Race, Ethnicity & Society

The big picture

Start with a puzzle that has no trick to it. A man born in Salvador, Brazil, is classified there as pardo, a broad category usually translated as brown or mixed. He moves to Chicago at nineteen. In the United States, on forms, in traffic stops, in the way strangers address him, he is Black. Nothing about his body changed on the flight. His parents did not change. His DNA did not change. What changed was the classification system he was standing inside.

That observation is the entire starting point of this course, and it is worth being precise about what it does and does not claim. It does not claim that human physical variation is imaginary. Skin pigmentation varies, hair texture varies, facial morphology varies, and populations that were geographically separated for many generations differ in the frequencies of many genetic variants. All of that is real and measurable. The claim is narrower and stronger: that the specific boxes societies sort people into, the number of boxes, the rules for assigning people to them, and the meanings attached to each one, are made by human beings in particular places at particular times, and they change.

By the end of this lesson you should be able to hold two things at once without strain. Human biological variation is real. Race, as a system of discrete categories with social meanings, is a human invention. Those statements are not in tension, and most of the confusion in public argument about race comes from treating them as if they were.

What social construction means

Sociologists call something a social construction when its existence and its properties depend on collective human agreement rather than on nature alone. Money is the standard teaching example. A twenty dollar bill is a piece of cotton and linen with ink on it. Its purchasing power exists because enough people, backed by institutions, treat it as having purchasing power. That does not make the money fake. Try telling a landlord that rent is a social construction.

Race works the same way. The categories are made and maintained by law, administration, custom, and everyday practice. They are also, and this is the part students most often miss, extremely consequential. A category can be invented and still determine whether you can vote, borrow, buy a house on a given street, or walk home without being stopped.

The sociologists W. I. Thomas and Dorothy Swaine Thomas gave us the formulation that captures this, usually called the Thomas theorem: if people define situations as real, they are real in their consequences. Apply it here. Whether or not the category "Black" or "white" or "Asian" describes a natural kind, people acted on those categories for four centuries, and those actions produced deeds, statutes, wage differences, and neighborhoods that still exist. The construction is in the category. The consequences are in the world.

Key idea: Calling race socially constructed is a claim about where the categories came from, not a claim that the categories have no effects. Constructed categories can be, and here plainly are, enormously consequential.

What the genetics actually shows

Now the part that deserves care, because it is the place where careless teaching in both directions has done real damage.

In 1972 the geneticist Richard Lewontin partitioned human genetic variation using the blood group and protein markers then available. He found that roughly 85 percent of the total variation sat within any single population, and only about 15 percent was accounted for by differences among the major geographic groupings people call races. Later work with far more markers has broadly confirmed the shape of that result: most human genetic diversity is found within any given population, not between populations. Two randomly chosen people from the same continent will typically differ from each other about as much as two people drawn from different continents.

That result is real, and it is often over-read. In 2003 the statistician A. W. F. Edwards published a careful objection, now usually referenced as the argument about Lewontin's fallacy. His point was technical and correct: even when each individual marker is a poor classifier, correlations across many markers can allow accurate assignment of individuals to ancestral populations. And in fact, in 2002, Noah Rosenberg and colleagues used 377 markers in more than a thousand people from 52 populations and found that a clustering algorithm, given no information about geography, produced groups that corresponded broadly to major geographic regions.

So which is it? Both, and the resolution is not a compromise but a clarification. Genetic ancestry is real, structured by geography and history, and detectable. What it is not is a set of sharp lines. Human variation across space is largely clinal, meaning it changes gradually as you move across the map rather than jumping at boundaries. The clusters that appear in Rosenberg's analysis appear because the sampling is of populations separated by long distances and major geographic barriers; add populations from the spaces in between and the boundaries blur. Where you draw the lines depends on how many clusters you ask the algorithm to find, which is a decision the researcher makes, not one the data makes for you.

Key idea: Ancestry is real and traceable, and most human genetic variation is within groups rather than between them. Neither fact supports the older idea of discrete biological races, because human variation is graded rather than partitioned.

Why racial categories are bad biology even when ancestry matters

Consider sickle cell trait, which many Americans learn about as a Black disease. The variant is common where falciparum malaria was historically common, because carrying one copy protects against it. That means it is found at elevated frequencies in West and Central Africa, and also in Greece, southern Italy, Turkey, Saudi Arabia, and parts of India. It is rare in southern Africa, where malaria pressure was lower. A South African of Xhosa descent is Black in American terms and unlikely to carry the variant. A Greek from Thessaly is white in American terms and may well carry it. The trait tracks malaria, not race.

Lactase persistence, the ability to digest milk sugar in adulthood, tells the same story from a different angle. It arose independently in northern Europe and in several East African pastoralist populations. It tracks dairying history, not any racial grouping.

These are not curiosities. They are the reason biological anthropologists and geneticists have concluded that racial categories are the wrong tool for describing human biological variation. The American Association of Biological Anthropologists put it plainly in its 2019 statement: race does not provide an accurate representation of human biological variation, and using it as a biological variable produces bad science. That is a professional consensus statement, not a political one, and it is worth reading in the original.

The awkward middle: race in medicine

Here is where students often want a clean answer and cannot have one. If race is not biology, why do clinicians use it?

Sometimes because it is a crude proxy for ancestry that is cheap to collect, sometimes because it is a proxy for exposure and treatment history, and sometimes out of inertia. The most instructive recent case is the estimated glomerular filtration rate, or eGFR, the number used to assess kidney function. For decades the standard equations included a coefficient that raised the estimated kidney function of anyone recorded as Black, based on old and thin evidence about muscle mass and creatinine. The practical effect was to make Black patients look healthier than they were, which delayed referrals and transplant listings. In 2021 a joint task force of the National Kidney Foundation and the American Society of Nephrology recommended a new equation without the race coefficient, and it was widely adopted.

Notice the structure of that story. It is not that race told doctors nothing. It is that a social category was smuggled in as a biological variable, the substitution was never carefully justified, and once examined it turned out to cause harm. The reform was not to ignore ancestry but to stop using a social classification as a stand-in for it.

Key idea: Race can correlate with health outcomes for many reasons, including exposure, access, and stress, so a correlation is not evidence that race is a biological cause. The eGFR case shows how expensive that confusion can be.

The same person, different countries

Return to the man from Salvador. Classification systems differ across societies in ways that would be impossible if they were reading nature off directly.

The United States developed, in law and custom, a version of hypodescent, popularly called the one-drop rule: a person with any known African ancestry was assigned to the subordinate category. Virginia wrote a version of it into statute in 1924. Brazil developed a system with many more terms, more sensitive to appearance and to social standing, in which siblings from the same parents can be described differently. Neither system is more scientific than the other. They are the residues of different histories of slavery, colonization, and law.

South Africa under apartheid ran a third system with a statutory board that reclassified thousands of people between categories, sometimes more than once in a lifetime. If race were simply read off the body, there would be nothing for a board to decide.

Key idea: Racial classification systems vary across societies in their number, their rules, and their criteria, which is the strongest single piece of evidence that they are social products.

What this means for the rest of the course

From here on, we will treat racial categories the way a careful researcher treats any measured variable: as something with a history, a definition, a purpose, and known error. When a study reports a Black-white gap, the first question is always what those labels were built from. Self-identification on a survey? An interviewer's judgment? An administrative record? Those are different measurements and they do not always agree, and later in the course we will look at research showing that a person's recorded race can shift after events such as incarceration or a change in income.

And we will keep one distinction sharp throughout. A descriptive claim reports a measured pattern. A causal claim says what produced it. A value judgment says whether it is acceptable. Most public argument about race collapses these three, and most of the real work of this course is keeping them apart.

Common misconceptions

• Saying race is socially constructed means race is not real. It means the categories are human products. Their effects on income, health, housing, and liberty are as real as anything sociology measures.
• Genetics has proven that races do not exist, so ancestry is meaningless. Ancestry is real, structured, and detectable. What the evidence rejects is the older claim of discrete biological races with sharp boundaries.
• Lewontin settled the question in 1972. His result stands, but Edwards showed that correlated markers permit accurate classification. The honest summary uses both findings, not one.
• Sickle cell is a racial disease. The variant tracks historical malaria exposure and is found in Mediterranean, Middle Eastern, and South Asian populations as well as West African ones.
• If race is constructed, differences in outcomes must be imaginary too. The opposite follows. Constructed categories were used to allocate land, credit, schooling, and punishment, which is exactly why measurable gaps exist.

Recap

• Race is a social construction: the categories, the rules for assignment, and their meanings are made by societies and change over time, while the consequences are entirely real.
• Most human genetic variation lies within populations rather than between them, and human variation is clinal, so it does not partition into discrete biological races.
• Ancestry is real and detectable, and correlated markers permit accurate assignment to ancestral populations, which is a different claim from the existence of biological races.
• Traits such as sickle cell trait and lactase persistence track specific historical pressures, not racial groupings, which is why racial categories are poor biological variables.
• Classification systems differ across the United States, Brazil, and apartheid South Africa, and the Thomas theorem explains how a made category becomes a real force.

Sources

1. Britannica. (2025). Race: Human. In Encyclopaedia Britannica. britannica.com
2. Rosenberg, N. A., Pritchard, J. K., Weber, J. L., Cann, H. M., Kidd, K. K., Zhivotovsky, L. A., and Feldman, M. W. (2002). Genetic structure of human populations. Science, 298(5602), 2381-2385. doi.org
3. American Association of Biological Anthropologists. (2019). AABA statement on race and racism. bioanth.org
4. Wikipedia contributors. (2025). Human genetic variation. In Wikipedia. wikipedia.org
5. National Institute of Diabetes and Digestive and Kidney Diseases. (n.d.). Estimating glomerular filtration rate. National Institutes of Health. niddk.nih.gov
6. American Anthropological Association. (n.d.). Race: Are we so different? understandingrace.org

The big picture

Every ten years the United States government asks every household to sort itself into racial categories. That sounds like a routine administrative act. It is actually one of the most revealing documents in American social history, because the categories change, and each change was a decision made by identifiable people for identifiable reasons.

If race were a fact of nature, the census would look roughly the same in 1790 and 2020. It does not. Categories appear, merge, split, and vanish. Rules for assigning people shift from an enumerator's judgment to the household's own report. Whole populations move from one box to another by administrative decision. Follow that record and you can watch a social construction being constructed.

Today has three tasks. Walk the census from 1790 to the standards revision announced in March 2024. Look at how the courts decided who counted as white when citizenship was at stake. And get the vocabulary straight, because race, ethnicity, and nationality are three different things that ordinary speech runs together, and confusing them will wreck your reading of data later in the course.

The census as a record of category change

The first census in 1790 did not ask about race in any modern sense. It counted free white males in two age brackets, free white females, all other free persons, and slaves. That is a taxonomy built for two purposes written into the Constitution: apportioning the House of Representatives and levying direct taxes. The three-fifths clause meant that counting enslaved people had direct consequences for the political power of slaveholding states, so the categories were, from the first count, instruments of political arithmetic.

What follows is not a smooth story of refinement. It is a series of lurches that track whatever the political and scientific anxieties of the moment happened to be.

Year	What changed	Why it matters
1790	Free whites, other free persons, slaves	Categories built for apportionment, not description
1850	Mulatto added	Reflects theories about racial mixture then current in American science
1890	Quadroon and octoroon added	An attempt at fractional ancestry that enumerators could not apply consistently and that was dropped
1930	Mexican listed as a race	Added under nativist pressure, removed after protest by the Mexican government and Mexican American groups
1960-1970	Self-identification replaces enumerator judgment	Race becomes what you say you are rather than what an official decides
1970-1980	Hispanic origin asked as a separate question	Creates a federal ethnicity distinct from race, a decision with long consequences
1977	OMB Statistical Policy Directive No. 15	Standardizes five race categories plus Hispanic ethnicity across all federal agencies
2000	Respondents may mark one or more races	Ends the assumption that every person belongs to exactly one category
2024	Revised standards: combined race and ethnicity question, new Middle Eastern or North African category	Recognizes that the separate Hispanic question confused respondents and that MENA respondents had no accurate option

Two entries deserve a closer look. The 1930 decision to make "Mexican" a race lasted exactly one census. It was reversed after diplomatic protest and organized objection, and Mexican Americans were returned to the white category, where they remained officially through the middle of the century even as they were segregated in practice in schools, theaters, and swimming pools. That gap between the official category and lived treatment is a recurring theme in this course.

The 2024 revision is the largest change since 1997. The Office of Management and Budget approved combining the race and ethnicity questions into one and adding a Middle Eastern or North African category. The reason was measurement quality: large numbers of Hispanic respondents had been selecting "some other race" because none of the listed races described them, and MENA respondents were being counted as white against their own understanding of themselves. Note what that tells you. The federal government revised its racial categories because respondents kept refusing to fit them.

Key idea: Federal racial categories are administrative instruments that have been repeatedly redesigned in response to politics, science, protest, and measurement failure, which is direct documentary evidence that they are made rather than discovered.

Who counted as white, and who decided

The stakes were never merely descriptive. The Naturalization Act of 1790 restricted naturalized citizenship to "free white persons." After the Civil War, eligibility was extended to persons of African nativity or descent. That left everyone else in an odd position: to become a citizen, you had to persuade a court that you were white.

Two Supreme Court cases decided a year apart show how that worked, and they are worth reading together because their reasoning is incompatible.

In Ozawa v. United States (1922), Takao Ozawa, a Japanese immigrant who had lived in the United States for twenty years, argued in part that his skin was as light as that of many Europeans. The Court rejected the claim and held that white meant Caucasian, a term drawn from the racial science of the day.

In United States v. Bhagat Singh Thind (1923), a Sikh immigrant from Punjab argued that anthropologists of the period classified North Indians as Caucasian, which by the reasoning of Ozawa should have settled it. The Court rejected that too. This time it held that the statutory word white must be read in accordance with the understanding of the common man, not the classifications of scientists.

Put the two together and the pattern is unmistakable. When science supported exclusion, the Court used science. When science supported inclusion, the Court used popular understanding. The category was doing political work, and its definition was adjusted to keep producing the desired result. Thind's citizenship was revoked, as was that of dozens of other South Asians, and some lost land they had bought under alien land laws that turned on citizenship eligibility.

Key idea: Legal whiteness was decided case by case with shifting criteria, which shows the category being manufactured in real time by institutions with something at stake.

Did the Irish and Italians become white?

Here is a genuine scholarly disagreement, and you should know both sides rather than the slogan.

A body of work sometimes called whiteness studies argued that European immigrant groups arrived as something less than fully white and achieved whiteness over generations. Noel Ignatiev's How the Irish Became White (1995) traced how Irish immigrants, arriving into intense hostility, secured position partly by aligning with the racial order against Black workers. David Roediger's The Wages of Whiteness (1991) argued that white racial identity gave workers a psychological and social wage that blunted class solidarity. Karen Brodkin made a parallel argument about Jewish Americans and the postwar expansion of housing and education benefits.

The critics are serious and their objection is empirical. The historian Thomas Guglielmo, in White on Arrival (2003), examined Chicago's Italians and concluded that they were legally and institutionally white from the moment they landed: they could naturalize, vote, serve on juries, buy in restricted areas, and were never subject to the exclusion laws aimed at Asians or the segregation laws aimed at Black Americans. What they faced was intense ethnic prejudice, which is a real thing and a different thing. Eric Arnesen argued that whiteness studies used the term so loosely that it explained everything and therefore nothing. Barbara Fields pressed a related objection: that treating whiteness as a possession people acquire obscures the political and economic interests actually at work.

Where does this leave you? With a distinction worth carrying. Legal and institutional standing is one variable. Social acceptance and everyday prejudice are another. Southern and Eastern European immigrants around 1900 scored low on the second and high on the first, and the difference between low social standing and legal exclusion is exactly the difference between their trajectory and that of Chinese immigrants under the exclusion laws.

Race, ethnicity, nationality

Now the vocabulary, which the previous sections have been quietly setting up.

Race, sociologically, is a category based on perceived physical characteristics to which a society attaches meaning and, historically, hierarchy. It is imposed from outside as much as it is claimed from inside.

Ethnicity is a category based on shared culture: language, religion, ancestry, custom, cuisine, and a sense of common descent. Ethnicity is more often claimed than imposed, and it is usually more voluntary. A third-generation Italian American can foreground or ignore that identity in a way that a Black American generally cannot ignore race, an asymmetry the sociologist Mary Waters documented in her work on what she called symbolic ethnicity among white Americans.

Nationality is legal membership in a state. It is documented, it can be changed by application, and it is the only one of the three with a passport office.

Now watch them come apart in real cases. A Nigerian doctor who immigrates to Houston has Nigerian nationality until she naturalizes, Yoruba or Igbo ethnicity, and in American terms is racialized as Black, a category she may not have used about herself at all before arriving. Federal statistics count Hispanic as an ethnicity and not a race, which is why a Dominican American may accurately mark Hispanic ethnicity and Black race, and why Hispanic respondents are found in every race category in census data. A Uyghur and a Han citizen of China share nationality and differ in ethnicity, and Chinese state classification treats that difference as consequential in ways that look racial to outside observers.

Key idea: Race is largely imposed and tied to perceived physical markers, ethnicity is largely claimed and tied to culture, and nationality is legal membership in a state; using them interchangeably will make you misread almost any dataset.

Why measurement decisions move the numbers

One last practical warning. Between 2010 and 2020 the population reporting more than one race in the census grew from about 9 million to about 33.8 million. That looks like an extraordinary demographic event. Much of it was not. The Census Bureau redesigned the race questions and the coding of write-in responses, and explicitly cautioned that changes to the questions and processing, along with real demographic change, both contributed to the shift.

So when you see a dramatic change in a racial statistic, your first question should be whether the instrument changed. That is not skepticism for its own sake. It is the ordinary discipline of reading measured data, and it applies with special force to a variable whose definition is set by committee.

Common misconceptions

• Census racial categories reflect scientific classification. They reflect administrative and political decisions, revised repeatedly, most recently in March 2024.
• Hispanic is a race in United States federal data. Under the standards in effect since 1977 it has been an ethnicity asked separately, which is why Hispanic respondents appear within every race category; the 2024 revision combines the questions.
• The courts applied a consistent scientific test for whiteness. Ozawa used scientific classification and Thind rejected it one year later, because the criterion was adjusted to reach the same exclusionary result.
• Everyone agrees the Irish and Italians were not white on arrival. Guglielmo and Arnesen argue they held legal and institutional whiteness immediately while facing severe ethnic prejudice, which is a different claim from racial exclusion.
• The multiracial population tripled between 2010 and 2020. Question redesign and coding changes account for much of the measured jump, as the Census Bureau itself warned.

Recap

• The census has revised its racial categories repeatedly since 1790, including adding and dropping mulatto, quadroon, and octoroon, briefly making Mexican a race in 1930, moving to self-identification by 1970, and allowing multiple races from 2000.
• OMB Directive No. 15 standardized federal categories in 1977, and the March 2024 revision combined race and ethnicity into one question and added a Middle Eastern or North African category.
• Naturalization was limited to free white persons from 1790, and Ozawa and Thind show courts switching criteria between science and common understanding to keep excluding Asian applicants.
• Whether European immigrants became white is genuinely contested: whiteness studies says yes over generations, while Guglielmo and Arnesen argue they held legal whiteness on arrival while facing intense ethnic prejudice.
• Race is imposed and physical, ethnicity is claimed and cultural, nationality is legal, and changes in measurement instruments can move racial statistics more than real demographic change does.

