In the 1930 United States census, the form offered "Mexican" as a race. It was the only census that ever did. By 1940 Mexican Americans were back to being counted as white, and since 1980 the census has asked a separate question about Hispanic origin. Nobody's ancestry changed between 1930 and 1940. The category did.
That is the central lesson of this subject. Race and ethnicity are among the most consequential facts in many people's lives, and the labels people use for them are, to a large extent, the product of laws, censuses, courts and history, and they differ from country to country. Saying so is not saying that race is "not real" in its effects: a category that governs where you may live, whom you may marry or how you are treated by a police officer has real consequences whatever its origin. It is saying that to study those consequences, you first have to ask who made the categories, and for what.
This piece also covers the other half of the subject, migration: how many people actually live outside the country where they were born, where they go, what economists and sociologists have found about the effects, and how different states have answered the question of how newcomers should fit in. It is a contested field. Each position is given in terms its holders would accept, and the evidence is kept apart from the arguments.
W. E. B. Du Bois opened The Souls of Black Folk (1903) with the line that "the problem of the twentieth century is the problem of the color-line", and spent a lifetime arguing that the line was drawn by societies, not by biology. Michael Omi and Howard Winant's Racial Formation in the United States (1986) turned that into a framework: racial categories are continually made and remade by states, social movements and everyday interaction. The census, the court and the classroom are where the line is written down.
Genetics bears on this in a particular way. In 1972 the geneticist Richard Lewontin found that about 85% of the variation in the human genes he studied lay within populations, and only a small share between continental groups. The geneticist A. W. F. Edwards replied in 2003 that Lewontin's point, while true gene by gene, does not mean that ancestry cannot be inferred: using many genes at once, people can be sorted by geographic ancestry with high accuracy. Both statements are correct. What they do not support is the nineteenth-century idea of a few discrete "races" with essences. Human variation is mostly gradual across geography. In 2023 a US National Academies report recommended that genetic research stop using race as a stand-in for genetic ancestry and describe people with measured ancestry instead. Whether "race" and "ancestry" should be kept apart in medicine and in social science is debated; the sociological claim is narrower: the social categories are not the genetic clusters.
Every classification needs a rule. Here is how a few rules from the history of the United States handle the same family. Choose each grandparent's ancestry, and see what each rule says about the grandchild. (This is historical illustration, not a way to classify anyone today.)
Tap each grandparent to switch between European-descended and African-descended ancestry (a simplification: real ancestry is often richer).
Rule sources: Virginia Racial Integrity Act (1924) defined "white" as having no trace of non-Caucasian blood, with an exception for up to one-sixteenth American Indian ancestry; the 1967 case Loving v. Virginia struck down its ban on marriage across the line. The census has allowed people to choose more than one race since 2000.
Without JavaScript: with one African-descended grandparent of four, the 1924 Virginia rule makes the grandchild "colored", a simple majority rule makes the grandchild "white", and the 2000-onward census lets the person tick both.
Other places use quite different rules. US states set their own rules at different times, which is why the same person could be one race in one state and another across the border. South Africa's Population Registration Act of 1950 sorted every person into White, Coloured, Bantu (later Black African) or Indian/Asian, and decided membership mostly by appearance and general acceptance in the community, with descent added in 1967; it was repealed in 1991. Brazil asks people to name their own colour, and for decades the answers have come from a wide vocabulary: when the 1976 household survey asked an open question, respondents gave more than 130 different colour terms.
Switch between the censuses below. Each shows what the latest national count looked like, which categories were offered, and what is not asked.
US: Census Bureau, 2020 census, race alone; Hispanic origin (18.7%) is a separate question and can be of any race. Brazil: IBGE, Census 2022, self-declared colour or race (rounded). South Africa: Statistics South Africa, Census 2022 (shares computed from the published counts; the census had an estimated undercount, which makes group shares uncertain). France: no ethnic or racial count; foreign-born share from the World Bank/UN DESA 2024 series.
Without JavaScript: US 2020, 61.6% White alone, 12.4% Black alone, 10.2% two or more races. Brazil 2022, 45.3% pardo, 43.5% white, 10.2% preto. South Africa 2022, 81.4% Black African, 8.2% Coloured, 7.3% White.
The United States has counted by race since 1790, when the first census sorted free people by sex and colour and counted enslaved people separately (under the Constitution's three-fifths clause, which counted each enslaved person as three-fifths of a person for representation). Since 1977 federal rules (OMB Directive 15) have set the standards for the whole government: race as a handful of categories, Hispanic origin as an ethnicity. The 2020 census counted 33.8 million people as multiracial, up from about 9 million in 2010, and the Census Bureau attributes most of that jump to improvements in how the questions were designed and coded, not to a sudden change in the population. In March 2024 the standards were revised: one combined question on race and ethnicity, and a new category for Middle Eastern or North African origin.
