive discovery, over which several disputes arose. Most important, Chrysler was not permitted to obtain some 20 EPA documents or to depose Mr. Eric Stork, head of the EPA team that certified the recall class. After four prehearing conferences and a lengthy public hearing in which Chrysler and EPA were the only parties, the ALJ issued an Initial Decision in favor of EPA ordering Chrysler to submit a recall plan in accordance with Section 207(c)(1). JA 111:1090-1196. The ALJ delineated two major disputed issues: first, whether the test vehicles that served as the basis for the Administrator’s notice of nonconformity were “properly maintained” within the meaning of Section 207(c)(1), and second, assuming that those vehicles had been “properly maintained,” whether there was sufficient evidence to support the Administrator’s decision that there were “a substantial number” of vehicles in the recall class that “do not conform” to the federal carbon monoxide emission standards. EPA urged the ALJ to adopt a definition of “properly maintained” that would refer solely to the actions of the vehicle owner in taking his car to a bona fide service station for Chrysler’s recommended maintenance at the recommended time intervals without regard for whether the maintenance was performed properly or not. The ALJ rejected this argument, holding that “it is the actual maintenance performed on the vehicle which is the determining factor.” JA 111:1102. He concluded that “the phrase ‘properly maintained’ has a technical meaning within the automobile industry and, when used within that industry, means ‘maintained in accordance with the manufacturer’s [instructions or] specifications.’ ” JA 111:1101 (brackets in original). However, the ALJ ruled in effect that vehicles logging less than 15,000 miles should automatically be considered “properly maintained” because Chrysler does not require readjustment of the idle setting within the first 15,000 miles of operation. Using this definition of “properly maintained,” the ALJ found substantial evidence supporting the conclusion that a substantial number of vehicles in the recall class do not conform to federal carbon monoxide emission standards. Relying on tests of low-mileage vehicles in the recall class conducted by four states and Chrysler, the ALJ determined that vehicles that “may be considered to have been properly maintained” under his definition—that is, vehicles logging less than 15,000 miles—demonstrated “gross conditions of idle misadjustment.” JA III :1107. Then, relying solely on the Olson and Emission Factor Programs, as analyzed in accordance with standard statistical techniques, he concluded that a “substantial number” of vehicles in the recall class failed to conform to the emission standards. For this purpose he disregarded the state and Chrysler investigations that'used the tailpipe exhaust, or “short,” method because he did not believe that the 1.0 percent carbon monoxide concentration cutoff point used by those tests was sufficiently reliable evidence of failure to meet the federal emission standards. The ALJ rejected all challenges to the validity of the Emission Factor and Olson Programs. Thus he reached a two-step conclusion: first, that idle misadjustments are characteristics of “properly maintained” vehicles, and second, that idle misadjustments have caused “a substantial number” of vehicles in the recall class to violate the carbon monoxide emission standards. Therefore, he found Chrysler liable to a recall action under the Act. In addition, the AU held that Chrysler should be held responsible for nonconformities caused by idle misadjustments by owners and mechanics in the field. Relying primarily on precedents in the products liability area, he held: The Clean Air Act imposes a statutory obligation upon the manufacturer to design and produce a vehicle emission system which will, its useful life, conform to the applicable emission standards. Any failure to do so is noncompliance with the Act. The standards set forth in the Act are intended for the purpose of cleaning up the nation’s air quality for the benefit of man and the environment. Any failure to accomplish such purpose by any means, including a “defect in design” carburetor, as here, constitutes a violation of the intended purposes of the Act for which liability attaches. * * * JA 111:1157-1158. EPA argued that Chrysler’s emission control system design was faulty because it contained no mechanism to limit idle mixture, was overly sensitive to small turns of the adjustment screw and to temperature and idling time, and contained no alternative means of providing excess air to the catalyst in the event of misadjustment. Because of these alleged design defects, and tests showing that a substantial number of the recall class vehicles were in a state of misadjustment, the ALJ concluded “that the design of the emission control system, including