Notice & Comment

Bostock and Conceptual Causation, by Guha Krishnamurthi & Peter Salib

In Bostock v. Clayton County,[1] the Supreme Court held that an employer who fires an individual merely for being gay or for being transgender violates Title VII.[2] In an opinion authored by Justice Neil Gorsuch, the Supreme Court explained that the Court’s decision was a matter of textualism:

An employer who fires an individual for being homosexual or transgender fires that person for traits or actions it would not have questioned in members of a different sex. Sex plays a necessary and undisguisable role in the decision, exactly what Title VII forbids.

Those who adopted the Civil Rights Act might not have anticipated their work would lead to this particular result. Likely, they weren’t thinking about many of the Act’s consequences that have become apparent over the years, including its prohibition against discrimination on the basis of motherhood or its ban on the sexual harassment of male employees. But the limits of the drafters’ imagination supply no reason to ignore the law’s demands. When the express terms of a statute give us one answer and extratextual considerations suggest another, it’s no contest. Only the written word is the law, and all persons are entitled to its benefit.[3]

Despite Justice Gorsuch’s avowedly textualist opinion in Bostock, the defendants’ core textual argument remains something of a loose thread. Their argument was that the plaintiffs were not being discriminated against “because of . . . sex,” since discrimination against homosexual individuals or transgender individuals is possible irrespective of the individual’s sex.[4]

Justice Alito’s dissent vigorously makes the case that discrimination based on sexual orientation is not discrimination based on sex — that it is a distinct characteristic which Title VII does not address, just as it does not prohibit discrimination on Zodiac sign.[5]

To see the dialectic, consider Bostock himself, a homosexual man. In the majority’s view, there are two relevant characteristics about Bostock:

  1. Bostock is a man.
  2. Bostock is attracted to men.

To determine whether Bostock was discriminated against “because of sex,” we employ the causal test: We counterfactually change one characteristic at a time to see whether Bostock would have been discriminated against. Changing Bostock’s sex, if Bostock were a woman, who was attracted to men, would Bostock have been discriminated against? The answer is seemingly no, because there was no allegation of discrimination against an individual based on being a heterosexual woman. Therefore, on this construction of the causal test, Bostock was discriminated against “because of sex.”

Imagine another construction, suggested by the dissent:

  1. Bostock is a man.
  2. Bostock is a homosexual.

Now, let’s again employ the causal test. We counterfactually change the one characteristic — Bostock’s sex — to determine whether Bostock would have been discriminated against. If Bostock were a woman, who is homosexual, would Bostock have been discriminated against? Perhaps, because it might be the case that homosexuals of any sex are discriminated against. On such a construction of the test, Bostock then would not be discriminated against “because of sex.”

So, which of these constructions is right? This is, indeed, a perplexing question. It critically depends on our understanding of causation.

Suppose a company has a (terrible) policy under which they refuse to hire Black practitioners of the Jewish faith. But, per the policy, you can be Black and belong to any other religion, and you can be Jewish and of any other race. You just cannot be Black and Jewish. Here we might say that the relevant characteristics are:

  1. Plaintiff is Black.
  2. Plaintiff is Jewish.

If we change characteristic (1) or (2), then the plaintiff may be hired under the policy. Thus, under the causal test, we should be able to conclude that the plaintiff is discriminated against “because of . . . race” and “because of . . . religion.” This seems straightforward regarding our use of the term “because.”[6]

Suppose another company has a (terrible) policy under which they will refuse to hire Asian applicants who have any typos in their written application materials. Here too, they are willing to hire Asian people who have typo-less written application materials, and they are willing to hire non-Asian people, whether or not they have typos in their written application materials. Asian applicants are evaluated under a different standard than applicants of any other race. Suppose then an Asian applicant with a few typos in their application materials is passed over for a position, in favor of a White person with some typos in their material — applying the policy. As a matter of causation, the Asian applicant was discriminated against, “because of . . . race” and because of the typos in their application materials.

So at this juncture, we think we can conclude the following point about causation: Whenever there is at least one way to change the outcome by changing at least one protected characteristic — e.g., being Asian, Black, or Jewish — the but-for causation test is satisfied. It does not matter that there might be more than one protected characteristic of this kind — such that changing the candidate’s race or religion gets him hired. Nor does it matter when there is an additional, non-protected characteristic — like making typos — that could change the outcome. You don’t have to be a perfect applicant or worker to be discriminated against.

Now, consider a variation on the religious discrimination described above. A company has a terrible policy of hiring only White Christians and non-White non-Christians. Perhaps, they have a segregationist view of the Christian faith — that it should only be accessible to White people. And suppose they have a term for people that fit the policy: they call them “proper.”

