West Virginia v. B.P.J., now before the Supreme Court, raises a question that lower courts are already struggling to answer. During oral argument, Chief Justice Roberts asked whether Bostock v. Clayton County bears on a challenge to a West Virginia state law barring transgender girls from girls’ sports teams in public schools: “In terms of Bostock, I understand that to say that discrimination on the basis of transgender status is discrimination on the basis of sex. But the question here is whether or not a sex-based classification is necessarily a transgender classification . . . .” That question captures a genuine doctrinal tension between statutory and constitutional frameworks for analyzing sex discrimination — B.P.J. itself raises both a Title IX1 and a Fourteenth Amendment challenge to that law.
Bostock interpreted Title VII’s prohibition on discrimination “because of . . . sex” to mean transgender discrimination is sex discrimination. Equal protection, by contrast, asks a different question: whether a law classifies on the basis of sex so as to trigger heightened scrutiny. United States v. Skrmetti answered that question with respect to a Tennessee law banning gender-affirming care for minors, holding that it was not a sex classification. The two inquiries can point in the same direction, but they need not.
Lange v. Houston County illustrates how courts wrongly collapse them. There, the Eleventh Circuit, sitting en banc, considered a Title VII challenge to an employer health plan that excluded coverage for gender-affirming surgery. The case combined a fact pattern closely analogous to Skrmetti with a statutory question that had binding precedent under Bostock. Because Skrmetti itself discussed a Bostock-style counterfactual, the Lange majority treated Skrmetti’s reasoning as effectively controlling the statutory analysis.2 That was the mistake.3 In Skrmetti, the Court’s Bostock counterfactual — asking whether changing the plaintiff’s sex would change the outcome — was run within a constitutional classification inquiry, not a statutory but-for causation question. Its logic does not transfer to Title VII. The two frameworks differ not just in source but in how they treat the role of intermediaries.
I. The Antecedent Question Bostock Does Not Answer
This Essay argues that the confusion arises because courts often treat Bostock’s counterfactual as if it resolves the analysis by itself. It does not. Before running the counterfactual, a court must first decide whether the trait linking sex to the challenged outcome should be treated as analytically independent of sex or as part of it. That antecedent choice is not merely causal; it is also classificatory and definitional. It determines whether sex appears to be a but-for cause of the plaintiff’s injury, whether the policy appears to classify on the basis of sex, and what counts as “sex” for purposes of the analysis in the first place. Because constitutional and statutory doctrines resolve that prior judgment differently, the same fact pattern can generate different answers in the two contexts. A policy may operate through a facially neutral intermediary and therefore evade heightened scrutiny under equal protection, even while sex remains a but-for cause of a plaintiff’s injury under Title VII.
This Essay then explains why those disagreements are structured rather than arbitrary. Scholars have long criticized the Bostock test as inherently indeterminate,4 producing different results depending on what variables are held constant. But that surface indeterminacy obscures a prior legal question: when may a seemingly neutral trait be treated as independent of sex, and when must it instead be understood as a way in which sex is operating? The answer is not uniform. Some intermediaries are defined through sex itself, as with sexual orientation and gender identity. Some are sex-exclusive but not sex-defined, as with pregnancy. Some appear neutral in the abstract but are not independent in application, as with sex stereotypes in Price Waterhouse v. Hopkins. Some become definitionally part of the protected class because Congress says so, as with pregnancy and religious practice. And some, like paramour preference, may genuinely function as independent intermediaries. The point of the discussion that follows is not simply to classify these intermediaries, but to identify the distinct reasons their apparent neutrality may or may not hold.
Professor Benjamin Eidelson’s account provides an important theoretical foundation for this analysis. On his view, disparate treatment law prohibits disfavoring individuals because of properties they possess partly in virtue of how they stand along the dimensions enumerated in the statute. Counterfactual tests are, in that framework, “an evocative tool for describing the role of a given attribute in a decision, but not the fundamental determinant of whether the attribute played a role.”5 Understood this way, the counterfactual does not determine whether a given trait is independent of sex; it gives causal expression to a prior judgment about whether that trait should be treated as a dimension of sex or as a variable to be held constant. The typology below makes that antecedent step explicit.
II. A Typology of Intermediaries
To visualize this framework, imagine each intermediary that can be used in a Bostock counterfactual as a “step” — holding all else constant, changing a step can change the outcome.
