Numbers, Facts and Trends Shaping Your World

Who Is Multiracial? Depends on How You Ask

Chapter 1: Estimates of Multiracial Adults and Other Racial and Ethnic Groups Across Various Question Formats

Standard Two-Question Measure

Standard Two-Question Measure

The first method used for measuring racial identity was the two-question race and ethnicity format typically used on all Pew Research Center surveys. It is similar to the two-question format presently used in U.S. Census Bureau surveys and by many other government and research organizations.4

In this format, respondents are first asked, “Are you of Hispanic, Latino or Spanish origin, such as Mexican, Puerto Rican or Cuban?” Then, they are asked to select as many as apply from a list including white, black or African American, and Asian or Asian American, or to specify some other race. Up to four responses are recorded and respondents who volunteer that they are “mixed race” or “biracial” are probed to provide their specific races.

This question was included on a computer-assisted telephone interviewing (CATI) random-digit dial (RDD) telephone survey of more than 10,000 adults conducted in English and Spanish Jan. 23-March 16, 2014. This survey was used to recruit panelists for Pew Research Center’s American Trends Panel (ATP), a probability-based, nationally representative online panel of adults in the United States.

Using the two-question approach, 3.7% of adults selected two or more races, regardless of whether or not they were of Hispanic origin. In addition, two-thirds (65.8%) said they were white only and not Hispanic, 11.8% were non-Hispanic black, and 2.6% non-Hispanic Asian. Some 13.3% said they were Hispanic alone or in combination with another race or races, including 0.4% who selected Hispanic and two or more races.5

Question wording: Standard Two-Question Race/Ethnicity Measure for a Telephone Survey

Are you of Hispanic, Latino, or Spanish origin, such as Mexican, Puerto Rican or Cuban?

  1. Yes
  2. No


  1. White (e.g., Caucasian, European, Irish, Italian, Arab, Middle Eastern)
  2. Black or African-American (e.g., Negro, Kenyan, Nigerian, Haitian)
  3. Asian or Asian-American (e.g., Asian Indian, Chinese, Filipino, Vietnamese or other Asian origin groups)
  4. Some other race (SPECIFY____ IF NEEDED: What race or races is that?)
  5. Native American/American Indian/Alaska Native (VOL.)
  6. Pacific Islander/Native Hawaiian (VOL.)
  7. Hispanic/Latino (VOL.) (e.g., Mexican, Puerto Rican, Cuban)
  8. Don’t know (VOL.)
  9. Refused (e.g., non-race answers like American, Human, purple) (VOL.)

This type of format is commonly used by survey researchers because it follows current Office of Management and Budget definitions of race and ethnicity used by the Census Bureau, in which Hispanic is considered an ethnicity and not a race. One of the major challenges with this format is that many Americans do not separate their race from their ethnic origin in the same way that some researchers often do, and this is particularly true among Hispanics.

In fact, a 2015 survey found that 67% of Hispanic adults consider being Hispanic part of their racial background or both their racial and ethnic backgrounds. This leaves many Hispanics perplexed as to how to answer this two-question format where Hispanic is not included among the racial response categories.

This is a challenge that the Census Bureau has also acknowledged. Their researchers found that some 37% of Latinos in the 2010 census reported their race as “some other race” instead of providing one of the given racial categories, and many wrote in responses such as “Mexican,” “Hispanic” or “Latin American.”6 This is also the case on Pew Research Center surveys of Latinos. In the 2014 National Survey of Latinos, 25% of Latinos volunteered their race as “Hispanic” or “Latino” instead of one of the standard racial classification groups, and 12% said they didn’t know how to answer or refused to answer.7

In preparation for the 2020 census, the Census Bureau has been testing alternative versions of their race and ethnicity measures, in large part to decrease the incidence of “some other race” reporting and low response rates for the race question among Hispanics. Instead of asking separate questions to measure Hispanic origin and race, the proposed single question allows people to select more than one category—and includes “Hispanic” as an option among the listed races and origins.8 A report released by the Census Bureau in early 2013 indicated that this new format did not reduce the estimates of Hispanics or minority race groups in the population even as it lowered item nonresponse. Furthermore, they concluded that the resulting data were more accurate and reliable, and better reflected the self-identity of respondents.