So, a White Buddhist (who is non-Christian) is deemed “improper,” under the policy, and therefore is barred from being hired. These are the relevant features of the plaintiff:

  1. Plaintiff is White.
  2. Plaintiff is non-Christian.
  3. Plaintiff is “improper.”

It seems similarly straightforward that the plaintiff is discriminated against “because of . . . race” and “because of . . . religion.” But one may contend that the plaintiff is just “improper” — and there are non-Whites, and Christians, who can be improper. Thus, changing these protected characteristics does not necessarily result in discrimination. Nevertheless, the presence of the third characteristic — propriety — does not seem relevant to whether the plaintiff is being discriminated against “because of . . . race” and “because of . . . religion.” As above, the causation test is satisfied because the employment outcome can be changed by changing at least one protected characteristic (here, race or religion).

One might object that this example is obvious because race and religion are both protected under Title VII and “impropriety” is just a composition of two protected classes. So let’s slightly alter the example: Suppose a company has a policy of hiring White locals and non-White non-locals, but not non-White locals (because they are segregationists) and not White non-locals (because they want to prefer their local Whites, of which there are many who need jobs). And suppose they call people who fit the policy “normal.” Notably, locality is not a protected category under Title VII.[7]

Suppose a White non-local and non-White local apply. The relevant features of the cases are:

  1. Plaintiff 1 is White.
  2. Plaintiff 1 is non-local.
  3. Plaintiff 1 is “abnormal.”
  1. Plaintiff 2 is non-White.
  2. Plaintiff 2 is local.
  3. Plaintiff 2 is “abnormal.”

So, Plaintiff 2 is different on both the racial and location categories from Plaintiff 1, but still remains abnormal. Abnormality essentially encompasses a particular pairing of traits (viewed binarily), and the pairing that inverts both of those traits. This is similar to sexual orientation, which encompasses a pairing of the characteristics of (i) sex and the (ii) sex that one is attracted to, and the pairing that inverts both of those characteristics.

Just as before, in both cases, it seems straightforward that the plaintiffs are discriminated against “because of . . . race” and because of locality. Changing one protected characteristic — race — changes the outcome. The presence of the “normality” condition is thus not relevant to whether the plaintiffs are discriminated against “because of . . . race.”

*          *          *

Consequently, we observe that the causation question in Bostock has been perplexing because it is structured in the same way as the above examples: Assume that a company’s policy is to discriminate against homosexuals. As with “improper” or “abnormal,” the characteristic “homosexual” might be decomposed into particular combinations of two other characteristics: “sex” and “sexual attraction.” If both characteristics of the plaintiff “sex” and “sexual attraction” are simultaneously changed, then there may be no change in the discriminatory behavior. That is, for example, if the plaintiff is a man, who is attracted to men, and we change that to a plaintiff being a woman, who is attracted to women, then the plaintiff will still be discriminated against.

But this is actually not a novel puzzle. Consider the two-switch problem. In one formulation of this problem, “two switches need to be turned off in order to avert a fire, and that X has a duty to turn off one, Y the other but neither does so and a fire which would have been averted had they both performed their duty breaks out.”[8]

Which caused the fire? Intuitively, it seems that both were causally responsible for the accident, even though neither’s conduct satisfies the but-for causation test.[9]

Similarly, the fact that both characteristics of “sex” and “sexual attraction” combine to create the grounds for discrimination does not mean that neither characteristic is causally relevant to the discrimination. Both are.

It’s important to acknowledge here that though the intuition on causation is relatively robust, a decisive analytic explanation of what constitutes causation in these types of examples has been elusive.[10] Our humble aim is not to comprehensively solve that problem, but rather to provide a sufficient condition for causation.[11]

One way of capturing this result on causation could be to formulate causation as follows: If there is any construction of the causal test such that the plaintiff is being discriminated against on the basis of a protected category, then the plaintiff has satisfied the showing. Thus, in Bostock, because the plaintiff can provide a construction of the causal test under which the plaintiff is discriminated against because of sex, plaintiff is then discriminated against “because of . . . sex” for purposes of the statute.

That may handle this case, but there’s something unsatisfying about that solution. It seems straightforward that discrimination based on sexual orientation is a type of discrimination that is based on sex. That is because whether someone is a homosexual depends on the particular individual’s sex; change their sex, while holding their attractions constant, and you change whether they are a homosexual. Thus, the very concept of sexual orientation takes into account the sex of the person. Similarly, once you hear the definition of “impropriety,” you know that a hiring policy employing it discriminates “because of . . . race” and “because of . . . religion” — because the concept of “impropriety” is just constructed from the religion and race of the individual.