Begin with the limiting case of a one-step framework [sex ↦ outcome] with direct dependence: for any individual, holding all else constant, changing sex always changes the outcome. In this setting, the counterfactual is straightforward, and sex is the but-for cause of the outcome. And because the rule assigns outcomes solely based on sex, individuals can be partitioned into groups by sex such that each group receives a distinct outcome — sex-based classification exists.
Under this one-step framework, sex is both the counterfactually determinative variable and the criterion by which the policy classifies. But-for causation and classification analysis therefore converge: sex discrimination is present. This logic is not specific to sex. Replace sex with any trait, protected or otherwise, and the same follows: where [trait ↦ outcome] exhibits direct dependence, the trait is both a but-for cause and the basis of classification.
Next, consider a one-step framework [sex ⇝ outcome] with partial rather than universal dependence: for some individuals, holding all else constant, changing sex will change the outcome; for others, it will not. An example is a policy that sets retirement eligibility at age 60 for women, and age 65 for men. Such a rule partitions the world into two groups based on sex and assigns different legal consequences — retirement-eligible ages — on that basis, which is paradigmatic sex classification under equal protection doctrine.6
At the individual level, but-for causation is not universal. A 62-year-old woman is eligible for retirement, but she would not be if she were a man. A 50-year-old woman is not eligible, and she would not be even if she were a man. But Title VII but-for analysis does not require universality across all individuals; it asks only whether sex is a but-for cause of the challenged differential treatment — that is, whether there exist individuals for whom the outcome would differ if their sex were different. That condition is satisfied here. For individuals between ages 60 and 65, changing sex changes the outcome. Within that range, the relationship reduces to the direct-dependence case: sex determines eligibility. Accordingly, sex is a but-for cause of the differential treatment, even though it is not determinative in every instance. In such cases, causation and classification still converge.
A. A-Type Intermediaries
Now imagine an intermediary A that stands between sex and the outcome. This produces a two-step framework [sex ↦ intermediary A ↦ outcome]. If changing sex will always change intermediary A, and changing intermediary A will always change the outcome, then the seemingly two-step chain effectively collapses into a one-step framework [sex (inclusive of intermediary A) ↦ outcome] with direct dependence. In that case, holding all else constant, changing sex necessarily changes the outcome, and sex is a but-for cause.
This is the logic of Bostock. Sexual orientation and transgender identity are sex-defined intermediaries: sexual orientation is defined as a function of (sex of person, sex of partner) and transgender identity is defined as a function of (biological sex assigned at birth, self-identified gender). Where sex is an input into the definition of the intermediary, holding all else constant, changing sex necessarily changes the intermediary. Once that is true, the intermediary does not break the causal chain; it becomes part of the way sex operates. The Bostock Court captured this when it reasoned that “discrimination based on homosexuality or transgender status necessarily entails discrimination based on sex” because the “first cannot happen without the second.”7 The chain [sex ↦ sexual orientation / transgender identity ↦ firing] therefore collapses into [sex (inclusive of sexual orientation and transgender identity) ↦ firing], reducing to the one-step case with direct dependence.
Is sex classification present? Yes. Once the analysis is collapsed into one-step with direct dependence, the rule is equivalent in operation to classifying individuals by sex, even if it is formally framed in terms of the intermediary. A canonical example is Loving v. Virginia, where the protected characteristic was race. Formally, the trigger for liability was the intermediary of interracial couple status, not the individual’s own race: [race ↦ interracial couple status ↦ legality of marriage]. But this is an A-type intermediary because an individual’s own race is an input into their status of being in an interracial relationship; just as Bostock’s outcome was a function of (sex of employee, sex of their partner), Loving’s outcome was a function of (race of person, race of their partner). Holding the partner’s race constant and changing the individual’s race changes the couple’s status and therefore the legality of the marriage. The chain thus collapses into [race (inclusive of interracial couple status) ↦ legality of marriage] and law is properly understood as classifying on the basis of race.
B. B-Type Intermediaries
Now imagine an intermediary B that also stands between sex and the outcome [sex ⇝ intermediary B ↦ outcome] where changing sex does not always change intermediary B, but changing intermediary B always changes the outcome. Because changing B will always change the outcome, intermediary B is necessarily both a but-for cause and a classification criterion in the sub-chain [intermediary B ↦ outcome]. The question is then whether sex remains a but-for cause or a classification criterion — and whether those inquiries yield the same answer.