the carburetor, encourages or fosters the gross misadjust-ments which have been shown to exist and therefore should have been foreseen.” JA IILllSS. The ALJ also concluded that Chrysler may be held responsible for misadjustments made by dealership mechanics because Chrysler dealers are agents of Chrysler in the performance of pre-delivery and warranty services and all emissions-related services. He stated that Chrysler has the “right to control” dealership performance of these services, and that members of the public reasonably rely on the dealers’ skill and Chrysler⅛ representations in servicing Chrysler cars. On the basis of these findings and conclusions the ALJ ordered Chrysler to submit a plan for remedying the nonconformity of the recall class, under Section 207(c)(1). C. The Administrator’s Decision and Order Chrysler appealed the Initial Decision of the ALJ to the EPA Administrator, Douglas M. Costle, on February 14,1978. On the basis of appellate and supplemental briefs, but without hearing oral argument, the Administrator issued the order under review on November 20, 1978. The Administrator acknowledged that the key issue in the proceeding was whether the vehicles in the recall class were “properly maintained” within the meaning of Section 207(c)(1). He reviewed the interpretations of “properly maintained” urged by EPA and Chrysler and also that adopted by the ALJ, but he found that “[t]he problem with all of these interpretations of the ‘properly maintained’ criterion is that they place undue emphasis on the text of the words ‘properly maintained’ without examining the purpose or function of the ‘properly maintained’ criterion in the overall statutory scheme.” JA III:1211-1212. He reasoned that the purpose of the “properly maintained” criterion of Section 207(c)(1) was “to allocate responsibility for emission nonconformities among the manufacturer, the vehicle owner and the mechanic.” JA 111:1212. The manufacturer should not have to bear the expense of a recall if the fault properly lies with the vehicle operator; on the other hand, if it is principally responsible for the nonconformity the manufacturer should be held accountable, even if the nonconformity relates to maintenance of the vehicle. Otherwise, the manufacturer would be able to avoid its duty to make vehicles that would conform to the emission standards during their useful life, and could frustrate the purpose of the Act. Thus the Administrator interpreted Section 207(c)(1) as imposing recall liability on the manufacturer for maintenance-related non-conformities only if EPA could establish that the vehicles would have been maintained properly “but for the actions of the manufacturer.” JA 111:1215 (emphasis in original). He describes EPA’s burden of proof as the presentation of “affirmative evidence” that the manufacturer is “primarily responsible for the condition of improper maintenance.” Id. In the case at bar the Administrator found that such affirmative evidence had been presented. He concluded, on the basis of the record, that “[w]hile that evidence also shows that the service industry does indeed contribute to the condition of improper maintenance * * *, it is clear that the service industry's contribution is the inevitable byproduct of Chrysler’s emission system design and service procedures.” Id. In reaching this conclusion the Administrator painstakingly reviewed the evidence in the record concerning the alleged design defects in the Chrysler emission control system, including the length and difficulty of the idle adjustment procedure, the inadequacy of the plastic limiter caps, the lack of reliable infrared exhaust analyzers, the sensitivity of the adjustment screws, and the failure to install an air pump. Focusing on the difficulty of sufficiently precise adjustment and resulting customer dissatisfaction, the Administrator found that these factors, “inherent in the recall class vehicles,” made the likelihood of proper adjustment “exceedingly small” for vehicles in actual use. JA III:1218. Having attributed responsibility for idle misadjustments to Chrysler, the Administrator easily found sufficient evidence in the record to support a finding that a substantial number of vehicles in the recall class were in violation of the emission standards as a result of such misadjustment. He relied primarily on the Olson Program to support his conclusion, but found that other testing programs, including those using the “short” test, could be used to corroborate the Olson Program result: namely, “that a high number of the recall class vehicles in the field are in a condition of misadjustment.” JA 111:1231. The Administrator found, on the basis of the evidence from these tests, that the misadjustments were uniformly prevalent throughout the country, and that they increase in frequency and severity as the mileage of the vehicles increases. Chrysler argued that the results