Opinion for the Court filed by Circuit Justice THOMAS.
Concurring opinion filed by Circuit Judge BUCKLEY.
Dissenting opinion filed by Chief Judge MIKVA.
THOMAS, Circuit Justice:When Barbara Driscoll Marmet applied for permission to build a radio station, the Federal Communications Commission, pursuant to policy, awarded her extra credit for being a woman. Jerome Thomas Lam-precht contends that the Commission’s policy deprived him of his constitutional right to the equal protection of the laws. We agree.
I
A
The Communications Act of 1934, Pub.L. No. 73-416, 48 Stat. 1064 (codified as amended in scattered sections of 47 U.S.C.), empowers the Federal Communications Commission to grant construction permits and operation licenses for radio and television stations when “public convenience, interest, or necessity will be served thereby.” 47 U.S.C. §§ 307(a), 309(a); see also id. § 303. In 1965, the Commission first set out the general policy that it follows when it entertains mutually exclusive applications. Policy Statement on Comparative Broadcast Hearings, 1 F.C.C.2d 393 (1965), modified, 2 F.C.C.2d 667 (1966). Two principal goals guide the Commission in its choice. In furthering the first objective, “a maximum diffusion of control of the media of mass communications,” the Commission examines each applicant’s interests in other media properties, taking into account the significance of the other media properties and the extent of the applicant’s interests. See 1 F.C.C.2d, at 394-95. In furthering the second objective, “the best practicable service to the public,” the Commission awards what it calls “quantitative-integration credit,” a term of art that describes the degree to which prospective owners will participate in their stations’ day-to-day management. See id. *384The Commission then “enhances” the quantitative-integration credit based on “qualitative” factors, such as an owner’s character and the service an owner proposes to offer, as well as (to the extent applicable) an owner’s local residence, involvement in civic affairs, and experience and education in broadcasting. See id., at 396-99.
In 1972, the Commission’s Review Board held that it was barred by statute from giving applicants qualitative-enhancement credit for being members of particular racial or ethnic groups. See Mid-Florida Television Corp., 33 F.C.C.2d 1, 17 (Rev. Bd.), review denied, 37 F.C.C.2d 559 (1972). This court disagreed. See TV 9, Inc. v. FCC, 495 F.2d 929 (D.C.Cir.1973) (reversing Mid-Florida), cert. denied, 419 U.S. 986, 95 S.Ct. 245, 42 L.Ed.2d 194 (1974). A later case clarified that in this circuit’s view the public-interest mandate of the Communications Act in effect requires the Commission to award applicants credit for being minorities. See Garrett v. FCC, 513 F.2d 1056, 1063 (D.C.Cir.1975); see also West Michigan Broadcasting Co. v. FCC, 735 F.2d 601, 609-11 (D.C.Cir.1984), cert. denied, 470 U.S. 1027, 105 S.Ct. 1392, 84 L.Ed.2d 782 (1985). In 1978, the Commission reacted to TV 9 and Garrett by expressly adopting three programs: the awarding of tax certificates, the holding of distress sales, and the giving of preferences in the comparative-licensing process. See Statement of Policy on Minority Ownership of Broadcasting Facilities, 68 F.C.C.2d 979, 982-83 (1978); WPIX, Inc., 68 F.C.C.2d 381, 411-12 (1978); see also Reexamination of the Commission’s Comparative Licensing, Distress Sales and Tax Certificate Policies Premised on Racial, Ethnic or Gender Classifications, 1 F.C.C. Red. 1315, 1315 (1986) (notice of inquiry), modified, 2 F.C.C. Red. 2377 (1987). Each of the three programs was meant to benefit members of only certain minority groups, specifically people of “Black, Hispanic Surnamed, American Eskimo, Aleut, American Indian and Asiatic American extraction.” Statement of Policy on Minority Ownership, 68 F.C.C.2d, at 980 n. 8.
Women fared differently. In June 1978, the Review Board decided “[u]pon further reflection,” but without explanation, to give preferences to women in its comparative-licensing program. Gainesville Media, Inc., 70 F.C.C.2d 143, 149 (Rev.Bd.1978). Early the next month, the Board offered reasons in support of its policy:
We hold that merit for female ownership and participation is warranted upon essentially the same basis as the merit given for black ownership and participation, but that it is a merit of lesser significance. The basic policy considerations are the same. Women are a general population group which has suffered from a discriminatory attitude in various fields of activity, and one which, partly as a consequence, has certain separate needs and interests with respect to which the inclusion of women in broadcast ownership and operation can be of value. On the other hand, it is equally obvious that the need for diversity and sensitivity reflected in the structure of a broadcast station is not so pressing with respect to women as it is with respect to blacks— women have not been excluded from the mainstream of society as have black people.
Mid-Florida Television Corp., 70 F.C.C.2d 281, 326 (Rev.Bd.1978), set aside on other grounds, 87 F.C.C.2d 203 (1981). Later that year, however, the Commission decided that women who are not also minorities may not participate in the tax-certificate or distress-sale programs. See National Telecommunications & Information Admin., 69 F.C.C.2d 1591, 1593 n. 8 (1978) (petition for notice of inquiry) (stating that “we have not concluded that the historical and contemporary disadvantagement [sic] suffered by women is of the same order, or has the same contemporary consequences, which would justify inclusion of a majority of the nation’s population in a preferential category defined by the presence of ‘minority groups’ ”); see also Wuenschel Broadcasting Co., 74 F.C.C.2d 389 (1979) (refusing to include women in a program that expedites the processing of minorities’ applications).
*385In 1983, James U. Steele was denied a construction permit in a case in which the sex of a competing applicant proved to be “decisively important.” Cannon’s Point Broadcasting Co., 93 F.C.C.2d 643, 656-57 (Rev.Bd.1983), review denied, No. 86-161 (Comm’n Apr. 13, 1984). Steele then challenged the Commission’s sex-conscious policy in this court. In Steele v. FCC, we struck the policy down. 770 F.2d 1192 (D.C.Cir.1985) (reversing Cannon’s Point). Noting first that the Review Board’s reasoning had been unclear, Judge Tamm, joined by Judge Scalia, accepted the Commission’s assertion that it sought to increase the quantity of women’s viewpoints on the air. Comparing the Commission’s ethnic-preference policy, the court then asked whether a station owner with ancestors from Italy, for example, “would primarily program Italian operas or would eschew Wagner in favor of Verdi,” id., at 1198, an assumption based in turn on two other presumed truths: that a station owner’s heritage will determine the owner’s interests, and that a station owner will indulge his or her own tastes and ignore the tastes of the members of the relevant programming audience.
Whatever the merit of these assumptions as applied to cohesive ethnic cultures, it simply is not reasonable to expect that granting preferences to women will increase programming diversity. Women transcend ethnic, religious, and other cultural barriers. In their social and political opinions and beliefs, for example, women in fact appear to be just as divided among themselves as are men. Therefore it is not reasonable to expect that a woman would manifest a distinctly “female” viewpoint.
Id., at 1199. We concluded that the policy violated the Communications Act. “Presumably, the Board thought that [its policy] was a Good Idea and would lead to a Better World. [But] a mandate to serve the public interest is not a license to conduct experiments in social engineering conceived seemingly by whim and rationalized by conclusory dicta.” Id.
A majority of the active judges in the circuit then voted to rehear the case en banc and vacated the panel’s opinion and judgment. Steele v. FCC, No. 84-1176 (D.C.Cir. Oct. 31, 1985) (en banc). After the court instructed the parties to file supplemental briefs, the Commission responded by admitting that it had assumed, with no factual support, a causal link between its preference schemes and increased diversity of viewpoints. See Brief for the Federal Communications Commission at 17-30, Steele v. FCC (D.C.Cir.) (No. 84-1176) (en banc). The Commission acknowledged that it thought its race- and sex-preference policies contrary to both the Communications Act and the Constitution, and it asked us to remand the Steele case for reconsideration. We granted the motion, and the Commission proceeded to call for comments on the wisdom and effectiveness of its policies. See Reexamination of the Commission’s Comparative Licensing, Distress Sales and Tax Certificate Policies Premised on Racial, Ethnic or Gender Classifications, 1 F.C.C. Red. 1315 (1986). (notice of inquiry), modified, 2 F.C.C. Rcd. 2377 (1987).
