Copies available from the authors or from:

Editor, BLS Statistical Notes
Bureau of Labor Statistics
Office of Research and Evaluation, OSMR 2
Massachusetts Avenue, NE
Room 4915 Washington, DC 20212

This paper is circulated for the information of members of the staff of the Bureau of Labor Statistics (BLS) and other professionals interested in the work of the Bureau. It has not been cleared for publication and should not be quoted without permission of the authors. The views expressed are those of the authors and do not reflect the policies of the BLS or the Bureau of the Census or the views of other BLS or Bureau of the Census members.

I. EXECUTIVE SUMMARY

In response to legislative and agency needs, the Office of Management and Budget (OMB) issued in 1977 the "Race and Ethnic Standards for Federal Statistics and Administrative Reporting" contained in Statistical Policy Directive No. 15. The basic racial categories set forth in the Directive are American Indian or Alaskan Native; Asian or Pacific Islander; Black; and White. The ethnic categories specified in the Directive are Hispanic origin, and Not of Hispanic origin. Although the standards in Directive 15 have been used for almost two decades throughout the Federal government for record keeping, collection, and presentation of data on race and Hispanic origin, they have come under increasing criticism from those who believe that the minimum categories do not reflect the increasing diversity of the nation's population. Some have also proposed changing the names of several categories.

In response to these concerns, OMB established an Interagency Committee for the Review of the Racial and Ethnic Standards. The committee created a Research Working Group charged with developing a plan to research a number of the important issues relevant to the review of the racial and ethnic categories. The first project of the research agenda was a Supplement to the Current Population Survey (CPS), which was conducted in May 1995, and was designed to collect information on several of the key issues under review.

Design of the Supplement

The CPS is a monthly survey of the population using a scientifically selected sample of close to 60,000 households in May, representative of the civilian, noninstitutional population of the United States. Households selected in the sample are interviewed for 4 consecutive months, are not interviewed for 8 months, and then are interviewed again for 4 consecutive months. The survey is administered by interviewers, who conduct a majority of the interviews by telephone. Race and ethnic origin of household members are routinely collected during the first month's interview. A CPS Supplement, a set of questions asked following the labor force survey, provides the opportunity to evaluate new panels of questions on race and ethnicity by comparing the results to answers given using the current CPS questions.

The English and Spanish versions of the Supplement were prepared in consultation with questionnaire design experts and subject matter experts. All questions were pretested in several rounds of cognitive testing with respondents representing the major racial and ethnic groups and geographical regions of the United States. The Supplement addressed the following important issues: (1) the effect of having a multiracial category among the list of races, (2) the effect of adding "Hispanic" to the list of racial categories, and (3) the preferences for alternative names for racial and ethnic categories (e.g., African-American for Black, and Latino for Hispanic). The Supplement was organized into four panels or versions representing a two-by-two experimental design for studying issues 1 and 2 above. Questions relating to issue 3 were the same for all panels. Each panel was given to one-fourth of the sample, or about 15,000, households. All respondents in a household received the same set of questions; household members 15 years and older were asked to respond for themselves, and parents answered for children too young to answer for themselves. The panels were:

Panel 1: Separate race and Hispanic-origin questions, no multiracial category

Panel 2: Separate race and Hispanic-origin questions, with a multiracial category

Panel 3: A combined race and Hispanic-origin question, no multiracial category

Panel 4: A combined race and Hispanic-origin question, with a multiracial category

In panels 1 and 2, the Hispanic-origin question was asked before the race question. The racial categories used in all 4 panels were (1) White, (2) Black, (3) American Indian, Eskimo, or Aleut, and (4) Asian or Pacific Islander. Besides these specific racial categories, respondents were given the opportunity to select "Something Else" (all panels), Hispanic (panels 3 and 4 only), or multiracial (panels 2 and 4 only). In addition, multiracial respondents were asked the reasons they identified as multiracial, and Hispanics were asked whether they preferred to identify themselves from a list of races that included Hispanic or through a separate Hispanic origin question.

Major Findings

Hispanic origin. A higher percentage of people identified themselves as Hispanic when they were asked a separate question than when "Hispanic" was included as a racial category. Even when respondents in panels 3 and 4, where "Hispanic" was a racial category, were given another opportunity to identify themselves as Hispanic, the proportion Hispanic remained below that in panels 1 and 2.

Racial distributions. The proportion identifying themselves as "White" is over 4 points higher in panels 1 and 2 when the Hispanic origin question is asked separately. The proportion of the samples identifying as Hispanic is about 8 percent in both panels 3 and 4, drawing from both the white category and the "Something Else" category. The multiracial category was included in panels 2 and 4, with very similar results in both panels--a little more than 1.5 percent identified as multiracial. When the multiracial category is included, the proportion of American Indian, Eskimo, or Aleut drops in both panels 2 and 4. The proportions for Blacks and Asian or Pacific Islanders were not affected by the introduction of either the Hispanic or multiracial options in the list of racial categories. A sizable percentage of respondents selecting the multiracial category chose only one race or gave both their race and their ethnic group when asked to specify their races (e.g., White and Irish). Respondents who specified two or more races were most likely to select a combination including American Indian, Eskimo, or Aleut.