Some B-type intermediaries are treated as neutral intermediaries that sever the causal chain. One example is paramour preference, where an employer discriminates in favor of a supervisor’s romantic paramour and against other employees: [sex ⇝ supervisor paramour status ↦ employment action]. Changing whether an employee’s paramour status always changes the outcome — partners receive favorable treatment; non-partners receive adverse treatment. But changing the employee’s sex only sometimes changes their paramour status. If a heterosexual male supervisor is romantically involved with a female employee, changing that employee’s sex from female to male would eliminate her paramour status; for other non-partner employees, changing sex would have no effect.
This result follows only if paramour status is treated as analytically independent of sex. On that characterization, paramour status is the only but-for cause, and sex plays no independent causal role [supervisor paramour status ↦ employment action]. The Ninth Circuit adopted this reasoning in Maner v. Dignity Health, joining every other circuit in rejecting paramour preference as sex discrimination under Title VII.8 There, the court explained that “[t]he employer discriminates in favor of a supervisor’s sexual or romantic partner and against all other employees because they are not the favored paramour, no matter the sex of the paramour or of the complaining employees.”9
Other B-type intermediaries, however, are not treated as independent of sex and therefore do not sever the causal chain. Even where sex does not always determine B, but-for causation is satisfied if, for the plaintiff, changing sex would change B and thus the outcome. In that setting, B does not function as an independent variable. This can occur either because the law treats B as part of sex; because B, though facially neutral, operates in a way that necessarily incorporates sex; or because B is conceptually inseparable from sex. In each case, intermediary B does not function independently of sex for the plaintiff.
But even where intermediary B is not independent of sex for purposes of causation, classification analysis may still characterize the policy as operating on B rather than on sex, meaning the same fact pattern can yield liability under Title VII while escaping heightened scrutiny under equal protection.
B-type intermediaries may fail to function independently of sex through three distinct, though non-exhaustive, mechanisms: statutory redefinition, where Congress expands the definition of the protected class to subsume a linked trait; sex-contingent operation, where a facially neutral intermediary incorporates sex in application for this plaintiff; and definitional entailment, where the intermediary is not analytically independent of the protected class at the level of the specific instance. Each produces the same result — sex remains a but-for cause.
C. Statutory Redefinition: Pregnancy and the PDA
The first mechanism is statutory redefinition of the protected class. A canonical example of this is pregnancy. In General Electric Co. v. Gilbert, an employee benefits plan excluded coverage for pregnancy-related disabilities while covering other conditions. The framework can be written as [sex ⇝ pregnancy ↦ coverage], with pregnancy as a B-type intermediary. Changing pregnancy status always changes the outcome: a disability plan that excludes pregnancy treats pregnant employees worse. But changing sex does not always change pregnancy status. Dividing the world into pregnant and non-pregnant populations, women appear in both groups — some women are not pregnant and may never be. The Gilbert Court characterized pregnancy as a physical condition independent of sex, importing the constitutional logic from Geduldig v. Aiello, which held that pregnancy classifications are not sex classifications for equal protection purposes.
But pregnancy is a physical condition experienced by only one sex. It is therefore a sex-exclusive intermediary: one that only members of one sex can occupy, even if not all members do. Not all women are pregnant, but only women can be. Congress overturned Gilbert by passing the Pregnancy Discrimination Act of 1978 (PDA), which expressly broadened the definition of Title VII sex discrimination to include pregnancy discrimination. That intervention altered the statutory meaning of sex without changing its constitutional meaning. As a result, pregnancy discrimination is sex discrimination under Title VII, but pregnancy classifications are still not sex classifications triggering heightened scrutiny under the Fourteenth Amendment.10
D. Sex-Contingent Operation: Price Waterhouse
Courts have also rejected intermediary independence on their own interpretive authority. Price Waterhouse v. Hopkins illustrates the second mechanism — sex-contingent operation — and how it differs from statutory redefinition: rather than Congress expanding the definition of the protected class, a court finds that the intermediary, though facially neutral, incorporates sex in its application for this plaintiff. There, the Court considered a partnership selection process that evaluated employees based on professional demeanor, including fit and interpersonal style: [sex ⇝ evaluation of professional demeanor ↦ partnership]. This intermediary is type-B because changing sex does not always change the employee’s evaluated demeanor,11 but the evaluation determines the outcome. Formally, the policy appeared to operate on a neutral criterion.