Soon after the Commission began to try to make a record, however, Congress ordered it to freeze. In a rider to the Continuing Appropriations Act for Fiscal Year 1988, Congress instructed that “none of the funds appropriated by this Act shall be used to repeal, to retroactively apply changes in, or to continue a reexamination of, the policies of the ... Commission with respect to comparative licensing, distress sales and tax certificates ... to expand minority and women ownership of broadcasting licenses ... other than to close [the pending reexamination] with a réinstatement of prior policy.” Pub.L. No. 100-202, 101 Stat. 1329,1329-1331 (1987). Congress has passed identical riders in each year since. See Pub.L. No. 101-515, 104 Stat. 2101, 2136-2137 (1990); Pub.L. No. 101-162, 103 Stat. 988, 1020-1021 (1989); Pub.L. No. 100-459, 102 Stat. 2186, 2216-2217 (1988). Obeying Congress’s order, the Commission continues to apply its preference policies. See, e.g., Cannon’s Point Broadcasting Co., 3 F.C.C. Rcd. 864 (1988) (reaffirming the original decision, including the sex preference, after the remand in *386Steele)-, see also James U. Steele, 4 F.C.C. Rcd. 4700 (1989), reaff’d, 5 F.C.C. Rcd. 4121 (1990).
Disappointed applicants meanwhile continued to challenge the Commission’s policies on both constitutional and statutory grounds. In Shurberg Broadcasting, Inc. v. FCC, 876 F.2d 902 (D.C.Cir.1989), a divided panel of this court struck down as unconstitutional the Commission’s distress-sale program, and in Winter Park Communications, Inc. v. FCC, 873 F.2d 347 (D.C.Cir.1989), a divided panel upheld on statutory and constitutional grounds the Commission’s comparative-licensing program for racial and ethnic minorities. In Metro Broadcasting, Inc. v. FCC, 497 U.S. 547, 110 S.Ct. 2997, 111 L.Ed.2d 445 (1990), the Supreme Court reversed Shurberg and affirmed Winter Park, thus upholding two of the Commission’s methods for preferring applicants on the basis of race, ethnicity, or surname. The Metro Broadcasting Court expressly refused to pass judgment on the Commission’s policy of preferring applicants on the basis of sex. See 110 S.Ct., at 3005 n. 7 (“[T]he Commission’s gender preference policy is not before us today.”).1 We had previously held this case in abeyance, and we reopened proceedings after the Supreme Court’s decision.
B
This case began in 1982, when Jerome Thomas Lamprecht, Barbara Driscoll Mar-met, Dragon Communications, Inc., and Port Royal Broadcasting, Inc. filed mutually exclusive applications to build a radio station that would broadcast on channel 276A (103.1 MHz), out of Middletown, Maryland. Discovery ensued and ended early in 1984. The record reveals the following facts.
Jerome Thomas Lamprecht, of Baltimore, Maryland, was twenty-seven when he applied for the construction permit. Lamprecht attended the University of Maryland full-time from 1974-1977 and part-time from 1977-1984. He graduated in *3871984 with a bachelor of science degree in radio, television, and film studies. After leaving full-time studies at the university in 1977, Lamprecht started working in broadcasting. He began his career as an announcer at a radio station in Towson, Maryland. After moving up to program director, Lamprecht soon became operations manager, then station manager, and eventually general manager and vice president. Lamprecht later became general manager and part owner of an AM station in Farm-ville, North Carolina. He recently sold the Farmville station; if awarded the construction permit for the station in Middletown, Lamprecht plans to move there immediately with his family.
Barbara Driscoll Marmet, who lived part of the time in Frederick County, Maryland, part of the time in Bethesda, Maryland, and part of the time on Amelia Island, Florida, was fifty-eight when she applied for the construction permit. In Frederick County, Marmet attended the Frederick County Symphony and workshops held by Community Commons, an organization of people who,.in Marmet’s words, are “worried about the population increase that was coming both into the county and into the city.” She received a plaque from the Frederick County Landmarks Foundation for restoring her house there. In Washington, Marmet served on a parents’ committee at the Saint Albans School for boys, as a sustaining member of the Washington Junior League, as a member of the Chevy Chase Country Club, and as fiscal officer of her garden club. Currently part-owner of Ogden & Marmet, which manages several of the real-estate properties that she controls, Marmet worked part-time from 1970-1975 as a bookkeeper for Thomas Jefferson, Inc., another Washington real-estate firm of which she was once part-owner. From 1967-1970 Marmet worked part-time as a bookkeeper for her husband, Robert A. Marmet, a communications lawyer now in partnership in Washington with their son-in-law, Harold K. McCombs, Jr., who has represented Barbara Marmet in this proceeding. Although Marmet had never worked in radio or television, she had twice before owned pieces of broadcast properties. In 1969, she bought an interest in Viking Television, one of her husband’s media clients, which was granted a permit (after Marmet sold her interest) to build a UHF television station in Minneapolis. In the mid-1970s, Marmet’s husband gave her as a gift his ownership interest in Montgomery Communications, Inc., a company that had applied for a permit to build an AM station in Gaithersburg, Maryland. Marmet’s relatives have also been involved in broadcasting. Her husband applied for a permit to build an FM station in Stamford, Connecticut, and her son-in-law applied at the same time for a permit to build an AM station there. Marmet’s son-in-law more recently applied to build an AM station in Claremont, Virginia.
Dragon Communications, Inc. was owned by Thomas Buffington of Washington, D.C.; Gary Hess of Tarpon Springs, Florida; George Christopher, Jr. of Bethesda, Maryland; and Christine DeWitt of Silver Spring, Maryland, as joint tenant with her son, H.M. Lee DeWitt. Lee DeWitt, who came up with the idea for Dragon, recruited the others, including his mother and Buffington, and remains the company’s moving force. He paid for his mother’s ownership interest in the firm. Christine DeWitt had worked as a secretary for Robert Thompson, Dragon’s lawyer, who was then at the law firm of Steptoe & Johnson. Hess lived and worked in Florida and owned quarter interests in three radio stations there. Buffington owns Thomas Buffington & Associates, an advertising and media consulting firm in Washington. He paid nothing for his interest in Dragon.
Port Royal Broadcasting, Inc. was owned by Richard Raymond Leverrier of Bethesda, Maryland; Mary Ellen Ferrel of Chevy Chase, Maryland; and Nancy Spruill South-mayd of Rockville, Maryland. Leverrier had previously worked at a television station, and his wife had filed an application competing with Dragon for a radio station in Ocean View, Delaware. Leverrier’s wife (represented by Southmayd’s husband) later was granted a construction permit for an FM station in Cfisfield, Maryland. Lev-errier’s wife also applied to build an FM *388station in Strasburg, Virginia, and the Lev-erriers applied together for permits to build FM stations in Long Beach, Washington, and Albany, Minnesota. Ferrel, who had once worked as an office assistant for a Washington architect, replaced her husband, at the time Leverrier’s boss, as part owner of Port Royal shortly before applications were due. Southmayd, once a secretary and later a real-estate agent, had also done some bookkeeping work for her husband, a communications lawyer who represented Port Royal in this proceeding. Like the Marmets and the Leverriers, the South-mayds are not unfamiliar with broadcast properties. Jeffrey Southmayd applied in his own right to build an FM station in New Market, Virginia, and Nancy South-mayd, represented by her husband, applied for permits to build an AM station in Virginia City, Nevada, and an FM station in Snowmass, Colorado.