Hispanic as a racial category.A substantial majority of Hispanics from each panel preferred the racial category option. Those who received the separate ethnicity question had a somewhat lower percentage favoring the racial category.

Preferred terminology. The majority of Hispanic respondents chose " Hispanic" as the term they preferred. A majority of whites chose "White". A large plurality of blacks preferred the term "Black," but almost as many chose "African-American" or "Afro-American." More than half of those identifying as American Indian or one of the classes of Alaska Native preferred either "American Indian" or "Alaska Native," but over a third chose the more generic "Native American." Almost 30 percent of those identifying as multiracial preferred the term "Multiracial," but about as many had no preference.

Comparison between CPS and Supplement measures for Hispanic Origin. A comparison between Hispanic classification in the CPS and Hispanic classification in the Supplement showed that the level of consistency differed across panels, and the differences can be attributed largely to the method for ascertaining Hispanic origin in the supplement. With a separate ethnicity question, the agreement is about 94 percent. If Hispanic is included as a race, the agreement drops to 80 percent.

Comparison between CPS and Supplement measures for Racial Identification. A comparison of racial identification in the CPS and the Supplement showed that regardless of the panel, almost 95 percent of those identifying as "Black" in the CPS question also identify as "Black" in the Supplement. The level of consistency for whites is about 96 percent in panel 1 and 2 but drops to below 91 percent in panels 3 and 4, where some whites select "Hispanic" instead. The agreement for Asian and Pacific Islander respondents ranges between 86 and 93 percent, but no detectable pattern emerges across panels. Most of those who identified as "Other" in the CPS question either selected "Something Else" (about 60 percent) or "White" (about 25 percent) in panels 1 and 2. In panels 3 and 4, however, over 70 percent of them chose "Hispanic." In contrast to the other racial categories, the agreement rate between the CPS and Supplement for American Indian, Eskimo, or Aleut, never reaches more than 75 percent in any panel. It is only 60 percent when multiracial is included as a category in panels 2 and 4. An analysis of racial identification by ancestry indicates that respondents with both White and American Indian ancestry may not consistently identify with one or the other.

Responses of Hispanics from different national origins. In panels 1 and 2, where Hispanic origin is determined by a separate question, Mexicans make up a smaller proportion of those identifying as Hispanic than in panels 3 and 4, where Hispanic origin is determined by identification with the Hispanic racial category. On the other hand , those identifying as "Cuban" or "Other Hispanic, Latino, or Spanish" make up a larger proportion of the Hispanics in panels 1 and 2. Cubans appear more likely to identify as "White" in all panels compared with the other nationalities.

Conclusions from Cognitive Research. For many Americans, race, ethnic group, and ancestry are overlapping concepts. When offered, the multiracial category appears to be a discretionary category for many respondents. Whether they identify as multiracial depends on community membership, family dynamics, and other aspects of their background not necessarily related to race. Hispanics' use of the multiracial category appears to be influenced by their level of identification with the existing racial categories.

II. BACKGROUND

The United States Government has long collected statistics on race and ethnicity in order to study changes in the social, demographic, health, and economic characteristics of various groups in our population. Federal data collections, through censuses, surveys, and administrative records, have provided an historical record of the Nation's population diversity and its changing social attitudes and policy concerns. Since the 1960's, data on race and ethnicity have also been used extensively in civil rights monitoring and enforcement, covering areas such as employment, voting rights, housing and mortgage lending, health care services, and educational opportunities. These legislatively-based priorities created the need among Federal agencies for compatible, nonduplicative data for specific population groups that historically had suffered discrimination and differential treatment on the basis of their race or ethnicity. In addition, ethnic categories were also needed to implement the requirements of Public Law 94-311 of June 16 1976, which called for the collection, analysis, and publication of economic and social statistics on persons of Spanish origin and descent.

In response to these needs, the Office of Management and Budget (OMB) issued in 1977 the "Race and Ethnic Standards for Federal Statistics and Administrative Reporting" contained in Statistical Policy Directive No. 15. The basic racial categories set forth in the Directive are: American Indian or Alaskan Native; Asian or Pacific Islander; Black; and White. The ethnic categories specified in the Directive are Hispanic origin, and Not of Hispanic origin. The population groups identified by the Directive No. 15 racial and Hispanic origin categories, therefore, reflected legislative and agency needs, and not efforts by population groups to be specifically identified.