Sources

1. United States Census Bureau. (n.d.). About the topic of race. census.gov
2. Wikipedia contributors. (2025). Race and ethnicity in the United States census. In Wikipedia. wikipedia.org
3. Ozawa v. United States, 260 U.S. 178 (1922). supreme.justia.com
4. United States v. Bhagat Singh Thind, 261 U.S. 204 (1923). supreme.justia.com
5. Britannica. (2025). Ethnic group. In Encyclopaedia Britannica. britannica.com
6. Pew Research Center. (n.d.). Race and ethnicity. pewresearch.org

The big picture

Three words do most of the work in public argument about race, and they are used almost interchangeably: prejudice, discrimination, racism. In sociology they name three different things, and the difference is not pedantry. Two people can agree completely on the facts of a case and disagree bitterly about whether it counts as racism, purely because they are using the third word differently. Most of the heat in American racial argument is generated at exactly this point.

So today we build the vocabulary carefully, and then we go straight at the hardest definitional fight in the field, which is over what racism means when no individual involved is prejudiced. You will not get a ruling from me on that fight. You will get the strongest version of each position and the evidence each side leans on, because that is what a sociology course owes you on a question the discipline has not settled.

Three words, three different things

A stereotype is a cognitive thing: an oversimplified generalization about a group, applied to individual members of it. Stereotypes can be positive in content and still cause harm, as with the model minority stereotype we will examine later. They can also be statistically defensible in the aggregate and still be unjust when applied to a person, which is a point worth sitting with, because it is where much of the analytical difficulty in this field lives.

A prejudice is an affective thing: a prejudgment, usually negative, about a group and its members, held in advance of and often in defiance of evidence. Gordon Allport's The Nature of Prejudice (1954) is still the standard starting point, and his central insight was that prejudice is not merely an error of fact but an emotionally invested error resistant to correction.

Discrimination is a behavioral thing: differential treatment of people because of their group membership. Note that discrimination is an action, which is why it is the only one of the three that laws can directly reach. You cannot legislate someone's attitude. You can legislate what they do with it at a hiring desk.

Key idea: Stereotype is a belief, prejudice is an attitude, and discrimination is a behavior; keeping them separate is what allows you to ask the interesting question of when and why they come apart.

Merton's typology, and why attitudes predict behavior badly

In 1949 Robert K. Merton published a small, powerful table that crosses attitude with behavior and yields four types. It remains one of the most useful teaching tools in the field.

Type	Prejudiced?	Discriminates?	Merton's label
1	No	No	The all-weather liberal, whose conduct matches their beliefs
2	No	Yes	The fair-weather liberal, who goes along with discrimination when it is costly not to
3	Yes	No	The timid bigot, restrained by law, cost, or social pressure
4	Yes	Yes	The active bigot, whose conduct matches their prejudice

Types 2 and 3 are the interesting ones, and they carry an enormous policy implication. If type 3 exists, then law can reduce discrimination without changing a single heart. If type 2 exists, then reducing prejudice is not sufficient to reduce discrimination, because people discriminate for reasons of conformity, incentive, and institutional routine.

The empirical backing for the gap between attitude and action goes back to a strange and famous 1934 study. The sociologist Richard LaPiere traveled the United States with a young Chinese couple, stopping at more than 250 hotels and restaurants. They were refused service exactly once. Months later LaPiere wrote to the same establishments asking whether they would accept Chinese guests. More than 90 percent of those who replied said no. The study has real methodological problems, including that the person answering the letter was probably not the person who had served them, but its central finding has held up in a century of subsequent work: stated attitudes are weak predictors of situated behavior.

Key idea: Attitude and behavior come apart in both directions, which means anti-discrimination policy and prejudice-reduction efforts are addressing different targets and neither substitutes for the other.

From individual acts to institutional patterns

In 1967 Stokely Carmichael and Charles Hamilton, in Black Power, drew a distinction that reorganized the field. Individual racism, they wrote, is the sort of thing a person does: an insult, a refusal, a bombing. Institutional racism operates through the ordinary functioning of established institutions, and it can produce the same harms with no identifiable villain and no one breaking a rule.

Their example was blunt. When white terrorists bomb a church and kill children, that is individual racism and the society condemns it. When the same city's Black infants die at higher rates because of housing, sanitation, and medical care allocated over decades, that is institutional racism, and no one is arrested.

American law absorbed a version of this distinction. Under Title VII of the Civil Rights Act, disparate treatment means intentionally treating a person worse because of a protected characteristic. Disparate impact means using a facially neutral practice that falls more harshly on a protected group and is not justified by business necessity. The Supreme Court established the disparate impact framework in Griggs v. Duke Power Co. (1971), where a high school diploma requirement and an aptitude test screened out Black applicants at much higher rates for jobs the tests did not predict performance in. Chief Justice Burger's opinion said the Act proscribes practices that are fair in form but discriminatory in operation.

Notice what Griggs did and did not do. It did not require proof of intent. It also did not treat a disparity by itself as a violation; the employer could defend the practice by showing it was job related and consistent with business necessity. That two-step structure, disparity establishes a case and justification can rebut it, is a good model for thinking about the whole area.

The fight over the structural definition

Here is the live argument. In much contemporary sociology, and in Eduardo Bonilla-Silva's influential 1997 formulation, racism is best understood not as a set of individual attitudes but as a property of a racialized social system: a society whose economic, political, and social structures are organized in part by racial categories, generating advantage and disadvantage regardless of anyone's intentions.

The strongest case for that move has four parts. First, it explains persistence: gaps in wealth and neighborhood quality continued shrinking slowly or not at all after open bigotry became socially costly, which an attitude-based account struggles with. Second, it directs research toward mechanisms that are measurable, such as inherited wealth, zoning, school finance, and network-based hiring. Third, it matches the law's own recognition in Griggs that harm can occur without intent. Fourth, it accounts for cases where every individual acts reasonably and the aggregate outcome is still stratified, which is a real and common pattern.

The strongest case against is not a denial that structures matter. It has three distinct strands, and they are worth separating.

The first is semantic. Racism, in ordinary English, carries a strong moral charge tied to intent. Using the same word for an intentional slur and for an unintended aggregate outcome guarantees that speakers will talk past one another, and it invites the accusation of bad faith where none was meant. Some scholars who fully accept structural analysis prefer terms like structural inequality or racial stratification precisely to keep the moral and the mechanical apart.

The second is methodological, and it is the sharper objection. If racism is defined as the existence of a racial disparity, then the question of what caused the disparity is settled by definition rather than by evidence. The economist Roland Fryer has called the move from disparity to discrimination the disparity fallacy, and the point holds regardless of one's politics: a gap is a description, and identifying its causes requires the same work as any other causal question. Serious structural scholars agree with this and specify mechanisms. The objection is to the loose usage that skips that step.

The third is about remedy. If the term covers everything from a hiring manager's animus to the compounding of nineteenth-century property law, then a single word is being asked to guide radically different interventions. Naming the specific mechanism, one critic's version goes, is more useful than naming the category.

There is also a distinction from the economist Glenn Loury that repays study because it cuts across the political lines. Loury separates discrimination in contract, unequal treatment in formal transactions like hiring, lending, and sentencing, from discrimination in contact, the unequal informal association that governs who mentors whom, who marries whom, who hears about the job opening. Contract discrimination is illegal and shrinking. Contact discrimination is legal, is protected by freedom of association, and reproduces advantage powerfully through networks and neighborhoods. If Loury is right, a society could eliminate every illegal act of discrimination and still transmit substantial inequality across generations. That framing is compatible with structural analysis and does not require the contested definition of racism to do its work.

Key idea: The structural definition of racism is defended as necessary to explain persistence without villains and criticized for merging a moral term with a causal claim; the productive move in either case is to name the specific mechanism rather than the category.

Prejudice plus power

One more contested definition, since you will meet it. A widely taught formula holds that racism equals prejudice plus power, from which it follows that members of a subordinated group can be prejudiced but not racist.

The case for it is that the word should track the capacity to impose consequences. A slur from someone with no institutional leverage does different damage than the same slur from someone who controls hiring, and a definition that treats them identically obscures the asymmetry the field exists to study.

The case against runs on two tracks. Descriptively, power is not a single quantity: a person can be a numerical minority nationally and hold considerable local institutional authority, so the formula is hard to apply. Morally and legally, American antidiscrimination law is symmetric on its face, and many people believe that a principle against judging by race should apply to everyone or it is not a principle. Sociologists who reject the formula generally still accept that power asymmetry is the central empirical fact; they object to writing it into the definition of the word.

Your job in this course is not to adopt one of these definitions. It is to notice which one a given speaker is using, and to say so before arguing.

Common misconceptions

• Prejudice and discrimination are the same thing. One is an attitude and one is a behavior, and Merton's types 2 and 3 exist precisely because they can occur separately.
• If no one intended harm, no discrimination occurred. Under Griggs, a neutral practice with disparate impact and no business justification can violate Title VII without any proof of intent.
• A racial disparity by itself proves discrimination. A disparity is a description; establishing its cause requires the same evidence any causal claim requires, which is a point made by structural and skeptical scholars alike.
• Structural accounts of racism deny individual agency or blame all white people. The standard formulation locates the cause in institutional arrangements rather than in individual character, which is close to the opposite claim.
• Everyone in sociology uses racism to mean the same thing. The definition is actively contested inside the discipline, and confusing the individual and structural senses is the single most common source of talking past each other.

Recap

• Stereotype is a belief, prejudice is an attitude, and discrimination is a behavior, and only the third is directly reachable by law.
• Merton's typology gives four combinations of prejudice and discrimination, and LaPiere's 1934 study is the classic demonstration that stated attitudes predict situated behavior poorly.
• Carmichael and Hamilton distinguished individual from institutional racism in 1967, and Griggs v. Duke Power (1971) built a legal version through disparate impact with a business necessity defense.
• The structural definition is defended for explaining persistence and criticized for fusing a moral term with an untested causal claim; naming the mechanism avoids most of the dispute.
• Loury's distinction between discrimination in contract and discrimination in contact identifies a legal, protected mechanism that can reproduce inequality even where illegal discrimination has ended.

Sources

1. OpenStax. (2021). Prejudice, discrimination, and racism. In Introduction to sociology 3e. Rice University. openstax.org
2. Britannica. (2025). Racism. In Encyclopaedia Britannica. britannica.com
3. Griggs v. Duke Power Co., 401 U.S. 424 (1971). supreme.justia.com
4. Wikipedia contributors. (2025). Institutional racism. In Wikipedia. wikipedia.org
5. U.S. Equal Employment Opportunity Commission. (n.d.). Title VII of the Civil Rights Act of 1964. eeoc.gov
6. Wikipedia contributors. (2025). Glenn Loury. In Wikipedia. wikipedia.org

The big picture

Everything else in this course is about groups. This lesson is about nations, and the difference is not rhetorical.

When the United States dealt with the Cherokee, the Lakota, or the Muscogee, it did so by treaty, the same instrument it used with France and Spain. Congress ratified roughly 370 of them between 1778 and 1871. Under Article VI of the Constitution, ratified treaties are the supreme law of the land. That means tribal nations entered American law not as a minority population but as sovereigns with a government-to-government relationship, and that status survives today in a body of law that has no parallel for any other group.

Getting this straight matters practically. Programs and preferences for tribal members have repeatedly been challenged as racial classifications and repeatedly upheld as political ones, because tribal citizenship is membership in a polity, not a racial designation. If you analyze Indian Country as though it were simply another minority-group question, you will get the law wrong, the history wrong, and the policy analysis wrong.

Today we follow the machinery of dispossession from the treaty era to the present, name each mechanism, and end with what the resulting legal structure produces now.

The Marshall Trilogy: sovereignty defined and diminished

Three Supreme Court decisions written by Chief Justice John Marshall between 1823 and 1832 still frame federal Indian law, and they pull in different directions.

Johnson v. M'Intosh (1823) imported the doctrine of discovery, holding that European discovery gave the discovering sovereign title to the land, leaving tribes with a right of occupancy that only the federal government could extinguish. Whatever one thinks of the reasoning, the practical effect was to make tribes unable to sell land to anyone but the United States, which removed the ordinary leverage of a landowner.

Cherokee Nation v. Georgia (1831) held that tribes are neither foreign nations nor states but domestic dependent nations, whose relation to the United States Marshall described as resembling that of a ward to a guardian. That phrase became the root of the federal trust responsibility, which cuts both ways in practice: it has justified paternalistic control, and it has also been the legal basis for successful tribal claims against the United States.

Worcester v. Georgia (1832) was the strong one. The Court held that the Cherokee Nation was a distinct community within which Georgia law had no force. It should have stopped removal. President Andrew Jackson declined to enforce it, and Georgia proceeded. The episode is a permanent lesson about the distance between a legal holding and an outcome.

Key idea: Federal Indian law rests on Marshall's framing of tribes as domestic dependent nations with a trust relationship, a status that is political rather than racial and that has been used both to constrain tribes and to protect them.

Removal, and the acreage record of allotment

The Indian Removal Act of 1830 authorized the president to negotiate the exchange of eastern lands for territory west of the Mississippi. In practice the negotiations were conducted with factions of dubious authority and enforced by the army. Between 1830 and 1850 the Cherokee, Muscogee, Choctaw, Chickasaw, and Seminole nations were removed from the Southeast. Roughly 16,000 Cherokee were forced west in 1838 and 1839; estimates of deaths on the route the Cherokee call the trail where they cried run to about 4,000, and the other nations suffered comparable losses. These are not contested facts. They are the documentary record of the Indian Office, the army, and the missions.

The second great mechanism was quieter and took more land. The General Allotment Act of 1887, usually called the Dawes Act, broke communally held reservation land into individual parcels, typically 160 acres to a head of household, and declared the remainder surplus and available for sale to non-Indians. Its supporters, including reformers who considered themselves friends of the Indian, argued that private property would produce prosperity and assimilation.

Here is what it produced. The tribal land base fell from roughly 138 million acres in 1887 to about 48 million by 1934, a loss of nearly two-thirds in fewer than fifty years. Much of the remainder was arid or otherwise marginal. Allotments passed to heirs undivided, so that a single 160-acre parcel today can have hundreds of fractional owners, making it nearly impossible to use or lease efficiently. Federal mismanagement of the resulting trust accounts produced Cobell v. Salazar, settled in 2009 for 3.4 billion dollars, one of the largest settlements against the United States government.

Key idea: Allotment took more land than removal did, and it did so through property law rather than force, which is a clean illustration of how an administrative mechanism can outperform an army.

The boarding schools

Beginning with the Carlisle Indian Industrial School in 1879, the federal government funded and operated a system of off-reservation boarding schools whose explicit purpose was to eliminate Native languages, religions, and family structures. Carlisle's founder, Richard Henry Pratt, summarized the philosophy in a phrase that requires no interpretation: kill the Indian, save the man. Attendance was compelled in many cases by withholding rations or by direct removal of children.

This is not disputed history recovered by activists. In 2022 the United States Department of the Interior published the first volume of its own investigative report, which identified 408 federal Indian boarding schools across 37 states operating between 1819 and 1969, documented marked and unmarked burial sites at dozens of them, and recorded the deaths of hundreds of children, a figure the report expected to rise with further investigation. A second volume followed. The department that ran the schools investigated and published the record.

Citizenship, reorganization, termination, self-determination

The twentieth century swung between opposite policies roughly every generation, and the swings themselves are part of the explanation for the conditions we will measure later.

Period	Policy	Effect
1887-1934	Allotment and assimilation	Land base falls from about 138 million to about 48 million acres
1924	Indian Citizenship Act	United States citizenship extended to all Native people born in the country, without ending tribal citizenship
1934	Indian Reorganization Act	Allotment ends, tribal governments are reconstituted, often on templates written in Washington
1953-1968	Termination and relocation	Congress terminates federal recognition of more than one hundred tribes; Public Law 280 transfers criminal jurisdiction to certain states; urban relocation programs move families to cities
1975-present	Self-determination	Tribes contract to run their own schools, clinics, and courts; ICWA (1978) and NAGPRA (1990) follow

Termination deserves a sentence of its own because it is so often forgotten. House Concurrent Resolution 108 in 1953 declared the policy of ending federal supervision. The Menominee of Wisconsin and the Klamath of Oregon, both relatively prosperous, were terminated; both lost their land base and their services, and both fought their way back to restoration, the Menominee in 1973. The episode is a controlled experiment of a grim kind: removing the trust relationship did not produce integration into prosperity.

Key idea: Federal policy reversed direction roughly every thirty years, and each reversal imposed new costs, so instability itself is one of the mechanisms producing present-day conditions in Indian Country.

Political, not racial: why the distinction has teeth

In Morton v. Mancari (1974) the Supreme Court considered a hiring preference for Indians in the Bureau of Indian Affairs. Non-Indian employees argued it was racial discrimination. The Court unanimously disagreed, holding that the preference was tied to membership in quasi-sovereign tribal entities and was therefore a political rather than a racial classification, subject to ordinary rational basis review rather than strict scrutiny.

Mancari is the hinge on which a great deal turns. It is why tribal gaming under the Indian Gaming Regulatory Act of 1988 is not an ethnic business preference, why the Indian Child Welfare Act of 1978 could be upheld in Haaland v. Brackeen in 2023, and why tribal citizenship rules are set by tribes rather than by federal racial standards. Some tribes use lineal descent, some use blood quantum thresholds inherited from federal rolls, and the disagreements about that are internal sovereign matters with real stakes.

What the structure produces now

There are 574 federally recognized tribes. The jurisdictional map they operate in is genuinely labyrinthine, and the complexity is itself a consequence of the history above. Oliphant v. Suquamish (1978) held that tribes lack criminal jurisdiction over non-Indians on tribal land. Since a large share of violent crimes against Native women are committed by non-Native men, that ruling created an enforcement gap that Congress has only partially closed through the 2013 and 2022 reauthorizations of the Violence Against Women Act. McGirt v. Oklahoma (2020) held that the Muscogee reservation was never disestablished, reshaping criminal jurisdiction across eastern Oklahoma, and Oklahoma v. Castro-Huerta (2022) narrowed it two years later.

The outcomes are measurable and severe. American Indian and Alaska Native life expectancy at birth was 67.9 years in 2022, the lowest of any group the National Center for Health Statistics reports and nearly a decade below the national figure. Poverty rates run well above the national average. And here a data caution matters: racial misclassification on death certificates causes AIAN mortality to be undercounted in official statistics, so the true disparities are probably somewhat larger than published figures show.

Key idea: Present-day conditions in Indian Country trace to identifiable legal mechanisms, land tenure, jurisdiction, and policy instability, which means they are matters of law and administration before they are matters of culture.

Common misconceptions

• Tribal programs are racial preferences. Morton v. Mancari held that classifications based on tribal membership are political, tied to citizenship in quasi-sovereign entities.
• Treaties are historical curiosities with no current force. Ratified treaties remain supreme law under Article VI, and modern cases on water, fishing, and land turn on their text.
• Removal was the main cause of land loss. Allotment took far more: from about 138 million acres in 1887 to about 48 million by 1934.
• The boarding school record is contested. The Interior Department's own 2022 investigation identified 408 schools, documented burial sites, and recorded child deaths.
• Federal Indian policy has moved steadily toward self-determination. It reversed direction repeatedly, including a termination era in the 1950s and 1960s that stripped recognition from more than one hundred tribes.