Brazil has asked about colour since the first census in 1872 (not in every census since). The answer is self-declared and the categories are colour words: branca, preta, parda, amarela and indígena. In 2022 the pardo category (roughly "brown", covering many mixed backgrounds) became the largest, at 45.3%, ahead of white at 43.5%, for the first time since 1991. Gilberto Freyre's idea of a Brazilian "racial democracy" (1930s) held that mixing had avoided American-style racism; research since the 1950s, including the UNESCO studies, documented large racial gaps in income and education. Since a 2012 law, federal universities reserve half their places for public-school graduates, with sub-quotas for Black, mixed and Indigenous students in proportion to each state's population; a 2014 law reserves 20% of federal civil-service posts for Black and mixed people. Because quota places are valuable, several universities added committees that check applicants' appearance, which is controversial in its own right. Defenders say a self-declared category invites misuse; critics say the state judging a person's looks reproduces the old classifying.
South Africa used the apartheid categories to assign where people could live, work and marry. After 1994 the same four labels were kept for a different purpose, to measure and redress disadvantage under laws such as the Employment Equity Act (1998) and broad-based black economic empowerment. The 2022 census counted about 62.0 million people. Defenders of keeping the categories say race is still the structure of disadvantage and you cannot repair what you do not measure. Critics, including some from within the redress tradition, say racial categories keep the apartheid classification alive, favour the better-off members of the groups and should give way to measures of actual disadvantage. The government's own policy debate on this is ongoing.
France has no census question on race or ethnicity. Article 1 of the 1958 Constitution commits the Republic to equality "without distinction of origin, race or religion", and the 1978 data-protection law (Informatique et Libertés) forbids, with exceptions, collecting data on racial or ethnic origin. In November 2007 the Constitutional Council struck down a clause that would have allowed studies measuring "diversity of origins" on ethnic and racial grounds. The statistical agencies instead use nationality, country of birth and the birthplace of parents; the INED/INSEE Trajectoires et Origines surveys (2008 to 2009, and 2019 to 2020) are the main source on discrimination. Supporters of the French approach argue that counting people by group creates and hardens the groups, and that the state should treat every citizen as a citizen. Critics, among them the sociologist Patrick Simon, argue that you cannot measure or prove discrimination you refuse to count, and that "colourblind" statistics leave minorities invisible. Both sides say they are defending equality. The disagreement is about whether equality requires counting.
The United Kingdom has asked an ethnic-group question in every census since 1991. Germany, with its history in mind, has been cautious about ethnic categories and uses "migration background". Mexico added a question on Afro-descendant identity in its 2020 census. India does not ask about race, and asks about caste only for Scheduled Castes and Tribes. Every one of these is a decision, and each has defenders and critics.
International migration looks vast in the news and is smaller than most people guess. The United Nations Population Division counts the migrant stock: people living in a country other than where they were born (or, for some countries, a country other than their citizenship). By its 2024 estimate there were about 304 million international migrants, about 3.7% of the world's people. In 1990 it was about 154 million, or 2.9%. In 2020, 281 million, or 3.6%. The stock includes refugees and people who moved decades ago.
UN DESA International Migrant Stock: 1990 (about 154 million, 2.9%), 2020 (281 million, 3.6%) and 2024 (304 million, 3.7%). Shares are of world population. Migrant stock is not the yearly flow: it counts everyone currently living abroad.
Without JavaScript: 2.9% in 1990, 3.6% in 2020, 3.7% in 2024, about 1 person in 27.
The largest destination in 2024 was the United States, with more than 52 million foreign-born residents, followed by Germany (about 16.8 million), Saudi Arabia (about 13.7 million), the United Kingdom (about 11.8 million) and France (about 9.2 million). Europe and North America together host about half of all migrants. The shares of populations vary much more than the totals do, as the next tool shows. In the Gulf, foreign workers outnumber citizens: the World Bank series puts the foreign-born at 74% of the UAE in 2024. In China and India it is a fraction of one per cent.
Surveys find that people in many countries overestimate the immigrant share of their own population by a wide margin. Try it for yourself: pick a country, set your guess with the slider, then reveal.
World Bank (from UN DESA), International migrant stock as % of population, 2024. Definitions differ by country (foreign-born or foreign citizens), and for Gulf states most migrants are temporary workers.