The standards in this Directive have been used for almost two decades throughout the Federal government for record keeping, collection, and presentation of data on race and Hispanic origin. The standards have been used in two decennial censuses, and in surveys of the population, data collections necessary for meeting statutory requirements associated with civil rights monitoring and enforcement, and in other administrative program reporting. During the past several years, the standards have come under increasing criticism from those who believe that the minimum categories set forth in Directive No. 15 do not reflect the increasing diversity of our Nation's population. Some have also proposed changing the names of some categories.

The review of Directive No. 15 began in 1993 when the then Chairman of the House Subcommittee on Census, Statistics, and Postal Personnel, Thomas C. Sawyer, held four hearings focusing primarily on the measurement of race and ethnicity in the decennial census. In testimony on July 29, 1993, OMB announced that it would undertake a comprehensive review of the categories. As a first step in its review, OMB asked the National Academy of Sciences' Committee on National Statistics to organize a workshop to discuss issues surrounding a review of the categories. Convened on February 17 - 18, 1994, the workshop included representatives of Federal agencies, academia, social science research institutions, interest groups, private industry, and a local school district.

In order to facilitate collaboration and cooperation with Federal agencies, OMB established and held the first meeting in March 1994 of the Interagency Committee for the Review of the Racial and Ethnic Standards, whose members represent the many and diverse Federal needs for racial and ethnic data, including statutory requirements for such data. Over thirty agencies are members of the Committee, ranging from the principal statistical agencies (such as the Bureau of the Census, Bureau of Labor Statistics, National Center for Education Statistics, and the Center for Disease Control) to monitoring and enforcement agencies (such as the Equal Employment Opportunity Commission, the Civil Rights Division in the Department of Justice, the Office of Federal Contract Compliance Programs in the Department of Labor, and the Office for Civil Rights in the Department of Education). The Interagency Committee is an integral part of this review process, assisting OMB in the development of a research agenda to evaluate and assess proposed changes on, for example, the quality of the resulting data and costs of implementation.

Among the first tasks undertaken by the Interagency Committee was the development of a set of general principles to govern the review process. This process entails not only evaluating suggestions received during the public comment period but also balancing statistical issues, data requirements, and social concerns.

In the June 9, 1994, Federal Register notice, OMB invited comment on three areas: (1) the adequacy of the Federal racial and ethnic categories currently used; (2) the suggested changes and criticism offered by the public during the past few years; and (3) the principles that had been developed to govern the review process. To provide additional opportunities to hear views from the public on Directive No. 15, OMB conducted a series of public hearings during July 1994 in Boston, Denver, San Francisco, and Honolulu.

In response to the 1994 notice, the public commented on the need for new categories, changes for the current categories, whether racial and ethnic data should be collected, legislative and programmatic needs for the data, and the issue of self-identification versus observed identification. OMB received nearly 800 letters in response to the 1994 Federal Register notice and heard the testimony of 94 witnesses during the four public hearings. OMB heard from a wide array of interested parties including individuals, data users, and data providers from within and outside the Federal Government. This included comment by Federal agencies on their requirements for racial and ethnic data. Another Federal Register notice, issued August 28, 1995, summarized these findings and also briefly described research activities in several areas.

The Interagency Committee for the Review of the OMB Racial and Ethnic Standards created a Research Working Group charged with developing a plan to research the more important issues relevant to the review of the racial and ethnic categories. This led to an interagency research initiative, within existing resources, to assess new racial and ethnic reporting categories as well as the potential effects which changes in the categories would have on the quality of the resulting data. In June 1994, after conducting a literature review, the Research Working Group proposed the first project for its research agenda. That project was a Supplement to the Current Population Survey (CPS), funded by the Bureau of Labor Statistics (BLS), which would collect information on several key issues under review.

III. THE CPS SUPPLEMENT ON RACE AND ETHNICITY

The Current Population Survey is the monthly national labor force survey of the population collected by the Bureau of the Census for the BLS which uses a scientifically selected sample of close to 60,000 households representative of the civilian noninstitutional population of the United States. The CPS routinely collects information on the race and ethnic origin of household members during the first month's interview. Thus, a CPS Supplement provided the opportunity to try out new versions of questions on race and ethnicity with CPS respondents, and provided comparative data on how these questions were answered under current and modified categories and classifications.

The research priorities for the Supplement were influenced by major demographic changes in the population (Evinger, 1995; Harrison and Bennett, 1995; Marks, 1994); recent research findings related to alternative question formats to reduce Hispanic undercoverage (Bates, Martin, DeMaio and de la Puente, 1994; Kamasaki, 1994 ); changing group preferences for racial and ethnic designations (Smith, 1992); and concerns about the validity, reliability, and popular understanding of racial and ethnic classifications (Hahn and Stroup, 1994; Elias-Olivares and Farr, 1992; Kissam, Herrera, and Nakamoto, 1993; Martin, De Maio, and Campanelli, 1990; Rodriguez, 1994; Snipp, 1986).