When discovery ended, the case came before an administrative law judge for resolution. Jerome Thomas Lamprecht, 99 F.C.C.2d 1229 (A.L.J.1984). The ALJ began by considering which of the applicants would further the diffusion of control over the means of mass communications. Lam-precht owned fewer than 1% (between $450 and $600 worth) of the shares in the company that owned the station where he worked, an amount considered irrelevant under Commission precedent. Marmet testified that she herself owned no significant interests in other media, aside from fewer than 1% of the shares of Cowles Communications (one of her husband’s clients). After noting that Marmet’s husband and son-in-law are both communications lawyers, the ALJ found, based at least in part on Marmet’s demeanor, that Marmet would spend her own money on the station and that she had no implicit understanding with her husband or son-in-law that would dilute her ownership. The AU also found that neither Port Royal nor its stockholders owned any parts of other media properties at that time. The AU then examined the media holdings of Hess and Buffington and gave Dragon “a healthy comparative demerit.”
Turning to the second of the Commission’s goals, service to the public, the AU reviewed the applicants’ proposals for integrating ownership and management. He gave Lamprecht and Marmet credit for 100% quantitative integration. He also gave Port Royal full credit, though he viewed skeptically Southmayd’s promise to commute from Rockville to Middletown and work forty or more hours a week for a mere 4% ownership interest in the station. Although Dragon claimed that it was entitled to 88% quantitative-integration credit, the AU awarded it only 76%, since Dragon had asked for 76% credit in its earlier submission, and since in any event there was no proof in the record that Christopher or either of the DeWitts would actually work if awarded the permit. Because Dragon was at so grave a comparative disadvantage, the AU excluded it from further consideration.
Turning to the qualitative factors, the AU enhanced Marmet’s score based on her having patronized the arts in Frederick County, her owning a home in Frederick County, and her sex. The AU enhanced Lamprecht’s score because of Lamprecht’s education in broadcasting, and his prior broadcasting experience. The AU denied Port Royal any enhancement for local residence or civic participation, since none of Port Royal’s principals lived or planned to live in or near Middletown, but he gave Port Royal a 52% enhancement because Ferrel and Southmayd are women. Finally, the AU gave Marmet and Port Royal credit for proposing auxiliary power in case of power failure. He concluded by awarding the permit to Marmet. Lamprecht finished second and Port Royal third.
Lamprecht, Dragon, and Port Royal appealed, focusing mainly on the AU’s decision to give Marmet enhancement credit for local residence and civic participation. The Review Board affirmed. Jerome Thomas Lamprecht, 99 F.C.C.2d 1229 (Rev.Bd.1984). The Board first held that Dragon should have been given only a slight diversification demerit, for Hess’s interest in one station in Florida but not for Buffington’s consulting company. It then held that because Dragon had not sub*389mitted its second integration proposal on time, the AU had properly awarded Dragon only 76% quantitative-integration credit and had appropriately eliminated Dragon from further consideration. The Board endorsed the AU’s decisions to award Mar-met 100% integration credit and to enhance her score for local residence, agreeing that Marmet had at least dual residence and that she had shown a sufficient, if less than consuming, interest in the affairs of Frederick County. The Board endorsed the AU’s decision to award Marmet credit for being a woman, and it agreed that Port Royal was entitled to less credit for having only 52% of its stock owned by women. The Board concluded by affirming the award of the permit to Marmet and holding that Port Royal should have finished second and Lamprecht third.
Port Royal decided to bow out. Lam-precht and Dragon petitioned for review of the Board’s order, and the Comission denied the requests. Jerome Thomas Lamprecht, 3 F.C.C. Rcd. 2527 (1988). Lam-precht then appealed to this court, and Dragon and Marmet intervened.
II
In his appeal, Lamprecht argues that the Commission’s sex-preference policy violates both the Communications Act and the equal-protection component of the Fifth Amendment. He also argues that the congressional rider that restrains the Commission from rethinking the wisdom of its policy violates the constitutional doctrine of separation of powers. The Commission vigorously defends its policy, and Marmet has joined in the Commission’s defense. Dragon, which filed its brief in this court the day before oral argument — twenty-nine days after it was due — contends that the AU, the Review Board, and the Commission mistakenly held Dragon to the original statement concerning its integration proposal, rather than accepting the statement that Dragon had filed after the deadline for submissions had passed. Dragon also argues that Metro Broadcasting, Inc. v. FCC, 497 U.S. 547, 110 S.Ct. 2997, 111 L.Ed.2d 445 (1990), obliges the Commission in comparative-licensing proceedings automatically to prefer minorities to women.
Before we address the merits of Lamprecht’s arguments, we explain briefly why we do not consider Dragon’s. Dragon, like Lamprecht, was denied the permit that went to Marmet. Lamprecht exercised his statutory right to appeal. See 47 U.S.C. § 402(b)(1). Dragon, in contrast, decided for reasons not revealed to intervene in Lamprecht’s case rather than appeal on its own. See id. § 402(e). Notwithstanding this unexplained (and unapologetic) lapse, in its brief in this court Dragon has raised arguments distinct from Lamprecht’s statutory and constitutional challenges to the Commission’s sex-preference policy. Except in extraordinary cases, however, see Synovus Fin. Corp. v. Board of Governors, 952 F.2d 426, 431-34 (D.C.Cir.1991), intervenors “may only join issue on a matter that has been brought before the court by another party,” Illinois Bell Tel. v. FCC, 911 F.2d 776, 786 (D.C.Cir.1990). They cannot expand the proceedings. See id. (noting that “[otherwise, the time limitations for filing a petition for review and a brief on the merits could easily be circumvented”); cf. Vinson v. Washington Gas Light Co., 321 U.S. 489, 498, 64 S.Ct. 731, 735, 88 L.Ed. 883 (1944) (“[A]n intervenor is admitted to the proceedings as it stands, and in respect of the pending issues, but is not permitted to enlarge those issues or compel an alteration of the nature of the proceeding.”). It would be grossly imprudent for us to address Dragon’s separate contentions, statutory, but especially constitutional, when Dragon, despite having had every incentive to raise its arguments in the proper fashion, not only failed to do so, but fails now to proffer an excuse. We therefore pass over Dragon’s contentions and proceed directly to Lamprecht’s.
A
When a federal court is asked to answer a constitutional question, basic tenets of judicial restraint and separation of powers call upon it first to consider alternative grounds for resolution. See, e.g., Ash*390wander v. TVA, 297 U.S. 288, 347, 56 S.Ct. 466, 483, 80 L.Ed. 688 (1936) (Brandéis, J., concurring). We therefore begin our discussion by determining whether the Commission has exceeded its statutory authority-
The Communications Act authorizes the Commission to grant construction permits for radio stations if “public convenience, interest, or necessity will be served thereby.” 47 U.S.C. §§ 307(a), 309(a). Relying on TV 9 and West Michigan Broadcasting, the Commission argues that its policy constitutes “a vital part of the FCC’s public interest mandate,” and is therefore consistent with the statute. Relying on Steele, Lamprecht argues that the policy violates the Communications Act if it violates the Constitution, since a Commission policy that is unconstitutional would inherently conflict with the public interest.
We have no occasion to decide whether the Communications Act bars the Commission from preferring women in comparative licensing. In the years since Steele, Congress by law has required a preference for women. See Pub.L. No. 101-515, 104 Stat. 2101, 2136-2137 (1990); Pub.L. No. 101-162, 103 Stat. 988, 1020-1021 (1989); Pub.L. No. 100-459, 102 Stat. 2186, 2216-2217 (1988); Pub.L. No. 100-202,101 Stat. 1329,1329-1331 (1987). That the laws at issue are appropriations riders does not change their status as the law. See, e.g., Action for Children’s Television v. FCC, 932 F.2d 1504, 1507-10 (D.C.Cir.1991). The Commission, in short, has not exceeded its authority — it has bowed to the will of Congress. Because we cannot decide this case on alternative grounds, we turn to the Constitution.
