Race as a Social Category

This is the lesson where 7.1.3's central distinction earns its place: constructed does not mean unreal.

Race has no basis as a system of natural biological kinds. It has organised property, law, policing, medicine, schooling and mortality for four centuries, and it predicts life outcomes in most societies that use it as powerfully as almost any variable social science has.

Holding both of those at once is the whole skill, and this lesson is about how.

One person, three countries, three races.

Consider a man with a mixed ancestry — some African, some European, some Indigenous — of a kind that describes a very large number of people in the Americas.

In the United States for most of its history, he is Black. The operative rule was hypodescent — assignment to the subordinate category — and in its extreme form it was codified: Virginia's legislature in 1924 defined as white only a person with no trace whatever of non-white ancestry. Any known African ancestry made a person Black, regardless of appearance.

In apartheid South Africa, he is most likely "Coloured" — a legally defined third category, distinct from both "Native" and "White", with its own residence rights, job reservations and identity documents. And the boundary was administered by a board. People applied to be reclassified; some succeeded and some did not; families were split across categories; and the tests used included appearance, associations, and — notoriously — whether a pencil pushed into the hair would fall out.

In Brazil, he is one of a dozen possible things, and which one depends on who is asking, in what setting, and on how he is dressed. When a national survey once asked Brazilians to describe their own colour in their own words, respondents produced well over a hundred distinct terms. Siblings with the same two parents are routinely classified differently. And the classification shifts with social position — the phenomenon summarised in the phrase that money whitens, which research finds to be real if modest in size.

Same ancestry. Same body. Three incompatible answers.

And the categories move within countries too. The racial categories on the United States census changed in almost every decennial count: "mulatto" appeared and disappeared; "Mexican" was a race for exactly one census; the option to select more than one category did not exist until 2000. A person could be one race in 1920, another in 1930, and another in 2000, without moving or ageing differently from anyone else.

If race were a property of bodies, this could not happen. Bodies do not change when a legislature votes.

If there is no natural basis, why does the category predict so much?

This is the question that makes people uncomfortable in both directions, and both discomforts are avoidable.

One temptation is to conclude that because racial classification is arbitrary, its effects must be exaggerated or the differences must be explained some other way.

The other temptation is to conclude that because the differences are large and stubborn, there must be something natural underneath after all.

Both fail on the same point. A category that is used to allocate property, credit, housing, policing, schooling and medical treatment will produce differences in property, credit, housing, safety, education and health — whether or not it corresponds to anything in nature. It does not need a natural basis to have consequences. It needs an institution.

This is Hacking's interactive kind (see 7.1.3) at maximum strength: a classification applied to people, with consequences attached, which then changes the people it classifies and thereby appears to confirm itself.

What the biology actually says

Three findings, stated carefully, because this is where imprecision does damage.

One — most human genetic variation is within, not between, conventionally defined racial groups.

The classic analysis found that the great majority of measured genetic variation — of the order of 85 per cent — occurs between individuals within populations, with only a small fraction distinguishing what were then called the major races. This finding has been replicated repeatedly and is not disputed.

Two — and there is a real technical criticism of the inference drawn from it, which should be stated.

A prominent statistical geneticist argued that the apportionment of variance at single locations in the genome does not establish that classification is impossible: when many locations are considered together, individuals can in fact be assigned to broad geographic ancestry groups with high accuracy, because the small differences are correlated across sites.

That criticism is technically correct — and it does not restore folk racial categories, for three reasons that are equally well established.

The clusters that emerge follow geography and migration history, and when populations are sampled evenly across a continent rather than at widely separated points, the variation appears largely continuous rather than clustered. The apparent clusters are substantially a product of where the samples were taken.

The boundaries are set by the analyst. Cluster analysis requires choosing how many groups to extract (see 8.2.2), and different choices produce different numbers of "populations".

And the resulting groupings do not correspond to the racial categories any society actually uses. African genetic diversity exceeds that of the rest of the world combined, so a single "Black" category groups together populations more different from each other than either is from Europeans.

Three — the disciplinary position. The major professional associations of anthropologists and biological anthropologists have issued formal statements over the past decades holding that human populations are not unambiguous, clearly demarcated biologically distinct groups, and that race as used socially is not a valid biological description.

The precise formulation to carry: human genetic variation is real, largely continuous, and structured by geography and migration. Racial categories are social classifications that sort people using visible traits and known ancestry, and they do not correspond to that structure.

Race — a social classification, usually based on selected visible traits and known ancestry, that allocates people to groups treated as fundamentally different.

Ethnicity — a classification based on claimed shared descent, culture, language, religion or origin. Usually more self-ascribed and more permeable than race, though the boundary between the two concepts is contested and differs by country.

Nationality — legal membership of a state.

Genetic ancestry — the actual population history of a person's genome, continuous and multi-dimensional. This is a biological quantity and it is not race. Two people assigned to the same racial category can have very different ancestry, and two people with similar ancestry can be assigned to different categories in different countries — which the story demonstrates.