Based on this research, the next method we tested to measure the race and ethnicity of respondents was a version of the proposed new census item, combining race and Hispanic origin into one question.

Though our research has shown that many consider their Hispanic origin to be a race, for the purposes of this report we will continue to use the current Census Bureau standards which consider race and ethnicity separately, and therefore Hispanics who also select one race would not be considered to be multiracial in our analysis.

Census Alternative Questionnaire Experiment Measure

Census Alternative Questionnaire Experiment Measure

The question derived from the Census Bureau’s Alternative Questionnaire Experiment (AQE) measure is similar to the standard two-question measure, but instead of asking a separate question to capture Hispanic origin, it asks a single question that allows people to select more than one race or origin and includes “Hispanic” as one of the options.

This question was included on Wave 5 of the American Trends Panel (conducted in English and Spanish July 7-Aug. 4, 2014), which included a mode experiment that randomly assigned the panel’s Web respondents to either their usual Web mode or a computer-assisted telephone interviewing (CATI) mode. The panel’s mail respondents (those who do not use the internet or who did not provide an email address) were assigned to CATI as well.

For Web respondents, the question read, “What is your race or origin? Mark one or more boxes,” and response categories were listed along with examples for each, such as “German, Irish, English, Italian, Lebanese, Egyptian and so on” for the “white” response option.

The phone respondents were read: “I’m going to read you a list of categories. Which of the following describes your race or origin? You can select as many as apply.” A random half sample of the phone respondents were read the entire list of response options in the question wording box below (without the examples) and the other half were read an abridged version that is more similar to the list used in the standard two-question measure—white, Hispanic, Latino or Spanish origin, Black or African American, Asian or Asian American, or some other race or origin.9

Altogether, 4.8% of adults selected two or more races using the AQE measure, regardless of whether or not they are Hispanic—slightly higher than the 3.7% share that did so using the standard two-question format. While this measure increases the reporting of two or more races, it does not significantly alter reporting of the other major races and ethnicities compared with the standard format. Two-thirds of adults were non-Hispanic white, 11.8% non-Hispanic black, and 2.8% non-Hispanic Asian. Some 12.9% said they were Hispanic alone or in combination with another race or races, including 0.9% who said they were Hispanic and two or more races.

Question wording: Census Alternative Question Experiment



Web: What is your race or origin? Mark one or more boxes.

  1. White Examples: German, Irish, English, Italian, Lebanese, Egyptian, and so on
  2. Hispanic, Latino, or Spanish origin Examples: Mexican or Mexican American, Puerto Rican, Cuban, Dominican, Salvadoran, Colombian, and so on
  3. Black or African American Examples: African American, Jamaican, Haitian, Nigerian, Ethiopian, Somalian, and so on
  4. Asian or Asian American Examples: Chinese, Filipino, Asian Indian, Vietnamese, Korean, Japanese, and so on
  5. American Indian or Alaska Native Examples: Navajo Nation, Blackfeet Tribe, Muscogee (Creek) Nation, Mayan, Doyon, Native Village of Barrow Inupiat Traditional Government, and so on
  6. Native Hawaiian or Other Pacific Islander Examples: Native Hawaiian, Samoan, Guamanian or Chamorro, Tongan, Fijian, Marshallese, and so on
  7. Some other race or origin List race(s) and/or origin(s)


  1. Yes
  2. No


  1. Yes
  2. No

* Respondents who selected only Hispanic for the AQE received the follow-up questions about their parents and grandparents but are not included in the analysis for this report because we are not counting someone who is Hispanic and a single race as multiracial.

