What the Census Can and Cannot Say

A Count That Keeps Redefining Itself

The decennial census has tracked Korean Americans continuously since 1970, and the headline trajectory is striking: roughly 70,000 people in that first post-migration count, rising past 350,000 by 1980, close to 800,000 by 1990, and past 1.7 million by 2010. Those numbers appear in civic histories, on municipal planning documents, and in the opening paragraphs of newspaper features about Koreatown. They are real numbers. They are also not quite the same kind of number from one decade to the next, and treating them as a clean series understates what the Census Bureau itself has changed along the way.

The largest single complication is the shift in how respondents are permitted to identify. Before 1980, "Korean" appeared as a listed Asian category on the census form, which was itself a change from earlier decades when the primary race categories offered no separate Asian listing at all. The 1980 and 1990 censuses listed Korean explicitly but required respondents to choose one race. The 2000 census introduced the option to select more than one race simultaneously, a methodological change the Census Bureau has documented as affecting population counts for all groups with significant multiracial populations. For Korean Americans — a population with rising rates of intermarriage by the late 1990s — the change was not trivial. A person who marked both Korean and white in 2000 would have had to choose a single race in 1990, and so would have appeared in only one group's count. Comparing the 1990 and 2000 figures as if they were measured the same way is therefore comparing unlike things.

The 2010 census compounded this by altering the detailed Asian-category write-in instructions, and the 2020 census went further still, adding a write-in field for all racial categories and running a separate question on Hispanic origin that interacts with the Asian categories in tabulation. Each reform reflected genuine attempts to capture a more accurate and more nuanced picture of the population. But each reform also introduced a discontinuity into the historical series. Demographers studying immigration and the Asian American population have noted that the multiracial option's introduction makes pre- and post-2000 totals structurally incomparable without adjustment.

What the Tract Can and Cannot Hold

Below the headline count, the census's spatial unit — the census tract — shapes what the data can say about residential concentration. Tracts are drawn to contain roughly 4,000 people and are redrawn every decade based on population change. A Korean American neighborhood that gains residents rapidly, as Koreatown in Los Angeles did through the 1970s and 1980s, will be subdivided at the next decennial boundary redraw. After subdivision, the group's concentration appears to drop, not because anyone moved, but because the denominator of the tract changed. The urban form is stable; the measuring unit is not.

After subdivision, the group's concentration appears to drop, not because anyone moved, but because the denominator of the tract changed.

Koreatown in Los Angeles occupies an area of roughly three square miles that has been sliced into differing numbers of tracts across successive censuses. Research using the census to trace the neighborhood's ethnic composition over time has had to deal with the fact that tracts whose borders shifted between 1980 and 1990 cannot be directly compared to those that existed in 1970. This is not a problem unique to Koreatown — it affects every urban district with a mobile or growing ethnic population — but it is a problem that sits invisibly behind confident statements about neighborhood change.

The series, plotted

Annandale, Virginia, and Flushing, Queens, illustrate a different version of the same problem. Both developed as commercial Korean American districts substantially through the 1980s and 1990s, but neither was ever a residential concentration in the sense the census easily captures. The businesses were there; the proprietors often lived in surrounding suburbs that the tract-level data assigned to general suburban categories. Counting the Korean American presence in those districts from residential census data alone would have understated it considerably. The American Community Survey, which the Census Bureau introduced as a rolling annual sample replacing the long-form decennial questionnaire after 2000, offers more frequent data but smaller sample sizes at the tract level, which makes small-group counts for a district like Albany Park in Chicago — dense enough to support an honorary street name but not a massive residential cluster — statistically unreliable in any single year.

The Self-Reporting Gap and the Country-of-Birth Proxy

The census has, at various points, offered a country-of-birth question alongside the race question, and researchers have used the two in combination to distinguish the immigrant generation from the American-born. Country-of-birth data for Korea appears in the census going back to 1910, when it captured a tiny number of early labor migrants and students. By 1970 it was rising; by 1990 it was essential for understanding the post-1965 cohort. But country-of-birth counts the foreign-born only, and the second generation — born in the United States to Korean immigrant parents — falls out of that count entirely. What the Immigration and Nationality Act of 1965 set in motion was a two-generation migration; the census can identify the first generation by birthplace and the whole group by race, but it cannot, in a single variable, capture the Korean-immigrant household that contains a Busan-born parent and a Los Angeles-born child and count them as a unit.

Language use offers a partial workaround. The American Community Survey asks about language spoken at home and English proficiency, and these variables have been used to approximate the concentration of first-generation immigrant households in particular tracts. But language use shifts rapidly across generations — the second generation skews heavily toward English dominance — and the variable captures linguistic behavior, not ancestry or identity. A third-generation Korean American who speaks only English at home is effectively invisible to any language-based proxy.

What Survives the Methodological Noise

None of this makes the census useless — it makes it a source that requires handling. The broad shape of the Korean American migration is legible in the data: a negligible pre-1965 count, a sharp acceleration through the 1970s and 1980s, a plateau in growth rates by the 2000s as the immigration wave from Korea itself slowed. The 2020 decennial census returned nearly 2 million people identifying as Korean alone or in combination, a figure that reflects both real population and the expanded identification options. The Korean-alone count, which most directly parallels earlier single-race tabulations, came in somewhat lower.

The spatial record — which tracts had significant Korean American populations, when, and how those populations moved toward suburbs like Duluth, Georgia, over the 1990s and 2000s — is recoverable with effort, provided a researcher corrects for boundary changes and acknowledges the limits of self-identification at small sample sizes. The rotating credit associations, the kye (a rotating pool of capital contributed by a fixed circle of participants), the press, the storefronts: those left records outside the census — in business licenses, newspaper archives, and community organization records — that the census alone cannot supply and that fill in what the count leaves dark.

The census is a ledger, not a portrait. It records what it was designed, in each decade, to record. Reading it well means reading the instrument alongside the numbers.