Recap

• Roughly 370 ratified treaties made tribal nations parties to supreme federal law, establishing a political relationship distinct from minority-group status.
• The Marshall Trilogy created the discovery doctrine, the domestic dependent nation category, and the trust responsibility, and Worcester showed that a favorable ruling can go unenforced.
• The Dawes Act of 1887 cut the tribal land base by roughly two-thirds by 1934 and created the fractionated heirship problem behind the 3.4 billion dollar Cobell settlement.
• The Interior Department documented 408 federal Indian boarding schools operating between 1819 and 1969, with burial sites and recorded child deaths.
• Morton v. Mancari makes tribal classifications political rather than racial, and jurisdictional rulings from Oliphant to McGirt and Castro-Huerta continue to shape enforcement and safety today.

Sources

1. National Archives. (n.d.). Indian Removal Act (1830). archives.gov
2. National Archives. (n.d.). Dawes Act (1887). archives.gov
3. National Park Service. (n.d.). Trail of Tears National Historic Trail. nps.gov
4. Bureau of Indian Affairs. (2022). Federal Indian Boarding School Initiative. U.S. Department of the Interior. bia.gov
5. Morton v. Mancari, 417 U.S. 535 (1974). supreme.justia.com
6. Worcester v. Georgia, 31 U.S. 515 (1832). supreme.justia.com
7. Britannica. (2025). Trail of Tears. In Encyclopaedia Britannica. britannica.com

The big picture

This lesson covers about two hundred and fifty years of slavery and roughly ninety years of legal segregation, which is obviously impossible. So the goal is not coverage. It is to give you the specific, checkable facts that the rest of the course depends on, and to show you exactly where the documented record ends and where genuine scholarly argument begins.

Those are different things, and I want you to be able to tell them apart. That slavery existed, that it was hereditary and racial, that Jim Crow was written into statute, that Black voters were removed from the rolls by identifiable legal devices: these are documented, uncontested, and available in primary sources you can read yourself. How much slavery contributed to American economic growth, and how to weigh it against other factors: that is contested among serious economic historians right now, and you should know why.

The scale and the law

Begin with numbers you can verify. The Trans-Atlantic Slave Trade Database, assembled from shipping records across several countries, documents roughly 12.5 million people embarked from Africa and about 10.7 million surviving the crossing. Of those, roughly 389,000 disembarked in mainland North America. The overwhelming majority went to Brazil and the Caribbean, a fact worth holding onto, because the enslaved population of the United States grew mainly through natural increase rather than importation.

By the 1860 census, 3.95 million people were enslaved in the United States, about one in eight of the national population and about a third of the population of the states that seceded.

The legal architecture is where the racial character of American slavery was made. In 1662 Virginia enacted the rule that a child's status followed the mother, partus sequitur ventrem, reversing English common law's rule of paternal descent. The effect was to make slavery inheritable through the mother and to convert the sexual exploitation of enslaved women into a source of property. Successive slave codes barred literacy, testimony against white persons, and legal marriage. The Constitution counted three fifths of the enslaved for apportionment, obliged the return of fugitives, and protected the international trade until 1808.

The end of legal importation in 1808 did not shrink slavery. It redirected it. Between 1808 and 1865 roughly one million people were sold from the Upper South to the cotton frontier of the Deep South, a forced migration that historians call the second middle passage. It separated an estimated one in three enslaved marriages and one in five enslaved children from a parent in the Upper South, and it was conducted by a professional, advertised, legal industry.

Key idea: American slavery was made racial and hereditary by specific statutes, and after 1808 it expanded through a domestic trade that broke families as a routine business practice.

The economics, and the argument historians are actually having

In 1860, raw cotton made up roughly 60 percent of the value of American exports. The commonly cited estimate of the market value of enslaved people that year is about 3 billion dollars, which would exceed the combined value of the nation's railroads and factories. Slavery was not a marginal southern institution; it was collateral for loans, a basis for insurance policies, and the supply chain for New England and British textile mills.

Now the disagreement. In 1974 Robert Fogel and Stanley Engerman published Time on the Cross, arguing with quantitative methods that southern slave agriculture was efficient and profitable and would not have collapsed on its own. The efficiency finding survived; several of their claims about material conditions and about the frequency of punishment drew sustained criticism, notably in Reckoning with Slavery (1976), and Fogel revised much of the argument later.

The current argument is between what is often called the new history of capitalism and a group of economic historians. Edward Baptist's The Half Has Never Been Told (2014) and Sven Beckert's Empire of Cotton (2014) argued that slavery was the engine of American and global capitalist development, and Baptist attributed rising cotton picking rates chiefly to escalating torture.

Alan Olmstead and Paul Rhode challenged both claims directly. Their work on cotton seed varieties shows that biological innovation, southern planters selecting and diffusing higher-yielding, easier-picking upland cotton strains, accounts for a large share of the productivity increase Baptist attributed to violence. They also argued that Baptist's calculation of slavery's share of GDP double counted, and Gavin Wright, no apologist for the institution, has argued that slave wealth was not the primary driver of national growth and that the South's economy stagnated relative to the free-labor North.

Notice what is and is not in dispute. No participant denies that slavery was violent, coerced, profitable, or central to the South. The argument is about causal weight in national economic growth and about the specific mechanism behind productivity gains. That is exactly the kind of question evidence can adjudicate, and it has not been settled yet.

Key idea: Slavery's brutality and profitability are documented and undisputed; its causal share in American economic growth and the source of rising cotton productivity are live quantitative disputes among economic historians.

Reconstruction: what was actually built

Emancipation came through the Emancipation Proclamation, the flight of hundreds of thousands of people to Union lines, and finally the Thirteenth Amendment in 1865. The Fourteenth followed in 1868 with birthright citizenship, due process, and equal protection, and the Fifteenth in 1870 barred denial of the vote on grounds of race.

What happened next is the part most students have never been taught. Black men registered and voted in enormous numbers. Roughly 2,000 Black men held public office during Reconstruction, from local justices to two United States senators from Mississippi, Hiram Revels and Blanche Bruce, and sixteen members of Congress. Reconstruction legislatures wrote the South's first public school systems, open to both races, and reformed tax and property law.

What was not built was an economic foundation. General Sherman's Special Field Order No. 15 in January 1865 set aside coastal land for freed families, the origin of the phrase forty acres and a mule. President Andrew Johnson reversed it within the year and returned the land to former owners. Without land, most freedpeople entered sharecropping and tenancy, arrangements that under crop lien laws produced chronic debt.

How it was overturned

Reconstruction ended through organized violence and then through law. Paramilitary groups, the Ku Klux Klan and its successors, attacked officeholders and voters. At Colfax, Louisiana, in 1873, a white militia killed scores of Black men, many after surrendering. In United States v. Cruikshank (1876) the Supreme Court overturned the federal convictions, holding that the Fourteenth Amendment restrained states rather than private individuals, which gutted federal enforcement.

Then came the legal machinery. Mississippi's 1890 constitutional convention produced the template: a poll tax, a literacy test, an understanding clause allowing registrars to require applicants to explain a passage of the constitution, and residency requirements. None mentioned race. All were administered by white registrars with unreviewable discretion. Grandfather clauses exempted men whose ancestors could vote before 1867, which is to say, white men.

The results were immediate and are documented in state registration records. In Louisiana, 130,344 Black voters were registered in 1896. By 1904 the number was 1,342. That is a 99 percent reduction accomplished without a single statute mentioning race.

Plessy v. Ferguson (1896) supplied the constitutional cover, holding that separate but equal accommodations satisfied the Fourteenth Amendment. Justice Harlan dissented alone.

Key idea: Disfranchisement was achieved by race-neutral statutes administered with discretion, which is the clearest historical demonstration that a law need not mention race to function as a racial instrument.

The regime that followed

Jim Crow was not merely custom. It was a dense body of statutes and ordinances governing schools, transport, hospitals, cemeteries, textbooks, drinking fountains, and marriage, enforced by criminal penalty and backed by violence. The Equal Justice Initiative's documentation project has recorded more than 4,400 racial terror lynchings between 1877 and 1950, and its research emphasizes that these were frequently public events, advertised in advance and unpunished afterward.

Two economic mechanisms deserve names. Convict leasing allowed states to arrest Black men on vagrancy and similar charges and lease them to private employers, a system Douglas Blackmon documented as slavery by another name that persisted into the 1940s. Debt peonage tied sharecroppers to land through crop liens and accounting they could not audit.

Between about 1916 and 1970, roughly six million Black Americans left the South in the Great Migration, a movement driven by both the pull of northern industrial wages and the push of terror and disfranchisement. It reshaped the racial geography of every large northern city, and it set up the housing history we take up in two lessons.

Common misconceptions

• Most enslaved Africans were brought to what is now the United States. Of roughly 10.7 million who survived the crossing, about 389,000 disembarked in mainland North America; most went to Brazil and the Caribbean.
• Ending the international trade in 1808 wound slavery down. It redirected it into a domestic trade that moved roughly one million people to the Deep South and broke families as ordinary business.
• Historians agree slavery caused American industrialization. That specific causal claim is contested, with Olmstead, Rhode, and Wright disputing the magnitude and the mechanism while agreeing on slavery's brutality and profitability.
• Reconstruction accomplished nothing. It produced the Reconstruction Amendments, the South's first public school systems, and roughly 2,000 Black officeholders, before being overturned by violence and law.
• Disfranchisement required explicitly racial laws. Poll taxes, literacy tests, and understanding clauses never mentioned race and cut Louisiana's Black registration from 130,344 to 1,342 in eight years.

Recap

• About 389,000 of the roughly 10.7 million survivors of the Atlantic crossing arrived in mainland North America, and the enslaved population reached 3.95 million by 1860 through natural increase.
• Virginia's 1662 rule that status followed the mother made American slavery hereditary and racial, and the post-1808 domestic trade moved about one million people south.
• Cotton was roughly 60 percent of exports in 1860 and slavery was deeply financialized, while the causal weight of slavery in national growth remains a live quantitative dispute.
• Reconstruction produced three amendments, public schools, and roughly 2,000 Black officeholders, but no land redistribution after Johnson reversed Special Field Order No. 15.
• Race-neutral disfranchisement devices, Cruikshank, Plessy, lynching, convict leasing, and debt peonage built a durable legal order that helped drive the six-million-person Great Migration.

Sources

1. SlaveVoyages. (n.d.). Trans-Atlantic Slave Trade Database: Estimates. Rice University. slavevoyages.org
2. National Archives. (n.d.). 13th Amendment to the U.S. Constitution: Abolition of Slavery (1865). archives.gov
3. Britannica. (2025). Reconstruction: United States history. In Encyclopaedia Britannica. britannica.com
4. Britannica. (2025). Jim Crow law. In Encyclopaedia Britannica. britannica.com
5. Equal Justice Initiative. (2017). Lynching in America: Confronting the legacy of racial terror (3rd ed.). eji.org
6. Plessy v. Ferguson, 163 U.S. 537 (1896). supreme.justia.com
7. Library of Congress. (n.d.). The African American odyssey: A quest for full citizenship. loc.gov

The big picture

The previous lesson followed one line of American racial history. This lesson follows three others, and they are grouped together because they share a mechanism: race and national origin written directly into immigration, citizenship, and property law, applied to people who were in many cases citizens.

These episodes are less often taught than slavery and Jim Crow, and their omission distorts the picture in a specific way. It makes American racial history look like a two-group story, when the legal record shows a multi-front system in which different populations were subjected to different instruments: exclusion for one, expulsion for another, incarceration for a third.

Every fact in this lesson comes from statutes, court records, census documents, or federal commission reports. Where numbers are genuinely uncertain, I will say so and give you the range.

The Asian exclusion regime

Chinese migration to the United States began in volume with the 1849 gold rush and grew with railroad construction. Chinese workers made up the great majority of the Central Pacific's labor force on the hardest western sections of the transcontinental line, blasting and grading through the Sierra Nevada at wages below those paid to white workers, and were largely absent from the photographs taken at the 1869 completion.

Legal exclusion came in stages, and it is worth seeing the sequence, because each step built on the last.

Year	Measure	Effect
1854	People v. Hall (California)	Chinese witnesses barred from testifying against white defendants, removing legal protection against violence
1875	Page Act	Barred Chinese women on the presumption of immorality, distorting family formation for decades
1882	Chinese Exclusion Act	First federal law barring a specific nationality; suspended labor migration for ten years
1892	Geary Act	Extended exclusion and required Chinese residents to carry certificates or face deportation
1913, 1920	California alien land laws	Barred land ownership by aliens ineligible to citizenship, which after the naturalization cases meant Asian immigrants
1917	Immigration Act	Created the Asiatic Barred Zone, extending exclusion across South and Southeast Asia
1924	Johnson-Reed Act	National origins quotas plus a bar on aliens ineligible to citizenship, closing Asian immigration almost entirely
1943	Magnuson Act	Repealed Chinese exclusion during the wartime alliance with China, with an annual quota of 105
1952	McCarran-Walter Act	Ended racial bars to naturalization while retaining the national origins quota system
1965	Hart-Celler Act	Abolished national origins quotas, transforming the sources of American immigration

Two cases deserve mention alongside the statutes. United States v. Wong Kim Ark (1898) held that a child born in the United States to Chinese parents was a citizen by birth under the Fourteenth Amendment, establishing birthright citizenship on facts the government had chosen deliberately to test. And the anti-Chinese violence of the period was severe: at Rock Springs, Wyoming, in 1885, white miners killed at least 28 Chinese workers and burned the district's Chinatown, with no convictions.

Key idea: Asian exclusion was constructed statute by statute over sixty years, tied directly to the naturalization cases that made Asian immigrants aliens ineligible to citizenship, and it was dismantled only between 1943 and 1965.

Mexican repatriation in the 1930s

During the Depression, federal, state, county, and private actors conducted a campaign to remove people of Mexican descent from the United States. Some were deported through formal proceedings. Far more left under pressure: raids designed to frighten a neighborhood, county relief offices that cut off aid and offered train fare south, and employers who fired Mexican workers first. A widely publicized February 1931 sweep of La Placita park in Los Angeles was staged for exactly this effect.

Here the numbers are genuinely uncertain and you should know why. There was no single program with a single set of records; removals were spread across many jurisdictions, and voluntary departures under duress leave little paper trail. Scholarly estimates of the total range from roughly 400,000 to more than a million. The historians Francisco Balderrama and Raymond Rodriguez estimated that around 60 percent of those removed were United States citizens, mostly children born in this country, an estimate other researchers consider high and none can verify precisely.

What is documented is the policy intent, the raids, the relief-office coercion, and the outcome. In 2005 California enacted the Apology Act for the 1930s Mexican Repatriation Program, formally acknowledging the unconstitutional removal of American citizens and legal residents.

Key idea: The 1930s repatriation removed a large number of people, including many United States citizens, through informal coercion rather than a single legal program, which is precisely why the totals remain contested and the record incomplete.

Japanese American incarceration

On February 19, 1942, President Franklin Roosevelt signed Executive Order 9066, authorizing military commanders to designate exclusion zones. Under that authority about 120,000 people of Japanese ancestry were removed from the West Coast and confined in ten War Relocation Authority camps in remote interior locations. Roughly two thirds were United States citizens by birth. Families were given days to dispose of homes, farms, and businesses, and property losses were extensive and permanent.

The legal record is important. In Korematsu v. United States (1944) the Supreme Court upheld the exclusion order, deferring to claims of military necessity. In Ex parte Endo, decided the same day, the Court held that the government could not continue to detain a concededly loyal citizen, and the camps began to close.

Then the record was reopened. In 1980 Congress created the Commission on Wartime Relocation and Internment of Civilians. Its 1983 report, Personal Justice Denied, concluded that the incarceration was not justified by military necessity and that the causes were race prejudice, war hysteria, and a failure of political leadership. Researchers working on the commission and on the legal team also found that government lawyers had suppressed intelligence assessments contradicting the necessity claim. On that basis, federal courts vacated the wartime convictions of Fred Korematsu, Gordon Hirabayashi, and Minoru Yasui in the early 1980s through the rarely used writ of coram nobis.

Congress then did something almost unique in American history. The Civil Liberties Act of 1988 issued a formal apology and authorized payments of 20,000 dollars to each surviving incarcerated person. More than 82,000 people received payments totaling roughly 1.6 billion dollars. In 2018, in Trump v. Hawaii, the Supreme Court stated that Korematsu was gravely wrong the day it was decided and had been overruled in the court of history.

Key idea: Japanese American incarceration is the clearest American case in which a government investigated its own racial wrong, found the official justification false, apologized, and paid individual compensation, which is why it anchors every subsequent reparations argument.

Why these three belong together

Set the three episodes side by side and the common structure appears. In each, a legal category did the work: alien ineligible to citizenship, person of Mexican descent, person of Japanese ancestry. In each, citizenship offered less protection than the Constitution promised, because the category overrode the status. And in each, an economic interest was served, whether by removing labor competition, cutting relief rolls, or acquiring farmland on the West Coast.

There is also a difference worth naming, because it becomes important in Module 6. These populations were later racialized in nearly opposite directions. The same Asian American population subjected to sixty years of exclusion became, in postwar discourse, the model minority, a shift we will examine as a case study in how racial meanings are made and remade rather than fixed.

Common misconceptions

• The Chinese Exclusion Act was a temporary wartime measure. It was a peacetime immigration statute passed in 1882, repeatedly extended, and not repealed until 1943.
• Only immigrants were affected by these policies. Roughly two thirds of those incarcerated under EO 9066 were United States citizens, and a large share of those removed in the 1930s repatriation were citizens as well.
• The number of people repatriated in the 1930s is precisely known. Estimates range from about 400,000 to more than a million, because most departures were coerced rather than formally processed.
• Korematsu was overturned in 1944 by Ex parte Endo. Endo barred continued detention of a concededly loyal citizen but did not overturn the exclusion holding, which the Court repudiated only in dicta in 2018.
• Alien land laws had no lasting effect. They prevented a generation of Asian immigrants from acquiring the farmland that was the main vehicle of asset accumulation available to them.

Recap

• Asian exclusion was built across sixty years, from People v. Hall and the Page Act through the 1882 Exclusion Act, the Asiatic Barred Zone, and the 1924 quota law, and dismantled between 1943 and 1965.
• The alien land laws worked by referencing the naturalization cases, converting a citizenship bar into a property bar without naming a race.
• The 1930s Mexican repatriation removed somewhere between about 400,000 and more than a million people, many of them citizens, largely through informal coercion; California apologized in 2005.
• Executive Order 9066 led to the incarceration of about 120,000 people, roughly two thirds of them citizens, upheld in Korematsu and limited by Endo.
• The 1983 commission found no military necessity, courts vacated the convictions, and the Civil Liberties Act of 1988 delivered an apology and 20,000 dollar payments to more than 82,000 people.

Sources

1. National Archives. (n.d.). Chinese Exclusion Act (1882). archives.gov
2. National Archives. (n.d.). Executive Order 9066: Resulting in Japanese American incarceration (1942). archives.gov
3. Densho. (n.d.). Densho Encyclopedia: Japanese American incarceration. densho.org
4. Library of Congress. (n.d.). Immigration and relocation in U.S. history: Chinese. loc.gov
5. Wikipedia contributors. (2025). Mexican repatriation. In Wikipedia. wikipedia.org
6. Korematsu v. United States, 323 U.S. 214 (1944). supreme.justia.com
7. Britannica. (2025). Japanese American internment. In Encyclopaedia Britannica. britannica.com

The big picture

Between 1933 and 1968 the federal government did two contradictory things about race and housing. First it built the modern mortgage market and wrote racial discrimination into its operating rules. Then, thirty-four years later, it made housing discrimination illegal.