III.1 Design and Content of the Supplement

Among the research issues considered for inclusion in the Supplement were: (1) The effect of having a multiracial category on the population counts for the current racial and ethnic categories; (2) The effect of adding "Hispanic" to the list of racial categories; (3) The effect of placing a question on Hispanic origin before the question on race; (4) Questions designed to explore the acceptability of alternative names for the racial and ethnic categories, e.g., African-American for Black, Latino for Hispanic; and (5) Conceptual questions designed to learn about the criteria which respondents to surveys use to categorize themselves and others into racial, ethnic, and ancestral groups. The final version of the Supplement did not include questions for researching conceptual issues because cognitive testing indicated further research was necessary for developing a useful set of questions for obtaining such information from respondents.

The Research Working Group prepared the first draft of the questions for the Supplement. The questionnaire was reviewed by panels of questionnaire design experts and subject matter experts; the latter panel was composed of academics who are authorities on the major racial and ethnic groups within the United States. The comments from the two expert panels were used to develop the version of the Supplement that was evaluated by means of cognitive interviews.

III.1.1 Experimental Design of the Supplement

In the Supplement, two factors were experimentally manipulated to gauge their effect on the respondent's racial and ethnic identification. The first factor was whether there was a separate or combined race and Hispanic origin question(s), and the second factor was the presence or absence of a multiracial category in the race question. The Supplement was organized into four panels or versions representing a two by two experimental design with the following experimental conditions for each panel:

Panel 1: Separate race and Hispanic origin questions, no multiracial category;

Panel 2: Separate race and Hispanic origin questions, with a multiracial category;

Panel 3: A combined race and Hispanic origin question, no multiracial category;

Panel 4: A combined race and Hispanic origin question, with a multiracial category.

This design allows for a number of comparisons of the individual panels and pairs of panels. The effect of having a separate versus combined race and Hispanic origin question can be examined by combining panels 1 and 2 (separate Hispanic origin question) and comparing the results with the combination of panels 3 and 4 (combined race and Hispanic origin questions). Likewise, the effect of adding a multiracial category can be examined by combining panels 2 and 4 (both contain a multiracial category) and comparing them with the combination of panels 1 and 3 (no multiracial category). Further contrasts and comparisons between individual panels can also be done. For example, a comparison of reporting in panels 1 and 2 will identify the effects of adding a multiracial category to the separate race question; a similar comparison of panels 3 and 4 will identify these effects in the context of a combined race and Hispanic origin question. Comparing panels 1 and 3, the effects of combining race and Hispanic origin can be identified independently of the multiracial category, and contrasting panels 2 and 4 allows for a comparison of the use of a multiracial option when Hispanic was not included in the list of races (panel 2) and when Hispanic was available (panel 4).

III.1.2 Supplement Measures

In panels 1 and 2, the first question on the Supplement is the Hispanic origin question and the next question is the race question. For panels 3 and 4 the first question is a combined race and Hispanic origin question. A listing of these and other major questions from the Supplement are given in Appendix A. Depending on their answers to these initial questions, respondents are given additional opportunities to identify their race(s) and ethnicities. For example, in panels 1 and 3, there was no multiracial category; however, a follow-up question asked respondents who had already reported their race if they would have liked a multiracial option to report their racial identity more satisfactorily. If they replied yes, they were asked what other races they were. From those who chose this option, something could be learned about the potential shifts from single racial categories to a multiracial category. Because the Supplement was to be collected in telephone and personal interviews, respondents in panels 2 and 4 could not anticipate that a multiracial category would be offered later in the list of racial options. There was, therefore, some concern that respondents would stop listening carefully after "their" race was read, and not be aware of the multiracial option which appeared later in the list. For this reason, a follow-up question for listing additional races was included in panels 2 and 4 for those who had not chosen the multiracial category on the initial race question. Thus, there is one opportunity to report as "multiracial" on panels 1 and 3 and two opportunities to report as "multiracial" on panels 2 and 4.

There are also different opportunities across the panels for respondents to report Hispanic origin. On panels 1 and 2, the first question is the respondents only opportunity to identify as Hispanic. Respondents in panels 3 and 4 receive the initial combined race/Hispanic origin question, and additional opportunities. Specifically, respondents who had not identified as Hispanic in response to the initial combined race/Hispanic origin question, were asked if they also considered themselves to be Hispanic. Panel 3 provides for a total of three opportunities to identify as Hispanic: first, by choosing Hispanic on the combined question; second, by answering "yes" to the follow-up question on Hispanic ethnicity; and third, by listing Hispanic on the follow-up question which asks respondents who would have liked a multiracial category about the additional groups they would list. Panel 4 provides one more opportunity to those described for panel 3 because the initial combined race/Hispanic origin question in panel 4 contains a multiracial category. Respondents who choose the multiracial category on this first question are able to list Hispanic as one of their groups.