In light of that, we suggest another sufficient test for whether a plaintiff encounters unlawful discrimination. If a putative non-protected basis for discrimination conceptually depends on the protected characteristics of the plaintiff, then the basis for discrimination is “because of” the relevant protected category.[12]

At this point, it may seem as we have answered the question of causation — what “because of” means — by replacing it with the synonym “conceptually depends on.” But conceptual dependence is more instructive than the amorphous notion of causation. We understand conceptual dependence as follows: A conceptually depends on B, if an analysis of A requires an analysis of B.

So, in our examples above, you cannot analyze whether someone is proper without analyzing their race and their religion; and you cannot analyze whether someone is abnormal without analyzing their race and their locality. Thus, propriety conceptually depends on race and religion, and normality depends on race and locality.

Indeed, the majority’s opinion suggests that this is the test it was employing: “There is no way for an applicant to decide whether to check the homosexual or transgender box without considering sex. To see why, imagine an applicant doesn’t know what the words homosexual or transgender mean. Then try writing out instructions for who should check the box without using the words man, woman, or sex (or some synonym). It can’t be done.”[13]

This understanding of causation also explains the Supreme Court’s gender stereotyping jurisprudence, beginning with Price Waterhouse v. Hopkins.[14] These cases hold that it is a Title VII violation to, for example, deny a woman a promotion because she does not wear make-up. Such discrimination has the same structure as the examples above.[15] Under one description, the employer hires only women who wear make-up and men who do not wear it. Although make-up wearing is not a protected characteristic, this is straightforward discrimination for the same reasons it is discrimination to refuse to hire Asian people with typos in their applications.

It does the employer no good to change the description of what is going on by constructing a third conceptual category. Suppose that the employer refuses to hire “gender non-conformists,” which the employer understands to include both women who do not wear make-up and men who do. Applying this new category requires analyzing both the non-protected characteristic of make-up wearing and the protected characteristic of sex.

Consider also the question of discrimination against individuals on the basis of pregnancy: Suppose a company has a policy that nobody who is pregnant will be hired or retained at the company. Our intuition is that this discriminates against females, and thus discriminates because of sex. But is that vindicated by the definition of “because of” as “conceptually depends on”? Arguably yes. Imagine there are two people who show similar changes that are consistent with pregnancy, one is female and pregnant, while one is male and not pregnant. In order for the company to effectuate their policy, they must analyze the sex of the two employees. Ergo, pregnancy is conceptually dependent on sex.

But suppose the company were to object that they don’t care whether the person is female or male — they just care about whether the person is carrying a child. (Perhaps, the company will claim that they are risk averse and do not want any added liability relating to a pregnant worker.) In this way, the company might argue that sex has nothing to do with it, they only care about the characteristic of “carrying a child.” Indeed, the company might say that they don’t rule out scientific miracles — if there were a male pregnancy, like in the hit movie Junior,[16] they would also not hire or retain that employee. This poses a difficult question, but one way of answering this is to examine the concept of sex. The first definition of “female” under Webster’s Dictionary is “of, relating to, or being the sex that typically has the capacity to bear young or produce eggs.”[17] Under that understanding of the female sex, discrimination based on carrying a child is discrimination for being female, and thus discrimination because of sex.

This definition of female, and the resulting analysis, presents a slight variation on conceptual dependence. Here, unlike with “homosexual,” the putative non-protected attribute, “carrying a child,” is not a concatenation of a protected attribute and another (protected or non-protected) attribute. The relationship is more straightforward. The putative non-protected attribute “carrying a child” is so conceptually entangled with the protected attribute “female” that the former logically implies the latter.

Now, we fully recognize that this conception of the female sex is not necessarily canonical. There are other conceptions of female, for example, biological definitions that may focus on genetic chromosomal type and appearance of genitals.[18] Or there may be social definitions that relate to behavior and self-conception.[19] We take no position on that; we recognize that these pose difficult questions beyond our ken. All we want to suggest is that the conceptual dependence understanding of causation provides a plausible answer to how pregnancy discrimination may be a type of discrimination based on sex.[20]

*          *          *

It is worth emphasizing that there is nothing anti-textualist about these arguments. As Professor Anita Krishnakumar shows, textualism does not lead to a single correct reading of this statute.[21] Title VII prohibits discrimination “because of . . . sex,” and the textual question is how to understand “because of.” The examples above simply draw out the observation — commonplace in both legal analysis and everyday life — that causes can be multifactorial. And sometimes we construct terms that encompass collections of causally sufficient factors — ”improper,” “abnormal,” “homosexual,” “gender non-conformist.” But even then, it remains eminently sensible — as a matter of both present linguistic practice and that of 1964 — to count each of the factors subsumed within a constructed term among an event’s causes.