Understanding both halves is the single most useful thing you can do to prepare for the wealth data in the next module, because home equity is where most American middle-class wealth sits, and the rules for acquiring it were racially specific for exactly the generation that built postwar suburbia.

This lesson has an unusual structure, and deliberately so. The first half is documented policy that no serious scholar disputes. The second half includes a live causal disagreement about which mechanism did the damage. And the third half asks the hardest question in twentieth-century American history: what did the civil rights revolution actually change?

The federal government builds a map

The Home Owners Loan Corporation, created in 1933, refinanced distressed mortgages during the Depression. Between 1935 and 1940 it produced Residential Security Maps for more than two hundred cities, grading neighborhoods A through D. Grade A was best. Grade D was outlined in red, the origin of the word redlining.

The grading criteria are not a matter of inference. The accompanying area description forms survive and are digitized in the Mapping Inequality project, where you can read them yourself. They record the racial and ethnic composition of neighborhoods explicitly and treat the presence of Black residents, and often of Jewish, Italian, or recent immigrant residents, as a negative factor. Phrases about infiltration and about undesirable population appear routinely.

The Federal Housing Administration, created in 1934, insured long-term amortized mortgages and thereby made the thirty-year loan possible for ordinary families. Its Underwriting Manual of 1936 and 1938 instructed appraisers that a neighborhood must be protected from adverse influences, warned against the infiltration of inharmonious racial groups, and recommended racially restrictive covenants and physical barriers such as highways to separate areas of different racial occupancy. This is a federal manual, not an allegation.

Racially restrictive covenants, private deed clauses barring sale to specified groups, spread widely with that encouragement. In Shelley v. Kraemer (1948) the Supreme Court held that courts could not enforce them, since judicial enforcement would be state action. The covenants themselves remained in deeds and continued to operate through private agreement and realtor practice until the Fair Housing Act of 1968.

The effects show up in the era's flagship developments. Levittown, New York, built from 1947, sold roughly 17,400 homes on terms that excluded non-white buyers by contract. In 1953 essentially none of its roughly 70,000 residents were Black. The GI Bill compounded the pattern, because its housing and education benefits were administered locally. Ira Katznelson's research documented that of tens of thousands of GI Bill mortgages issued in the New York and northern New Jersey suburbs, fewer than one hundred went to non-white buyers.

Key idea: The federal government did not merely fail to prevent housing discrimination; it wrote discrimination into the underwriting standards of the mortgage system it created, during the exact decades when suburban home equity became the main American middle-class asset.

What Black buyers faced instead

Excluded from insured mortgages, Black families in cities like Chicago bought on contract. In a contract sale the buyer made a large down payment and monthly payments at inflated prices and interest, but received no deed and built no equity until the final payment. Miss one payment and the seller could evict and resell, keeping everything paid.

Beryl Satter's research reconstructed the system in Chicago, and a 2019 Duke University study estimated that contract selling extracted between 3 and 4 billion dollars in wealth from Black Chicago families in the 1950s and 1960s alone. The Contract Buyers League, organized in the late 1960s, is one of the better documented episodes of collective resistance to it.

Urban renewal added a second mechanism. Federally funded clearance and highway construction displaced roughly a million people nationally between the late 1940s and the early 1970s, with Black neighborhoods disproportionately targeted, which is why the programs acquired the bitter nickname negro removal.

The causal argument scholars are actually having

Now a genuine dispute, and it is a good example of how causal questions differ from documentary ones.

Daniel Aaronson, Daniel Hartley, and Bhashkar Mazumder compared neighborhoods on either side of HOLC grade boundaries, using the fact that map lines sometimes split otherwise similar areas. They found that a D grade caused lower homeownership, lower house values, higher segregation, and effects persisting for decades. Their 2021 paper is the strongest evidence that the maps themselves mattered.

Price Fishback and colleagues pushed back with archival work showing that HOLC maps were closely held and not distributed to private lenders in the way the standard account assumes, and that the grades largely encoded neighborhood characteristics that lenders already observed. On this reading, the maps are better understood as a record of discrimination than as its cause, and the causal agent was the FHA underwriting rules and private practice.

Be precise about what is at issue. Both camps agree that federal housing policy discriminated by race, that the FHA manual said what it said, and that Black families were largely excluded from the insured mortgage market. The dispute is narrower: whether those specific color-coded maps were a causal instrument or a symptom. Nothing about the historical record turns on the answer, and a great deal about how we model policy effects does.

Key idea: Documented policy and causal attribution are different claims requiring different evidence, and it is entirely coherent to hold that federal housing discrimination is established while the specific causal role of the HOLC maps remains contested.

The civil rights transformation

Between 1954 and 1968 the legal foundation of American racial hierarchy was dismantled with remarkable speed. Brown v. Board of Education (1954) rejected separate but equal in public education. The Civil Rights Act of 1964 barred discrimination in public accommodations, in programs receiving federal funds, and in employment, and created the Equal Employment Opportunity Commission. The Voting Rights Act of 1965 suspended literacy tests and required jurisdictions with a history of discrimination to preclear voting changes with the federal government. Loving v. Virginia (1967) struck down bans on interracial marriage in the sixteen states still enforcing them. The Fair Housing Act of 1968 made housing discrimination illegal.

The measurable effects in some domains were fast and enormous. Black voter registration in Mississippi rose from about 6.7 percent of eligible adults in 1964 to about 59.8 percent by 1967. Public accommodations desegregated across the South within a few years, a change that had seemed unimaginable a decade earlier. The share of Black students in the South attending majority-white schools rose from roughly 2 percent in 1964 to about 44 percent by 1988, making the South, for a period, the least segregated region in the country by that measure. Economists John Donohue and James Heckman found that Title VII enforcement produced real gains in Black relative wages, concentrated in the South.

Now the other side of the ledger. The wealth gap did not close, and by most measures has not narrowed appreciably in fifty years. Residential segregation declined slowly and remains high. Southern school desegregation peaked around 1988 and reversed as courts released districts from supervision. Milliken v. Bradley (1974) barred most interdistrict busing remedies, which meant that in metropolitan areas where white families had moved to separate suburban districts, the remedy could not reach them. And the Fair Housing Act as passed in 1968 had famously weak enforcement machinery, relying on individual complaints, until it was strengthened in 1988.

Key idea: The civil rights statutes ended legally mandated exclusion quickly and effectively, and they did comparatively little about accumulated assets, residential geography, and the jurisdictional boundaries that separate school systems.

Why the pattern makes sense

The asymmetry is not a puzzle once you name what each kind of barrier was made of. A statute requiring separate lunch counters can be repealed on a Tuesday. A century of differences in accumulated home equity cannot, because it is not a rule, it is a stock. Segregation maintained by thousands of private decisions and by municipal boundaries is not a rule either, and antidiscrimination law reaches individual acts far better than it reaches aggregate patterns.

That distinction, between rules you can repeal and stocks and structures you cannot, is the hinge of the entire policy debate you will meet in the final module. Keep it.

Common misconceptions

• Redlining was private-sector bigotry that government failed to stop. The FHA Underwriting Manual instructed appraisers to avoid inharmonious racial groups and recommended restrictive covenants.
• Shelley v. Kraemer eliminated racial covenants in 1948. It made them judicially unenforceable; they persisted in deeds and in practice until the Fair Housing Act of 1968.
• Scholars agree the HOLC maps caused later segregation. Aaronson, Hartley, and Mazumder find causal effects at grade boundaries; Fishback and colleagues argue the maps were not widely circulated and largely recorded existing patterns.
• The GI Bill was race neutral because its text was. It was administered locally through banks and universities that discriminated, so its benefits flowed overwhelmingly to white veterans.
• The civil rights acts failed. They produced very large, measurable changes in voting, public accommodations, and southern schooling, and comparatively little change in wealth and residential segregation, which is a pattern with an explanation rather than a verdict.

Recap

• HOLC Residential Security Maps graded neighborhoods with explicit reference to racial composition, and the area description forms are publicly readable today.
• The FHA Underwriting Manual recommended restrictive covenants and warned against racial mixing, shaping the insured mortgage market that built suburban wealth.
• Excluded buyers turned to contract sales that built no equity, a system estimated to have extracted 3 to 4 billion dollars from Black families in Chicago alone.
• Whether the HOLC maps were a cause or a record of discrimination is contested, while federal housing discrimination itself is documented and undisputed.
• The 1964-1968 statutes transformed voting, public accommodations, and southern schooling quickly, and left wealth, segregation, and district boundaries largely intact.

Sources

1. Nelson, R. K., Winling, L., Marciano, R., and Connolly, N. D. B. (n.d.). Mapping inequality: Redlining in New Deal America. University of Richmond. dsl.richmond.edu
2. Britannica. (2025). Redlining. In Encyclopaedia Britannica. britannica.com
3. Aaronson, D., Hartley, D., and Mazumder, B. (2021). The effects of the 1930s HOLC redlining maps. American Economic Journal: Economic Policy, 13(4), 355-392. doi.org
4. National Archives. (n.d.). Civil Rights Act (1964). archives.gov
5. National Archives. (n.d.). Voting Rights Act (1965). archives.gov
6. Shelley v. Kraemer, 334 U.S. 1 (1948). supreme.justia.com
7. U.S. Department of Housing and Urban Development. (n.d.). History of fair housing. hud.gov

The big picture

If you remember one number from this course, make it this one. In the 2022 Survey of Consumer Finances, the Federal Reserve's flagship wealth survey, median net worth was about 285,000 dollars for white non-Hispanic families, about 44,900 dollars for Black families, about 61,600 dollars for Hispanic families, and about 132,900 dollars for families in the residual other category, which includes Asian, American Indian, Pacific Islander, and multiracial respondents.

That is roughly a six to one ratio at the median between white and Black families. Now put it next to the income figures. Median household income differs by a ratio closer to one and a half or one and two thirds to one. The wealth ratio is about four times as large as the income ratio.

Why the enormous difference between the two measures? That question is the whole lesson, and answering it well requires you to be careful, because it is a place where sloppy argument runs in both directions.

Stock and flow

Start with the distinction that makes the rest intelligible. Income is a flow: money arriving over a period, wages, salary, benefits, interest. Wealth, or net worth, is a stock: everything you own minus everything you owe, measured at a moment.

Stocks and flows behave differently in three ways that matter here. Stocks accumulate, so a difference in flows that persists for decades produces a much larger difference in stocks. Stocks compound, so an early advantage grows on its own. And stocks are transferable at death and during life, so a difference in one generation becomes a starting condition for the next. Income is earned fresh each year. Wealth is inherited.

This is why the housing history in the previous lesson is not merely background. Home equity is the largest single asset for most American families below the top decile. The generation that bought FHA-insured suburban homes between 1934 and 1968 accumulated equity that appreciated for fifty years and then passed to their children as down payments and inheritances. The families excluded from that market by underwriting rules did not simply lose one house. They lost the compounding and the transfer.

Key idea: Wealth accumulates, compounds, and transfers across generations, so policies that ended decades ago continue to shape today's balance sheets in a way that current income differences alone cannot explain.

The numbers, with their caveats

Measure (2022 SCF unless noted)	White non-Hispanic	Black	Hispanic
Median family net worth	About 285,000 dollars	About 44,900 dollars	About 61,600 dollars
Homeownership rate (Census, recent years)	About 74 percent	About 45 to 46 percent	About 49 percent
Share reporting any inheritance	Roughly 30 percent	Roughly 10 percent	Roughly 7 to 10 percent

Three cautions before you use these figures. First, medians hide enormous variation within every group; there are poor white families and wealthy Black families in large numbers, and a median is not a description of any individual. Second, the other category is a statistical artifact of small sample sizes, and it conceals real differences: Asian American families in particular span an extremely wide wealth distribution by national origin. Third, household surveys undercount wealth at the very top, which affects means far more than medians.

The documented mechanisms

Four mechanisms are well established, and they interact.

Inheritance and family transfers. Roughly three times as many white families as Black families report having received an inheritance, and conditional on receiving one, the amounts are larger. Transfers during life matter as much as bequests: help with a down payment, a car, tuition, or a rent deposit in a bad month. These transfers are largely invisible in income statistics and enormously consequential for asset building.

Homeownership rates and timing. The roughly thirty point gap in homeownership is the single largest proximate contributor. Timing compounds it: buying earlier means more years of appreciation and amortization. Families that entered the market after 1970 bought at higher price-to-income ratios than those who bought in 1950.

Portfolio composition and returns. Black and Hispanic families hold a larger share of their wealth in housing and vehicles and a smaller share in equities and business assets. Since equities have historically returned more than housing, identical savings rates produce different growth. This also creates asymmetric exposure to shocks. Between 2005 and 2009, Pew Research Center analysis found median wealth fell by roughly two thirds for Hispanic households and roughly half for Black households, against a much smaller decline for white households, because the crash hit housing hardest and subprime lending had been concentrated in minority neighborhoods.

Debt, especially student debt. Black college graduates borrow more and, four years after graduation, owe substantially more than white graduates with comparable degrees, in significant part because they receive less family help. Negative net worth in early adulthood delays every subsequent asset purchase.

Key idea: The wealth gap is produced by identifiable mechanisms, transfers, homeownership, portfolio composition, and debt, each of which can be measured, which means the explanation does not have to rest on inference from the gap itself.

The honest complications

Now the part that a course committed to evidence has to include, because several common claims about the wealth gap are more complicated than they sound.

First, income differences do not explain the whole gap, but neither are they irrelevant. Substantial wealth gaps remain when you compare families at similar income levels, which is a real finding. But some of that residual reflects age structure, since a younger population has had fewer years to accumulate; some reflects household composition, since married-couple households hold much more wealth and marriage rates differ; and some reflects differences in returns and transfers. Decomposing the residual is exactly the hard work, and honest researchers report ranges rather than a single number.

Second, family structure is genuinely contested rather than settled in either direction. That two-earner married households accumulate more wealth is arithmetic. Whether differences in marriage rates are better understood as a cause of the wealth gap or as a consequence of the labor market and incarceration conditions that preceded them is a live empirical question, and researchers of different politics read the same data differently. Treat anyone who tells you it is obvious with suspicion.

Third, the pattern is not a simple racial hierarchy. Median wealth in the other category sits well above Black and Hispanic medians and below white medians, and within Asian America the distribution is extremely wide. Nigerian American and Indian American households have among the highest median incomes in the country, while Hmong, Burmese, and Bhutanese American households have poverty rates well above the national average. Any explanation that treats race as a single sufficient variable will fail on these cases, which is one reason immigration history, national origin, and selection into migration get a full lesson later in this course.

Fourth, a finding that complicates everything in a productive way. Raj Chetty and colleagues, using tax and census records covering essentially the entire population, found that Black and white children raised in families with the same income have very different adult outcomes, and that the gap is driven almost entirely by men. Black and white women raised at the same parental income level have similar individual earnings in adulthood. That result is difficult for any theory that runs purely through family resources, since sisters and brothers share those. It points instead toward mechanisms specific to men, including neighborhood conditions, incarceration, and the labor market, and it is one of the most cited findings in the field.

Would income equality close the gap?

Several research teams have simulated what would happen to the wealth gap under different interventions. The results converge on an uncomfortable finding: closing income gaps would narrow the wealth gap slowly, over generations rather than years, because the current gap reflects accumulated stocks and ongoing transfer flows rather than only current earnings. Federal Reserve Bank of Cleveland researchers modeling the dynamics concluded that even equalizing incomes would leave a large gap in place for a very long time.

That finding is politically ambiguous, which is a good sign that it is being reported honestly. Some read it as the case for direct wealth-based remedies, including reparations proposals we take up in the final lesson. Others read it as evidence that policy should target the flows it can actually influence, education, employment, and family stability, and let accumulation follow. The simulation does not choose between those readings. It tells you the arithmetic that either position has to work with.

Key idea: Because wealth is a stock built by decades of flows and transfers, no plausible change in current income alone closes the gap quickly, a finding that constrains every policy argument without settling any of them.

Common misconceptions

• The wealth gap is just the income gap restated. The median wealth ratio is roughly four times the median income ratio, because wealth accumulates, compounds, and is inherited.
• Median figures describe individuals. They describe the middle of a distribution with enormous internal variation, including many poor white families and many wealthy Black families.
• Wealth gaps disappear once you control for income. Large gaps remain at similar income levels, though age structure, household composition, transfers, and returns account for part of the residual.
• All non-white groups show the same pattern. Median wealth in the other category is far above Black and Hispanic medians, and Asian American outcomes vary enormously by national origin.
• Chetty's work shows family income explains the gap. It shows the opposite in an important respect: at equal parental income the gap appears among men and largely not among women.

Recap

• The 2022 SCF reports median family net worth of roughly 285,000 dollars for white families, 44,900 for Black families, 61,600 for Hispanic families, and 132,900 for the other category.
• Wealth is a stock that accumulates, compounds, and transfers, which is why gaps created by past policy persist long after the policy ends.
• The main measured mechanisms are inheritance and family transfers, the roughly thirty point homeownership gap, portfolio composition and returns, and debt burdens including student loans.
• Age structure, household composition, national origin, and immigration history all shape the numbers, and the family structure question is genuinely contested.
• Simulations indicate that equalizing income alone would close the wealth gap only over generations, a constraint every policy position has to accommodate.

Sources

1. Board of Governors of the Federal Reserve System. (2023). Survey of Consumer Finances. federalreserve.gov
2. United States Census Bureau. (n.d.). Housing vacancies and homeownership. census.gov
3. Chetty, R., Hendren, N., Jones, M. R., and Porter, S. R. (2020). Race and economic opportunity in the United States: An intergenerational perspective. Quarterly Journal of Economics, 135(2), 711-783. doi.org
4. Pew Research Center. (n.d.). Race and ethnicity research. pewresearch.org
5. Federal Reserve Bank of Cleveland. (n.d.). Research on the racial wealth gap. clevelandfed.org
6. Federal Reserve Bank of St. Louis. (n.d.). Institute for Economic Equity. stlouisfed.org

The big picture

This lesson does two jobs. The first half reports what the primary federal data show about work, schooling, and health. The second half teaches the toolkit researchers use to figure out what causes those patterns, and, just as importantly, what each tool cannot do.

I have put them together on purpose. Students who learn the gaps without the methods end up believing that any measured difference is evidence of discrimination. Students who learn only the methodological cautions end up believing that nothing can be shown. Both conclusions are wrong, and the way to avoid each is to hold the numbers and the methods in the same hand.

Work

Two patterns in Bureau of Labor Statistics data are worth memorizing because they are so durable.

The first is the two-to-one unemployment ratio. Across expansions and recessions for more than half a century, the Black unemployment rate has run roughly twice the white rate. It falls in booms and rises in busts, but the ratio is remarkably stable. In April 2023 the Black unemployment rate reached a record low of 4.8 percent, and even at that record the white rate was lower. A pattern that survives fifty years of very different labor markets is telling you something structural rather than cyclical.

The second is the earnings ordering. Among full-time wage and salary workers, median usual weekly earnings run highest for Asian workers, then white, then Black, then Hispanic workers. That ordering matters because it immediately falsifies any simple white-on-top model, and because it forces the analytic questions that occupy the rest of this course: what varies with education, occupation, region, immigration cohort, and national origin, and what remains after those are accounted for.

School

Educational patterns have moved more than most people realize, in both directions.