AQE Results by Mode and Response Categories

Some of the difference between the multiracial estimates for the AQE and the standard measure may have been due to differences in mode (AQE measure in a mixed-mode survey vs. standard measure in a phone survey) and response options (long list vs. short list). Some 5.2% of respondents who answered the AQE via the Web mode reported two or more races, significantly higher than the 3.7% who gave two or more races using the standard measure. By contrast, 4.5% of respondents who answered the AQE via phone reported two or more races, not significantly higher than the share for the standard two-question phone measure. However, this finding hides variation among the phone respondents based on the list of response options read to them. Among phone respondents who were read the short list of response categories, the share reporting two or more races—3.6%—is nearly identical to the share that did so for the standard race question, which included a similar list of response options. Among phone respondents who were read the full list of response options, the share selecting two or more races—5.4%—looks more similar to the share for Web respondents, who also saw the full list of races.

This suggests that the difference in the share reporting two or more races between the standard and AQE measures we tested isn’t due to including “Hispanic” in the race question, but rather due to providing most of the respondents with a more extensive list of race categories. If this is the case, it is possible that a version of the standard two-question measure with more extensive response categories would also register higher levels of adults reporting two or more races.

An increase in the share of people saying that they are American Indian in addition to one other race accounted for 79% of the difference between the long- and short-response option versions of the phone questions.10 “American Indian or Alaska Native” was one of the response options read for the longer list of responses, but not for the abridged version.

All in all, though the share of adults indicating a multiple-race background was about 30% higher in the AQE than in the standard race question, this method is still a very similar “mark one or more” type of race question. It differs only in the placement of Hispanic origin and, in this case, response options. However, some researchers have argued that the share of people who have a multiracial background is likely higher than the share that reports two or more races when asked to identify their own race by selecting one or more categories from a list. Therefore, we tested several other measures to see if they would capture a broader population of mixed-race adults.

What Are the Races of Your Parents and Grandparents?

Census Alternative Questionnaire Experiment Measure, With Parents’, Grandparents’ Race/Origin

The first approach we took to explore the possibility of capturing a broader multiracial population was asking about the racial backgrounds of respondents’ parents and grandparents.

Along with the AQE measure, we tested a follow-up question that asked single-race respondents if either of their parents was a different race or origin than the one they reported for themselves. Those who said “no” were then asked if any of their grandparents were some other race or origin than their own.

Altogether, we found that 6% of adults chose one race for themselves but said they had a parent who was a different race or origin, which roughly doubled the multiracial estimate, from 4.8% based only on the respondents’ own races to 10.8% when the respondents’ parents were taken into account. And when extended to include single-race adults with grandparents of a different race or origin, the share of the population that could be considered multiracial rose by another 6 percentage points, to 16.6%.

[insert respondent’s AQE race selection]

Since these questions asked about a different race or origin, some respondents may have been thinking about their ethnic or even national origin in their responses. This may have been the case particularly among respondents with a Hispanic parent or grandparent, since “Hispanic” was explicitly listed when they were asked about their own race or origin in the AQE. But it could also have affected those who were thinking of a relative of Irish or Jewish or Middle Eastern origin, for example. Or a Chinese respondent who self-identified as “Asian” may have been thinking of a Japanese relative, who, while of a different national origin, would also be considered “Asian.”

It is also important to note that respondents who selected only one race for the AQE were not asked the follow-up questions about their parents or grandparents if they also selected “Hispanic” or “some other race.” Altogether, 89 respondents, or about 3% of all single-race respondents, fell in this category.

With this in mind, in order to more accurately find those who identify themselves as only one race but who may have had a parent, grandparent or earlier ancestors of a different race, the 2015 Pew Research Center survey of multiracial Americans asked all respondents to describe their parents’, grandparents’ and earlier ancestors’ races using the same Census AQE question they were given when asked to classify their own race.11 Using this refined method, we found that 2.9% of adults who chose only one race or no races12 for themselves could be considered to have two or more races in their background because of their parents’ races, and another 2.6% could be considered to have two or more races in their background because of their grandparents’ races. Altogether, using this method, the multiracial share of the adult population based on races reported for themselves, their parents and their grandparents was estimated to be 6.9% of the U.S. adult population.

