Discrimination is a problem. It is a blight on society and a blemish on personal conduct. During the last one hundred or so years, the fight against discrimination has played an increasingly important role in political discourse, particularly on the left: against racism, sexism, homophobia, transphobia, and white privilege. Nowadays this discourse has its own name: identity politics. We both recognize and repudiate more kinds of discrimination than ever before.

This is as it should be. Undeniably many forms of discrimination exist; and discrimination—depriving people of rights and privileges without legitimate reason—is the enemy of equality and justice. If we are to create a more fair and open society, we must fight to reduce prejudice and privilege as much as we can. Many people are already doing this, of course; and identity politics is rightly here to stay.

And yet, admirable as the goals of identity politics are, I am often dissatisfied with its discourse. Specifically, I think we are often not clear about why certain statements or ideas are discriminatory. Often we treat certain statements as prejudiced because they offend people. I have frequently heard arguments of this form: “As a member of group X, I am offended by Y; therefore Y is discriminatory to group X.”

This argument—the Argument from Offended Feelings, as I’ll call it—is unsatisfactory. First, it is fallacious because it generalizes improperly. It is the same error someone commits when they conclude, from eating bad sushi once, that all sushi is bad: the argument takes one case and applies it to a whole class of things.

Even if many people, all belonging to the same group, find a certain remark offensive, it still is invalid to conclude that the remark is intrinsically discriminatory: this only shows that many people think it is. Even the majority may be wrong—such as the many people who believe that the word “niggardly” comes from the racial slur and is thus racist, while in reality the word has no etymological or historical connection with the racial slur (it comes from Middle English).

Subjective emotional responses should not be given an authoritative place in the question of prejudice. Emotions are not windows into the truth. They are of no epistemological value. Even if everybody in the world felt afraid of me, it would not make me dangerous. Likewise, emotional reactions are not enough to show that a remark is discriminatory. To do that, it must be shown how the remark incorrectly assumes, asserts, or implies something about a certain group.

In other words, we must keep constantly in mind the difference between a statement being discriminatory or merely offensive. Discrimination is wrong because it leads to unjust actions; offending people, on the other hand, is not intrinsically wrong. Brave activists, fighting for a good cause, often offend many.

Thus it is desirable to have logical tests, rather than just emotional responses, for distinguishing discriminatory responses. I hope to provide a few tools in this direction. But before that, here are some practical reasons for preferring logical to emotional criteria.

Placing emotions, especially shared emotions, at the center of any moral judgment makes a community prone to fits of mob justice. If the shared feelings of outrage, horror, or disgust of a group is sufficient to condemn somebody, then we have the judicial equivalent of a witch-hunt: the evidence for the accusation is not properly examined, and the criteria that separate good evidence from bad are ignored.

Another practical disadvantage of giving emotional reactions a privileged place in judgments of discrimination is that it can easily backfire. If enough people say that they are not offended, or if emotional reactions vary from outrage to humor to ambivalence, then the community cannot come to a consensus about whether any remark or action is discriminatory. Insofar as collective action requires consensus, this is an obvious limitation.

What is more, accusations of discrimination are extremely easy to deny if emotional reactions are the ultimate test. The offended parties can simply be dismissed as “over-sensitive” (a “snowflake,” more recently), which is a common rhetorical strategy among the right (and is sometimes used on the left, too). The wisest response to this rhetorical strategy, I believe, is not to re-affirm the validity of emotions in making judgments of discrimination—this leads you into the same trap—but to choose more objective criteria. Some set of non-emotional, objective criteria for determining whether an action is discriminatory is highly desirable, I think, since there is no possibility of a lasting consensus without it.

So if these emotional tests can backfire, what less slippery test can we use?

To me, discriminatory ideas—and the actions predicated on these ideas—are discriminatory precisely because they are based on a false picture of reality: they presuppose differences that do not exist, and mischaracterize or misunderstand the differences that do exist. This is important, because morally effective action of any kind requires a basic knowledge of the facts. A politician cannot provide for his constituents’ needs of she does not know what they are. A lifeguard cannot save a drowning boy if he was not paying attention to the water. Likewise, social policies and individual actions, if they are based on a false picture of human difference, will be discriminatory, even with the best intentions in the world.

