dissenting.
States should have latitude to experiment in their handling of prisoners. No one has found a satisfactory way to deal with offenders, and only an exceptionally strong reason should induce courts to interfere with the search. Courts should accept reasonable judgments by state officials in the ordinary course of prison management. The security and privacy concerns of prison life should cause judges to doubt that making prisons more like civilian life is wise. Female guards may make Taycheedah both a more humane place for its inmates and a more secure place in which to confine them. The judges in the majority deploy these contentions in reversing the district court.
If this were a suit directly under the fourteenth amendment, the state would receive the benefit of these observations. E.g., Turner v. Safley, 482 U.S. 78, 107 S.Ct. 2254, 2259-62, 96 L.Ed.2d 64 (1987); O’Lone v. Shabazz, 482 U.S. 342, 107 S.Ct. 2400, 2404-05, 96 L.Ed.2d 282 (1987). It is not, so the state does not. This suit invokes Title VII of the Civil Rights Act of 1964. Exercising its power under § 5 of the fourteenth amendment, Congress altered the usual burdens. Fitzpatrick v. Bitzer, 427 U.S. 445, 96 S.Ct. 2666, 49 L.Ed.2d 614 (1976); cf. Katzenbach v. Morgan, 384 U.S. 641, 86 S.Ct. 1717, 16 L.Ed.2d 828 (1966). The employer in a Title VII suit must show that its use of gender as a ground of decision is not only reasonably related to legitimate objectives (the constitutional inquiry) but also “necessary”. Dothard v. Rawlinson, 433 U.S. 321, 97 S.Ct. 2720, 53 L.Ed.2d 786 (1977). The employer’s burden is less onerous when the plaintiff protests the disparate consequences of a policy that is neutral with respect to sex, see Watson v. Ft. Worth Bank & Trust, — U.S. -, 108 S.Ct. 2777, 2787-91, 101 L.Ed.2d 827 (1988) (plurality opinion), but Wisconsin uses gender as an explicit ground of decision and so must satisfy the *1536most exacting standard. The statute grants rights to employees, not to employers. The district court held a trial and concluded that the state’s evidence did not satisfy its burden.
The allocation of the burden makes all the difference when uncertainty predominates — a fair description of arguments pro and con about the operation of prisons. Do the privacy cards protect privacy at the expense of security? Surely yes, but equally surely the sacrifice is not great. Taycheedah retained its privacy cards even after banishing men from the corridors. No one knows the extent of the cards’ effects, so we must accept the district court’s view of the subject. Anderson v. Bessemer City, 470 U.S. 564, 105 S.Ct. 1504, 84 L.Ed.2d 518 (1985). Women’s prisons throughout the nation employ male guards; other women’s prisons in Wisconsin do so; four other courts of appeals have held that unless prisons entail grave risks of violence to members of one sex only, as in Dothard, sex is not a bona fide occupational qualification for the position of prison guard, see Harden v. Dayton Human Rehabilitation Center, 779 F.2d 50 (6th Cir.1985), affirming 520 F.Supp. 769 (S.D.Ohio 1981); Hardin v. Stynchcomb, 691 F.2d 1364 (11th Cir.1982); Forts v. Ward, 621 F.2d 1210 (2d Cir.1980); Gunther v. Iowa State Men’s Reformatory, 612 F.2d 1079 (8th Cir.1980).
Do the male guards hinder rehabilitation of the female prisoners? Maybe, but the expert testimony in the case conflicts, which mirrors the conflict in society at large about rehabilitation. Some use this word to refer to restoring (or creating) a sense of normal social interactions, including respect for privacy. One expert testified along these lines that male guards hinder rehabilitation by reminding the prisoners of their past, often unhappy, encounters with men; another expert testified that male guards promote rehabilitation because their presence reflects a part of reality to which prisoners must be accustomed if they are to function normally after release. Who is right? When the burden is on the defendants, and the trier of fact has found the defendants’ explanations shal-
low, an appellate court must accept that decision. We ought not take it as indisputable that women are different from men and use that premise to justify sex differences in employment. Congress thought otherwise; the district court found otherwise. Under the court’s treatment of “necessity” a district judge may (perhaps is compelled to) accept an employer’s bare opinion that segregation by sex is helpful. Rarely are we so deferential even in litigation filed by prisoners directly under the Constitution, the source of the “deference” cases to which the court refers, see maj. op. at 1527-29. Dothard held that a level of violence (including sexual violence) exceptional for American prisons supplied a BFOQ; the majority today says that sex may be a BFOQ in every female prison (maybe every prison) in America.
There is another sense of rehabilitation: the ability of a prison to “cure” persons of their criminal habits, measured by the recidivism rate of the former inmates. The rate of crime falls with age, D.P. Farring-ton, Age and Crime, 7 Crime and Justice: An Annual Review of Research 189 (1986), and with the threat of punishment (deterrence), so any prison may “rehabilitate” its inmates by holding them longer. No solid evidence shows that the conditions at a prison affect recidivism. Neither educational programs nor confinement arrangements nor any other measurable attribute seems to reduce the rate of crime after release, holding other things (such as age and personal characteristics) constant. See Lee Sechrest, Susan O. White & Elizabeth D. Brown (eds.), The Rehabilitation of Criminal Offenders: Problems and Prospects (1979). “The view that ‘nothing works’ has ... become the conventional wisdom among most corrections practitioners and researchers.” Peter W. Greenwood, Controlling the Crime Rate through Imprisonment, Crime and Public Policy 251, 253 (James Q. Wilson ed. 1983). The evidence in this record suggests that none of the changes in Taycheedah’s operation affects the recidivism rate of its inmates. 639 F.Supp. 271, 280 (E.D.Wis.1986).