of the Olson Program tests performed on vehicles after maintenance in accordance with Chrysler instructions supported its contention that most vehicles in the recall class pass the emission standards when properly maintained. As we have noted, after maintenance . about three-quarters of the recall class vehicles satisfied the standards. The Administrator, however, concluded that the maintenance performed on the vehicles during the Olson Program was not the type of maintenance contemplated by Section 207(c)(1) of the Act. He said that the recall determination must be based upon a finding that a substantial number of vehicles exceed the applicable emission standard when “in actual use,” although properly maintained and used. * * * Maintenance in actual use is where all of the factors that have been previously identified as contributing to the cause of misadjustments will combine and represent the true state of maintenance being performed on the recall class vehicles. The sensitivity of the adjustment screws, the inconvenience involved in attaching the exhaust gas analyzer at the tap in front of the catalyst, the pressure to satisfy the customer, and the incentives to complete the task in an expeditious manner all interact in a way that is totally unlike the maintenance performed in the Olson Program. JA 111:1244.. Thus the Administrator relied on the “as received” portion of the Olson tests in reaching his determination. The Administrator analyzed at length the expert testimony by Chrysler and EPA witnesses concerning the foreseeability of mis-adjustments of the idle mixture, given the design of the recall class. He discounted the testimony of three main Chrysler witnesses and credited the testimony of EPA witnesses that the carburetor design necessarily encouraged or fostered misadjustment and should have been known by Chrysler to do so. He stated: * * * The evidence presented by EPA in the form of expert opinion and other proof convinces me that Chrysler should have foreseen the possibility of widespread misadjustments occurring in the recall class vehicles due to the combined effects of the various factors described above and further that such mis-adjustments would result in a substantial number of the recall class vehicles exceeding the Federal emission standard for carbon monoxide. * * * JA 111:1233. Chrysler also argued that requiring a recall in this case is tantamount to retroactive rulemaking because EPA has issued new regulations that require manufacturers to construct all vehicles to meet emission standards even when misadj usted. 44 Fed.Reg. 2960 (1979). Chrysler suggested that the order now under review is, in effect, an application of this prospective rule to a past model year. The Administrator rejected this argument, saying, “The recall order is based on the fact that the defective design and service procedures induced misadjust-ments, not on the fact that the design merely allowed misadjustments.” JA 111:1257 (emphasis in original). He also rejected Chrysler’s allegations of material evidentia-ry and discovery errors on the part of the ALJ, and declined to rule on EPA’s theory—adopted by the ALJ—that Chrysler was responsible for the misadjustments made by Chrysler dealers because such dealers are agents of Chrysler for the purpose of maintaining emission equipment. The Administrator therefore affirmed the ALJ’s order and required Chrysler to submit a plan for correcting the nonconformities of the recall class. D. The Denial of Chrysler’s Motion for Reconsideration On December 17, 1978, about one month after the Administrator released his Decision and Order in this case, EPA published a study entitled “Motor Vehicle Tampering Survey” concerning the problem of tampering with emission control systems of vehicles made by all major manufacturers. Claiming that the survey introduced new evidence relevant to this case, Chrysler petitioned the Administrator for reconsideration of his Decision and Order. Chrysler contended that the survey showed that tampering with emission control systems was an industry-wide phenomenon, in no way encouraged or caused by defects in Chrysler’s carburetor design. The Administrator denied Chrysler’s motion for reconsideration. He analyzed the underlying raw data of the “Motor Vehicle Tampering Survey” and found that Chrysler vehicles of the 1975 model year averaged 3.87 percent carbon monoxide at idle. In contrast, 1975 General Motors vehicles averaged 1.72 percent and Ford vehicles averaged 0.88 percent. He therefore found no reason to depart from his earlier findings and conclusions. This petition for review followed. IV. THE MEANING OF THE “PROPERLY MAINTAINED” CRITERION IN SECTION 207(C)(1) A. Scope of Review This case is the first contested recall action under Section 207(c)(1) to reach the courts, and we have little precedent to guide us in deciding it. The only authoritative interpretation of the