I am not arguing that discrimination is wrong purely because of this factual deficiency. Indeed, if I falsely think that all Hungarians love bowties, although this idea is incorrect and therefore discriminatory, this will likely not make me do anything immoral. Thus it is possible, in theory at least, to hold discriminatory views and yet be a perfectly ethical person. It is therefore necessary to distinguish between whether a statement is offensive (it upsets people), discriminatory (it is factually wrong about a group of people), and immoral (it harms people and causes injustice). The three categories do not necessary overlap, in theory or in practice.

It is obvious that, in our society, discrimination is usually far more nefarious than believing that Hungarians love bowties. Discrimination harms people, sometimes kills people; and discrimination causes systematic injustice. My argument is that to prove any policy or idea is intrinsically discriminatory requires proving that it asserts something empirically false.

Examples are depressingly numerous. Legal segregation in the United States was based on the premise that there existed a fundamental difference between blacks and whites, a difference that justified different treatment and physical separation. Similarly, Aristotle argued that slavery was legitimate because some people were born slaves: they were intrinsically slavish. Now, both of these ideas are empirically false. They assert things about reality that are either meaningless, untestable, or contrary to the evidence; and so any actions predicated on these ideas will be discriminatory—and, as it turned out, horrific.

These are not special cases. European antisemitism has always incorporated myths and lies about the Jewish people: tales of Jewish murders of Christian children, of widespread Jewish conspiracies, and so on. Laws barring women from voting and rules preventing women from attending universities were based on absurd notions about women’s intelligence and emotional stability. Name any group which has faced discrimination, and you can find a corresponding myth that attempts to justify the prejudice. Name any group which has dominated, and you can find an untruth to justify their “superiority.”

In our quest to determine whether a remark is discriminatory, it is worth taking a look, first of all, at the social categories themselves. Even superficial investigation will reveal that many of our social categories are close to useless, scientifically speaking. Our understanding of race in the United States, for example, gives an entirely warped picture of human difference. Specifically, the terms “white” and “black” have shifted in meaning and extent over time, and in any case were never based on empirical investigation.

Historically speaking, our notion of what it means to be “white” used to be far more exclusive than it is now, previously excluding Jews and Eastern Europeans. Likewise, as biological anthropologists never tire of telling us, there is more genetic variation in the continent of Africa than the rest of the world combined. Our notions of “white” and “black” simply fail to do justice to the extent of genetic variation and intermixture that exists in the United States. We categorize people into a useless binary using crude notions of skin color. Any policy based on supposed innate, universal differences between “black” and “white” will therefore be based on a myth. Similar criticisms can be made of our common notions of gender and sexual orientation

Putting aside the sloppy categories, discrimination may be based on bad statistics and bad logic. Here are the three errors I think are most common in discriminatory remarks.

The first is to generalize improperly: to erroneously attribute a characteristic to a group. This type of error is exemplified by Randy Newman’s song “Short People,” when he says short people “go around tellin’ great big lies.” I strongly suspect that it is untrue that short people tell, on average, more lies than taller people, which makes this an improper generalization.

This is a silly example, of course. And it is worth pointing out that some generalizations about group differences are perfectly legitimate. It is true, for example, that Spanish people eat more paella than Japanese people. When done properly, generalizations about people are useful and often necessary. The problem is that we are often poor generalizers. We jump to conclusions—using the small sample of our experience to justify sweeping pronouncements—and we are apt to give disproportionate weight to conspicuous examples, thus skewing our judgments.

Our poor generalizations are, all too often, mixed up with more nefarious prejudices. Trump exemplified this when he tweeted a table of statistics of crime rates back in November of 2015. The statistics are ludicrously wrong in every respect. Notably, they claim that more whites are killed by blacks than by other whites, when in reality more whites are killed by other whites. (This shouldn’t be a surprise, since most murders take place within the same community; and since people of the same race tend to live in the same community, most murders are intra-racial.)

The second type of error involved in prejudice is to make conclusions about an individual based on their group. This is a mistake even when the generalizations about the group are accurate. Even if it were statistically true, for example, that short people lied more often than tall people, it would still be invalid to assume that any particular short person is a liar.