Why the distinction matters practically: research and policy that use "race" as a proxy for ancestry, or ancestry as a proxy for race, will misfire in both directions — and the medical example below is what that costs.

How the categories were made

Racial classification is a historical product with datable moments and identifiable purposes.

Slavery and inheritance law. A colonial Virginia statute of 1662 established that a child's status followed the mother's — reversing English common law, under which status followed the father. The effect was to make the children of enslaved women the property of the enslaver, and to remove any inheritance claim arising from paternity. This is a property rule that produced a racial rule, and it illustrates the general pattern: the categories were built where they were needed to settle questions of ownership, inheritance and citizenship.

Colonial administration. The census, the classification schedule, the identity document (see 6.1.1). Categories devised for governing were adopted by the governed, became the basis of political organisation, and outlived the arrangements that created them — the six-point mechanism of 6.1.1, and Rwanda's 1933 identity cards are where that stops being academic.

Immigration and citizenship law, which required courts to decide who was white for the purposes of naturalisation — producing a body of case law in which the criterion shifted between "scientific" classification and "common understanding" depending on which produced the desired result.

And apartheid's Population Registration Act of 1950, which required every person to be classified, established boards to adjudicate disputed cases, and reclassified hundreds of people each year — the clearest demonstration available that racial categories are administrative decisions, because in that system there was literally an office where they were made.

The rules of assignment differ, and each does specific work.

Hypodescent — assignment to the subordinate category — maximises the size of the subordinate group and keeps the dominant category closed. Where property and inheritance run through the dominant group, this is exactly what a boundary must do to protect them.

Continua — as in much of Latin America — allow movement and intermediate positions, which softens the boundary and, critics argue, has historically made collective mobilisation harder while leaving the gradient of advantage intact.

And a legally administered tripartite or quadripartite system allocates rights by category directly.

Three different rules, three different societies, one common feature: the rule was made by the people it advantaged.

Why it predicts so much anyway

Because the classification is attached to consequences, and the consequences accumulate.

A category with no natural basis becomes consequential the moment institutions act on it — which is the Thomas theorem (see 2.4) operating at the scale of a state.

Housing and credit. Where mortgage lending, insurance and public housing allocation were organised by racial classification — as they were, explicitly, in living memory in several countries — the resulting distribution of property persists for generations, because housing wealth compounds and is inherited (see 8.3.3). A discriminatory rule abolished fifty years ago is still in the wealth distribution today.

Policing, prosecution and sentencing, which determine records, which determine employment (see 7.4.2).

Schooling, through catchment and residential segregation.

Medicine, through access, treatment and — as below — through classification embedded in clinical tools.

Labour markets, through the mechanisms 8.4.2 measures directly.

And here is the finding that closes the loop most sharply. Longitudinal American research has found that individuals' racial classification is not stable, and that it changes with social position. Following the same people over years, researchers found that being unemployed, imprisoned or in poverty raised the probability that a person was subsequently classified — by interviewers, and to some extent by themselves — as Black rather than white, controlling for prior classification.

Read what that means. Not only does race shape status. Status shapes race. The category and the outcome are mutually constituting, which is the looping effect quantified, and it means that any analysis treating racial classification as a fixed prior characteristic is missing part of the process.

What goes wrong when the category is treated as biology: the clinical case.

For decades, several routine medical tools incorporated a racial adjustment.

The most widely used estimate of kidney function applied a race coefficient that raised the estimated function for patients identified as Black, based on old and small studies suggesting differences in creatinine production. The practical effect was that Black patients' kidneys were scored as healthier than the same measurements would indicate in others — with consequences for referral, medication dosing and, critically, eligibility for transplant waiting lists.

Lung function testing has applied similar corrections, with a history that runs back to nineteenth-century measurements taken on enslaved people and interpreted as evidence of innate difference.

And a widely used obstetric calculator included race as a factor lowering the predicted success of vaginal birth after caesarean for Black and Hispanic women, which fed into recommendations for repeat surgery.

In each case the same error was made: a social classification was entered into a clinical model as though it were a biological variable, on the basis of correlations that reflected the accumulated effects of the classification itself — differences in exposure, nutrition, stress, access and treatment.

All three have since been revised or abandoned, following sustained argument within the professions concerned, and the kidney equation was formally replaced with a race-free version in 2021.

The general lesson is 7.2.2's: a variable's presence in a model is a claim about what it is. Race entered as biology encoded the consequences of racism as though they were properties of patients — and the resulting decisions then produced more of the differences that had justified including it.

And the opposite error: refusing to measure it.

The colour-blind argument says that if race has no natural basis, recording it reifies a fiction and the right response is to stop.

The objection is decisive and empirical: you cannot detect discrimination in a category you do not record. Every audit study, every gap analysis, every enforcement action against discriminatory lending or hiring requires the classification. A society that stops counting does not stop discriminating; it stops being able to know.

Several countries do prohibit or heavily restrict the collection of ethnic and racial statistics, generally on principled grounds rooted in their own histories — and the documented consequence is that discrimination there is much harder to demonstrate, which is a real cost that the principle has to be weighed against.

The resolution, and it follows from everything above: measure race as what it is — a social assignment — and say which operationalisation you used.