All respondents who identified as Hispanic, either on the initial combined race/Hispanic origin question or the later follow-up question, were asked to name their Hispanic origin group, e.g., Mexican, Puerto Rican, Cuban. This permits comparisons of reporting race versus Hispanic origin among the various Hispanic origin groups. In addition, respondents who identified as Hispanic were asked if they preferred to identify as Hispanic through a separate question on Hispanic origin or to choose Hispanic in the list of races. The exact wording of the question depended upon the panel (see Appendix A). In panels 3 and 4, Hispanic respondents, who answered a combined race and Hispanic origin question, were asked if they would have preferred separate questions on Hispanic origin and race to the combined version. In panels 1 and 2, Hispanic respondents, who answered separate Hispanic origin and race questions, were asked if Hispanic should be included as a racial category instead of a separate question.

On all four panels, respondents who chose the multiracial reporting category were asked about their reasons for using the multiracial category. These options included: (1) because their parents are from different racial groups; (2) because their ancestors before their parents were from different racial groups; (3) because the specific group to which they belong is mixed; (4) because of some other reason. In addition, all of the panels concluded with a question about the language(s) the respondent spoke at home.

All four panels also asked the same questions on other aspects of race and ethnicity, such as ancestry and preferred racial terms. All respondents were asked to give their ancestry or ethnic origin in a free-response format. Based on their racial identification earlier in the interview, respondents other than Asian or Pacific Islanders were also asked to choose their preferred racial term from a list of terms for that group. They could also provide another term they preferred or indicate no preference for any of the terms. Respondents identifying as American Indian, Alaska Native, or Aleut were given the following terms to choose from: American Indian, Alaska Native, or Native American. Respondents identifying as Black were given the following terms to choose from: Black, African-American, Afro-American, Negro, or Colored. Respondents identifying as Hispanic were given the following terms to choose from: Hispanic, Latino, or of Spanish origin. Respondents who identified as multiracial were given the following terms to choose from: Multiracial, More than one race, Biracial, or Mestizo/Mestiza. Respondents who identified as White were given the following terms to choose from: White, Caucasian, European-American, or Anglo.

III.1.3 Race and Hispanic Origin from the CPS

During the first month's CPS interview, respondents are asked the race and origin of each member of the household. The actual CPS questions are shown in Appendix A. In May 1995, all CPS respondents were also asked the questions from the Supplement on race and ethnicity. Most of the respondents had given their race and origin during a previous month's interview, but respondents who were in the sample for the first time in May 1995 (1/8 of the sample) were asked the CPS race and origin questions and the Supplement questions on race and ethnicity in the same interview (the supplement questions were asked after the basic CPS interview was completed).

The availability of racial and ethnic origin data from the CPS allowed comparisons between how respondents answered the CPS questions and how they responded to the questions in different panels of the Supplement. In addition, other demographic characteristics collected as part of CPS can be used to analyze the characteristics of people more affected by the different ways of determining racial and ethnic identity in the different panels of the Supplement.

III.2 Cognitive Research in the Development of the Supplement on Race and Ethnicity

The research protocol to test successive versions of the Supplement was developed by a multiracial and multi-ethnic team of behavioral scientists from several Federal agencies. The protocol called for individual, face-to-face interviews in which the respondent completed one of the four panels to be tested. The respondent was to answer all of the questions on one of the four panels. After responding to each question, the respondent was asked to paraphrase the question, i.e., to tell the interviewer what the question meant in his or her own words. For questions containing terms of special interest to the research, e.g., race, ethnicity, Latino, the respondent was also asked to provide a definition of these terms. (See Royston, et al. 1986, for a more complete description of paraphrasing and probes in cognitive interview research.)

All of the materials used in the cognitive research, including the four panels of the Supplement and the research interview protocols, were translated into Spanish for use with Hispanic respondents. The protocol for developing the Spanish translation of the Supplement called for independent Spanish translations of the English questionnaire by translators of Mexican origin, Puerto Rican origin, and Cuban origin. Differences among the three translations were negotiated in a reconciliation conference attended by all three translators. In addition, a "decentering" approach was used in developing the Spanish version whereby any terms that were not equally intelligible in both languages would not be used.

The research plan called for matching the race and Hispanic origin of the respondent and researcher. This was achieved for all groups except for the American Indian respondents who were interviewed by an Asian-American researcher. The cognitive interviews were carried out in three phases between November 1994 and January 1995. Each successive phase tested a version of the questionnaire that had been revised in light of problems found in earlier rounds of cognitive research.

A total of 82 cognitive interviews were carried out in the following locations with the groups indicated: Albuquerque (American Indians); Chicago (Blacks); Houston (Hispanics, Whites); New Orleans (Creoles); New York (Hispanics, Whites); Rural California (Hispanics); Rural Mississippi (Blacks); Rural West Virginia (Whites); San Francisco (Asians and Pacific Islanders, Hispanics, multiracials); Washington, DC (Asians and Pacific Islanders, Blacks, Hispanics, multiracials, Whites). For each of the racial groups, it was possible to interview respondents who had a high school education or less, as well as respondents with some years of college.