relevant portions of the Act is that adopted by the Administrator in the order under review. We must accord his interpretation “important but not controlling significance.” Batterton v. Francis, 432 U.S. 416, 424, 97 S.Ct. 2399, 2404, 53 L.Ed.2d 448 (1977); see also Griggs v. Duke Power Co., 401 U.S. 424, 433-434, 91 S.Ct. 849, 854-855, 28 L.Ed.2d 158 (1971). As the Supreme Court has said, “It is the settled rule that the practical interpretation of an ambiguous or doubtful statute that has been acted upon by officials charged with its administration will not be disturbed except for weighty reasons.” Brewster v. Gage, 280 U.S. 327, 336, 50 S.Ct. 115, 117, 73 L.Ed. 981 (1930). Although this court has the duty under the Administrative Procedure Act, 5 U.S.C. § 706 (1976), to “decide all relevant questions of law,” we recognize that the special expertise of EPA in interpreting the legislation which it is called upon to administer requires that we defer to the judgment of the Agency where that judgment is reasonable and is consistent with the language and purpose of the legislation. See generally Wilderness Society v. Morton, 479 F.2d 842, 864-870 (D.C.Cir.) (en banc), cert. denied, 411 U.S. 917, 93 S.Ct. 1550, 36 L.Ed.2d 309 (1973). B. The Statute Chrysler argues that the language of Section 207(c)(1), the recall provision, is clear on its face, and that this court should reject the .Administrator’s interpretation even without resort to interpretive aids such as legislative history. We cannot agree. Section 207(c)(1), 42 U.S.C. § 7541(cXl) (Supp. II 1978), reads in relevant part: If the Administrator determines that a substantial number of any class or category of vehicles or engines, although properly maintained and used, do not conform to the [emission standards], when in actual use throughout their useful life * * * he shall require the manufacturer to submit a plan for remedying the nonconformity ***.*** Chrysler argues—and the ALJ agreed —that the phrase “properly maintained and used” must be interpreted in accordance with Section 207(b)(2)(A), 42 U.S.C. § 7541(b)(2)(A) (Supp. II 1978), the warranty provision of the Act. That provision requires the manufacturer to repair or replace the emission control systems of its vehicles under certain conditions. One of those conditions is that the vehicles must have been “maintained and operated in accordance with instructions under subsection (c)(3) of this section.” Id. Subsection (c)(3) requires manufacturers to furnish vehicle purchasers with “written instructions for the proper maintenance and use of the vehicle” in accordance with EPA regulations. 42 U.S.C. § 7541(c)(3)(A) (Supp. IÍ 1978). Thus Chrysler concludes that “proper maintenance ” for the purpose of a recall action under Section 207(c)(1) must be defined as maintenance strictly in accordance with the written instructions provided to vehicle purchasers as required by Section 207(c)(3). We agree that Chrysler has propounded one plausible interpretation of the Act, but we cannot agree that no other interpretation is possible. First, Chrysler’s interpretation does not account for the difference in language between Section 207(c)(1), which limits a recall to classes of vehicles that had been “properly maintained and used,” and Section 207(b)(2)(A), which limits warranty claims to vehicles that had been “maintained and operated in accordance with [the manufacturer’s written] instructions[.]” “Properly” in this context may mean the same thing as “in accordance with [the manufacturer’s written] instructions,” as Chrysler says. On the other hand, the difference in language may indicate a subtle difference in meaning. In a warranty action the individual vehicle owner seeks to shift his cost of repairing the emission control system to the manufacturer; he has every incentive to produce documentation to prove that he had maintained his vehicle in accordance with the manufacturer’s instructions. In a recall action, however, the Administrator seeks to require a remedy for an entire class of vehicles; he may not be able to obtain precise documentation of the maintenance that was performed. His emphasis must be on systemic, class-wide defects in the design or construction of the emission controls, rather than on individual breaches of the warranty. Such systemic defects may be revealed in vehicles deviating somewhat from the precise specifications of the manufacturer. It is possible, therefore, that Congress intended the “properly maintained” requirement of Section 207(c)(1) to be more flexibly interpreted than the “maintained and operated in accordance with [the manufacturer’s written] instructions” requirement of Section 207(b)(2)(A). Second, Chrysler’s proposed interpretation fails to give full effect to the phrase “when in actual use” in' Section 207(c)(1). This language might be taken to indicate that Congress was concerned with everyday