Point Allocation Measure

Point Allocation Measure

Another method we tested as a potential way to capture a broader concept of race and ethnicity is called the point allocation measure. This is an experimental measure developed by Berkeley political scientist Taeku Lee in which respondents are given 10 “identity points” and asked to allocate them across different racial and ethnic categories they consider to be part of their background. This allows for a more nuanced reporting of secondary races or ethnicities, wherein a respondent can assign different weight to different categories by allocating a different number of points to each.

In his work, Lee argues that the share of respondents reporting more than one race will be greater using this measure than with a typical “mark one or more” measure. In the latter, each category is given equal importance, so someone who feels mostly of one race but has a less salient secondary race might not mark their secondary race. However, Lee hypothesizes that by allowing the allocation of points to different races, this same respondent may then acknowledge this secondary racial identity by, for example, allocating nine points to his or her primary race and a single point to a secondary race.13

The point allocation exercise was included in wave 7 of the American Trends Panel (conducted in English and Spanish Sept. 9-Oct. 3, 2014), which was conducted almost entirely on the Web, with the remainder completing mail-in questionnaires.

Using this measure, 12.7% of respondents reported a multiracial background, defined as allocating at least one point to two or more racial groups. An additional 61.2% of respondents reported that they were non-Hispanic white only, 7.0% reported that they were non-Hispanic black only and 3.4% said they were non-Hispanic Asian only. Some 13.3% said they were Hispanic, alone or in combination with another race or races, including 2.3% who selected Hispanic and two or more races.14

The point-allocation approach increases the share of adults reporting two or more races, but decreases the share of adults reporting that their racial and ethnic background is non-Hispanic whites or non-Hispanic black compared with the standard two-question format (see Appendix A, Table A1).

Question wording: Point Allocation Measure

In thinking about their background, people often will describe which racial or ethnic groups best describe them. Imagine if we used a 10 point system where points are allocated to whichever racial or ethnic groups we think accurately describe a person.

For example, if you think of someone as half-white and half-Asian, you might allocate 5 points to each. Or if you think of someone as mostly black but with some Hispanic heritage, you might allocate 8 or 9 points for African American and 1 or 2 points for Latino.

Now think of your own background in racial and ethnic terms. How would you describe your race and ethnicity using this 10-point system?


__ White __ Black or African American __ Asian or Asian American __ Hispanic, Latino, or Spanish origin __ Native American/American Indian/Alaska Native __ Native Hawaiian or other Pacific Islander __ Other 10 TOTAL POINTS


Incorrect Response and Nonresponse in Point Allocation Method

But the point allocation measure faced some major challenges. In particular, this question was difficult to implement and seems to have caused confusion among some respondents. Altogether, about one-in-ten respondents either did not give a valid response to the item or skipped it altogether. Among the 343 panelists who received mail-in surveys, half did not give a valid response, including 138 respondents who checked boxes instead of inserting numbers as instructed. The share giving non-valid responses or refusals on the Web was much lower (4%), likely because the Web version included a running total showing the points used and a prompt that informed respondents if their responses did not add to 10 before they moved onto the next question. Despite these aids, some 2% of Web respondents still entered numbers that did not add up to 10. In the analysis for this report, the 204 Web and mail respondents who gave non-valid responses are excluded.15

Attitudinal Measure: Are You Multiracial?

Attitudinal Measure

Finally, we asked an attitudinal question to directly capture the share of adults who say they consider themselves to be mixed race—that is, belonging to more than one racial group.

This question was asked on wave 7 of the American Trends Panel, along with the point allocation measure. A random half of respondents were asked if they considered themselves to be multiracial before they were asked to allocate points to different races, and the other half were asked this question after the point allocation measure.

Overall, 12.0% of adults said they are mixed race, including very similar shares of those who were asked the attitudinal question first (11.7%) and of those who were asked this question second (12.3%). But our analysis shows that to consider oneself “mixed race” is not necessarily equivalent to selecting more than one race when given a list of possible races. And the extent to which responses to the attitudinal question were consistent with the selection of two or more races varied depending on the question used to identify a respondent’s race.