This report summarizes the cognitive research findings relevant to interpreting the results of the May 1995 CPS Supplement on Race and Ethnicity. They will be discussed under the following headings: Respondent sensitivity to the Supplement questions; Conceptual difficulties distinguishing race, ethnicity/ethnic origin, and ancestry; and Multiracial category. (See McKay and de la Puente, 1995, for a more detailed account of the cognitive research on the May 1995 CPS Supplement.)

III.2.1 Respondent Sensitivity to the Supplement Questions

Many respondents were uncomfortable answering questions about race, finding it to be an emotionally-charged subject. Concern was expressed that the survey was about racial attitudes or racism, rather than about racial classifications. This concern led some respondents to offer what might be termed "politically correct" answers. Thus, in response to an early version of the question about wanting a multiracial category to be added to the list, several respondents enthusiastically agreed that it should be added for a variety of "politically correct" reasons. The question was reworded to ask more clearly if the respondents would like a multiracial category to describe themselves better or others in their households.

III.2.2 Conceptual difficulties distinguishing race, ethnicity/ethnic origin, and ancestry

Early versions of the Supplement contained several conceptual questions designed to learn about the criteria which respondents used to define such terms as race, ethnicity/ethnic origin, and ancestry. Despite several attempts to make these questions less abstract and easier to answer, the overwhelming majority of respondents found the questions too difficult. For all but a few, highly-educated respondents, it appeared that the terms represent overlapping concepts which draw on a single semantic domain. The relative lack of differentiation among the three concepts was also evident in the answers given to less abstract questions which asked respondents to name their ancestry or ethnic origin. In some cases, respondents gave the same answer, e.g., Black or White, to questions about their race and ancestry or ethnic origin. Respondents belonging to the Asian or Pacific Islander category tended neither to find the Supplement questions sensitive nor have any difficulty distinguishing among the terms. For some groups who have been in the United States for several generations, such as the rural Whites in West Virginia, the question about their ancestry or national origin was met with a long pause, followed either by recalling that they had an American Indian ancestor some way back, or that they were not quite sure (All of the respondents interviewed in this community had British surnames.). Of interest is the spontaneous comment, volunteered by several urban white respondents, that they "never think about (their) race."

III.2.3 Multiracial category

The term "multiracial" was understood across all racial and ethnic groups surveyed. The cognitive interviews revealed that, for some respondents, multiracial was a discretionary category. Thus, one man of Hispanic and Black parentage identified himself as "Black," and did not use the multiracial category because "society sees me as Black." Another man, of American Indian and Hispanic parents, identified only with the former race because his father was Hispanic and the respondent did not get along well with his father. Respondents in a rural Black Mississippi community did not use the multiracial category even though, as their interviews revealed, they "knew that they had some other races in them." This fact did not matter, however, for their membership in that Black community.

Earlier it was mentioned that some respondents expressed confusion about the difference between race and ethnic origin. This confusion affected the way in which some single-race respondents used the multiracial category. One suburban, college-educated woman in her forties, who had initially identified as "White," chose the multiracial reporting option when it was offered later in the interview. The woman, who was of Italian and Irish parentage, said that she chose the multiracial category because "her parents were of different races."

III.3 Cognitive Research conducted during the May 1995 CPS Supplement

Once the final Supplement instrument was constructed, additional cognitive research was conducted to help interpret the results from the Supplement. During the CPS collection week of May 14th through May 20th, cognitive researchers monitored Supplement interviews in the Hagerstown and Tucson CATI facilities on one day, and conducted focus groups with interviewers in both facilities on the following day to learn about their experiences in conducting the Supplement interviews. Also during that week, researchers accompanied CPS interviewers in Tucson and in Miami to observe how the interview was conducted in the field. Four hundred CATI interviews, two hundred in each CATI facility, were taped for subsequent behavior coding. The findings of this research, summarized below, will help in interpreting the results from the statistical analyses of the CPS Supplement data.

III.3.1 Monitoring CATI interviews

The monitoring of CPS interviews conducted during the evening hours revealed that fatigue seemed to hinder respondents from always paying careful attention to the questions. This appeared to be especially true for respondents who had reported labor force information for several household members. Respondents had a tendency to interrupt the reading of the list of races by volunteering their race after it was read aloud. Most interviewers explained that they were required to read the entire list and did. The question about wanting to have a multiracial category, or the opportunity to report as more than one race, caused problems for some respondents. Unlike the other, more factual questions, this was asking them about a hypothetical situation. One woman thought that she was being asked if she would like to be more than one race. Some interpreted it as an opportunity to report their ethnicity, e.g., Irish and Italian, although some regarded ethnicities as races and indicated that they are multiracial because their parents are of different races.