The logical mistake is obvious: even if a group has certain characteristics on average, that does not mean that every individual will have these characteristics. On average, Spaniards are shorter than me; but that does not mean that I can safely assume any Spaniard will be shorter than I am. On average, most drivers are looking out for pedestrians; but that doesn’t make I can safely run into the road.

Of course, almost nobody, if they had a half-second to reflect, would make the mistake of believing every single member of a given group had whatever quality. More often, people are just wildly mistaken about how likely a certain person is to have any given quality—most often, we greatly overestimate.

It is statistically true, for example, that Asian Americans tend to do well on standardized math and science exams. But this generalization, which is valid, does not mean you can safely ask any Asian American friend for help on your science homework. Even though Asian Americans do well in these subjects as a group, you should still expect to see many individuals who are average or below average. This is basic statistics—and yet this error accounts for a huge amount of racist and sexist remarks.

Aside from falsely assuming that every member of a group will be characterized by a generalization, the second error also results from forgetting intersectionality: the fact that any individual is inevitably a member of many, intersecting demographic groups. Race, gender, income bracket, sexual orientation, education, religion, and a host of other categories will apply to any single individual. Predicting how the generalizations associated with these categories—which may often make contradictory predictions—will play out in any individual case, is close to impossible.

This is not even to mention all of the manifold influences on behavior that are not included in these demographic categories. Indeed, it is these irreducibly unique experiences, and our unique genetic makeup, that make us individuals in the first place. Humans are not just members of a group, nor even members of many different, overlapping groups: each person is sui generis.

In sum, humans are complicated—the most complicated things in the universe, so far as we know—and making predictions about individual people using statistical generalizations of broad, sometimes hazily defined categories, is hazardous at best, and often foolish. Moving from the specific to the general is fairly unproblematic; we can collect statistics and use averages and medians to analyze sets of data. But moving from the general to the specific is far more troublesome.

The third error is to assert a causal relationship where we only have evidence for correlation. Even if a generalization is valid, and even if an individual fits into this generalization, it is still not valid to conclude that an individual has a certain quality because they belong to a certain group.

Let me be more concrete. As we have seen, it is a valid generalization to say that Asian Americans do well on math and science exams. Now imagine that your friend John is Asian American, and also an excellent student in these subjects. Even in this case, to say that John is good at math “because he’s Asian” would still be illogical (and therefore racist). Correlation does not show causation.

First of all, it may not be known why Asian Americans tend to do better. And even if a general explanation is found—for example, that academic achievement is culturally prized and thus families put pressure on children to succeed—this explanation may not apply in your friend John’s case. Maybe John’s family does not pressure him to study and he just has a knack for science.

Further, even if this general explanation did apply in your friend John’s case (his family pressures him to study for cultural reasons), the correct explanation for him being a good student still wouldn’t be “because he’s Asian,” but would be something more like “because academic achievement is culturally prized in many Asian communities.” In other words, the cause would be ultimately cultural, and not racial. (I mean that this causation would apply equally to somebody of European heritage being raised in an Asian culture, a person who would be considered “white” in the United States. The distinction between cultural and biological explanations is extremely important, since one posits only temporary, environmental differences while the other posits permanent, innate differences.)

In practice, these three errors are often run together. An excellent example of this is from Donald Trump’s notorious campaign announcement: “When Mexico sends its people, they’re not sending their best. … They’re sending people that have lots of problems, and they’re bringing those problems with us [sic.]. They’re bringing drugs. They’re bringing crime. They’re rapists.”

Putting aside the silly notion of Mexico “sending” its people (they come of their own accord), the statement is discriminatory because it generalizes falsely. Trump’s words give the impression that a huge portion, maybe even the majority, of Mexican immigrants are criminals of some kind—and this isn’t true. (In reality, the statistics for undocumented immigrants can put native citizens to shame, as demonstrated here.)