B
In Metro Broadcasting, Inc. v. FCC, 497 U.S. 547, 110 S.Ct. 2997, 111 L.Ed.2d 445 (1990), the Supreme Court considered equal-protection challenges to two of the programs through which the Commission favors certain minorities. Together with the Supreme Court’s cases involving classifications by sex, Metro Broadcasting compels us to strike down the program through which the Commission favors women. We do not reach Lamprecht’s argument that Congress’s appropriations rider violates the separation of powers.
(1)
Metro Broadcasting concerned two programs through which the Commission chooses the owner of a broadcast station based in part on an applicant’s race, ethnicity, or surname. In sustaining the programs against equal-protection attack, the Supreme Court applied an intermediate standard of scrutiny, holding that “benign race-conscious measures mandated by Congress ... are constitutionally permissible to the extent that they serve important governmental objectives within the power of Congress and are substantially related to achievement of those objectives.” 110 S.Ct., at 3008-09 (footnote omitted). The Court thus considered first the stated aim of the government’s programs — “enhancing broadcast diversity” — and held that this interest qualifies as a sufficiently “important” government goal. Id., at 3010.
The Court then examined the Commission’s programs to determine whether they were “substantially related” to achieving diversity of viewpoints. In its inquiry, the Court relied heavily on both the expertise of the Commission, see, e.g., id., at 3011, and the factfinding of Congress. For proof of congressional factfinding, the Court focused closely on the debate over the Commission’s statutorily authorized lottery scheme, which, in order “[t]o further diversify the ownership of the media of mass communications,” grants “an additional significant preference” to “Blacks, Hispanics, American Indians, Alaska Natives, Asians and Pacific Islanders,” 47 U.S.C. § 309(i)(3)(A), (i)(3)(C)(ii) (though not to women, see Amendment of the Commission’s Rules, 102 F.C.C.2d 1401 (1985)). See 110 S.Ct., at 3012-16 & nn. 17-25. The Court found especially significant a report prepared by the Congressional Research Service entitled Minority Broadcast Station Ownership and Broadcast Programming: Is there a Nexus? (June 29, 1988) [hereinafter Minority Programming ].
*391The Court found a “host of empirical evidence” supporting the judgment that there is, in fact, a link between the ownership of stations by members of minority groups and the stations’ programming practices. Id.; see id., at 3017 nn. 31-33 (citing eight studies). Examining Minority Programming, for example, the Court noted that “[wjhile only 20 percent of stations with no Afro-American ownership responded that they attempted to direct programming at Afro-American audiences, 65 percent of stations with Afro-American ownership reported that they did so.” Id., at 3017 n. 31 (citing Minority Programming at 13). The Court explained that “[ejvi-dence suggests that an owner’s minority status influences the selection of topics for news coverage and the presentation of editorial viewpoint.” Id., at 3017. Other evidence suggested that “Afro-Americán-owned, Afro-American-oriented radio stations have more diverse playlists than white-owned, Afro-American-oriented stations.” Id., at 3017 n. 31. In light of all this evidence, the Court held,. Congress’s “predictive judgment” — that “there may be important differences between the broadcasting practices of minority owners and their non-minority counterparts” — was sufficiently well-grounded in fact so as not to be a constitutionally “impermissible stereotype.” See id., at 3016-17.
(2)
Metro Broadcasting, together with the Supreme Court’s sex-discrimination cases, has ended debate on several matters. The first is our standard of scrutiny. The Metro Broadcasting Court held that “benign race-conscious measures mandated by Congress” do not violate the Fifth Amendment if “they serve important governmental objectives within the power of Congress and are substantially related to achievement of those objectives.” 110 S.Ct., at 3008-09 (footnote omitted). The Supreme Court previously has used the same standard with respect to all sex-based classifications, whether enacted by Congress or by a state legislature. See, e.g., Mississippi Univ. for Women v. Hogan, 458 U.S. 718, 724, 102 S.Ct. 3331, 3336, 73 L.Ed.2d 1090 (1982) (“[Tjhe party seeking to uphold a statute that classifies individuals on the basis of their gender must carry the burden of showing an ‘exceedingly persuasive justification’ for the classification. The burden is met only by showing at least that the classification serves ‘important governmental objectives and that the discriminatory means employed’ are ‘substantially related to achievement of those objectives.’ ” (citations omitted)) (reviewing a state statute); Califano v. Westcott, 443 U.S. 76, 85, 99 S.Ct. 2655, 2661, 61 L.Ed.2d 382 (1979) (reviewing a congressional statute under the same standard). We see no reason after Metro Broadcasting to examine sex-conscious measures mandated by Congress any more or less strictly than before, and no one seriously contends that we should. Metro Broadcasting also establishes that the promotion of diversity of viewpoints in general qualifies as an “important” objective within the government’s power. The sole question remaining, therefore, is whether the government’s methods are “substantially related” to the goal that it hopes to achieve.
Metro Broadcasting confirms that although we are “to give ‘great weight to the decisions of Congress and to the experience of the Commission,’ ” 110 S.Ct., at 3011 (citation omitted), we are still obliged in the end to review the government’s policy — both the judgment of law that the policy is constitutional and the findings of fact that underlie it. The Court explained: “[W]e do not ‘ “defer” to the judgment of the Congress and the Commission on a constitutional question,’ and would not ‘hesitate to invoke the Constitution should we determine that the Commission has not fulfilled its task with appropriate sensitivity’ to equal protection principles.”. Id., at 3011 (quoting Columbia Broadcasting Sys. v. Democratic Nat’l Comm., 412 U.S. 94, 103, 93 S.Ct. 2080, 2087, 36 L.Ed.2d 772 (1973)); cf. Action for Children’s Television, 932 F.2d, at 1509 (Mikva, C.J.) (“While ‘we do not ignore’ Congress’ apparent belief that a total ban on broadcast indecency is constitutional, it is ultimately the judiciary’s task ... to decide whether *392Congress has violated the Constitution.” (citation omitted)). In Justice Brandeis’s formulation, “where a statute is valid only in case certain conditions exist, the enactment of the statute cannot alone establish the facts which are essential to its validity.” Whitney v. California, 274 U.S. 357, 374, 47 S.Ct. 641, 648, 71 L.Ed. 1095 (1927) (concurring opinion).2 We examine, then, the relationship between achieving diversity on the airwaves and the Commission’s policy of preferring women owners, and in doing so we bear in mind Justice O’Con-nor’s warning for the Court in Mississippi University for Women: “Although the test for determining the validity of a gender-based classification is straightforward, it must be applied free of fixed notions concerning the roles and abilities of males and females.” 458 U.S., at 724-25, 102 S.Ct. at 3336.
