How the Census Measures Race and Ethnicity

Race and ethnicity data trips up more readers than any other part of a demographic profile, and almost always for the same reason: people expect the categories to work like slices of a pie that sum to a neat 100%, and they don’t. The Census Bureau measures race and ethnicity in a specific — and frequently misunderstood — way. Once you see how the questions are built, the numbers stop looking contradictory.

Two questions, not one

The single most important fact about American race and ethnicity data is this: race and Hispanic origin are two separate questions.

The Census Bureau, following federal standards set by the Office of Management and Budget, treats “Hispanic, Latino, or Spanish origin” as an ethnicity, not a race. Every person is asked both whether they are of Hispanic origin and what their race is. The Bureau’s own guidance is explicit: Hispanic people may be of any race.

This is why a place can be, say, 60% White and 40% Hispanic at the same time without anything being wrong. A large share of Hispanic residents also select White (or, increasingly, “some other race”) on the race question, so the two figures overlap by design. They are answers to different questions layered on top of each other, not competing slices of one total. Add the race percentages and the Hispanic percentage together and you’ll sail past 100% — as intended.

Nationally the picture looks like this:

Group	Share
White alone	58.2%
Hispanic or Latino (any race)	19.0%
Black alone	12.0%
Asian alone	5.8%
Two or more races	3.9%
Native American / Alaska Native alone	0.5%
Native Hawaiian / Pacific Islander alone	0.2%

The race rows plus “two or more” describe the race question; the Hispanic row sits alongside them, cutting across all of them.

What “alone” means

You’ll notice the word alone attached to most race categories — “White alone,” “Black alone.” It’s doing real work.

“Black alone” counts people who selected only one race. People who selected more than one — say, Black and White — are pulled out into the separate two or more races category rather than being counted in each. That keeps the race categories mutually exclusive so they can be added without double-counting. The trade-off is that the single-race numbers slightly understate how many people have any connection to a given group, because multiracial respondents are counted apart.

The multiracial category is also the fastest-changing part of the data. Revised question wording in recent years made it easier for people to record more than one race, and the share reporting two or more races rose sharply as a result. Some of that is real demographic change; some is simply that the form finally let people describe themselves accurately. When you see a jump in the multiracial share, it’s worth remembering that measurement changed, not just the population.

A category we corrected

Categories like these are easy to get subtly wrong, and getting them right takes care. On this site, the “some other race” figure once behaved as a residual — whatever was left after the named groups — and it quietly double-counted people already tallied as multiracial or Native. The symptom was unmistakable: a place’s shares would sum to about 104%, which is impossible for the race question on its own.

We rebuilt that figure to draw the “other” category directly from the Census table designed for it, rather than backing into it by subtraction. Now the race shares sum cleanly to 100%, and Hispanic origin sits beside them as the separate dimension it’s meant to be. It’s a small fix with a general lesson: whenever demographic shares don’t add up the way the definitions say they should, the problem is usually in how the categories were combined, not in the underlying count. Our full approach is written up on the methodology page.

Reading a place’s diversity honestly

Because the categories overlap and vary so much in size, comparing two places by eye is unreliable. A place that’s 70% one group and 30% another is less diverse than one split evenly four ways, even if both have “two major groups.” To compare fairly you need a single measure that accounts for the whole distribution.

The standard tool is a diversity index — the probability that two residents chosen at random belong to different groups. A perfectly uniform mix scores near the top; a place where nearly everyone shares one background scores low. It collapses the whole racial and ethnic makeup into one comparable number, which is why we compute it for every place and rank states and cities by it. You can see how places stack up in our most diverse states and most diverse cities rankings.

A last word of caution on precision. These figures come from a survey sample, and for smaller places the estimate for any single group can carry a wide margin of error — a town listed as 3% one group might really be anywhere in a band around that. Read the broad composition, not the second decimal place, and you’ll rarely go wrong. The categories are complex because the country is; the trick is remembering that race and ethnicity are two questions, that “alone” means single-race, and that the numbers overlap on purpose. Definitions for every term are in our glossary.