conditions, not with an ideal world in which the technical instructions of the manufacturer are obeyed to the letter. If—as the Administrator found—the design of Chrysler’s emission control system and maintenance procedures encouraged or fostered improper maintenance by ordinary owners and mechanics, we cannot say that Section 207(b)(2)(A) or any other section of the Act would preclude the Administrator from requiring a recall. We therefore must look to the legislative history of the Act and to the function and purpose of the “properly maintained” language in the statutory scheme to see whether the Administrator’s interpretation can be sustained. C. Legislative History Prior to the Clean Air Amendments of 1970, Pub.L. No. 91-604, 84 Stat. 1676, there was no system of in-use testing of vehicles. Thus Section 207 represented “a significant departure from prior provisions for control of vehicle emissions.” S.Rep. No. 1196, 91st Cong., 2d Sess. 29 (1970), reprinted in Library of Congress Congressional Research Service Environmental Policy Division, 93d Cong., 2d Sess., 1 A Legislative History of the Clean Air Amendments of 1970 at 429 (Committee Print 1974) (hereinafter cited as Leg.Hist.). Under the earlier scheme compliance with national emission standards was judged by tests of prototype models and samples of production-line vehicles. If a class of vehicles satisfied this test, then emission standards were presumed met. As data became available on the performance of these certified vehicles in actual use, however, federal officials found that emission control performance seriously deteriorated after sale. A study by the National Air Pollution Control Administration concluded that more than half of the previously-certified vehicles in actual use in the spring of 1968 failed to satisfy either the hydrocarbon or the carbon monoxide standard. One model registered an 80 percent failure rate. Faced with this problem, Congress resolved to require manufacturers to construct vehicles that would satisfy performance tests as well as design and production tests. Senator Muskie, principal spokesman for the bill, said, “So what we are concerned about is not only the tests or the standards that the cars meet while they are in the factory, but also whether or not they continue to meet these standards afterward.” 116 Cong.Rec. 33093 (1970), reprinted at Leg.Hist. 330. As we have already noted, the performance requirements were to be enforced in two ways: through warranty provisions and recall actions. Congress recognized, however, that some responsibility for deterioration of emission control equipment in actual use must be borne by the vehicle owners, whose negligence or abuse in maintenance and operation can frustrate or disable the emission controls. Thus the warranty and recall provisions of the Act included a requirement that owners maintain and operate their vehicles properly. As Senator Muskie explained: We understand that it is not presently possible to build maintenance-proof, clean car, but that it is possible—with the use of a system that is built with some durability in it and some responsibility imposed upon the operator—to assure reasonably clean operation of such an automobile. We have to have the two. One without the other is like a one-legged man. Id. at 33094-33095, Leg.Hist. at 332. To promote better maintenance Congress required manufacturers to provide vehicle purchasers with written instructions on proper maintenance. These instructions should be “reasonable and uncomplicated,” S.Rep. 1196, supra, at 30, Leg.Hist. at 430, so that they can be understood and followed by vehicle owners. In order to take advantage of the warranty protection owners must do their part to keep the emission system in good working order. Both the explicit language of the statute and the legislative history make clear that owners facing sanctions for emission control failures cannot shift the expense of remedying the nonconformity to the manufacturer unless they can prove that they maintained the system in accordance with the written instructions of the manufacturer. Chrysler contends that this clear language and legislative history with respect to the warranty provisions must apply with equal force to the recall provisions. However, the legislative history does not support this view. Rather, that history indicates that Congress realized that vehicle owners are often unable to find mechanics to service their vehicles competently, and that Congress therefore placed the burden on the auto manufacturers to design an emission control system that would effectively reduce auto emissions despite the poor performance of the maintenance industry. This is evident from a colloquy between Senators Allott and Muskie. Senator Allott complained that poor maintenance facilities represented a “weak spot” in the bill. He said that “to secure competent maintenance on a car at the present time in any respect is almost an impossibility.” 