High school completion gaps have narrowed dramatically over five decades and are now comparatively small. College enrollment rose substantially for Black and Hispanic students, while gaps in completion, and especially in completion at selective institutions, remain wide.

On the National Assessment of Educational Progress, the Black-white and Hispanic-white gaps narrowed substantially through the 1970s and 1980s and then largely stalled. Explaining both the narrowing and the stall is one of the discipline's harder problems, and there are several live accounts.

Sean Reardon added a finding that reframes the question. Over the same decades in which racial achievement gaps narrowed, the achievement gap between high-income and low-income students grew substantially, by roughly 40 percent between cohorts born in the 1970s and those born around 2000. Income and race are correlated but not identical, so this means the story is not simply that inequality got worse in every dimension.

Roland Fryer and Steven Levitt, using a large early-childhood cohort study, found that the raw Black-white test score gap at kindergarten entry was substantially reduced by controlling for a handful of observable characteristics such as family socioeconomic status and birth weight, but that the gap grew as children progressed through school. That widening pattern points attention toward school quality, segregation, and neighborhood rather than toward endowment at entry.

Then there is the cultural hypothesis, which deserves a fair hearing because it is often either endorsed or dismissed without evidence. Signithia Fordham and John Ogbu proposed in 1986 that some Black students face peer sanctions for behaviors coded as acting white. Karolyn Tyson, William Darity, and Domini Castellino tested it with school-based fieldwork and found little support for a distinctively racial burden of high achievement; they found instead that achievement-related stigma appeared among students of all races and was strongest where tracking made achievement visible. Fryer and Paul Torelli, using national friendship network data, found evidence of a social cost to high grades that was concentrated in racially integrated public schools and largely absent in predominantly Black or private schools. Those results are not identical, and taking them together supports a narrow conclusion rather than a sweeping one: the effect, where it exists, appears to depend on school context rather than on culture in general.

Key idea: Achievement gaps narrowed and then stalled while income-based gaps grew, and the leading explanations point toward school context and segregation more than toward differences present at kindergarten entry.

Health

The health data are the starkest in this lesson. Life expectancy at birth in 2022, from the National Center for Health Statistics, ran roughly as follows: about 84.5 years for Asian Americans, about 80.0 for Hispanic Americans, about 77.5 for white Americans, about 72.8 for Black Americans, and about 67.9 for American Indian and Alaska Native people. The national figure was about 77.5.

Maternal mortality shows one of the largest disparities in American health statistics. In 2022 the rate for Black women was about 49.5 deaths per 100,000 live births, against about 19.0 for white women and about 16.9 for Hispanic women. Infant mortality for Black infants runs roughly twice the white rate. These gaps do not disappear at higher education or income levels, which is one of the strongest arguments against a purely socioeconomic explanation.

And then the finding that should stop any simple story in its tracks. Hispanic Americans have longer life expectancy than white Americans despite lower average income, lower insurance coverage, and less access to care. This is called the Hispanic paradox or the immigrant health advantage, and it is real and replicated. The explanations under active investigation include selection, since migration favors the healthy; the salmon bias, since some migrants return home when seriously ill and their deaths are recorded elsewhere; lower smoking rates in some origin populations; and social support. Notice that the paradox weakens across generations, which is itself informative.

Key idea: Health disparities are large, persist across income and education levels, and do not follow a single ordering, since the group with the second-longest life expectancy has among the lowest average incomes.

The evidence problem

Now the methods. Suppose you observe a gap. How would you establish that discrimination produced it?

Audit and correspondence studies are the closest thing to an experiment. Researchers create applicants who are identical except for a signal of race, then measure responses. Marianne Bertrand and Sendhil Mullainathan sent about 5,000 resumes to Boston and Chicago job ads in 2001 and 2002, randomly assigning names that sounded distinctively white or Black. White-sounding names received about 50 percent more callbacks, a gap equivalent to roughly eight additional years of experience. Devah Pager sent matched pairs of testers to Milwaukee employers with and without a criminal record and found that Black applicants with no record received callbacks at about the same rate as white applicants who disclosed a felony conviction.

Lincoln Quillian and colleagues meta-analyzed field experiments conducted between 1989 and 2015 and found no decline in discrimination against Black applicants over that quarter century, with some decline against Latino applicants. Patrick Kline, Evan Rose, and Christopher Walters then ran the largest study of its kind, sending roughly 83,000 applications to 108 of the largest United States employers, and found an overall contact gap of a few percentage points that was heavily concentrated: a minority of firms accounted for most of the measured discrimination, while many showed none.

The limits are real and researchers state them. These studies measure callbacks, not hires or wages. Distinctively raced names may also signal social class, which Fryer and Levitt showed is entangled with name choice. In-person audit studies raise the problem James Heckman and Peter Siegelman identified, that matching testers on observable traits does not match them on the unobserved ones employers respond to. And a study of entry-level applications tells you little about promotion.

Regression decomposition takes a gap and splits it into a portion explained by measured characteristics and an unexplained residual. It is indispensable and it is routinely misread. The residual is not discrimination; it is everything not measured, which includes unmeasured skills, preferences, and errors, in both directions. Worse, the technique has a trap: controlling for a variable that is itself a product of discrimination, such as current occupation or prior job experience, absorbs the discrimination into the explained portion and makes it disappear from the residual. So a careful analyst reports both what the residual is and what it might contain.

Natural experiments exploit a change in rules or conditions that resembles random assignment. The best of them are powerful. They can also surprise you. Amanda Agan and Sonja Starr studied ban-the-box policies, which prohibit asking about criminal records early in hiring. The intent was to help applicants with records. The measured effect was to widen the racial gap in callbacks among young men without records, because employers deprived of the specific information appeared to fall back on group averages. That is a finding no one wanted, produced by a good design, and it is exactly why designs like this matter.

Outcome and threshold tests ask not who was selected but what happened afterward. If police search Black drivers at a lower evidentiary threshold, then searches of Black drivers should yield contraband at a lower rate. Emma Pierson and colleagues applied this logic to about 95 million traffic stops and found exactly that pattern, alongside a veil-of-darkness test showing that stop disparities shrink after sunset, when race is harder to observe.

Key idea: No single method establishes discrimination on its own; the standard of evidence in this field is convergence across designs with different weaknesses, and an unexplained residual is a question rather than an answer.

Common misconceptions

• Any measured gap is discrimination. A gap is a description; establishing its cause requires a research design, and the residual from a regression is not a measurement of discrimination.
• Audit studies prove discrimination in hiring outcomes. They demonstrate differential callbacks under controlled conditions, which is powerful but is not the same as measuring hires, wages, or promotion.
• Controlling for more variables always improves an estimate. Controlling for a variable that is itself downstream of discrimination hides the effect you are trying to measure.
• Discrimination in hiring is uniform across employers. Kline, Rose, and Walters found it heavily concentrated in a minority of large firms, with many firms showing none.
• Health gaps are simply income gaps. Black maternal and infant mortality gaps persist at higher income and education levels, and Hispanic life expectancy exceeds white life expectancy despite lower income.

Recap

• The Black unemployment rate has run about twice the white rate for over fifty years, and median weekly earnings run highest for Asian workers, then white, Black, and Hispanic workers.
• Achievement gaps narrowed through the 1980s and stalled, income-based gaps grew by roughly 40 percent, and the acting-white hypothesis finds support only in specific school contexts.
• Life expectancy in 2022 ranged from about 84.5 years for Asian Americans to about 67.9 for American Indian and Alaska Native people, and the Hispanic paradox contradicts any simple income-based account.
• Audit studies, from Bertrand and Mullainathan through Kline, Rose, and Walters, show real callback discrimination that is concentrated among a minority of employers and has not declined for Black applicants since 1989.
• Decomposition residuals, natural experiments, and outcome tests each carry specific weaknesses, so credible claims rest on convergence across designs rather than on any one study.

Sources

1. U.S. Bureau of Labor Statistics. (n.d.). Current Population Survey: Labor force statistics. bls.gov
2. National Center for Education Statistics. (n.d.). The Nation's Report Card (NAEP). nces.ed.gov
3. National Center for Health Statistics. (n.d.). Life expectancy. Centers for Disease Control and Prevention. cdc.gov
4. Bertrand, M., and Mullainathan, S. (2004). Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. American Economic Review, 94(4), 991-1013. doi.org
5. Pager, D. (2003). The mark of a criminal record. American Journal of Sociology, 108(5), 937-975. doi.org
6. Quillian, L., Pager, D., Hexel, O., and Midtboen, A. H. (2017). Meta-analysis of field experiments shows no change in racial discrimination in hiring over time. PNAS, 114(41), 10870-10875. doi.org
7. Kline, P., Rose, E. K., and Walters, C. R. (2022). Systemic discrimination among large U.S. employers. Quarterly Journal of Economics, 137(4), 1963-2036. doi.org

The big picture

In 1998 a team of psychologists published a reaction-time task that would become one of the most famous instruments in the history of social science. The Implicit Association Test asks you to sort words and faces into paired categories as fast as you can, and infers the strength of your mental associations from how much slower you are when the pairings run against them. Project Implicit has since administered tens of millions of these tests online.

The IAT entered public life as a device that could reveal hidden bigotry in people who sincerely denied it. Corporate training programs, police departments, and universities built curricula around it. And then, over about fifteen years, a large body of careful research complicated nearly every strong claim made on its behalf.

This lesson is a case study in intellectual honesty. Understanding what the implicit bias literature does and does not support is more valuable to you than either the enthusiastic version or the dismissive one, and it will make you a better reader of every contested finding in this course.

What the test does

The mechanics are simple. You press one key for faces from one group and words with pleasant meanings, and another key for the other group and unpleasant words. Then the pairings switch. Most people are somewhat slower in one configuration than the other, and the difference in average response time, standardized, is the score.

Here is what is not in dispute. On average, across millions of test takers, most participants respond faster when white faces are paired with pleasant words than when Black faces are. This includes a substantial share of Black test takers, which is important, because it tells you the test is picking up culturally available associations rather than personal animus.

That aggregate finding is real, replicated, and interesting. Associations learned from a culture saturated with racial imagery show up in reaction times. Almost everything contested lies downstream of that.

Key idea: The IAT reliably detects culturally learned associations at the group level, including among members of the group the associations disfavor, which is a claim about culture rather than about individual prejudice.

The reliability problem

For a measure to tell you something stable about a person, it has to give roughly the same answer when you take it twice. Test-retest reliability for the race IAT sits in the neighborhood of 0.4 to 0.5. For comparison, a well-constructed personality or cognitive measure typically runs above 0.7, often above 0.8.

What that means concretely is that an individual's score bounces around across sittings. Someone can take the test on Tuesday and score in the moderate-preference range, and take it again on Friday and score near zero. If you have taken the test and were unsettled by your score, this is the most useful thing you can know about it: the instrument was not designed to be, and does not perform as, an individual diagnostic.

The predictive validity problem

The second question is whether IAT scores predict discriminatory behavior. This is where the literature has moved the most.

A 2009 meta-analysis by Anthony Greenwald and colleagues reported a correlation around 0.24 between IAT scores and behavior in racial domains. A 2013 reanalysis by Frederick Oswald, Gregory Mitchell, Hart Blanton, James Jaccard, and Philip Tetlock, using stricter criteria for what counts as a discrimination outcome, found correlations closer to 0.15 or lower, and found that IAT scores did not outperform straightforward self-report measures of prejudice.

Then, in 2019, Patrick Forscher and colleagues meta-analyzed 492 studies of interventions that attempted to change implicit measures. They found that implicit measures could be shifted, at least briefly, but that changes in implicit measures did not produce corresponding changes in explicit attitudes or in behavior. That is a severe result for the theory of change underlying most implicit bias programming, which assumes that moving the implicit measure moves conduct.

To their credit, the test's own developers have said much of this publicly. Greenwald and Mahzarin Banaji have stated that the IAT's psychometric properties do not support using it to diagnose individuals or to predict individual behavior, and that its appropriate use is in aggregate research.

Key idea: The race IAT has low test-retest reliability and weak individual-level predictive validity, and its own developers say it should not be used to diagnose or predict for individuals.

What the strongest defense actually is

Do not conclude that implicit attitudes are nothing. There is a serious defense of their importance, and it does not depend on individual prediction.

The argument runs like this. A correlation of 0.15 is small for predicting one person's behavior in one situation and is not small when the same small effect operates across millions of decisions: resumes glanced at, patients triaged, students called on, split-second judgments made by armed officers. Small biases aggregated across a system can produce substantial disparities even when no individual case is predictable. Greenwald has made this argument explicitly.

The counterargument, pressed by Mitchell and Tetlock among others, is that this reasoning has not itself been tested. Moving from a small laboratory correlation to a claim about aggregate real-world outcomes requires assumptions about how effects compound that no one has demonstrated, and small effects can also wash out rather than accumulate.

You should hold both of these. The aggregation argument is plausible and unproven. That is a legitimate scientific state of affairs, and it is where the field currently sits.

Does training work?

The practical question most organizations care about has an uncomfortable answer.

Elizabeth Levy Paluck and Donald Green's review of prejudice-reduction research found that the enormous popular literature rested on remarkably few rigorous field evaluations. A meta-analysis of diversity training by Katerina Bezrukova and colleagues found that effects on attitudes were modest and tended to decay, with effects on behavior weaker still. A large field experiment by Edward Chang and colleagues found that a one-shot online diversity training produced some attitude change in some subgroups and very little detectable behavior change over the following weeks.

The strongest positive evidence comes from Patricia Devine and colleagues, whose prejudice habit-breaking intervention taught participants specific strategies over time rather than delivering a single session, and produced effects that persisted for months. Note the design difference: extended practice with concrete strategies, not a lecture about bias.

Meanwhile, aggregate implicit attitudes have in fact been changing. Tessa Charlesworth and Banaji, analyzing more than a decade of Project Implicit data, found that implicit racial attitudes moved roughly a quarter of the way toward neutrality between 2007 and 2020, while implicit attitudes in some other domains did not improve at all. Whatever is producing that change, it is not implicit bias training, which reached only a fraction of the population.

Key idea: Evidence for attitude-focused training changing behavior is weak, while aggregate implicit attitudes have shifted substantially over time for reasons that lie outside training programs.

What to do with all this

Return to Merton's typology from Module 1. If the goal is less discriminatory behavior, you can try to change attitudes or you can change the situations in which decisions are made. The evidence for the second is considerably better.

Structured interviews with predetermined questions and scoring rubrics outperform unstructured ones. Removing identifying information from initial review changes who advances. Requiring decision makers to record reasons before seeing candidate identity, and holding them accountable for the criteria they named, constrains discretion. None of these require anyone to change a belief, which is why they work on the fair-weather liberal and the timid bigot alike.

That is not a claim that attitudes are irrelevant. It is a claim about where the evidence is strongest, which is the only claim a course like this can responsibly make.

Common misconceptions

• Your IAT score reveals whether you are a racist. Test-retest reliability near 0.4 to 0.5 means individual scores are unstable, and the developers say the test should not be used to diagnose individuals.
• The IAT has been debunked and implicit attitudes do not exist. The aggregate finding that culturally learned associations show up in reaction times is robust and replicated.
• Reducing implicit bias reliably reduces discriminatory behavior. Forscher and colleagues, meta-analyzing 492 studies, found changes in implicit measures did not translate into behavior change.
• Diversity training is well supported by evidence. Rigorous field evaluations are scarce, effects on attitudes are modest and decay, and effects on behavior are weaker still.
• Nothing works. Extended habit-breaking interventions show durable effects, and procedural changes such as structured interviews and blind initial review have better evidence than attitude change does.

Recap

• The IAT measures reaction-time differences in category pairings and reliably detects culturally learned associations at the aggregate level, including among members of disfavored groups.
• Test-retest reliability of roughly 0.4 to 0.5 makes individual scores unstable, and meta-analyses place predictive correlations around 0.15 to 0.24 depending on criteria.
• Forscher and colleagues found that shifting implicit measures did not shift behavior, which undermines the standard theory of change behind implicit bias programming.
• The aggregation defense, that small effects across millions of decisions produce large disparities, is plausible and remains untested.
• Procedural interventions that constrain discretion have stronger evidence than attitude-change training, and aggregate implicit racial attitudes moved about a quarter toward neutrality between 2007 and 2020.

Sources

1. Project Implicit. (n.d.). About the Implicit Association Test. Harvard University. implicit.harvard.edu
2. Oswald, F. L., Mitchell, G., Blanton, H., Jaccard, J., and Tetlock, P. E. (2013). Predicting ethnic and racial discrimination: A meta-analysis of IAT criterion studies. Journal of Personality and Social Psychology, 105(2), 171-192. doi.org
3. Forscher, P. S., Lai, C. K., Axt, J. R., Ebersole, C. R., Herman, M., Devine, P. G., and Nosek, B. A. (2019). A meta-analysis of procedures to change implicit measures. Journal of Personality and Social Psychology, 117(3), 522-559. doi.org
4. Charlesworth, T. E. S., and Banaji, M. R. (2019). Patterns of implicit and explicit attitudes: Long-term change and stability. Psychological Science, 30(2), 174-192. doi.org
5. Wikipedia contributors. (2025). Implicit-association test. In Wikipedia. wikipedia.org

The big picture

Here is the question this lesson exists to answer. Openly expressed racial prejudice has fallen dramatically on every survey measure since the 1950s. Explicit legal exclusion ended. And yet the gaps we measured in the last module are large and, in the case of wealth, barely moving. How?

The answer is that several mechanisms produce racial inequality without requiring anyone to hold a racial animus at all. They are not hypothetical. Each is measurable, most are legal, and some are actively defended on grounds that have nothing to do with race. That combination is exactly what makes them durable.

We will take four: two theories of discrimination from economics, then segregation, then networks and hoarding. By the end you should be able to look at an unequal outcome and generate several candidate mechanisms rather than one.

Two economies of discrimination

In 1957 Gary Becker published the first formal economic model of discrimination. In it, a discriminating employer has what Becker called a taste for discrimination: an actual preference for not hiring certain workers, which he treats as a cost the employer is willing to pay. The prediction that follows is striking. A discriminating employer pays more for labor than a non-discriminating competitor, so in a competitive market the discriminator should lose ground over time. Discrimination should survive best where competition is weak, in monopolies, in regulated industries, in the public sector.

There is evidence for the model. Kerwin Charles and Jonathan Guryan found that the prejudice of the marginal white person in a state, measured by attitude surveys, predicts the Black-white wage gap in that state in the way Becker's model implies.

Fifteen years later, Edmund Phelps and Kenneth Arrow developed a very different account. In statistical discrimination, the decision maker has no preference about race at all. She has imperfect information about an individual applicant, gathering more is costly, and she knows or believes something about group averages. So she uses group membership as a proxy for the trait she actually cares about.

Four features of this model deserve attention, because it is the harder case in every way.

It can be individually rational. It does not erode under competition, since a firm using an informative proxy is not paying a penalty. It harms individuals precisely to the extent that they differ from their group average, which means the most qualified members of a disadvantaged group bear the largest cost. And in Arrow's version it can be self-fulfilling: if employers expect lower returns from a group, members of that group get lower returns on investments in skill, so they invest less, so the belief becomes accurate. An equilibrium of that kind requires no prejudice and no error to persist.

One legal point that students frequently get wrong. Statistical discrimination, however rational, is illegal under American antidiscrimination law when it involves a protected characteristic. Rationality is not a defense. The law requires individual assessment on legitimate criteria.