Trump then falls into the third error by treating people as inherently criminal—the immigrants simply “are” criminals, as if they were born that way. Even if it were proven that Mexican immigrants had significantly higher crime rates, it would still be an open question why this was so. The explanation might have nothing to do with their cultural background or any previous history of criminality. It might be found, for example, that poverty and police harassment significantly increased criminality; and in this case the government would share some of the responsibility.

Donald Trump committed the second error in his infamous comments about Judge Gonzalo Curiel, who was overseeing a fraud lawsuit against Trump University. Trump attributed the Curiel’s (perceived) hostility to his Mexican heritage. Trump committed a simple error of fact when he called Curiel “Mexican” (Curiel was born in Indiana), and then committed a logical fallacy when he concluded that the judge’s actions and attitudes were due to his being of Mexican heritage. Even if it were true (as I suspect it is), that Mexican-Americans, on the whole, don’t like Trump, it still doesn’t follow that any given individual Mexican-American doesn’t like him (2^nd error); and even if Curiel did dislike Trump, it wouldn’t follow that it was because of his heritage (3^rd error).

These errors and mistakes are just my attempt at an outline of how discrimination can be criticized on logical, empirical grounds. Certainly there is much more to be said in this direction. What I hoped to show in this piece was that this strategy is viable, and ultimately more desirable than using emotional reactions as a test for prejudice.

Discourse, agreement, and cooperation are impossible when people are guided by emotional reactions. We tend to react emotionally along the lines of factions—indeed, our emotional reactions are conditioned by our social circumstances—so privileging emotional reactions will only exacerbate disagreements, not help to bridge them. In any case, besides the practical disadvantages—which are debatable—I think emotional reactions are not reliable windows into the truth. Basing reactions, judgments, and criticisms on sound reasoning and dependable information is always a better long-term strategy.

For one, this view of discrimination provides an additional explanation for why prejudice is so widespread and difficult to eradicate. We humans have inherited brains that are constantly trying to understand our world in order to navigate it more efficiently. Sometimes our brains make mistakes because we generalize too eagerly from limited information (1^st error), or because we hope to fit everything into the same familiar pattern (2^nd error), or because we are searching for causes of the way things work (3^rd error).

So the universality of prejudice can be partially explained, I think, by the need to explain the social world. And once certain ideas become ingrained in somebody’s worldview, it can be difficult to change their mind without undermining their sense of reality or even their sense of identity. This is one reason why prejudices can be so durable (not to mention that certain prejudices justify convenient, if morally questionable, behaviors, as well as signal a person’s allegiance to a certain group).

I should say that I do not think that discrimination is simply the result of observational or logical error. We absorb prejudices from our cultural environment; and these prejudices are often associated with divisive hatreds and social tension. But even these prejudices absorbed by the environment—that group x is lazy, that group y is violent, that group z is unreliable—always inevitably incorporate some misconception of the social world. Discrimination is not just a behavior. Mistaken beliefs are involved—sometimes obliquely, to be sure—with any prejudice.

This view of prejudice—as caused, at least in part, by an incorrect picture of the world, rather than pure moral depravity—may also allow us to combat it more effectively. It is easy to imagine a person with an essentially sound sense of morality who nevertheless perpetrates harmful discrimination because of prejudices absorbed from her community. Treating such a person as a monster will likely produce no change of perspective; people are not liable to listen when they’re being condemned. Focusing on somebody’s misconceptions may allow for a less adversarial, and perhaps more effective, way of combating prejudice. And this is not to mention the obvious fact that somebody cannot be morally condemned for something they cannot help; and we cannot help if we’re born into a community that instructs its members in discrimination.

Even if this view does not adequately explain discrimination, and even if it does not provide a more effective tool in eliminating it, this view does at least orient our gaze towards the substance rather than the symptoms of discrimination.

Because of their visibility, we tend to focus on the trappings of prejudice—racial slurs, the whitewashed casts of movies, the use of pronouns, and so on—instead of the real meat of it: the systematic discrimination—economic, political, judicial, and social—that is founded on an incorrect picture of the world. Signs and symptoms of prejudice are undeniably important; but eliminating them will not fix the essential problem: that we see differences that aren’t really there, we assume differences without having evidence to justify these assumptions, and we misunderstand the nature and extent of the differences that really do exist.