A combination of fatigue plus political sensitivity or social desirability led a respondent who had identified his race and his daughter's race as white, to give an affirmative answer to the question about wanting to have had a multiracial category for reporting his daughter's race. "I want her to grow up in a world where there are multiracial categories." When he was asked about the additional racial categories he would report for his daughter, he paused and said that he knew that there was some American Indian in his family, but he wasn't sure how far back. The interviewer recorded "American Indian, Eskimo, Aleut," for the daughter. In another interview, a respondent chose the multiracial reporting option but gave only one race: African American (Something else). This respondent believed that African American was a multiracial term and it sounded better (than Black).

Monitoring revealed that multiracial reporting could change over the generations, and could change in either direction. One young woman in panel 2, who reported her mother as "multiracial," (Black, American Indian, Hispanic), her father as "Black," and her own race as "Black." In panel 4, a mother who reported her race as Black, and ancestry as American Indian, and her daughter's race as Black, later took advantage of the "more than one race" question to add Cherokee for her daughter.

Interviewers were occasionally asked to explain what the "ancestry or national origin" question was asking. In one interview, the respondent said he didn't know. The interviewer asked, "You don't know where your relatives came from?" The respondent replied, "Well, its been about 200 years." Some interviewers told respondents that they need only report one ancestry.

III.3.2 Field observation of CAPI interviews

Cognitive researchers observed May CPS interviews conducted in Miami and Tucson to learn about "Hispanic" and "multiracial" reporting. These observations pointed to problems with the interpretation of concepts and terms relating to these categories. The following is a summary of some of these observations.

In Miami, two of the three households used Hispanic origin and multiracial in responding. In a household interviewed with panel 1, a Mexican-American man reluctantly chose "White" from the list of racial categories, gave an affirmative response to the question about wanting a multiracial reporting category, and said "Mexican-American," when asked which additional groups he would like to add to "White." This response was coded as "Something else." This respondent answered "No," to each of the reasons offered for choosing the multiracial category, and did not provide one when asked if he had another reason for choosing the multiracial category. In another household interviewed with panel 2, a non-Hispanic white woman answered the questions for her 15 month-old daughter whose father is Mexican-American. The woman said that her daughter is Hispanic, chose "multiracial" for her race, and listed " White" and "Hispanic" as her daughter's racial groups. The woman gave an affirmative answer to the first three reasons for choosing the multiracial category, i.e., parents and ancestors were from different racial groups and she belonged to a mixed group. In a third household which was in panel 3, an elderly man who listed his race as White, and declined the multiracial option, included "American Indian" in his ancestry/ethnic origin response.

The CPS interviewer in Miami made the following observations on responses to the Supplement: "Some Black respondents give 'English' for their ancestry and I think that it means 'American.' Some respondents will say 'White' for race and 'White' for ancestry. Some said, 'It's White for everything.' If people think they're Mexican-American, it's 'Mexican-American' for everything. On preferred racial term, a 72 year-old woman said, 'Not African American. We're through with all that business. Don't want to go back to being African American, being slaves. My ancestry is 'Black'."

Two of the three households visited in Tucson were non-Hispanic white; the Mexican origin respondent in the third household was interviewed in Spanish with panel 1 of the Supplement. In reporting her race, and that of her husband and children, the respondent shook her head to indicate "No" for each of the racial categories, leading them to be placed in the "Something else" category. This respondent later indicated that she would like to see "Hispanic" added to the list of racial categories.

III.3.3 Behavior coding of questions on multiracial status in taped CATI interviews

The questions on multiracial status in the taped CATI interviews were behavior coded for a range of interviewer and respondent behaviors. The interviewer behaviors coded in this analysis included: Major or minor changes in the wording of questions; verifying vague answers; probing for incomplete answers, and correct or incorrect coding of responses. Respondent behaviors which were coded included: Requests for clarification, interrupting the reading of a question to give a response, offering "don't know" or refusal responses, providing inadequate or adequate answers, and any comments regarding the difficulty or sensitivity of the question. The coding data revealed that respondents found the race follow-up question on panels 1 and 3, which asks "Would you have liked a 'Multiracial' category on the list to better describe yourself?" to be the most problematic. The CPS usually asks only factual questions. To be asked a hypothetical question about an unfamiliar subject was confusing to some respondents. There were relatively few problems in the way the interviewers read and coded the questions, other than occasionally not reading the entire list of races for the race questions.

One of the problems in the questions about multiracial reporting is the use of the phrase "choose more than one group on the list to better describe yourself" instead of "more than one race." Respondents did not always pay attention to the phrase "on the list," and seemed to think they were being asked for additional ways they could describe their identity. Also, because the respondent could not see the layout of the race and ancestry/origin questions, they could not anticipate that they would have a separate question on their ethnic origin. Some respondents said that everyone would want a multiracial category so that they could report everything they were, e.g., Dutch, German, etc. When the ancestry/ethnic origin question was asked later in the interview, some respondents were uncertain about how to report their ancestry, e.g., "We've been here so long, I'm not sure I know."