Guha Krishnamurthi is a Climenko Fellow and Lecturer in Law at Harvard Law School and an incoming Assistant Professor at South Texas College of Law.

Peter Salib is a Climenko Fellow and Lecturer in Law at Harvard Law School.

[1] 140 S.Ct. 1731 (2020).

[2] Id. at 1737.

[3] Id.

[4] We use “sex” here as understood by the courts to encompass the notion of biological sex, as well as the concept of gender. Schwenk v. Hartford, 204 F.3d 1187, 1202 (9th Cir. 2000) (“Thus, under Price Waterhouse, ‘sex’ under Title VII encompasses both sex—that is, the biological differences between men and women—and gender.”). Courts have generally treated “sex” as a binary. We recognize that this is problematic, but follow suit for purposes of interpreting the statute.

[5] Id. at 1762 (Alito, J., dissenting).

[6] Indeed, the majority suggests a similar intuition with the example of restricting hiring based on whether someone is Black or Catholic: “Suppose an employer’s application form offered a single box to check if the applicant is either black or Catholic. If the employer refuses to hire anyone who checks that box, would we conclude the employer has complied with Title VII, so long as it studiously avoids learning any particular applicant’s race or religion? Of course not: By intentionally setting out a rule that makes hiring turn on race or religion, the employer violates the law, whatever he might know or not know about individual applicants.” Bostock v. Clayton Cty., Georgia, 140 S. Ct. 1731, 1746 (2020).

[7] 42 U.S. Code § 2000e.

[8] David A. Fischer, Insufficient Causes, 94 Ky. L.J. 277, 313 (2006) (quoting Richard W. Wright, Once More into the Bramble Bush: Duty, Causal Contribution, and the Extent of Legal Responsibility, 54 Vand. L. Rev. 1071, 1127 (2001)).

[9] Id. (citing survey responses to the two-switch case).

[10] See, e.g., Michael S. Moore, Causation and Responsibility: An Essay in Law, Morals, and Metaphysics (2009); Richard W. Wright, Once More into the Bramble Bush: Duty, Causal Contribution, and the Extent of Legal Responsibility, 54 Vand. L. Rev. 1071, 1128 (2001); H.L.A. Hart & Tony Honoré, Causation In The Law 235–49 (2d ed. 1985).

[11] For example, the conceptual definition of causation does not obviously solve the causation question posed by the real-world two-switch problem. One explanation of why each of the switch omissions caused the fire is that neither has causal priority, but both switches would have avoided the fire. David A. Fischer, Insufficient Causes, 94 Ky. L.J. 277, 313 (2006) (quoting Richard W. Wright, Once More into the Bramble Bush: Duty, Causal Contribution, and the Extent of Legal Responsibility, 54 Vand. L. Rev. 1071, 1128 (2001)).

[12] In general, the understanding of causation may relate to a more generalized notion of dependence. In some cases, statistical dependence may ground the notion of causation and in others physical process may ground the notion of causation, and still other accounts remain viable. For more on this, see Jonathan Schaffer, The Metaphysics of Causation, Stan. Ency. Phil. (July 5, 2016), available at For this question, we think conceptual dependence does the best work.

[13] Bostock v. Clayton Cty., Georgia, 140 S. Ct. 1731, 1746 (2020).

[14] 490 U.S. 228 (1989).

[15] Id. at 228–30.

[16] Junior (Northern Lights Entertainment 1994).

[17] (last accessed July 17, 2020).

[18] See, e.g., “sex”, Stedmans Medical Dictionary.

[19] See, e.g., Mari Mikkola, Feminist Perspectives on Sex and Gender, Stan. Ency. Phil. (Oct. 25, 2017), available at (discussing, in Section 3.2, whether sex is purely a biological category).

[20] Indeed, any confusion on whether pregnancy is covered was cleared up by Congress when it passed the Pregnancy Discrimination Act of 1978. That Act amended Title VII to explicitly say, “The terms “because of sex” or “on the basis of sex” include, but are not limited to, because of or on the basis of pregnancy, childbirth, or related medical conditions.” 42 U.S. Code § 2000e. [7/28/2020: The scope of the coverage of the Pregnancy Discrimination Act is potentially an open question. For example, it is unclear whether the following language in the Act that “women affected by pregnancy, childbirth, or related medical conditions shall be treated the same for all employment-related purposes . . . ” would be read to cover trans men affected by pregnancy or childbirth. Id. (emphasis added). However, in light of Bostock, there are strong arguments that it should. We thank Michael Dorf for calling our attention to these points.]

[21] Anita S. Krishnakumar, Three Lessons About Textualism from the Title VII Case, Yale J. Reg. Notice & Comment (June 24, 2020), available at