Key idea: Taste-based discrimination is a preference that competition should erode; statistical discrimination is an inference that competition does not erode, is often individually rational, can be self-fulfilling, and is illegal anyway.

Telling them apart, and why it matters

Distinguishing the two empirically is genuinely difficult, and one of the cleanest pieces of evidence comes from the ban-the-box research we met last module. When employers were barred from asking about criminal records early in the process, the racial gap in callbacks among young men without records widened. That is the signature of statistical discrimination: deprived of individual information, decision makers fell back on group averages.

The policy implications diverge sharply. If discrimination is mostly taste-based, remedies point toward competition, enforcement, and penalties. If it is mostly statistical, penalties alone will not fix it, and the leverage is in better individual information, credentials, certifications, structured assessments, and work histories that let people distinguish themselves from an average. Getting the mechanism wrong means prescribing the wrong intervention, which is the practical reason this distinction is not merely academic.

Segregation, and three explanations of it

The standard measure is the index of dissimilarity, which reports the share of one group that would have to move to achieve an even distribution across neighborhoods. It runs from 0 to 1, and values above 0.6 are conventionally called high.

Black-white dissimilarity in American metropolitan areas averaged about 0.79 in 1970 and has fallen to roughly 0.55 to 0.60 today. That is real progress and it is also, after half a century, still high. Douglas Massey and Nancy Denton's American Apartheid (1993) introduced the concept of hypersegregation, applying to metros that score high on several distinct dimensions of segregation at once, a condition that concentrates disadvantage far more than any single measure suggests.

Three explanations compete, and the honest answer is that all three contribute.

Preferences. Thomas Schelling showed with a simple model that even mild preferences about neighbors can produce near-total segregation, because individual moves cascade. His agents were content in a mixed neighborhood and moved only if a modest threshold was crossed, and the result was extreme sorting. Survey evidence complicates the picture in an important direction: Black respondents consistently express preference for substantially mixed neighborhoods, while white respondents' expressed comfort declines as the minority share rises past modest levels. So preferences are asymmetric, not symmetric.

Income. Groups with different income distributions will sort differently across a housing market priced by income. But this cannot be the whole story, and the reason is decisive. Segregation persists at every income level. Research by Sean Reardon, Patrick Sharkey, John Logan and others has repeatedly found that high-income Black households live in neighborhoods with poverty rates and school quality resembling those of much lower-income white households. Money does not buy the same neighborhood.

Discrimination. HUD's paired-testing studies, most recently the 2012 Housing Discrimination Study, send matched testers of different races to inquire about the same advertised units. Outright refusal is now rare. What persists is quieter: minority testers are told about and shown fewer units, which reduces choice without producing a single refusal anyone could complain about. Steering by agents toward particular neighborhoods has also been documented.

Key idea: Segregation is produced jointly by asymmetric preferences, income sorting, and ongoing discrimination in the form of differential information rather than refusal, and the persistence of segregation at every income level rules out any purely economic account.

Why segregation matters: the neighborhood effects story

This is one of the best examples in social science of a question that got a clearer answer over time.

The Moving to Opportunity experiment, launched in the 1990s, randomly assigned housing vouchers to families in high-poverty neighborhoods. Early evaluations found improvements in mental health and safety but disappointing effects on adult employment and earnings. For years this was read as evidence that neighborhoods matter less than people assumed.

Then Raj Chetty, Nathaniel Hendren, and Lawrence Katz revisited the experiment with tax data on the children, who had by then grown up. Children who moved to lower-poverty areas before about age 13 had substantially higher adult earnings, roughly 31 percent higher, along with higher college attendance rates and better family outcomes. Children who moved as teenagers showed no such gains and by some measures did slightly worse, consistent with the disruption of moving late.

The lesson is about exposure time. Neighborhood effects operate through years of accumulated childhood exposure, which is why an experiment measuring adults a few years out found little, and an experiment measuring their children twenty years out found a great deal. Patrick Sharkey's work extends this across generations, showing that the neighborhood a child's parents grew up in predicts outcomes over and above the child's own neighborhood.

Networks and opportunity hoarding

The last mechanism is the most legal and the least visible.

Mark Granovetter's Getting a Job established that a large share of jobs are found through personal contacts, and that weak ties, acquaintances rather than close friends, are especially productive because they reach into different information pools. Now combine that with segregation and with the fact that American social networks remain highly racially homogeneous. Referral hiring, which no one considers discriminatory and which employers favor because referred workers perform well, mechanically reproduces the existing racial composition of a workforce.

Deirdre Royster demonstrated this with unusual clarity. She studied Black and white graduates of the same Baltimore vocational program, with comparable grades, credentials, and instructors. The white graduates found skilled trade work at much higher rates, and the reason was not credentials. It was that instructors, union members, and employers activated networks on their behalf. Equal training produced unequal access because access ran through relationships.

Charles Tilly gave the general theory in Durable Inequality (1998). Categorical inequality, he argued, is reproduced primarily at organizational boundaries through two mechanisms: exploitation, where an in-group captures value produced by an excluded group, and opportunity hoarding, where a network secures access to a valuable resource and restricts it to members. Hoarding does not require hostility toward outsiders. It requires only that insiders prefer insiders.

Look at the American examples and notice how each is defended on non-racial grounds: legacy admissions preserve institutional tradition and alumni giving; exclusionary single-family zoning preserves neighborhood character and property values; unpaid internships are voluntary; school district boundaries reflect local control; referral bonuses improve hiring quality. Every one of these has a coherent defense that never mentions race. Every one of them, applied within racially homogeneous networks, transmits advantage along racial lines.

Key idea: Opportunity hoarding reproduces categorical inequality through practices that are legal, defensible on non-racial grounds, and effective precisely because insiders prefer insiders rather than because they dislike outsiders.

Common misconceptions

• Discrimination requires prejudice. Statistical discrimination requires only imperfect information and beliefs about group averages, and it can persist in equilibrium with no prejudice and no error.
• Rational discrimination is legal. Using a protected characteristic as a proxy is unlawful under American antidiscrimination law regardless of how well it predicts.
• Segregation today is just income sorting. Segregation persists at every income level, and high-income Black households live in neighborhoods resembling those of much lower-income white households.
• Moving to Opportunity proved neighborhoods do not matter. The early adult results were modest; the follow-up found children who moved before about age 13 earned roughly 31 percent more as adults.
• Referral hiring is racially neutral because it does not consider race. Applied within homogeneous networks it mechanically reproduces the existing composition of a workforce.

Recap

• Becker's taste-based discrimination is a costly preference that competition should erode; Phelps and Arrow's statistical discrimination is an inference that competition does not erode and that can be self-fulfilling.
• Ban-the-box results, in which callback gaps widened when criminal history was hidden, are a clear empirical signature of statistical discrimination.
• Black-white dissimilarity fell from about 0.79 in 1970 to roughly 0.55 to 0.60 today, and is produced jointly by asymmetric preferences, income sorting, and differential information from housing providers.
• Moving to Opportunity showed neighborhood effects operate through childhood exposure time, with large adult earnings gains for children who moved young.
• Tilly's exploitation and opportunity hoarding, operating through referral hiring, legacy admissions, zoning, and district boundaries, transmit advantage along racial lines through legal and defensible practices.

Sources

1. Chetty, R., Hendren, N., and Katz, L. F. (2016). The effects of exposure to better neighborhoods on children: New evidence from the Moving to Opportunity experiment. American Economic Review, 106(4), 855-902. doi.org
2. U.S. Department of Housing and Urban Development, Office of Policy Development and Research. (n.d.). Housing discrimination studies. huduser.gov
3. Wikipedia contributors. (2025). Schelling's model of segregation. In Wikipedia. wikipedia.org
4. Wikipedia contributors. (2025). Residential segregation in the United States. In Wikipedia. wikipedia.org
5. Britannica. (2025). Racial segregation. In Encyclopaedia Britannica. britannica.com
6. United States Census Bureau. (n.d.). Housing topics and data. census.gov

The big picture

The last two lessons dealt with mechanisms you can measure. This one deals with three arguments you cannot settle with a regression, because each involves a genuine disagreement about which principles should govern, which evidence counts, and how to weigh competing goods.

That does not make them unserious. It means they need a different kind of treatment. My commitment for the next few pages is to state each position in the form its most capable defender would recognize, and to tell you honestly where I think the evidence constrains the argument and where it does not.

Colorblindness

The principle is that individuals should be treated without regard to race, by government especially. Its lineage in American law is distinguished. Justice John Marshall Harlan, dissenting alone in Plessy in 1896, wrote that the Constitution is colorblind. Martin Luther King's most quoted sentence expresses the aspiration in moral terms. And Chief Justice John Roberts, in Parents Involved in 2007, gave it its sharpest modern formulation: the way to stop discrimination on the basis of race is to stop discriminating on the basis of race.

The strongest case for it has four parts. The moral argument is that a principle against sorting people by race must apply generally or it is not a principle at all; it is a preference for particular outcomes wearing a principle's clothes. The historical argument is that government classification by race has an appalling track record, and that the machinery built for benign purposes does not stay confined to them. The practical argument, pressed recently by writers such as Coleman Hughes, is that class-based or need-based policies reach disadvantaged people more precisely, avoid stigmatizing beneficiaries, and do not require the state to decide who belongs to which group. The political argument is that colorblind principles command broad and stable public support across demographic lines, which matters for whether a policy survives.

The strongest case against is not a defense of racial hierarchy, and treating it as one is the most common failure in this debate. It has three parts. The first is logical: if past discrimination produced present differences in wealth, neighborhood, and network, then a rule of ignoring race from now on preserves those differences rather than remedying them. The second is empirical, from Eduardo Bonilla-Silva, who argues that colorblind language functions in practice as a set of frames, abstract appeals to liberal principle, treating segregation as natural, cultural explanations for inequality, and minimizing the significance of discrimination, that let people account for stark inequality without ever expressing prejudice. The third is practical: research by Evan Apfelbaum, Michael Norton, and Samuel Sommers finds that strong colorblind norms in organizations can make people less willing to notice or name discrimination when it occurs, because acknowledging it requires acknowledging race.

Justice Sonia Sotomayor's answer to Roberts is the compact version: the way to stop discrimination on the basis of race is to speak openly and candidly on the subject of race.

Here is the thing worth noticing, because students rarely see it. Both camps overwhelmingly share the destination: a society in which race does not determine life chances. The disagreement is about the route, and specifically about whether race-conscious remedies are a bridge to that destination or an obstacle to it. That is a disagreement about empirical consequences and about moral priority, and it is genuinely hard.

Key idea: Colorblindness is defended on grounds of principle, history, precision, and durability, and criticized on grounds that ignoring race locks in differences that race produced; both sides generally share the goal and disagree about the path.

Critical race theory, described accurately

Critical race theory is a body of legal scholarship that emerged in American law schools in the late 1970s and 1980s, initially from scholars dissatisfied with both mainstream civil rights liberalism and with critical legal studies. Its founding figures include Derrick Bell, Kimberle Crenshaw, Richard Delgado, Mari Matsuda, Cheryl Harris, and Patricia Williams, and the first workshop under the name was held in 1989.

Its principal claims, stated as its own scholars state them, are these. Racism is ordinary rather than aberrational, a routine feature of American institutions rather than an occasional deviation. Race is socially constructed. Formal legal equality is insufficient to remedy inequality produced by centuries of formal legal inequality. Bell's interest convergence thesis holds that gains for Black Americans have tended to arrive when they also served the interests of white elites, and he applied it controversially to Brown, arguing that Cold War competition for international opinion shaped its timing. Harris's whiteness as property argues that legal doctrine has treated white identity as a form of property interest deserving protection. Crenshaw's intersectionality, introduced in 1989, began as a legal observation: Black women plaintiffs could lose discrimination cases because courts required them to show either race discrimination that also affected Black men or sex discrimination that also affected white women, leaving the specific intersection unremediable.

The tradition also uses narrative and counterstorytelling as method, on the argument that legal doctrine encodes a perspective and that competing perspectives need articulation.

Now the scholarly criticism, which is substantive and comes largely from within legal academia rather than from outside it. Randall Kennedy, writing in the Harvard Law Review in 1989, criticized the claim that scholars of color possess a distinctive voice conferring special authority, arguing that it was empirically unsupported and risked a form of essentialism the field elsewhere rejects. Daniel Farber and Suzanna Sherry, in Beyond All Reason (1997), argued that storytelling as a method sidesteps ordinary standards of evidence and that a general critique of merit criteria as instruments of racial hierarchy has implications the field has not confronted, since some minority groups outperform the majority on precisely those criteria. Others have pressed the falsifiability objection: if a theory can accommodate both progress, by reading it as interest convergence, and stagnation, by reading it as ordinary racism, it is unclear what evidence would count against it.

Kennedy, Farber, and Sherry are liberals writing in mainstream legal journals. That is worth stating because it shows the critique is not reducible to a political position.

Finally, a fact about language that you need in order to read the news. Since about 2020, critical race theory in political usage has come to refer to a much broader set of things: elementary and secondary curricula, workplace diversity programming, and general discussion of racism in schools. Legislation passed in a number of states between 2021 and 2023 restricted instruction using language that generally does not track the academic literature. Whether one supports or opposes those laws, the descriptive point stands: the phrase now names two different objects, and arguments about it routinely fail because participants are discussing different things.

Key idea: Critical race theory is a specific legal-academic tradition with identifiable claims and serious scholarly critics, and the contemporary political usage of the term refers to something substantially broader, which makes naming the referent a precondition for any productive argument.

Culture and structure

The third argument concerns explanation. When a group shows persistently different outcomes, how much weight goes to structural conditions such as labor markets, discrimination, and neighborhood, and how much to cultural patterns such as norms, expectations, and practices?

The modern history of this question in American sociology begins badly. Daniel Patrick Moynihan's 1965 report on the Black family argued that family structure had become an independent source of disadvantage. The reception was ferocious, much of it reading the report as blaming victims for conditions imposed on them. One lasting effect was that many sociologists avoided cultural explanation of racial inequality for a generation, which several scholars now regard as a serious cost to the discipline.

William Julius Wilson did more than anyone to reopen it responsibly. The Truly Disadvantaged (1987) argued that concentrated urban poverty resulted primarily from structural forces: the loss of manufacturing employment, the spatial mismatch between where jobs moved and where Black workers lived, and the out-migration of middle-class families once civil rights law permitted it. Two decades later, in More Than Just Race (2009), Wilson argued that a complete account requires cultural factors as well, while insisting on a definition of culture as shared frames, meanings, and repertoires that people develop in response to conditions, rather than as inherited values or deficiency.

Mario Small, David Harding, and Michele Lamont formalized that program in 2010, showing how concepts such as frames, narratives, and cultural repertoires can be measured and tested without carrying the moralizing baggage that made the earlier debate toxic.

The other side deserves its strongest statement. Thomas Sowell has argued across many books that cultural patterns have independent causal force, that they travel with populations across borders and institutions, and that they explain differences among groups facing similar formal treatment. Glenn Loury has argued that purely structural accounts risk unfalsifiability, and has developed the concept of racial stigma, the way a negative social meaning attached to a group shapes both how others read behavior and how members respond, in a loop that neither pure discrimination nor pure culture captures. Orlando Patterson, a sociologist of the left, has argued that the discipline's avoidance of cultural explanation was a scientific failure with real costs.

Two observations should discipline your thinking here. First, culture and structure are not independent variables; culture is produced by and responds to structure, and structure is sustained by cultural understandings. Second, culture is frequently used as a residual, a name for whatever the measured variables failed to explain, which is the same error we identified with regression residuals in Module 3.

So the productive question is not which one, but which specific mechanism, measured how. Does the effect run through expectations of return on schooling? Through information about how institutions work? Through peer networks in a particular kind of school? Each of those is testable. The general question of culture versus structure is not.

Key idea: Culture and structure are entangled rather than competing, and the question becomes tractable only when a specific mechanism is named and measured instead of a general causal share being argued about.

Common misconceptions

• Opponents of race-conscious policy want racial hierarchy. The mainstream case for colorblindness rests on universalist principle, historical caution about state racial classification, and targeting precision.
• Critics of colorblindness reject equal treatment. Their claim is that a rule of ignoring race preserves differences that race-conscious policy produced, which is an argument about consequences.
• Critical race theory is a K-12 curriculum. It is a legal-academic tradition from the 1970s and 1980s; the term's contemporary political usage refers to a substantially broader set of practices.
• Criticism of critical race theory comes only from the political right. Randall Kennedy, Daniel Farber, and Suzanna Sherry published influential critiques in mainstream legal journals from liberal positions.
• The culture-versus-structure debate is a choice between two explanations. They are entangled, and the tractable question is which specific mechanism operates and how it can be measured.

Recap

• Colorblindness is defended on moral, historical, practical, and political grounds and criticized on the ground that ignoring race entrenches differences race produced; the camps largely share the goal.
• Bonilla-Silva's frames and the Apfelbaum research supply the empirical case that colorblind norms can suppress recognition of discrimination.
• Critical race theory's core claims include the ordinariness of racism, interest convergence, whiteness as property, and intersectionality, developed by Bell, Crenshaw, Harris, Delgado, and others.
• Kennedy, Farber, and Sherry raised serious scholarly objections concerning essentialism, evidentiary standards, and falsifiability, from liberal positions within legal academia.
• Wilson, Small, Harding, and Lamont rehabilitated cultural analysis as frames and repertoires, while Sowell, Loury, and Patterson press the case for culture's independent force, and the entanglement of the two makes mechanism the tractable question.

Sources

1. Britannica. (2025). Critical race theory. In Encyclopaedia Britannica. britannica.com
2. Crenshaw, K. (1989). Demarginalizing the intersection of race and sex. University of Chicago Legal Forum, 1989(1), Article 8. chicagounbound.uchicago.edu
3. Parents Involved in Community Schools v. Seattle School District No. 1, 551 U.S. 701 (2007). supreme.justia.com
4. Britannica. (2025). Intersectionality. In Encyclopaedia Britannica. britannica.com
5. Wikipedia contributors. (2025). Critical race theory. In Wikipedia. wikipedia.org
6. Plessy v. Ferguson, 163 U.S. 537 (1896), Harlan, J., dissenting. supreme.justia.com

The big picture

No institution in this course generates more heat with less shared ground than criminal justice. So we are going to do something specific here: walk the process stage by stage, and at each stage give you the best study supporting a discrimination finding and the best study or critique cutting the other way.

By the end you should be able to say, for each stage, what the evidence supports strongly, what it supports weakly, and what remains genuinely open. You will notice that the answer differs by stage, which is itself one of the most useful things to know.

Start with the aggregate. Bureau of Justice Statistics data show Black adults imprisoned at roughly five times the rate of white adults. That is the fact everything else in this lesson is trying to explain, and the explanations available range from differences in offending, to differences in enforcement, to accumulated differences at each decision point, and the honest answer involves all three in proportions that vary by offense and jurisdiction.

Stage one: stops and searches

Emma Pierson and colleagues assembled roughly 95 million traffic stop records from state patrols and municipal departments and applied two clever tests.

The first is the veil of darkness. Compare stops in the same location at the same clock time before and after sunset across the year, when a driver's race is harder to observe. If disparities shrink after dark, the stop decision is responding to race. They shrank.

The second is a threshold test for searches. If officers apply a lower evidentiary bar to some drivers, searches of those drivers should turn up contraband less often. The inferred search thresholds were lower for Black and Hispanic drivers than for white drivers.