Without JavaScript: foreign-born share in 2024 was about 74% in the UAE, 22% in Canada, 21% in Sweden, 20% in Germany, 15% in the US, 3% in Japan, under 1% in Brazil and India.
The nineteenth-century geographer E. G. Ravenstein described migration as shaped by "push" and "pull" factors; later theory added that wage gaps are not enough. Douglas Massey and colleagues argued that families send members abroad to diversify risk, that networks of earlier migrants cut the cost and risk of coming, and that migration then feeds itself. The rich-poor wage gap across a border is large; most people nonetheless stay home, because moving is costly and people prefer the places and families they know.
What the evidence shows (United States, 2016). A National Academies of Sciences, Engineering, and Medicine panel reviewed the research and concluded that, over the long run, immigration has little effect on the overall wages of native-born workers, with some possible losses for the least-educated natives and for earlier immigrants. On public finances, first-generation immigrants cost more at the state and local level (mostly schooling for their children), while second-generation Americans are among the strongest net contributors.
Specific findings are disputed. When about 125,000 Cubans arrived in Miami in the 1980 Mariel boatlift, David Card (1990) found no visible effect on wages of low-skilled workers there. George Borjas (2017) re-analysed the same events with a narrower comparison and found a large wage drop for high-school dropouts. Michael Clemens and Jennifer Hunt (2019) argued that a change in the survey's sample composition explained Borjas's result. Researchers still disagree on how large any short-term effect is on the least-skilled, though few now find large effects on the average native worker.
Positions are a separate matter. Joseph Carens (1987) argued that borders are like feudal privilege and that liberal principles imply something close to open borders. Michael Walzer (1983) argued that communities have a right to decide who joins, like clubs and families, though they owe duties to refugees. David Miller (2016) defends limits on the grounds that national communities have legitimate shared obligations. Others stress humanitarian duty (the 1951 Refugee Convention) or economic nationalism. Evidence about wages does not settle any of these positions; each rests on a view about what a state owes to citizens and strangers.
The older sociology of immigration, the Chicago school (Robert Park, 1920s) and Milton Gordon (1964), expected newcomers to move steadily toward the mainstream, in language first, then jobs, then marriage. Later work found more varied paths. Alejandro Portes and Min Zhou (1993) described segmented assimilation: some immigrant children rise into the middle class, some are pushed into poor and disadvantaged neighbourhoods, and others do best by holding on to their community's networks and values. Richard Alba and Victor Nee (2003) argued instead that the mainstream itself changes, so assimilation is the blurring of boundaries and not a one-way absorption.
The empirical question under these debates is whether diversity affects trust. Robert Putnam (2007) found that, in the US, people in more diverse neighbourhoods reported lower trust, including of their own group; he argued that this was a short-run effect that societies overcome. A 2014 review by Tom van der Meer and Jochem Tolsma found a negative association in the literature that was typically weak and varied by measure. A 2006 meta-analysis of 515 studies by Thomas Pettigrew and Linda Tropp found that contact between groups under favourable conditions generally lowers prejudice. These are findings, and none of them tells a state whether to pursue assimilation or multiculturalism.
Racial and ethnic categories are made by laws and censuses, and they differ by country, so the first question about any group statistic is how the group was defined. Migration is smaller and more varied than the news suggests, its measured effects are real but modest and disputed at the margins, and how states answer the integration question is a matter of values that evidence can inform but not settle.
Try this: find your own country's census or official form, count the categories, and note one that was added or removed in your lifetime.
Sources. US Census Bureau, 2020 census results on race and Hispanic origin and the 2021 note on the multiracial count; OMB Statistical Policy Directive 15 revision (March 2024). IBGE, Censo 2022 (December 2023 release). Statistics South Africa, Census 2022. Population Registration Act 1950 (repealed 1991). Loi 78-17 of 6 January 1978; Conseil constitutionnel decision 2007-557 DC (15 November 2007); INED/INSEE TeO surveys. UN DESA, International Migrant Stock 2024, via the Migration Policy Institute; World Bank indicator SM.POP.TOTL.ZS (2024). Du Bois 1903; Omi and Winant 1986; Lewontin 1972; Edwards 2003; National Academies 2023; Brubaker 2004; Massey and Denton 1993; Wilson 1978. National Academies 2016; Card 1990; Borjas 2017; Clemens and Hunt 2019. Carens 1987; Walzer 1983; Miller 2016. Portes and Zhou 1993; Alba and Nee 2003; Kymlicka 1995; Barry 2001. Putnam 2007; van der Meer and Tolsma 2014; Pettigrew and Tropp 2006.