And the operationalisations differ, materially.

Self-identification — how a person describes themselves. Standard in censuses and surveys. Observer classification — how others categorise them, which is what determines treatment. For some outcomes this predicts better, because discrimination operates on perception, not on self-concept. Ancestry — a different quantity again. And skin tone, which produces a gradient within racial categories. Colourism is measurable and consequential: research consistently finds differences in earnings, employment, sentencing and marriage outcomes by skin tone among people classified in the same racial category, which no binary classification captures.

Which one to use depends on the question — and using the wrong one is 7.2.2's operationalisation error with unusually large consequences.

Because both errors are common and they are made by people trying to be careful.

Four things to carry.

"Socially constructed" is a claim about origin and contingency, not about reality or importance (see 7.1.3). Race was made, by identifiable people, at datable moments, for stateable purposes, and could have been made differently. It is also one of the most powerful predictors in social science. Both.

Genetic ancestry is not race, and treating one as a proxy for the other misfires in both directions — with the clinical algorithms as the concrete cost.

Classification is not stable, either across societies or within a person's life, and it responds to social position as well as producing it.

And measurement requires a choice — self-identification, observer classification, ancestry or skin tone — which should be stated, because they answer different questions.

One closing point, which connects this Topic to the rest of Part 8. Everything in 8.3 was about transmission within families. Race operates partly through that channel — because a family's racial position determines the housing, wealth, schooling and treatment it can transmit — and partly through a channel that families cannot mediate at all, because it is applied to the child directly, by strangers, from outside.

That difference is why race is not reducible to class, and it is what 8.4.3 and 8.7 have to account for.

One person's ancestry produces three different racial classifications in three societies — Black under American hypodescent, "Coloured" under an apartheid board that reclassified hundreds of people a year, and one of over a hundred self-described colours in Brazil, where siblings are classified differently and classification shifts with social position. And American census categories changed in almost every decade. Bodies do not change when a legislature votes.

The biology, stated precisely. The great majority of human genetic variation — around 85 per cent — is within populations rather than between conventional racial groups. The technical criticism that many loci considered together permit accurate assignment to geographic ancestry groups is correct — and does not restore folk races, because the variation is largely continuous when sampling is even, the number of clusters is an analyst's choice, and the resulting groups do not match any society's categories. African genetic diversity alone exceeds that of the rest of the world. The professional associations' formal position is that race is not a valid biological description of human populations.

Race, ethnicity, nationality and genetic ancestry are four different things, and using one as a proxy for another misfires.

The categories were made: by inheritance law that made children follow the mother's status; by colonial censuses and identity documents; by naturalisation courts; and by an apartheid registration office where the decisions were literally made. Hypodescent maximises the subordinate group and keeps the dominant category closed, which is what protects property boundaries.

It predicts outcomes because institutions act on it — housing and credit, whose effects persist through inherited property long after the rules were abolished; policing; schooling; medicine; labour markets. And the loop closes: longitudinal evidence finds that unemployment, imprisonment and poverty change how a person is subsequently classified. Status shapes race as well as the reverse.

Treating the category as biology has a cost: clinical algorithms for kidney function, lung function and obstetric decisions encoded the consequences of racism as patient properties, and have since been revised.

And refusing to measure has the opposite cost: discrimination cannot be detected in a category that is not recorded. The resolution is to measure race as a social assignment and state the operationalisation — self-identification, observer classification, ancestry or skin tone, the last of which reveals colourism gradients within categories.

Hypodescent — assignment of mixed ancestry to the subordinate category; the "one-drop" rule in its extreme form.

Racialisation — the process by which a group comes to be treated as a race.

Genetic ancestry — the continuous, multi-dimensional population history of a genome; not race.

Clinal variation — variation that changes gradually across geography rather than in discrete clusters.

Interactive kind / looping effect — a classification of people that changes the people classified.

Racial fluidity — the instability of racial classification within a person's life, responsive to social position.

Self-identification / observer classification — how a person describes themselves; how others categorise them, which determines treatment.

Colourism — advantage graded by skin tone within a racial category.

Race correction in clinical algorithms — the embedding of racial classification as a biological variable in medical tools.

Colour-blind measurement — the position that racial data should not be collected; its cost is undetectable discrimination.

One — find your country's categories. Look up the ethnic or racial categories used in your last census, and the ones used fifty years earlier. List what changed, and find out why.

Two — separate the four terms. Take a news story using "race", "ethnicity", "nationality" or "ancestry" and check whether the word used is the one the evidence supports.

Three — check an operationalisation. For any study reporting racial gaps, find out whether race was self-identified or observer-classified. Then ask which one the mechanism being claimed would operate on.

Four — trace a rule forward. Take one historical rule that allocated property or residence by racial classification. Work out how its effects would appear in a wealth distribution three generations later, using 8.3.3's channels.

Five — hold both. Write two sentences: one stating that racial categories have no natural biological basis, one stating that racial classification predicts life outcomes powerfully. Then write the sentence that explains why both are true. If you can do that, this lesson has worked.