The New York City stop-and-frisk record supplies a concrete case. Between 2004 and 2012 the department recorded roughly 4.4 million stops, of which about 83 percent involved Black or Hispanic New Yorkers, with roughly 6 percent producing an arrest. In Floyd v. City of New York (2013), a federal court found the program unconstitutional as implemented.

What weakens the stop findings? Mainly that stop rates depend on driving exposure, patrol allocation, and calls for service, and that patrol allocation often follows crime reports, which are not distributed evenly. The veil-of-darkness design is powerful precisely because it holds location and time constant, which is why it carries more weight than raw stop rate comparisons.

Key idea: The strongest stop evidence does not compare raw rates but uses designs holding location and time fixed, and those designs find disparities consistent with race entering the stop and search decision.

Stage two: use of force, and the benchmarking problem

This is the most contested area in the field, and the dispute is fundamentally about denominators.

Roland Fryer analyzed several datasets, including detailed Houston records, and reported two findings. On non-lethal force, Black and Hispanic civilians were substantially more likely than white civilians to experience hands-on force during an interaction. On officer-involved shootings, conditional on an interaction having occurred, he found no statistically significant racial difference in the data he examined. The second finding drew enormous attention.

Dean Knox, Will Lowe, and Jonathan Mummolo published the decisive methodological response. If the police-civilian encounter is itself racially selective, then conditioning the analysis on encounters introduces what statisticians call post-treatment bias. In plain terms: if officers stop Black civilians on weaker grounds, then the pool of stopped Black civilians includes more people who did nothing, and comparing outcomes within that pool understates bias, potentially severely. Their conclusion was not that Fryer's numbers were miscalculated but that no analysis conditioned on administrative encounter records can identify the quantity people care about.

The alternative benchmark is population. Databases maintained by news organizations and researchers consistently find Black Americans killed by police at roughly two and a half times the white per capita rate. Critics respond that per capita benchmarks ignore differences in encounter rates and in the circumstances producing encounters, and that the right denominator would be something like violent crime arrests or armed encounters, which changes results substantially.

You should also know that this literature has been volatile. A 2019 paper in a major journal reporting no anti-Black bias in fatal shootings was retracted after a dispute over how its own results had been characterized. That is a caution about drawing confident conclusions from any single study here.

The honest summary: non-lethal force disparities are well supported. Lethal force conclusions depend heavily on the choice of denominator, no available design cleanly identifies the causal quantity, and confident claims in either direction outrun the evidence.

Key idea: The central dispute over police use of force is a benchmarking dispute, and because every available denominator is defective in a known way, this is a genuinely open empirical question rather than a settled one.

Stage three: charging and bail

Here the evidence is stronger and more consistent, and it points somewhere specific.

M. Marit Rehavi and Sonja Starr traced federal cases from arrest through sentencing and found that Black male defendants received sentences roughly nine to ten percent longer than comparable white male defendants arrested for the same conduct, and that most of that gap originated not with judges but at the charging stage, particularly in prosecutors' decisions to file charges carrying mandatory minimums. That finding relocates the problem from the courtroom to the prosecutor's office, which matters for reform design.

David Arnold, Will Dobbie, and Crystal Yang studied bail decisions using an outcome test on marginal defendants, exploiting random assignment of cases to judges of varying strictness. They found evidence of racial bias against Black defendants, concentrated among less experienced judges and consistent with reliance on inaccurate stereotypes rather than deliberate animus. Note the mechanism: this looks like statistical discrimination with an inaccurate prior, which is the Module 4 model in operation.

Stage four: sentencing

The United States Sentencing Commission, an official federal body, has repeatedly examined demographic differences in federal sentencing. Its 2017 report found that Black male offenders received sentences on average about 19 percent longer than similarly situated white male offenders over the period studied, with much of the difference attributable to departures and variances not sponsored by the government.

The most famous sentencing disparity was written into statute. The Anti-Drug Abuse Act of 1986 set a 100 to 1 ratio between the quantity of powder cocaine and crack cocaine triggering the same mandatory minimum, and crack prosecutions fell overwhelmingly on Black defendants. The Fair Sentencing Act of 2010 reduced the ratio to 18 to 1, and the First Step Act of 2018 made that change retroactive.

Now a fact that complicates the standard narrative and that you should know. The 1986 legislation was supported by a substantial number of Black elected officials, including members of the Congressional Black Caucus, responding to constituents in neighborhoods devastated by crack-related violence. James Forman Jr., whose book on this won the Pulitzer Prize for general nonfiction in 2018, documents how Black officials in Washington, D.C., supported tougher enforcement for reasons rooted in protecting their own communities. This does not make the resulting disparity acceptable. It does mean any account that presents mass incarceration as something simply done to Black communities from outside is incomplete.

Key idea: Sentencing disparities are documented by the government's own commission, the largest statutory disparity was reduced in 2010 and made retroactive in 2018, and the political origins of these laws included substantial support from Black officials responding to constituent demand.

The other half: victimization and what communities want

A course that discussed only enforcement would mislead you badly.

Black Americans are roughly 13 percent of the population and account for a very disproportionate share of homicide victims; homicide has been among the leading causes of death for young Black men for decades, according to CDC mortality data. Under-policing is a documented harm as well as over-policing: unsolved homicides, slow response times, and unreturned calls are concentrated in the same neighborhoods as aggressive enforcement.

Survey evidence bears this out. Gallup polling has repeatedly found that a majority of Black Americans want police to spend the same amount of time or more time in their area, alongside strong majorities favoring major reform in how police operate. Those two findings are not in tension; they describe a demand for better policing rather than less of it, and any policy analysis that ignores either half is not describing the people it claims to represent.

What the policy evidence shows

The most instructive recent finding cuts both ways at once. Aaron Chalfin, Benjamin Hansen, Emily Weisburst, and Morgan Williams found that adding police officers reduces homicides, and that because Black Americans are disproportionately homicide victims, the lives saved per officer are disproportionately Black lives. The same study found that larger forces also increase low-level quality-of-life arrests, and that this burden falls disproportionately on Black civilians.

That is one study finding both a benefit and a cost distributed along the same racial line. It does not tell you what to do. It tells you that the trade-off is real and that arguments treating police staffing as purely harmful or purely beneficial are both ignoring half the result.

Two other bodies of evidence are worth carrying. Hot-spots policing, concentrating enforcement in the small number of micro-locations where crime clusters, has among the strongest evidence bases in criminology for reducing crime without simply displacing it. And federal consent decrees produce mixed results, with meaningful improvements in some departments and little measurable change in others, which suggests implementation matters more than the instrument.

Common misconceptions

• Fryer's study proved there is no racial bias in police shootings. Knox, Lowe, and Mummolo showed that conditioning on encounters cannot identify that quantity if encounters are themselves racially selective.
• Per capita comparisons settle the use-of-force question. They ignore differences in encounter rates, which is why the benchmarking problem has no agreed solution.
• Sentencing disparities originate mainly with judges. Rehavi and Starr found most of the federal gap arises at the charging stage, particularly in mandatory minimum filing decisions.
• Mass incarceration was imposed entirely from outside Black communities. Forman documents substantial support from Black officials responding to constituents facing severe violence.
• Black Americans want less policing. Gallup finds majorities wanting the same or more police presence alongside strong support for major reform in how policing is done.

Recap

• Black adults are imprisoned at roughly five times the white rate, and explaining that requires attention to offending, enforcement, and accumulation across decision points.
• Veil-of-darkness and threshold tests across about 95 million stops indicate that race enters stop and search decisions.
• Non-lethal force disparities are well supported, while lethal force conclusions turn on an unresolved benchmarking problem that no current design solves.
• Most of the federal sentencing gap originates at charging rather than at sentencing, and bail research finds bias concentrated among less experienced judges.
• Black Americans are disproportionately victims of violence and majorities want equal or greater police presence with major reform, and added officers reduce homicides while increasing low-level arrests, both effects falling disproportionately on Black civilians.

Sources

1. Bureau of Justice Statistics. (n.d.). Prisoners series and corrections statistics. U.S. Department of Justice. bjs.ojp.gov
2. Pierson, E., Simoiu, C., Overgoor, J., Corbett-Davies, S., Jenson, D., Shoemaker, A., ... Goel, S. (2020). A large-scale analysis of racial disparities in police stops across the United States. Nature Human Behaviour, 4, 736-745. doi.org
3. Fryer, R. G. (2019). An empirical analysis of racial differences in police use of force. Journal of Political Economy, 127(3), 1210-1261. doi.org
4. Knox, D., Lowe, W., and Mummolo, J. (2020). Administrative records mask racially biased policing. American Political Science Review, 114(3), 619-637. doi.org
5. Arnold, D., Dobbie, W., and Yang, C. S. (2018). Racial bias in bail decisions. Quarterly Journal of Economics, 133(4), 1885-1932. doi.org
6. United States Sentencing Commission. (n.d.). Research on demographic differences in sentencing. ussc.gov
7. Centers for Disease Control and Prevention. (n.d.). Leading causes of death and injury. cdc.gov

The big picture

Five institutions in one lesson, which means we move fast. For each I will give you the strongest measured finding, the strongest complication, and the single mechanism that best repays attention. The point is not encyclopedic coverage; it is to show you that the analytic moves you have learned transfer across domains.

Schools

School segregation today is mostly not the segregation of 1954. The dominant form is between districts rather than within them, which is a direct legacy of Milliken v. Bradley. Researchers at the UCLA Civil Rights Project have documented that the share of intensely segregated schools, those enrolling 90 to 100 percent students of color, has grown since about 1990 as districts were released from court supervision.

Funding is more contested than most people realize, and it is worth being careful. An influential 2019 EdBuild analysis found that predominantly non-white school districts received about 23 billion dollars less than predominantly white districts serving the same number of students. Other analyses, accounting for state and federal aid that flows toward higher-need districts, find much smaller gaps, and in some states a reversal, while noting that equal dollars do not mean equal adequacy since educating higher-need students costs more. Both sets of numbers are correct about what they measure. What they measure differs.

The strongest causal evidence in this area is more encouraging than the descriptive picture. Rucker Johnson used the staggered timing of court-ordered desegregation to show that Black students who attended desegregated schools for more years completed more education, earned more as adults, and had better adult health, with no detectable harm to white students' outcomes. Kirabo Jackson, Rucker Johnson, and Claudia Persico used school finance reform litigation as a natural experiment and found that increased per-pupil spending raised educational attainment and adult earnings, with the largest effects for low-income children. Together these are among the most robust positive findings in the field: both integration and money, done at sufficient scale and duration, appear to work.

The discipline gap is genuinely disputed. Federal civil rights data show Black students suspended at roughly three times the rate of white students. Russell Skiba and colleagues find the gap is not explained by differences in the behavior recorded, pointing to differential response to similar conduct. Other researchers, using teacher-reported prior behavior measures, find that much of the gap is accounted for once those measures are included. The dispute turns on whether the behavior ratings are themselves affected by the perceptions under study, which is the bad control problem from Module 3 in a new setting.

Key idea: School segregation today operates mainly across district lines, funding comparisons depend heavily on what is counted, and the strongest causal studies find that sustained desegregation and increased spending both improve long-run outcomes.

Housing and lending

Home Mortgage Disclosure Act data, which lenders must report, consistently show higher denial rates for Black and Hispanic applicants than for white applicants. The interesting question is what survives adjustment for credit characteristics.

Federal Reserve research by Neil Bhutta and Aurel Hizmo found that, conditional on observable credit characteristics, differences in the interest rates offered were small, while differences in denial remained. Robert Bartlett and colleagues found that Latino and Black borrowers paid modestly higher rates on purchase mortgages, and reported a finding worth pausing on: algorithmic lenders discriminated in pricing roughly 40 percent less than face-to-face lenders, but still discriminated. Automation reduced the effect without eliminating it, which is informative about mechanism.

Appraisal is an active area. A 2021 Freddie Mac analysis found that appraisals came in below the contract price more often for properties in Black and Latino census tracts. Brookings researchers estimated substantial average devaluation of homes in majority-Black neighborhoods after adjusting for structural and neighborhood characteristics. Critics argue that unmeasured amenity differences, school quality, and expected appreciation could account for part of the residual. That is the decomposition residual argument again, and it is a legitimate objection that has not been fully resolved.

On enforcement, the Supreme Court held in Texas Department of Housing and Community Affairs v. Inclusive Communities Project (2015) that disparate impact claims are available under the Fair Housing Act, subject to safeguards including a robust causality requirement. The disparate impact framework from Module 1 therefore remains live in housing law.

Health care

Three findings here are unusually clean, and each identifies a mechanism rather than a correlation.

Ziad Obermeyer and colleagues examined a widely used commercial algorithm that identified patients for extra care management. It used predicted health costs as a proxy for health need. Because less is spent on Black patients at any given level of illness, the algorithm systematically rated equally sick Black patients as lower risk. Correcting the target variable would have raised the share of Black patients flagged for additional help from about 17.7 percent to about 46.5 percent. Note the structure: no programmer chose race, and the proxy carried the history.

Kelly Hoffman and colleagues surveyed medical students and residents and found that a substantial share endorsed false beliefs about biological differences between Black and white patients, such as differences in skin thickness or nerve sensitivity, and that those who did made less accurate pain treatment recommendations for Black patients.

Michael Sjoding and colleagues found that pulse oximeters, which estimate blood oxygen optically, missed hypoxemia roughly three times more often in Black patients than white patients. This is a device calibration problem with life-and-death consequences that came to broad attention during the COVID-19 pandemic, when oxygen levels governed treatment decisions.

Key idea: The best-documented health care mechanisms are structural in a very literal sense: a proxy variable in an algorithm, a false belief carried through training, and a device calibrated on light skin, none of which requires a prejudiced clinician.

Media

Content analyses from the Annenberg Inclusion Initiative and the UCLA Hollywood Diversity Report track speaking roles, directing, and executive positions, and show representation for Black performers roughly at population share in film leads in recent years, with Latino representation persistently far below population share and behind-the-camera roles lagging in every category.

The older and more consequential line of research concerns news. Travis Dixon and Daniel Linz found that local television news in Los Angeles overrepresented Black individuals as perpetrators relative to arrest statistics and underrepresented them as victims. Replications in other markets have produced mixed results, with some finding smaller or reversed patterns, so the safest statement is that distortion is documented in specific markets and periods rather than as a universal law.

Effects research is genuinely mixed. Some studies find that exposure to crime coverage shifts punitive policy attitudes; others find small or null effects. This is a place to resist confident claims in either direction.

Voting

Two Supreme Court decisions bracket the current landscape. Shelby County v. Holder (2013) struck down the coverage formula determining which jurisdictions needed federal preclearance, effectively ending that requirement. Allen v. Milligan (2023) upheld a Section 2 challenge to Alabama's congressional map, confirming that the Voting Rights Act's remaining provisions retain force. Anyone who tells you the Act is either intact or dead is oversimplifying.

Voter identification laws are the best example in this course of a contested empirical literature that moved. An influential 2017 study reported large negative turnout effects for minority voters in strict ID states. A 2018 reanalysis by Justin Grimmer and colleagues found that the results were sensitive to coding and data problems and did not hold up. Enrico Cantoni and Vincent Pons, using a large national voter file panel, found no significant negative effect of strict ID laws on turnout overall or by race, and no detectable effect on fraud either. The current weight of evidence suggests turnout effects are small; researchers still disagree about magnitude, and reasonable people disagree about whether small measured effects settle the normative question.

Two other facts matter. Black turnout exceeded white turnout nationally in 2012, according to Census survey data, before declining in 2016, which shows that turnout gaps are not fixed. And felony disenfranchisement excluded roughly 4.6 million Americans as of 2022 by Sentencing Project estimates, with rates for Black adults several times the national rate, a restriction that Florida voters moved to narrow by referendum in 2018.

Key idea: Voting rights law is partially intact, the empirical literature on voter ID effects corrected itself toward smaller estimates, and felony disenfranchisement remains the largest single race-differentiated restriction on the franchise.

Common misconceptions

• School segregation today happens mainly inside districts. Most of it is between districts, which is why Milliken limited the available remedies so consequentially.
• Non-white districts are uniformly underfunded. The 23 billion dollar figure counts local revenue; analyses including state and federal aid find smaller gaps, though equal dollars are not equal adequacy.
• The health care algorithm problem was caused by racist programmers. The algorithm used cost as a proxy for need and inherited the pattern that less is spent on equally sick Black patients.
• Voter ID laws have been shown to substantially depress minority turnout. The influential early finding did not survive reanalysis, and the best current national evidence finds small effects.
• Automation removes discrimination from lending. Bartlett and colleagues found algorithmic lenders discriminated about 40 percent less than face-to-face lenders, which is a reduction rather than an elimination.

Recap

• Intensely segregated schools have grown since 1990, mostly through between-district separation, and Johnson's and Jackson's natural experiments show desegregation and increased spending both improve long-run outcomes.
• Lending gaps persist mainly in denial rather than pricing, appraisal disparities are documented and their residual contested, and disparate impact remains available under the Fair Housing Act after Inclusive Communities.
• Obermeyer's algorithm study, Hoffman's false-beliefs study, and Sjoding's pulse oximetry finding identify concrete mechanisms requiring no prejudiced clinician.
• Media representation has improved unevenly, news distortion is documented in specific markets, and effects research is mixed enough to warrant caution.
• Shelby County ended preclearance while Allen v. Milligan preserved Section 2, voter ID turnout effects appear small in the best current evidence, and felony disenfranchisement excluded about 4.6 million people in 2022.

Sources

1. UCLA Civil Rights Project. (n.d.). Research on school segregation and diversity. civilrightsproject.ucla.edu
2. Obermeyer, Z., Powers, B., Vogeli, C., and Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447-453. doi.org
3. Hoffman, K. M., Trawalter, S., Axt, J. R., and Oliver, M. N. (2016). Racial bias in pain assessment and treatment recommendations. PNAS, 113(16), 4296-4301. doi.org
4. Sjoding, M. W., Dickson, R. P., Iwashyna, T. J., Gay, S. E., and Valley, T. S. (2020). Racial bias in pulse oximetry measurement. New England Journal of Medicine, 383(25), 2477-2478. doi.org
5. Cantoni, E., and Pons, V. (2021). Strict ID laws don't stop voters: Evidence from a U.S. nationwide panel, 2008-2018. Quarterly Journal of Economics, 136(4), 2615-2660. doi.org
6. Federal Financial Institutions Examination Council. (n.d.). Home Mortgage Disclosure Act data. ffiec.gov
7. U.S. Department of Education, Office for Civil Rights. (n.d.). Civil rights data collection. ocrdata.ed.gov

The big picture

Every category in this course is under strain right now, and the strain comes from demography. Immigration has remade the ethnic composition of the country twice in a century. Intermarriage has produced a large and growing population that the classification system handles poorly. Panethnic labels invented in the 1970s contain populations with almost nothing in common. And one group in this picture is not a racial category at all.

This lesson is about what happens to a classification system when the people it classifies stop cooperating with it.

Immigration and the remaking of American ethnicity

The foreign-born share of the United States population is now around 14 percent, approaching the historical peak of about 14.8 percent recorded in 1890. What changed after the Hart-Celler Act of 1965 was not primarily the volume but the origins. Before 1965 the national origins quota system channeled immigration from Europe. After it, the largest flows came from Latin America and Asia, and in recent years Asia has been the largest source region for new arrivals in several years.

That shift is the direct cause of most of what follows in this lesson. The categories Asian American and Hispanic barely functioned as social facts in 1960. They are central now.