Many Who Allocate Points to More Than One Race Don’t Consider Themselves “Mixed Race”

A majority of adults who gave two or more races for themselves using the standard two-question format or the Census AQE format said they consider themselves mixed race—71% of each group said this. Still, about a quarter of these multiracial adults did not consider themselves mixed race.

Those who chose only one race for themselves in the AQE measure but indicated that they had a parent or a grandparent of a different race or origin were among the least likely to say they consider themselves mixed race. Just 27% of single-race adults who said at least one parent was a different race or origin and 15% with a grandparent with a different background than their own said they considered themselves multiracial.

Among those who gave at least one point to two or more races using the point allocation approach, some 46% said they consider themselves mixed race, while 54% said they do not think of themselves this way.

On the other side, some adults who selected only one race or origin say they consider themselves mixed race when asked the attitudinal question. Looking at the Census AQE measure for example, adults who say they are only Hispanic (22%) or only black (15%) are more likely to say they consider themselves mixed race than adults who are only white (4%). This finding for Hispanics is consistent with findings in the 2014 Pew Research Survey of Latinos, which shows that many Hispanics consider themselves mixed race, indigenous or Afro-Latino, even if this isn’t reflected in the races they select for themselves in a standard race question.

Ultimately, the inconsistency between selecting more than one race on a race question and identifying as mixed race informed our decision to use the attitudinal question as a dependent variable to measure the salience of a respondent’s multiracial identity in our survey of multiracial adults, rather than as a way to define our multiracial population.

  1. The Census Bureau introduced the option to choose more than one race in 2000. Prior to that, only one race selection was permitted.
  2. Analysis of the different race questions will report figures for Hispanics alone or in combination with another race or races; while this group is not mutually exclusive from the two or more races category, it is useful because when most survey organizations (including the Census Bureau) report numbers for Hispanics, they include all Hispanics regardless of their racial background.
  3. See Humes, Jones and Ramirez (2011) and Tafoya (2004).
  4. For more on the survey’s methodology, see
  5. The bureau is also looking into several other modifications to the race question that are not discussed at length here and were not included in our tests, including whether to include a new “Middle East and North Africa” race category and whether to collect specific origins for all races (currently, this information is only collected for Hispanics, American Indian or Alaska Natives, Asians and Pacific Islanders).
  6. To keep question wording as close as possible to the standard two-question measure, the question with the short list of response options asked respondents about their race (not “race or origin”), but the Hispanic option was still included as “Hispanic, Latino or Spanish origin.”
  7. Throughout this report, references to “American Indians” include Alaska Natives.
  8. For full question wording, see
  9. Because of the way this question was revised, respondents who gave no races or a Hispanic origin only but selected multiple races for their parents or grandparents could also be counted as having two or more races. In the original, because Hispanic respondents were asked if they had relatives who were “some other race or origin than Hispanic” it was not possible to determine whether these respondents had multiple races in their backgrounds. Furthermore, this revision allowed us to analyze the parents and grandparents of respondents who gave a single race in addition to Hispanic or “some other race.”
  10. Lee tested this hypothesis in a 2003 survey of Californian adults and found that 26% identified as multiracial. By comparison, just 5% of Californian adults identified as two or more races on the 2000 census.
  11. “Other” responses were not counted toward one’s racial composition, so a respondent who splits points between a single response category such as white and “other” would be counted as just white. However, “other” responses were backcoded into the standard racial categories when possible (for example, “Mexican” into “Hispanic”).
  12. We did not test this format on the phone. However, one can imagine that the added challenge of requiring mental math could make this task difficult for many phone respondents. Taeku Lee conducted this experiment over the phone in California and found that one-in-six respondents did not provide a valid response.
Icon for promotion number 1

Sign up for our weekly newsletter

Fresh data delivery Saturday mornings

Icon for promotion number 1

Sign up for The Briefing

Weekly updates on the world of news & information