In applying that test, the Supreme Court has repeatedly denounced “unsupported generalizations about the relative interests and perspectives of men and women,” Roberts v. United States Jaycees, 468 U.S. 609, 628, 104 S.Ct. 3244, 3255, 82 L.Ed.2d 462 (1984), regardless of which sex the generalizations purport to favor. See, e.g., Mississippi Univ. for Women, 458 U.S., at 723-31, 102 S.Ct. at 3335-40 (striking down a women-only admissions policy at a state nursing school; condemning the stereotype that nursing is a woman’s profession); Caban v. Mohammed, 441 U.S. 380, 99 S.Ct. 1760, 60 L.Ed.2d 297 (1979) (striking down a statute that permitted mothers, but not fathers, to prevent the adoption of their children born out of wedlock; condemning the stereotype that a biological mother always bears a more intimate relationship with a child than does a father); Craig v. Boren, 429 U.S. 190, 97 S.Ct. 451, 50 L.Ed.2d 397 (1976) (striking down a statute that permitted women to buy low-alcohol beer at a younger age than men; condemning the stereotype that women mature earlier than men and are more responsible at an earlier age); Weinberger v. Wiesenfeld, 420 U.S. 636, 95 S.Ct. 1225, 43 L.Ed.2d 514 (1975) (striking down a congressional statute that allowed widows, but not widowers, to collect Social Security benefits without proving dependency; condemning the stereotype that men rather than women are the primary supporters of spouse and family). As the Court has recognized time and again, “[Discrimination based on archaic and overbroad assumptions about the relative needs and capacities of the sexes forces individuals to labor under stereotypical notions that often bear no relationship to their actual abilities.” Jaycees, 468 U.S., at 625, 104 S.Ct. at 3253; see also Orr v. Orr, 440 U.S. 268, 283, 99 S.Ct. 1102, 1114, 59 L.Ed.2d 306 (1978) (“Legislative classifications which distribute benefits and burdens on the basis of gender carry the inherent risk of reinforcing sexual stereotypes about the ‘proper place’ of women and their need for special protection.”); Frontiero v. Richardson, 411 U.S. 677, 686-87, 93 S.Ct. 1764, 1770, 36 L.Ed.2d 583 (1973) (plurality opinion) (“[Sjtatutory distinctions between the sexes often have the effect of invidiously relegating the entire class of females to inferior legal status without regard to the actual capabilities of its individual members.”).
*393Applying its standard of intermediate scrutiny, the Court has therefore required not only that sex-based generalizations be “supported,” but that the support be strong enough to advance “substantially” the legitimating governmental interest. Craig v. Boren, for example, concerned an Oklahoma statute that allowed women, but not men, to buy low-alcohol beer at age 18. The state showed (among other evidence) that 2% of 18- to 20-year old men, but only .18% of women that age, were arrested for driving under the influence. Rejecting the correlation between maleness and drunk driving as “unduly tenuous [a] ‘fit,’ ” 429 U.S., at 202, 97 S.Ct. at 459, the Court also noted that “[w]hile [the] disparity [between arrest rates for men and women] is not trivial in the statistical sense, it hardly can form the basis for employment of a gender line as a classifying device,” id., at 201, 97 S.Ct. at 459. Justice Brennan concluded: “In sum, the principles embodied in the Equal Protection Clause are not to be rendered inapplicable by statistically measured but loose-fitting generalities concerning the drinking tendencies of aggregate groups.” Id., at 208-09, 97 S.Ct. at 463.
Similarly, in Weinberger v. Wiesenfeld, the Court reviewed the constitutionality of an act of Congress that gave to widows Social Security benefits not available to widowers. Congress had proceeded on the presumption that women are less likely than men to be the family breadwinners, and the Court acknowledged that Congress’s presumption was “not entirely without empirical support.” Nevertheless, as Justice Brennan explained for the Court, “such a gender-based generalization cannot suffice to justify the denigration of the efforts of women who do work and whose earnings contribute significantly to their families’ support.” 420 U.S., at 645, 95 S.Ct. at 1232; see also Califano v. Westcott, 443 U.S., at 88, 99 S.Ct. at 2662 (striking down a part of the Aid to Families with Dependent Children program because there was “no evidence, in the legislative history or elsewhere,” supporting Congress’s assumption that fathers are likelier than mothers to desert their families, and thus concluding that “it does not appear that the [sex-based] classification is substantially related to the achievement of [Congress’s] goal”); Califano v. Goldfarb, 430 U.S. 199, 97 S.Ct. 1021, 51 L.Ed.2d 270 (1977) (plurality opinion) (relying on Wiesenfeld to strike down another section of the Social Security Act, and stating that Congress had “not verified” the assumption that men are less likely to depend on women for support than vice versa); cf. Arizona Governing Comm. v. Norris, 463 U.S. 1073, 103 S.Ct. 3492, 77 L.Ed.2d 1236 (1982) (holding that title VII of the Civil Rights Act of 1964 prohibits differences in the amount of pension benefits paid to men and women, even though sex-based actuarial tables show that women in the aggregate live longer than men).
Metro Broadcasting thus reinforces the lessons of cases such as Craig and Wiesenfeld: Any “predictive judgments” concerning group behavior and the differences in behavior among different groups must at the very least be sustained by meaningful evidence.3 In concluding in *394Metro Broadcasting that the government’s race-based predictions were “a product of ‘analysis’ rather than a ‘stereotyped reaction’ based on ‘[hjabit,’ ” the Supreme *395Court examined a “host” of studies supporting the government’s judgment that diversity of viewpoints increases substantially when station owners are selected in part on the basis of race or ethnicity. See 110 S.Ct., at 3018 (citation omitted and internal quotation marks amended). We now examine, in light of these principles, the evidence supporting the government’s judgment that diversity increases substantially when owners are selected in part on the basis of sex.
(3)
Implicit in the government’s judgment are at least three assumptions: first, that there exists such a thing as “women’s programming” (or “Alaskan programming,” say, or “suburban teenage easy listener programming”); second, that these distinct types of programming are underrepresented on the airwaves; and third, that women who own radio or television stations are likelier than white men to broadcast these distinct types of programming. Lamprecht challenges each of these assumptions.
We decline to address Lamprecht’s first argument. The Supreme Court assumed in Metro Broadcasting that there are such things as “minority programming” of different kinds, and we assume the same with respect to “women’s programming” and other distinct programming types. We also decline to address Lamprecht’s second argument. Although the Commission might risk raising other constitutional questions if it tried to set out some “correct” mix of information or viewpoints, we follow the Supreme Court’s lead and assume that all the types of programming that women might put on the air are underrepresented. We assess, then, the last of the assumptions implicit in the Commission’s sex-conscious policy. But having considered the evidence offered to demonstrate a link between ownership by women and any type of underrepresented programming, we are left unconvinced.4
The Commission’s brief cites nothing that might support its predictive judgment that women owners will broadcast women’s or minority or any other underrepresented type of programming at any different rate than will men. Nor is there any proof in the administrative record, a point that the Commission’s lawyer confirmed repeatedly at oral argument.5 We consider then the *396only study, so far as we know,6 either inside or outside the legislative record, that actually did examine the possibility: the Congressional Research Service report, Minority Broadcast Station Ownership and Broadcast Programming: Is There a Nex-us7
Whatever the study’s methodological flaws,8 it does answer its own question, at least with respect to women. The answer it gives is “no.” Consider the following facts, which we summarize in text and show in tables in the appendix.
*397* Of stations owned primarily by women, slightly more than a third, or 35%, broadcast what the study calls “women’s programming.” Of stations owned entirely by “non-women,” a slightly lower portion, or 28%, broadcast women’s programming. For others given preferences in comparative licensing, the difference is dramatic. Of stations owned primarily by Blacks, 79% broadcast Black programming, while of stations owned entirely by people who are not Black, 20% broadcast Black programming; of stations owned primarily by Hispanics, nearly three-quarters, or 74%, broadcast Hispanic programming, while of stations owned entirely by people who are not Hispanic, 10% broadcast Hispanic programming; of stations owned primarily by Asians or Pacific Islanders, 25% broadcast Asian or Pacific Islander programming, while of stations owned entirely by people who are not Asians or Pacific Islanders, 3% broadcast Asian or Pacific Islander programming; of stations owned primarily by Indians or Alaskans, 46% broadcast Indian or Alaskan programming, while of stations owned entirely by people who are not Indian or Alaskan, 4% broadcast Indian or Alaskan programming. To put it another way: Stations owned primarily by Indians or Alaskans are more than eleven times as likely to broadcast Indian or Alaskan programming as are stations with no Indian or Alaskan owners. For Asians or Pacific Islanders, the comparable multiplier is more than eight; for Hispanics, more than seven; for Blacks, almost four. In contrast, stations owned primarily by women are just one and one quarter times as likely to broadcast women’s programming as are stations owned entirely by men. See app. table 1, at 35.