The plurality nonetheless concluded that the intermediary was not independent of sex in application. Hopkins was denied partnership because her behavior was deemed “aggressive” and insufficiently “feminine.” For her, the evaluation of demeanor was sex-contingent: changing her sex would have changed the assessment and thus the outcome. The sex-contingency here may in fact run deeper — a less aggressive woman might equally have been penalized for failing to project the authority expected of a partner, suggesting that no woman could satisfy the evaluation on terms independent of sex. But Title VII’s but-for test does not require that sex-contingency be universal across all members of the protected class; it requires only that it hold for this plaintiff. Here, the mechanism was judicial recognition that the intermediary’s evaluative content was defined by sex-based norms: the standard for “appropriate” professional demeanor was itself a function of the plaintiff’s sex, making its application sex-contingent for her. The plurality therefore treated this B-type intermediary not as a break in the causal chain, but as a constitutive element of what sex discrimination looks like in practice. A facially neutral intermediary that is operationally sex-dependent for the plaintiff was treated as a component of sex discrimination itself.
E. Definitional Entailment: EEOC v. Abercrombie & Fitch Stores, Inc.
The third mechanism — definitional entailment — arises where the intermediary, even if facially neutral and not sex-contingent in application, is not analytically independent of the protected class at all. Consider my hypothetical of an employer who refuses to hire a Muslim woman because she wears a hijab, citing a facially neutral “no-headwear” policy. To identify possible B-type intermediaries, consider what criteria the policy facially classifies by. Two such intermediaries are possible: religious practice and wearing headwear [religion ⇝ religious practice ⇝ wearing headwear ↦ adverse outcome].
Religious practice is a B-type religion-exclusive intermediary: not all religious individuals are engaged in religious practices, so changing religious status does not always change whether someone engages in religious practices. But only religious people engage in religious practices. This is structurally identical to pregnancy: only members of the protected class can be in the affected subgroup, but not all members are.
Congress collapsed the first step by statute. Section 2000e(j) defines religion to include “all aspects of religious observance and practice, as well as belief,” reducing the chain to [religion (inclusive of religious practice) ⇝ wearing headwear ↦ outcome]. Religious practice is therefore itself a protected characteristic, not an independent intermediary. The mechanism is again statutory definition, parallel to the PDA.
The harder question is whether the remaining intermediary — wearing headwear — breaks the causal chain. At a general level, it appears neutral. It is similarly a B-type intermediary because not all individuals who engage in religious practices wear headwear, and not all headwear is religious. Dividing the world into headwear-wearers and non-headwear-wearers, religious people and religious practices appear on both sides. On that characterization, the policy could be described as classifying on a facially neutral basis.
But that description fails at the level of the specific act. Wearing a hijab is not merely correlated with Islam for the employee who wears it as a religious obligation — it is, by definition, an act of religious observance for Muslims. The intermediary does not statistically track the protected class; it is not religion-exclusive, but instead instantiates the protected class in this specific case. Holding all else constant, one cannot vary religion while also holding the fixed act of wearing a hijab for religious reasons. The two are not analytically independent.
This distinguishes the case from pregnancy, which can be defined without reference to the category “woman” — it is a biological condition that happens to be sex-exclusive but is not constituted by sex at the definitional level. It is also distinguishable from generic headwear, which carries no inherent religious content. Causal independence is rejected not because all headwear-wearing is religious, but because this act of wearing this headwear for this reason is, by its nature, an exercise of the protected characteristic.
This illustrates a third way in which an ostensibly neutral intermediary may fail to break the causal chain even where neutral classification criteria may be present: definitional entailment at the level of the specific instance. In EEOC v. Abercrombie & Fitch Stores, Inc., the Court held that such a facially neutral no-headwear look policy provided no defense to a Title VII claim brought by a Muslim applicant who wore a hijab, because “it is no response that the subsequent ‘fail[ure] . . . to hire’ was due to an otherwise-neutral policy.”12 Treating such conduct as a neutral criterion misunderstands its character: the act at issue is not separate from the protected trait and constitutes its exercise.
III. Applying the Framework: Lange and the Medical Indication Intermediary
Each mechanism reflects a distinct reason Title VII refuses to treat an intermediary as independent of sex — and Lange implicates all three. The relevant chain there is [sex ↦ transgender status ⇝ medical indication ↦ exclusion]. Under Bostock, transgender status is an A-type intermediary, so the chain shortens to [sex (inclusive of transgender status) ⇝ medical indication ↦ exclusion].