*1537The court concludes, maj. op. at 1530-32, that the district judge erred in requiring “objective” (that is, empirical) evidence to support any defense of bona fide occupational qualifications. If the district judge had used such an inflexible standard, he would have been in error. Digital computers are not the only source of facts. Employers often must act on widely held beliefs even if data remain to be collected. When there' are no data, when the impressions and judgments of the witnesses conflict — witnesses for both sides were experienced in studying or managing prisons — the district court is entitled to credit one side’s approach over the other’s. The district judge did no more than that. He simply pointed to the lack of one kind of evidence as additional support for his conclusion. We review judgments, not opinions; an isolated sentence in a thoughtful opinion is not a sufficient basis on which to reverse a judgment.
The court’s opinion masks a deeper question. Why does the need to employ female guards to promote rehabilitation establish the “necessity” of doing so? It cannot be that whenever the employer gets any benefit, however small, from using gender as a ground of decision, it is therefore “necessary” to use this criterion. Rehabilitation as a justification for confinement has all but vanished from American penology. The women’s prison movement adopted in the last century and abandoned in this one a rehabilitative model of confinement. Nicole Hahn Rafter, Prisons for Women, 1790-1980, 5 Crime and Justice: An Annual Review of Research 129, 146-66 (1983). Congress jettisoned the rehabilitative justification for imprisonment when adopting the current criminal code in 1984. The arguments presented by Taycheedah’s administrators reflect some persons’ attitudes, but they are minority views among legislators and social scientists. To say, as the court does, maj. op. at 1530, that because Wisconsin believes in trying to rehabilitate prisoners, then “rehabilitation, no matter how elusive, must be attempted”, is to assign no weight to the interests protected by Title VII. No state may define a trivial— or impossible — objective that requires the use of gender for its achievement and then invoke its objective as justification for the discrimination. If a state wants to pursue the impossible dream, it may not do so at the expense of race or sex discrimination.
One way or another, the arguments offered by the majority are a plea for a special rule for prisons. My colleagues believe not that the state has demonstrated the necessity of hiring guards by sex, but that the state should have leeway to experiment in the operation of its prisons even though it has not so far established the need to run them in a particular way. Congress might come to this conclusion, but it did not do so in 1972, when it extended Title VII to the states. Dothard rejects any argument that states should be treated differently as employers. The Court said:
... [The prison officials] contend that the establishment of the minimum height and weight standards by statute requires that they be given greater deference than is typically given private employer-established job qualifications. The relevant legislative history of the 1972 amendments extending Title VII to the States as employers does not, however, support such a result. Instead, Congress expressly indicated the intent that the same Title VII principles be applied to governmental and private employers alike_ Thus for both private and public employers, “[t]he touchstone is business necessity” ...; a discriminatory employment practice must be shown to be necessary to safe and efficient job performance to survive a Title VII challenge.
Dothard, 433 U.S. at 331 n. 14, 97 S.Ct. at 2728 n. 14. Dothard was about the qualifications of prison guards. We cannot distinguish a prison guard case on the ground that prison officials should have special latitude to experiment. So Wisconsin does not prevail by establishing that its decision is rationally related to legitimate penological considerations — the standard for the constitutionality of most state action that the majority applies to Taycheedah’s policy. A rational basis does not carry the day when the statute demands a “bona fide occupational qualification.” See Western *1538Air Lines, Inc. v. Criswell, 472 U.S. 400, 422-23, 105 S.Ct. 2743, 2756, 86 L.Ed.2d 321 (1985) (decided under the Age Discrimination in Employment Act, but construing the defense applicable to many anti-discrimination laws).
A jailer’s lot is not a happy one. David K. v. Lane, 839 F.2d 1265, 1278-80 (7th Cir.1988) (concurring opinion); Walker v. Rowe, 791 F.2d 507 (7th Cir.1986). (Apologies to W.S. Gilbert.) If the prison takes sex into account in choosing and assigning guards, the rejected applicants will sue under Title VII. If the prison hires guards of both sexes, the prisoners will sue on the ground that being observed and searched by someone of the opposite sex invades privacy. Whether to tighten this vise is a legislative decision, and the statute now in force prefers the interests of guards (as employees) over the interests of prisoners. At all events, the panel’s opinion shows that the use of male guards at Taycheedah does not violate the privacy interests of the prisoners, 838 F.2d at 951-53, if they have any that the state is bound to respect after Hudson v. Palmer, 468 U.S. 517, 524-30, 104 S.Ct. 3194, 3199-3200, 82 L.Ed.2d 393 (1984).
When judges in constitutional litigation defer to prison officials’ decisions, this is not because they know the decisions to be right. It is a contradiction in terms to decide which are the best ideas about prison management and then “defer” to their proponents. Prison cases reflect not rhetorical manipulation but genuine deference, based on a lack of knowledge. Certainly judges lack it; often so does everyone else. When the plaintiff bears the burden of persuasion, ignorance, combined with the judiciary’s professional skepticism, implies leaving well enough alone. Uncertainty is not a good reason to remake prisons. They don’t need reform; things are bad enough as they are. When information does not permit confident separation of right from wrong, courts cannot mark the “better” path. Only someone with the answer, and not just a point of view, may insist that everyone else fall into line. When Congress requires the state to justify its position, uncertainty cuts in favor of plaintiffs. Judges cannot be skeptics in suits under the Constitution and confident that they (or prison officials) have the right answers in suits under Title VII. Even-handed skepticism implies that the party with the burden loses.