* Stations in which women own anywhere from 1% to 50% of the equity are just as likely to broadcast women’s programming as are stations owned principally or entirely by women. In contrast, stations in which Indians or Alaskans own from 1% to 50% are only half as likely to broadcast Indian or Alaskan programming as are stations owned principally by Indians or Alaskans. See app. table 2, at 36.
In five large cities, New York, Los Angeles, Chicago, Dallas, and Atlanta, stations with any owners of Black, American Indian, Alaskan, Hispanic, Asian, or Pacific Islander heritage are likelier to broadcast women’s programming than are stations with any owners who are women: Only one-third of stations with any women owners broadcast women’s programming, while almost three-fifths of stations with Hispanic owners, half of stations with Asian, Pacific Islander, Indian, or Alaskan owners, and more than two-fifths of stations with Black owners, broadcast women’s programming. In all cities combined, barely more than one-third of stations with women owners broadcast women’s programming, while half of stations with Indian or Alaskan owners, and more than 40% of stations with Hispanic or Black owners, do so. For stations in which the relevant group holds between one and fifty percent ownership interests, women are less likely to broadcast women’s programming than are Indians, Alaskans, Hispanics, Blacks, Asians, and Pacific Islanders. See app. table 3, at 37; app. table 4, at 37; app. table 5, at 38;
* Of the ten most-used programming formats, nine are as popular in almost precisely the same order for stations owned by “non-minorities" (again, that could include both women and men) as they are for stations owned by women. See app. table 10, at 40.
* Nor are stations owned by women much likelier to engage in minority programming than are stations owned by men, as Minority Programming reveals. Stations with women owners are barely, if at all, likelier to broadcast assorted types of minority programming than are stations owned by anyone else, and are in fact much less likely to broadcast minority programming than members of most of the relevant minority groups. Of stations owned in any part by women, just 5% broadcast Asian or Pacific Islander programming, for example, while of stations owned by “non-Asian/Pacific Islanders” (which may or may not include women) only 3% broadcast Asian or Pacific Islander programming. Stations with Hispanic owners, in contrast, are three times as likely to broadcast Asian or *398Pacific Islander programming as are stations with any women owners, and stations with Black owners are more than twice as likely to broadcast Asian or Pacific Islander programming as are stations with any owners who are women. The data show similar results for Hispanic programming: stations with Hispanic owners are nearly five times as likely, and stations with Asian or Pacific Islander owners more than twice as likely, to broadcast Hispanic programming as are stations owned in any part by women. Of stations with women owners, only 5% broadcast Indian or Alaskan programming, barely a higher proportion than the 4% of stations owned entirely by “non-Indian/Alaskans” that do so. For Black programming the comparable percentages are 26% for stations with women owners and 20% for stations owned by non-Blacks. See app. table 6, at 38; app. table 7, at 39; app. table 8, at 39; app. table 9, at 40.
To summarize: The data in Minority Programming fail to establish any statistically meaningful link between ownership by women and programming of any particular kind. The study, in short, highlights the hazards associated with government endeavors like this one. As Justice Brennan has written for the Court, “[P]roving broad sociological propositions by statistics is a dubious business, and one that inevitably is in tension with the normative philosophy that underlies the Equal Protection Clause.” Craig v. Boren, 429 U.S., at 204, 97 S.Ct. at 460.9
When the government treats people differently because of their sex, equal-protection principles at the very least require that there be a meaningful factual predicate supporting a link between the government’s means and its ends. In this case, the government has failed to show that its sex-preference policy is substantially related to achieving diversity on the airwaves. We therefore hold that the policy violates the Constitution.
Ill
We turn finally to the question of remedy. Lamprecht urges us to order the Commission to award him the permit for the Middletown station. It is well settled, however, that once we correct an agency’s error of law, we must remand for the agency to exercise its discretion, assuming, of course, that the agency retains any discretion to exercise. See generally FCC v. Pottsville Broadcasting Co., 309 U.S. 134, 141-46, 60 S.Ct. 437, 440-43, 84 L.Ed. 656 (1940). In deciding which of several applicants best served the public interest, the Commission did not assign precise numerical values for each of the factors that it considered. Cf. Omaha TV 15, Inc., 4 F.C.C. Rcd. 730 (1989). The Commission’s ultimate decision is thus not susceptible to mathematical adjustment by this court; hence, we cannot allow “the contingencies of judicial review and of litigation, rather than the public interest, [to] be decisive factors in determining which of several pending applications [should] be granted.” *399Pottsville Broadcasting, 309 U.S., at 145-46, 60 S.Ct. at 443.
We have held on the record before us that the Commission’s sex-preference policy violates the Fifth Amendment. On remand, therefore, the agency must determine who, in the absence of this unconstitutional policy, should receive the permit to build the station in Middletown.
IV
As our dissenting colleague has written, “[J]ust as the FCC may not ignore the dictates of the legislative branch, neither may the judiciary ignore its independent duty to check the constitutional excesses of Congress.” Action for Children’s Television, 932 F.2d, at 1509-10. Congress and the Commission have failed to meet the standards specified by the Supreme Court in its intermediate-scrutiny cases. We therefore vacate the decision under review and remand this case for further proceedings consistent with our opinion.
It is so ordered.
Appendix
[[Image here]]
[[Image here]]
*400[[Image here]]
[[Image here]]
[[Image here]]
*401[[Image here]]
[[Image here]]
[[Image here]]
*402[[Image here]]
[[Image here]]
(Source: Minority Programming table 3, at 38.)
. Our dissenting colleague suggests throughout that the Supreme Court already has decided that the Commission’s sex-based policy passes constitutional muster. "[A]s a matter of law," he writes, “the constitutionality of this affirmative action program is clear — at least until the Supreme Court overturns Metro." Post, at 415. “In striking down the preference policy,” he writes, “my colleagues have done precisely what the Supreme Court forbids them to do.” Post, at 404. And "it strikes me as impossible,” our colleague writes, “to reconcile the Supreme Court’s decision in Metro Broadcasting with [this court’s] decision today.” Post, at 404. With all due respect, we think it impossible to reconcile our dissenting colleague's suggestion with the unambiguous reservation of Metro Broadcasting itself: "[T]he Commission’s gender preference policy is not before us today.” 110 S.Ct., at 3005 n. 7.
The Commission's gender-preference policy is, however, before us today, and though its subject may challenge certain articles of faith, this case deserves the fair, careful, and dispassionate treatment that we try to accord all of the cases we decide. Our dissenting colleague nonetheless accuses us of "telling the first branch how to make national policy,” post, at 415, and of “pay[ing] lip service to Justice Brennan’s majority opinion” in Metro Broadcasting while "applying] in practice ... Justice O’Connor’s dissent,” post, at 404 thereby showing disrespect not only to a coordinate branch of government, see post, at 415 (”[J]udges who are devoted to the original intention of the framers of the Constitution ... ignore the original intentions of selected representatives in Congress.”); post, at 406 (“There is not even a pretense of deference to Congress anywhere in their opinion.”), but also to our own branch of government, see post, at 415 ("Today my colleagues thwart not only the intentions of Congress and the executive, but also the intentions of the Supreme Court.”); post, at 404 (“[A]s appellate judges, our duty is to follow Supreme Court precedents, not to anticipate them.”), and, for good measure, to the lawyers for one of the parties, in their efforts before and at oral argument, see post, at 412 ("[M]y colleagues have belittled their efforts at every turn_ [T]he questioning judge repeatedly cut off the lawyer’s attempt to explain the government’s reasoning.”). We welcome vigorous debate, of course, but with all due respect, we think our colleague’s overheated approach discouraging: not so much an invitation to invigorating debate as a bid to provoke a shouting match. One hopeful note, however: if taking seriously the responsibility of judicial review is a vice, it is a vice that fortunately is shared, at least at times, across the jurisprudential spectrum. See, e.g., Action for Children’s Television v. FCC, 932 F.2d 1504, 1509-10 (D.C.Cir.1991) (”[J]ust as the FCC may not ignore the dictates of the legislative branch, neither may the judiciary ignore its independent duty to check the constitutional excesses of Congress.").