The dispute therefore turned on whether medical indication functions as an independent B-type intermediary. Changing it always changes the outcome: if Lange had sought hormone therapy for menopause instead of gender dysphoria, the plan would have covered it. But changing sex or transgender status does not always change medical indication; some transgender people do not have gender dysphoria, just as some women are not pregnant. The Lange majority, drawing on Skrmetti, treated the exclusion as operating on a neutral medical criterion that severs the causal chain.
Skrmetti itself, while involving no Title VII challenge, walked through a Bostock analysis. There, the Court collapsed only the A-type intermediary of transgender status into sex [sex (inclusive of transgender status) ⇝ medical indication ↦ outcome] and treated medical indication as a neutral B-type intermediary: transgender people fall on both sides of the medical line, just as women fall on both sides of the pregnancy line. Because changing medical indication changes the outcome, the discrimination is because of medical indication, not sex.
Under Title VII, however, that characterization fails. Medical indication in this context does not operate as an independent intermediary, and it fails in three distinct ways already reflected in the statute’s doctrinal structure.
First, at the level of definition, gender dysphoria is not an independent medical condition that merely correlates with transgender individuals. Unlike a neutral biological condition like cancer, gender dysphoria is sex-defined at the level of the diagnosis itself. The DSM-5 defines gender dysphoria as a marked incongruence between one’s experienced or expressed gender and one’s sex assigned at birth — variables that Bostock itself treated as dimensions of sex. A medical indication whose very formulation requires those variables cannot be held constant while toggling sex in the Bostock counterfactual; the two are not independent variables. It is therefore not a freestanding B-type intermediary that merely tracks sex; it is constituted by sex and gender identity. In this respect, it resembles the hijab example above: for this plaintiff, the act or condition at issue is not merely associated with the protected class, but is an expression of it by definition. The intermediary therefore does not sit between sex and the outcome as an independent step; it instantiates sex at the level of the specific case. This matters for statutory causation even if, under a classification framework, the intermediary appears facially neutral.
Second, even if medical indication is treated as analytically distinct, it does not operate independently of sex in application. As in Price Waterhouse, where ostensibly neutral evaluation criteria like interpersonal skills were found to incorporate sex in their application for the specific plaintiff, the diagnosis and treatment of gender dysphoria depend on how sex is societally understood and evaluated. The criteria for diagnosing and treating gender dysphoria are sex-contingent for Lange: the very definition of the diagnosis references the patient’s sex assigned at birth, meaning that changing sex would change whether the diagnosis attaches and therefore whether coverage is triggered. The intermediary is therefore not causally independent in operation, even if it can be described in neutral terms.
Third, Title VII’s statutory history and precedents reflect a broader refusal to treat condition-based distinctions as severing causation where they track protected characteristics. Geduldig’s reasoning — that pregnancy is a condition independent of sex — was imported into Title VII in Gilbert and then repudiated by Congress through the PDA. Section 2000e(j) performs similar work for religion. These interventions reflect a broader pattern: Congress and courts do not treat facially neutral intermediaries as dispositive when they are not genuinely independent of the protected trait. Lange reverses that pattern by importing constitutional classification logic into a statutory framework that has repeatedly rejected it.
IV. The Error the Court Should Not Repeat
The underlying error is thus not simply a misapplication of Bostock, but a failure to resolve the prior question on which Bostock depends: whether the intermediary can be treated as independent of sex. The but-for test does not answer that question; it inherits it. And on that antecedent issue, Title VII’s text, history, and precedents point away from Lange’s conclusion.
The characterization of intermediaries is dispositive. And once that structure is made explicit, it becomes clear that causation and classification do not necessarily travel together: a policy may not classify on the basis of sex for constitutional purposes and yet still treat an individual worse because of sex under Title VII. Chief Justice Roberts is therefore correct to separate the two inquiries. B.P.J. may present a statutory violation even if it does not trigger heightened scrutiny under the Equal Protection Clause. The relevant question under Title VII is not whether an intermediary maps symmetrically onto sex categories, but whether for this plaintiff, sex is a but-for cause. Answering that question requires courts to confront, rather than obscure, the role of intermediaries: to determine whether they are truly independent of the protected trait, or instead define it, incorporate it, or express it in application. When courts import constitutional neutrality reasoning without undertaking that inquiry, they collapse statutory causation into constitutional classification. That is the mistake in Lange — and it is one the Court should not repeat in B.P.J.