Three theories of assimilation

Sociology has revised its account of what happens to immigrant groups three times, and the revisions are instructive.

The classical account, associated with Robert Park and elaborated by Milton Gordon in 1964, treated assimilation as a broadly linear sequence. Gordon's most durable contribution was distinguishing acculturation, adopting language and cultural practices, from structural assimilation, entering the primary group relationships of the host society through friendship and marriage. His key claim was that the first can occur without the second, and that only the second dissolves group boundaries.

The critique was straightforward: the model treated the mainstream as fixed and assimilation as one-directional, and it did not fit groups excluded by law and by race regardless of how thoroughly they acculturated.

Richard Alba and Victor Nee's new assimilation theory (2003) reframed it. Assimilation, on their account, is a largely unintended outcome of individual decisions made in pursuit of ordinary goals, and it is two-directional: the mainstream itself is altered by the groups entering it. American food, music, vocabulary, and religious composition are the evidence. Assimilation on this view does not require abandoning ethnicity; it requires that ethnicity stop determining life chances.

Alejandro Portes and Min Zhou's segmented assimilation (1993) argued that there is no single mainstream to assimilate into. They identified three paths: upward assimilation into the middle class; downward assimilation into disadvantaged urban sectors, where second-generation youth adopt the outlooks of the marginalized native-born peers around them; and selective acculturation, in which families deliberately retain language and community ties as protective resources. Which path a family takes depends on the context of reception, the human capital they arrive with, and the strength of the co-ethnic community.

The evidence broadly supports optimism about integration alongside real variation. A 2015 National Academies consensus report found that immigrants are integrating on language, education, and occupation about as fast as or faster than earlier waves, with English acquisition proceeding rapidly across generations. Ran Abramitzky and Leah Boustan, using linked historical and modern records, found that children of poor immigrants achieve higher upward mobility than children of similarly poor United States-born parents, and that this was true a century ago and remains true now. They also found that first-generation immigrants catch up more slowly than folklore suggests, which is why the second generation matters so much to the story.

Key idea: Assimilation is best understood as a two-way, largely unintended process with multiple possible paths, and the strongest evidence shows rapid integration across generations with outcomes that depend heavily on context of reception.

Panethnicity: how Asian American and Hispanic were built

Neither category was inherited. Both were made, and recently.

Yen Le Espiritu documented how Asian American panethnicity was constructed by activists in the late 1960s and consolidated through state classification, funding categories, and shared political struggle. The 1982 killing of Vincent Chin in Detroit, beaten to death by two autoworkers who reportedly blamed Japanese competition for job losses, became a defining case precisely because Chin was Chinese American and his assailants did not distinguish. Being lumped together produced a reason to organize together.

G. Cristina Mora traced the parallel construction of the Hispanic category in the 1970s and 1980s through the interaction of activists seeking political weight, the Census Bureau seeking a workable classification, and Spanish-language media seeking a national advertising market. All three had reasons to want one large category.

Now the internal diversity those labels conceal, which is substantial. Among Asian Americans, income inequality is the widest of any major racial grouping in the country: Indian American and Taiwanese American households have among the highest median incomes in the United States, while Hmong, Burmese, and Bhutanese American households have poverty rates far above the national average. Among Hispanics, people of Mexican origin are roughly six in ten, followed by Puerto Rican, Salvadoran, Cuban, Dominican, and Guatemalan populations with very different migration histories, legal statuses, and economic profiles.

Survey research from Pew consistently finds that most Hispanic adults prefer to describe themselves by national origin rather than by the panethnic term, and that the label Latinx is used by only a small share, in the low single digits. Most Asian Americans likewise identify first by national origin. The panethnic labels are more useful to institutions than to the people inside them, which is a finding worth remembering whenever you see data reported at that level.

The model minority stereotype deserves separate mention. It entered wide circulation through mid-1960s magazine journalism praising Japanese American success, in an era when it was explicitly deployed to contrast with Black political demands. Its costs are documented: it obscures the enormous internal variation just described, discourages provision of services, sets expectations that harm students who struggle, and coexists with substantial underrepresentation of Asian Americans in senior leadership positions relative to their professional representation.

Key idea: Panethnic categories were assembled in the 1970s by activists, agencies, and media with converging interests, and they conceal internal variation large enough that group averages can mislead badly.

Multiracial identity

The 2000 census was the first to let a respondent mark more than one race. The measured multiracial population has since grown enormously, though as we saw in Module 1, much of the 2010 to 2020 jump reflects changes in questions and coding rather than change in the population.

Two research findings matter more than the raw counts. First, identity is contextual: Pew surveys find that many adults with mixed ancestry do not describe themselves as multiracial, and that whether they do depends on appearance, upbringing, and which relatives they knew. Second, recorded race is fluid over time. Aliya Saperstein and Andrew Penner found that a person's race as recorded in longitudinal survey data changes with life events, with people who experience incarceration, unemployment, or poverty more likely to be classified as Black in a later wave and less likely to be classified as white. Race, in other words, is not simply an input to social outcomes; it also responds to them.

Intermarriage is the engine underneath. In 1967, when Loving v. Virginia was decided, about 3 percent of newlyweds married someone of a different race or ethnicity. That figure is now roughly 19 percent, with about one in ten married people overall in an intermarriage.

Sovereignty is a different question

One reminder before we leave the topic of categories, because it is the most common analytic error people make about Indian Country.

Census counts of people identifying as American Indian or Alaska Native have grown substantially, driven partly by increased identification among people with distant ancestry. Tribal citizenship is a different thing entirely. It is membership in a polity, determined by each nation according to its own criteria, and it is what triggers the legal relationship established by treaty. As Morton v. Mancari held and Haaland v. Brackeen reaffirmed in 2023 when it upheld the Indian Child Welfare Act, classifications based on tribal membership are political rather than racial.

So a rising census number and a tribal citizenship roll are measuring different things, and policy that confuses them will be incoherent.

The majority-minority projection and why it is contested

Census Bureau projections indicate that the non-Hispanic white share of the population will fall below 50 percent around the middle of this century. That projection is widely reported and frequently misunderstood.

Richard Alba's The Great Demographic Illusion (2020) makes the sharpest critique. The projection depends heavily on how people of mixed ancestry are counted, and the standard method assigns a child with one white parent and one Hispanic or Asian parent to the minority column. Alba's research indicates that many such young people live, marry, and are socially treated much like the white majority, meaning the projection may overstate the demographic transformation it appears to describe. He argues that the majority-minority framing encourages a zero-sum politics that the underlying demography does not support.

There is also evidence about the framing's effects. Maureen Craig and Jennifer Richeson found in experiments that exposing white respondents to information about majority-minority projections shifted their expressed political attitudes in a more conservative direction. That finding is about the effect of a message, not about the demography, but it explains why how the projection is described is not a neutral choice.

Key idea: The majority-minority projection is a product of counting conventions applied to a rapidly intermarrying population, and both its accuracy and its political effects are contested by serious researchers.

Common misconceptions

• Assimilation means immigrants give up their culture to join a fixed mainstream. Alba and Nee describe a two-way process in which the mainstream itself changes and ethnicity persists without determining life chances.
• Panethnic categories reflect shared identity. Asian American and Hispanic were constructed in the 1970s by activists, agencies, and media, and most people within them identify first by national origin.
• The model minority stereotype is harmless praise. It obscures very large internal variation, discourages service provision, and coexists with underrepresentation in senior leadership.
• A person's race is a fixed characteristic in data. Saperstein and Penner found recorded race shifts with life events such as incarceration and unemployment.
• The majority-minority projection is a straightforward demographic fact. It depends on counting conventions for mixed-ancestry children, and Alba argues those conventions overstate the transformation.

Recap

• The foreign-born share is near its 1890 peak, and the post-1965 shift in origins from Europe to Latin America and Asia produced the categories this lesson examines.
• Classical, new, and segmented assimilation theories differ over whether the mainstream is fixed and whether there is one path, and evidence shows rapid integration with substantial contextual variation.
• Espiritu and Mora documented the deliberate construction of Asian American and Hispanic panethnicity, which conceals the widest internal income inequality of any grouping.
• Intermarriage rose from about 3 percent of newlyweds in 1967 to roughly 19 percent, and recorded race is fluid rather than fixed.
• Tribal citizenship is political and distinct from census self-identification, and the majority-minority projection depends on contested counting conventions with measurable political effects.

Sources

1. Pew Research Center. (n.d.). Race and ethnicity, Hispanic, and Asian American research. pewresearch.org
2. United States Census Bureau. (n.d.). Foreign-born population. census.gov
3. National Academies of Sciences, Engineering, and Medicine. (2015). The integration of immigrants into American society. nationalacademies.org
4. Britannica. (2025). Assimilation: Society. In Encyclopaedia Britannica. britannica.com
5. Wikipedia contributors. (2025). Model minority. In Wikipedia. wikipedia.org
6. Wikipedia contributors. (2025). Multiracial Americans. In Wikipedia. wikipedia.org

The big picture

This is the last lesson, and it is the one where the course's method has to prove itself. Four policy debates, each of which you have probably encountered as a shouting match. My job is to give you each position in the form its ablest defender would recognize, tell you what evidence bears on it, and be explicit about where evidence runs out and values take over.

I am not going to tell you what to conclude. That is not modesty; it is the correct division of labor. Sociology can tell you what a policy is likely to do. Whether that is worth doing depends on what you think a society owes people, and that question is yours.

Affirmative action after SFFA

In June 2023, in Students for Fair Admissions v. Harvard and the companion case against the University of North Carolina, the Supreme Court held that the universities' race-conscious admissions programs violated the Equal Protection Clause and Title VI. Chief Justice Roberts wrote that the programs lacked measurable objectives, used race as a negative and as a stereotype, and had no logical end point. The opinion left two openings: applicants may still discuss how race has affected their lives, so long as the discussion is tied to a quality of character or a specific ability, and the Court declined to address the military service academies.

The case for race-conscious admissions. Zachary Bleemer studied California's Proposition 209, which banned racial preferences in 1998, using detailed applicant records. He found that the ban caused underrepresented minority applicants to cascade into less selective campuses, reduced their degree completion, and lowered their early-career wages by roughly 5 percent, with no offsetting gains for other applicants. William Bowen and Derek Bok's earlier study of admitted students found high graduation rates and disproportionate later civic and professional leadership. And there is evidence that professional representation itself has downstream effects: Marcella Alsan and colleagues found in a randomized study that Black male patients assigned to Black physicians agreed to more preventive services, a difference with real health implications.

The case against. The mismatch hypothesis, developed by Richard Sander, holds that admitting students with substantially weaker academic preparation into highly competitive environments can reduce their persistence, particularly in law and in STEM fields, so that a policy meant to help produces worse outcomes for intended beneficiaries. Peter Arcidiacono and Michael Lovenheim's review found evidence consistent with mismatch on some margins and inconsistent on others, and concluded the net effect depends on which margin you examine. There is also a distributional objection with strong empirical support: race-conscious admissions at selective institutions disproportionately benefit relatively advantaged members of minority groups, including immigrants and children of professionals, rather than the descendants of the specific historical harms usually cited. A third objection is one of principle, that state classification by race is wrong regardless of purpose. And there is public opinion: in multiple major surveys, majorities of Americans, including substantial shares of Black and Hispanic respondents, said race should not be a factor in admissions.

Two things complicate both sides. Class-based preferences, proposed by Richard Kahlenberg as a substitute, do help low-income students of every background, but at highly selective institutions they produce less racial diversity than race-conscious policies did, because of the wealth gap documented in Module 3. And legacy and athletic preferences, which are far larger in number at some institutions and overwhelmingly benefit white and affluent applicants, were untouched by the ruling, which is why complaints against them followed immediately.

Key idea: The affirmative action debate involves a genuine empirical dispute about mismatch and cascading, a genuine distributional dispute about who benefits, and a genuine dispute about principle, and no one of these settles the others.

Reparations

The case for. William Darity and Kirsten Mullen's From Here to Equality (2020) makes the most fully developed version. Their argument has three parts: the harms were inflicted or enforced by government, from slavery through legally mandated segregation to federal housing policy; the harms are documented and their economic consequences measurable; and the wealth gap is the appropriate metric because it is the accumulated residue of those harms. They propose eligibility based on documented descent from people enslaved in the United States combined with a period of self-identification, and estimate the cost of closing the racial wealth gap in the range of 10 to 14 trillion dollars.

The precedent argument is strong and specific. The United States has paid reparations before, most clearly under the Civil Liberties Act of 1988 for Japanese American incarceration, and at the state level in the Rosewood settlement in Florida in 1994. Evanston, Illinois, adopted a housing-based program in 2021, and California's state task force issued extensive recommendations in 2023.

The case against. The administrative objection is the most concrete: eligibility determination is genuinely hard. Roughly one in ten Black Americans is foreign born, with a larger share having at least one immigrant parent, and multiracial identification is rising and fluid. A lineage-based rule requires documentation that Reconstruction-era record keeping often makes impossible, and a self-identification rule invites the problems we saw with census fluidity.

The moral objection is that liability for wrongs committed by people long dead should not attach to a population that includes tens of millions descended from post-1865 immigrants, and that this differs categorically from the Japanese American case, where the government compensated the living people it had itself wronged.

The opportunity-cost objection comes from the left as often as the right. William Julius Wilson has argued for decades that race-neutral programs with disproportionate benefit to Black Americans, such as universal health coverage, early childhood investment, and full-employment policy, are more politically durable and reach poor people of all backgrounds. On this view the question is not whether the debt is owed but what the same political capital and dollars would accomplish deployed differently.

And the political fact: Pew found in 2021 that about 68 percent of American adults opposed cash reparations for descendants of enslaved people while about 77 percent of Black adults supported them. HR 40, a bill to study the question, has been introduced repeatedly since 1989 and has never passed.

Policing reform

Module 5 gave you the evidence. Here are the positions built on it.

More police, better trained. Rests on the finding that additional officers reduce homicide with disproportionately Black lives saved, combined with survey evidence that most Black Americans want equal or greater police presence. Must answer the same study's finding that added officers increase low-level arrests borne disproportionately by Black civilians.

Reallocate specific calls. Denver's STAR program, which sends behavioral health responders rather than police to certain calls, was evaluated with a design exploiting its staggered rollout and found substantial reductions in reported minor crimes at modest cost. Must answer the question of scale, since the calls suitable for reallocation are a limited share of the total.

Accountability and oversight. Consent decrees, civilian review, discipline reform, and changes to qualified immunity. Must answer that consent decree results are mixed across departments, suggesting implementation matters more than the instrument.

Community-based violence interruption. Programs employing credible messengers to interrupt retaliatory violence. Evaluations are genuinely mixed, with some strong local results and some null findings, so the honest position is promising but unproven at scale.

Key idea: Each policing position rests on real evidence and must answer a real objection from the same literature, which is why the productive question is which specific intervention in which specific context rather than more or less policing in general.

Immigration

The labor market debate has a famous center. David Card studied the 1980 Mariel boatlift, which added roughly 125,000 Cubans to Miami's labor force in months, and found no measurable effect on the wages or employment of low-skilled Miami workers. George Borjas reanalyzed the episode with a narrower sample of high school dropouts and reported large negative wage effects. Giovanni Peri and Vasil Yasenov disputed the sample restrictions. The exchange is unusually instructive because all three teams are competent and the disagreement is about specification, not arithmetic.

The National Academies' 2016 consensus report gives the field's best summary: little to no effect on the overall wages or employment of native-born workers over the long run; negative effects concentrated among prior immigrants and, in some studies, native workers without a high school diploma; and a fiscal picture in which the first generation costs state and local governments more than it contributes while the second generation is among the strongest net fiscal contributors in the population.

One dispute belongs specifically in this course. Borjas, Jeffrey Grogger, and Gordon Hanson found that immigration was associated with reduced employment and higher incarceration among Black men with low levels of education. Other researchers dispute the identification strategy and find smaller or no effects. This is a live disagreement, and it matters because it is a case where the interests of two disadvantaged groups may genuinely conflict, which is something coalition politics tends to obscure.

Public opinion is also more varied than commonly assumed. Pew surveys consistently find that Hispanic and Asian American respondents hold a wide range of positions on enforcement and legalization, and that partisanship, generation, and national origin predict those positions better than ethnicity does.

How to argue about all of this

You now have the whole toolkit. Use it in this order.

First, separate the claims. What is the descriptive claim, and what data would settle it? What is the causal claim, and what design would test it? What is the value judgment, and does the speaker acknowledge that it is one? Most bad arguments about race are two people asserting different kinds of claim at each other.

Second, name the mechanism. Not racism or culture, but which specific process, operating through what, measurable how.

Third, ask who disagrees and why. If you cannot state the strongest version of the opposing position, you do not yet understand your own.

Fourth, distinguish what is documented from what is contested. Slavery, Jim Crow, exclusion, incarceration, and redlining are documented. The causal weight of any one of them in a specific 2026 outcome is usually contested, and treating the second as though it inherited the certainty of the first is the most common error in this entire field.

Key idea: The discipline of separating description from causation from valuation, naming mechanisms, and stating opposing positions at full strength is what this course was actually teaching; the content was the occasion for it.

Common misconceptions

• SFFA banned all consideration of race in admissions. The Court held these programs unconstitutional while permitting applicants to discuss how race affected their lives when tied to character or ability, and declined to address the military academies.
• Ending affirmative action has no measurable effects. Bleemer found California's ban reduced degree completion and early-career wages for affected applicants.
• Mismatch has been refuted. Arcidiacono and Lovenheim found the evidence consistent with mismatch on some margins and not others, which is a contested finding rather than a settled one.
• Reparations have no American precedent. The Civil Liberties Act of 1988 compensated surviving incarcerated Japanese Americans, and Rosewood, Evanston, and California provide state and local precedents.
• Immigration's labor market effects are settled. Card, Borjas, and Peri reached different conclusions from the same episode, and the National Academies report describes effects that vary sharply by subgroup.

Recap

• SFFA held race-conscious admissions at Harvard and UNC unlawful while preserving discussion of race in an applicant's own account of character and ability.
• Bleemer's Proposition 209 findings and the mismatch literature pull in opposite directions, and the distributional objection about who benefits is independent of both.
• Darity and Mullen's reparations proposal rests on documented state-inflicted harm and the 1988 redress precedent, and faces serious administrative, moral, and opportunity-cost objections.
• Each policing position rests on real evidence and must answer a real objection from the same body of research, which makes context-specific questions more productive than general ones.
• Immigration's measured effects vary sharply by subgroup, the Mariel dispute remains unresolved, and the question of effects on Black workers with low education is genuinely contested.

Sources

1. Students for Fair Admissions v. President and Fellows of Harvard College, 600 U.S. 181 (2023). supremecourt.gov
2. Bleemer, Z. (2022). Affirmative action, mismatch, and economic mobility after California's Proposition 209. Quarterly Journal of Economics, 137(1), 115-160. doi.org
3. Britannica. (2025). Affirmative action. In Encyclopaedia Britannica. britannica.com
4. Wikipedia contributors. (2025). Reparations for slavery in the United States. In Wikipedia. wikipedia.org
5. National Academies of Sciences, Engineering, and Medicine. (2016). The economic and fiscal consequences of immigration. nationalacademies.org
6. Pew Research Center. (n.d.). Public opinion on race, reparations, and admissions. pewresearch.org
7. California Department of Justice. (2023). California Task Force to Study and Develop Reparation Proposals for African Americans. oag.ca.gov