. Our dissenting colleague reads Metro Broadcasting for the rule that "although courts should not defer to Congress on constitutional questions, we should defer ... to Congress on empirical questions.” Post, at 406. Of course, the point about which we disagree with our colleague and with Congress — whether sex-based preferences will advance "substantially" the goal of increased programming diversity for intermediate scrutiny purposes — is a mixed question of law and fact, neither purely constitutional nor purely empirical. We nonetheless agree with our colleague insofar as he suggests that we must review Congress’s judgment deferentially, without reweighing the evidence de novo. At times, however, our colleague hints at the far different hypothesis that deference means not just limited factual review, but none at all. See, e.g., post, at 406. We know of no support — and our colleague cites none — for the proposition that if the constitutionality of a statute depends in part on the existence of certain facts, a court may not review a legislature’s judgment that the facts exist. If a legislature could make a statute constitutional simply by "finding” that black is white or freedom, slavery, judicial review would be an elaborate farce. At least since Marbury v. Madison, 5 U.S. (1 Cranch) 137, 2 L.Ed. 60 (1803), that has not been the law.
. Our dissenting colleague asserts that a sex-based classification might survive intermediate scrutiny even if it rests upon unsupported generalizations about men or women as a group. See post, at 406-408. He notes correctly that "the purpose of intermediate scrutiny” is to ensure that sex-based classifications are based on "reasoned analysis” rather than "archaic stereotypes,” post, at 408, possibilities that he appears rightly to regard as mutually exclusive. But an "analysis” that rests upon unsupported factual premises cannot possibly be "reasoned,” and an untrue and widely-held generalization about men or women is by definition a "stereotype.” See, e.g., U.S. Civil Rights Comm’n, Characters in Textbooks 5 (1980) (explaining that a member of a group is “stereotyped” when she is "portrayed in a stylized manner that conforms to widely accepted, but often untrue, ideas of what members of the group are like”). More to the point, unless a generalization about men or women asserted in defense of a sex-based classification is grounded in some degree of fact, the classification cannot possibly advance any legitimate state interest, much less an important one. Our dissenting colleague, however, insists that a relevant generalization is presumed true (even if unsupported) unless proved otherwise. See post, at 408, 414. On this point, our colleague is mistaken — for it is a "firmly established principle[ ]” that "the burden of showing an 'exceedingly persuasive justification’" falls on "the party seeking to uphold a statute that *394classifies individuals on the basis of their gender." Mississippi Univ. for Women, 458 U.S., at 723-24, 102 S.Ct. at 3335-36 (citation omitted); see also Heckler v. Mathews, 465 U.S. 728, 744-45, 104 S.Ct. 1387, 1398, 79 L.Ed.2d 646 (1984) (imposing the same burden with respect to federal statutes).
In our view, Craig v. Boren establishes beyond doubt that a sex-based classification cannot survive unless the “sex-centered generalization” asserted in the law’s defense "actually comport[s] with fact.” 429 U.S., at 199, 97 S.Ct. at 458. Craig struck down a sex-based differential in Oklahoma’s drinking age precisely because of the "ultimate unpersuasiveness of th[e] eviden-tiary record” compiled by the State in order to prove that young men were likelier to drive drunk than were young women. Id., at 201, 97 S.Ct. at 459; see also id., at 204, 97 S.Ct. at 460 (”[T]he showing offered by [the State] does not satisfy us that sex represents a legitimate, accurate proxy for the regulation of drinking and driving.”). Because so few men in the aggregate drive drunk, as the statistics themselves indicated, the Court concluded that "the relationship between gender and traffic safety” was "far too tenuous” to meet the intermediate scrutiny test. See id., at 204, 97 S.Ct. at 460. Our dissenting colleague contends that the Court "struck down the law only after pointing out that the different drinking ages were based on ‘social stereotypes.’ ” Post, at 408 (quoting 429 U.S., at 202 n. 14, 97 S.Ct. at 459 n. 14). But the point of the Court’s observation was that the stereotypes (themselves unsupported generalizations) were likely "to distort the accuracy of the [ ] comparative statistics” asserted to justify them, because "‘reckless’ young men who drink and drive are transformed into arrest statistics, whereas their female counterparts are chivalrously escorted home.” 429 U.S., at 202 n. 14, 97 S.Ct. at 459 n. 14. This and other “methodological problems” with the statistics further supported the Court’s judgment that the necessary link was "far too tenuous” empirically. See id., at 202-04, 97 S.Ct. at 459-60. Thus, the Court’s observation that "proving broad sociological propositions by statistics is a dubious business," id., at 204, 97 S.Ct. at 460, underscores only that some statistically-observed correlations are insufficient to validate a sex-based classification— and not, as our dissenting colleague suggests, see post, at 408, that a correlation "between gender and the characteristic or trait that gender purports] to represent,” 429 U.S., at 199, 97 S.Ct. at 458, is unnecessary. Weinberger v. Wiesenfeld is to the same effect as Craig v. Boren. The Court struck down the statute at issue because it rested on an '“archaic and overbroad’ generalization” — that is, on assumptions that "in view of the large percentage of married women working (41.5% in 1971)” and "other current statistics” bore "little relationship to present reality." 420 U.S., at 642 & n. 11, 95 S.Ct. at 1231 & n. 11; see also Califano v. Westcott, 443 U.S., at 88-89, 99 S.Ct. at 2663 (concluding that because no '“statistical evidence’” supported Congress’s sex-based distinction, the classification must rest on some “ baggage of sexual stereotypes’ ” and therefore "is not substantially related to the attainment of any important and valid statutory goals”).
Our dissenting colleague cites' no case in which a sex-based classification was upheld despite the absence of a demonstrable and relevant difference between men and women. Neither Michael M. v. Superior Court, 450 U.S. 464, 101 S.Ct. 1200, 67 L.Ed.2d 437 (1981) (plurality opinion), nor Rostker v. Goldberg, 453 U.S. 57, 101 S.Ct. 2646, 69 L.Ed.2d 478 (1981), see post, at 407-408, establishes this possibility. Michael M. upheld a statute punishing only men for having sexual relations with minors. The statute was designed to prevent teenage pregnancy, and it reflected the obvious biological fact that only women can become pregnant. Addressing the question whether that relevant difference was sufficient to justify the classification, a plurality rested on the ground that a sex-neutral statute might be less effective because “a female is surely less likely to report violations of the statute if she herself would be subject to criminal prosecution.” 450 U.S., at 473-74, 101 S.Ct. at 1207 (plurality opinion). The plurality's subsequent characterization of that proposition as merely a plausible one of “differing speculations," see id., at 474 n. 10, 101 S.Ct. at 1207 n. 10, might have suggested that sex-based classifications are presumed supported until proven otherwise, but the Court’s later pronouncements squarely reject this reading. See, e.g., Mathews, 465 U.S., at 744, 104 S.Ct. at 1398 (“‘[T]he party seeking to uphold a statute that classifies individuals on the basis of their gender must carry the burden of showing ... that the "discriminatory means employed” are “substantially related to the achievement of [important governmental] objectives." ’ ’’ (quoting Mississippi Univ. for Women, 458 U.S., at 724, 102 S.Ct. at 3336 (internal quotations omitted))). Rostker upheld a statute requiring only men to register for the draft. Another statute, unchallenged in Rostker, prohibited women from combat. Considering whether that relevant difference was sufficient to justify another sex-based classification with respect to draft registration, the Court rested upon Congress’s factual finding that "any future draft ... would be characterized by a need for combat troops.” 453 U.S., at 76, 101 S.Ct. at 2657. After itself reviewing the evidence supporting that finding, the Court criticized the district court for not conducting an “appropriately deferential examination” of the evidence, id., at 83, 101 S.Ct. at 2661—a limited review (as is always the case when courts review legislative factfinding), but a review nonetheless. In short, neither Michael M. nor Rostker insulates Congress’s empirical judgments from scrutiny, and neither even remotely supports the proposition for which they are cited — that the constitutionality of a sex-based distinction does not depend upon the degree of correlation between .sex and the attribute for which sex is used as a proxy.