116 Cong.Rec. 33095, Leg. Hist. 332-333. Senator Muskie agreed with Senator Allott that the “greatest problem the industry faces is the shortage of mechanics across this country,” but stated that “[w]e cannot by legislation remake the automobile industry.” Id., Leg.Hist. 334. He attributed the maintenance problems to the manufacturers’ insistence on using the internal combustion engine and said, “We cannot solve the problem of whatever technology the industry chooses to put its bets on. All we can do is set the standards. The automobile industry has created all of the problems from the top to the bottom.” Id., Leg.Hist. 335. He concluded: So, if the bill is weak in not providing for the solution of the maintenance problem, I would welcome an amendment that would cure that weakness. But I do not think there is any way of writing a law that will create maintenance capability all across the country. Only the automobile industry can do that. Id. at 33096, Leg.Hist. at 335. If Congress thus believed that “only the automobile industry” could solve the maintenance problem, it was logical to pass an Act placing primary responsibility on the design capability of the manufacturers rather than on the maintenance capability of mechanics and owners. Thus, despite Senator Allott’s objections, Section 207 was left unchanged. The primary responsibility for emission control was left with the manufacturers; they were expected to solve the maintenance problem by designing a system that would be less susceptible to faulty maintenance. Admittedly, Congress did not intend liability to rest on the manufacturer if the condition of poor maintenance could be attributed to owners or mechanics. If, however, design defects of the manufacturer proved to be responsible for the condition of poor maintenance, the legislative history would indicate that Congress did indeed intend responsibility for remedy to rest on the manufacturers. D. Function of the “Properly Maintained” Criterion in Achieving the Purposes of the Act We cannot interpret Section 207 “in a manner which runs counter to the broad goals which Congress intended it to effectuate.” See FTC v. Fred Meyer, Inc., 390 U.S. 341, 349, 88 S.Ct. 904, 908, 19 L.Ed.2d 1222 (1968). The Administrator warned that Chrysler’s interpretation was in “evident disregard for the goals of the Clean Air Act.” JA 111:1213. The broad purpose of the Clean Air Amendments of 1970 is plain: “to protect and enhance the quality of the Nation’s air resources so as to promote the public health and welfare and the productive capacity of its population[.]” Section 101(b)(1) of the Act, 42 U.S.C. § 7401(b)(1) (Supp. II 1978). The purpose of the recall provisions seems equally plain: to give auto manufacturers an incentive to design and construct their emission control systems to meet standards in actual use for the estimated life of the vehicle. The theory underlying the provisions is that the manufacturers are better equipped than Congress or EPA to design a system that would meet federal emission standards. By setting standards, and leaving the auto manufacturers free to design and construct vehicles that would satisfy them, EPA can harness the forces of American industrial ingenuity to solving the serious national problem of vehicle-produced air pollution. From this perspective, it would make little sense to impose liability on the manufacturers for conditions of nonconformity caused by the actions of vehicle owners or mechanics. Unless the cause of the nonconformity is within the manufacturer’s control, an imposition of liability would be an unwarranted financial burden on the manufacturers, unrelated to the strategy of forcing technological progress. Thus the “properly maintained and used” criterion in Section 207(c)(1) is best understood as an ac-knowledgement of the limited ability of manufacturers to prevent those noncon-formities primarily caused by intentional or negligent faulty maintenance by owners and mechanics. Chrysler’s contention—that manufacturers have no responsibility for nonconformities if the owners failed to attain maintenance in accordance with the manufacturer’s precise written specifications—would far exceed the purpose of the “properly maintained” criterion and would undermine the broad objectives of the Act. Such an interpretation would strip the manufacturers of a large part of the incentive to design emission systems that would operate effectively while in actual use. The manufacturer could prescribe maintenance that is difficult to perform or incompatible with smooth operation of the vehicle, but bear no responsibility for the natural consequence of such design—so long as the vehicles passed the initial certification tests. We could expect little or no progress in development of better, more maintenance-free emission controls. The Administrator’s construction—m