. Our dissenting colleague tells us that we identify "women's or minority programming” as "programming specifically targeted at women or minorities,” post, at 409. We do not. Neither "women’s viewpoint” nor "women’s programming” necessarily means viewpoints or programming targeted to women or of special interest to women or concerning subjects that some may suppose women are likely to put on the airwaves, and the same is true for "minority viewpoint" or "minority programming.” Rather than presume to give content to these terms, cf. post, at 411 (calling "innocuous to the point of being obvious” the assumption that "some female programmers will choose to emphasize different subjects — breast cancer, say, or glass ceilings in the workplace”); infra note 8, we remain agnostic. We do assume, however, as did Congress, that these terms necessarily have some meaning — but whatever "women’s programming" and "minority programming” are, there is no evidence showing that women owner/managers are substantially likelier to program them than are men.
. This exchange was typical:
Judge Thomas: ... I am asking you, in the aggregate, what difference does it make, other than [that] you have a male owner or a female owner?
Mr. Pash: I think that what difference it makes, ... is that there is a reasonable expectation—
Judge Thomas: Based on what?
Mr. Pash: —a reasonable expectation based on the fact that women are a significant group, the fact that they are significant in size, they are a significant group in the society.
Judge Thomas: I understand that. I understand that women are a significant group, and I understand that they are significant in size. But what difference does it make, if a woman owns a station or if women owned all the stations, other than that they owned all the stations? Does it make a difference in programming, does it make a difference in content of the points of view, does it make a difference in the editorials? What does it make a difference?
Mr. Pash: It makes a difference ... it makes a difference in viewpoint and perspective presented on the station, and the expectation is that—
Judge Thomas: Okay. Now, what is that based on?
*396Mr. Pash: That is based on the expectation that, because women and other groups — but in this case we are talking about women — have been historically subject to societal prejudices and particular societal attitudes that they—
Judge Thomas: So, does that mean they will have a different point of view as a result of prejudice, or they don’t own stations as a result of prejudice?
Mr. Pash: No, that means that there is an expectation that, in the aggregate, they will have a different perspective on questions, not on every question, not in every case, but—
Judge Thomas: Based on what, though? I mean how can you conclude that [women] will have a different perspective?
Mr. Pash: Well, I guess this is going to sound circular, but you can conclude they will have a different perspective, because, historically, they have been subject to prejudice and different societal attitudes, and this has led to their playing a different role in society, they are being treated differently, subject to stereotypes and so forth....
Judge Thomas: But how is this expectation any different from a stereotype, if it doesn’t have any basis in fact?
Mr. Pash: Well, I am not saying it does. I am saying it doesn’t have any basis in fact. I am saying that this record doesn’t provide factual evidence for it....
Judge Thomas: Okay. Fine. But is there any evidence that there is a difference between the stations owned by women and owned by men?
Mr. Pash: No, there is no evidence in the record.
Chief Judge Mikva: You said that three times now....
Tr. of Oral Arg. at 33-36; see also id., at 32, 38, 41-42, 44.
.Our dissenting colleague terms us “less than sporting" for failing to “call[ ] for amicus briefs on the statistical question.” Post, at 414. Any interested party could have filed an amicus brief had it chosen to, however, and we doubt that our colleague seriously means what he says — that a court generally should ask of its own accord for amicus briefing whenever (as is often the case) additional paper might prove helpful. To the extent that our colleague still disagrees with our decision not to accept the amicus brief of American Women in Radio and Television, Inc., we note for the record only that the organization submitted its brief out of time, but that neither the organization nor our colleague offered an explanation for why we should forgive the group for violating the rules of the court.
. Although he maintains that “if my colleagues want more statistics, there are plenty to support Congress’s position,” post at 414, the anecdotes that our colleague relates do not sustain his argument. Our colleague relies upon an amicus brief filed in Metro Broadcasting, by American Women in Radio and Television, that he says shows that of the recipients of the 1986 “National Commendation Awards” for presenting women in a "positive and realistic light," women made up 58.5% of the producers, 84.2% of the writers, and 92.9% of the reporter/hosts. This is interesting, but irrelevant, for reasons that also apply to our colleague’s admittedly “quick skim of the best-seller lists,” post, at 415. Congress has given a preference to women who own and manage radio and television stations, not women who write or produce or report for them. Our dissenting colleague’s statistics thus matter only insofar as they rest on certain stereotypes about how women behave. "Managers” are not necessarily “programmers"; as the Commission has made clear since it first set out its comparative-licensing scheme, an owner is entitled to integration credit if she serves as "program director” or as "station manager,” “business manager,” or "sales manager," among other jobs. Policy Statement on Comparative Broadcast Hearings, 1 F.C.C.2d, at 395. And even if we were to assume for the sake of argument that women owners are likelier than men to become programmers, rather than business managers, we would have to assume that women owner/programmers are likelier than men to hire women producers, women writers, and women reporters. Either or both of these may well be so, but since our colleague offers no evidence to support them, we think it appropriate to hesitate before giving too much credence to these assumptions about sex roles in the marketplace.
. Unfortunately, Minority Programming does not define terms such as “women’s programming" (or “minority programming,” for that matter), but rather, relied on the reporting stations to characterize themselves. The study itself warns readers of this and other problems. See Minority Programming at 1-8; see also Spitzer, Justifying Minority Preferences in Broadcasting, 64 S.Cal.L.Rev. 293, 342-45, 345-46 (1991); Winter Park, 873 F.2d, at 358-61 (Williams, J., concurring in part and dissenting in part).
. Our dissenting colleague declares that "judges have no basis, except their own policy preferences,” for concluding that the degree of correlation shown here is not "enough," since the Fifth Amendment does not identify the “mystical" point at which a correlation satisfies the dictates of equal protection. Post, at 413. Our colleague is not the first to criticize the intermediate-scrutiny test for its indeterminacy. See Craig v. Boren, 429 U.S., at 221, 97 S.Ct. at 469 (Rehnquist, J., dissenting) ("[T]he phrases used are so diaphanous and elastic as to invite subjective judicial preferences or prejudices relating to particular types of legislation masquerading as judgments whether such legislation is directed at ‘important’ objectives or whether the relationship to those objectives is ‘substantial.’ ”). While rational basis or strict scrutiny review would make for fewer close cases, however, we have no authority for departing from the Supreme Court’s decision to follow the middle course. Intermediate scrutiny, then, which Metro Broadcasting requires of us, calls for judgment — for drawing a line. The line we necessarily draw today, holding that the degree of correlation in this case is not enough, is different from the line drawn in Metro Broadcasting (that a much stronger degree of correlation was enough), the line drawn in Craig (that an arguably weaker degree of correlation was not enough), and the line proposed today by our colleague in dissent (that the degree of correlation in this case is enough). Despite what our colleague states, the line that we draw is neither more nor less principled than the line that he draws; our lines are merely grounded in different exercise of judgment.