Interviewers would not always read the term "multiracial" accurately. Interviewers occasionally substituted such terms as "multicultural," "multinational," "multiple-racial," and "multalracal." One interviewer pronounced "origin," on the Hispanic origin, and the ancestry and ethnic origin questions, as "eurogin," which few respondents could understand.

The vast majority of the respondents of all racial groups seemed surprised by the question on their "preferred racial term." Many said that they had never given it any thought.

III.3.4 Focus group interviews with CATI interviewers

Interviewers reported no problem in respondents' comprehension of the Hispanic origin question on panels 1 and 2. A number of the non-Hispanic respondents seemed to be surprised at being asked this question first, that is, it was not what they were expecting. Some non-Hispanics also seemed either confused or offended to be asked this question first. They would say, "Oh, no. Nothing like that." Respondents would also interrupt the list of races being read when they heard their race, especially if this was the second or third proxy report. Some of the Tucson interviewers reported that some Hispanics wanted to report their race as "Brown." Although Hispanics did not find the follow-up questions, e.g., want Hispanic as a racial category, to be difficult, some Hispanics felt they were being singled out, and that Hispanic was being emphasized too much. Some non-Hispanics also asked why there were so many questions on Hispanic.

Interviewers in both CATI centers found the most problematic question to be the one asking respondents, who had already reported their race, if they would have preferred having a multiracial category to report their race. (The regular CPS asks only factual questions.) This question frequently elicited requests to have the question repeated, and some respondents said that they did not understand what they were being asked. Interviewers believed that some respondents would say "Yes" without really meaning "Yes."

III.3.5 Conclusions from Cognitive Research

Respondents' use of the multiracial category was influenced by a number of circumstances, including their perceptions of how they were viewed by their communities, their confusion about the differentiation of "race" and "ethnicity," their personal attachments to a parent of one race, and their concerns for "political correctness," among others. Thus, some respondents with membership in more than one race will not choose the multiracial category while others belonging to a single race will identify as multiracial. Some single race reporters will list a second race in response to the question on ancestry. Hispanics' use of the multiracial category, in panels where "Hispanic" was not offered as a racial category, appeared to be influenced by their lack of identification with the other racial categories. Some respondents will ignore the racial response categories being read to them after they hear the category with which they usually identify.

IV. STATISTICAL DESIGN

In the CPS, approximately 60,000 eligible households are sampled each month in a two-stage, clustered design. Households selected in the sample are interviewed for four consecutive months, are not interviewed for the next 8 months, and then are interviewed again for 4 consecutive months. In any given month, one-eighth of the sample is composed of households participating for the first-time (month-in-sample 1), one-eighth the second-time, etc.. The first and fifth interviews are done in person and the other interviews are typically done over the phone by a field interviewer or an interviewer in a central facility. Since January 1994, all data collection for the CPS has been conducted using computer-assisted methods, with laptop computers in the field and desktop computers in the central telephone facilities.

The Race and Ethnicity Supplement contained four different panels of questions on race and ethnicity as described earlier. Each household was randomly assigned to receive only one panel of questions with all persons within the household answering the same set of questions. The four panels were further randomly assigned to households within clusters, which means that each household in a cluster received a different panel. This design results in each panel of questions being administered equally in every part of the country.

IV.1 Response Rates and Weighting Adjustments

The CPS had a 6.5 percent nonresponse rate in May. There was also a 10.6 percent nonresponse rate to the Race and Ethnicity Supplement, in addition to the nonresponse to the CPS. This level of nonresponse is typical of many CPS supplements and did not differ by panel. Normally, CPS base weights go through an adjustment for household nonresponse. Because of a concern about the impact of the additional nonresponse on the Supplement, the Census Bureau did an additional nonresponse adjustment to the weights to help compensate for the Supplement's nonresponse rate. All statistical analyses were conducted using this weight that was adjusted for nonresponse to the CPS itself and the Race and Ethnicity Supplement.

The weight adjusted for nonresponse in the CPS and the Supplement is not the same as the final weight that is typically used in analyzing data from the CPS. The CPS final weights are inflated to independent estimates of the civilian noninstitutional population of the United States. The analyses of the May Supplement did NOT use the weight with these adjustments because they are based on data collected in the basic CPS including race and Hispanic origin and would distort the effects of the Supplement's experimental design. Therefore, racial and ethnic proportions from the Supplement must be interpreted within the context of the experimental design. The percentages reported may not correspond with other sources and are only intended for comparisons among the four panels.

IV.2 Statistical Methodology

The CPS has a complex, two-stage clustered sample design. All statistical analyses on the effects of the different panels were conducted using CPLX, a program written by Robert Fay of the U. S. Bureau of the Census for log-linear model analysis of complex survey data that takes into account the complex sample design (Fay, 1989). This program uses a generalized replication method to calculate standard errors, and it also computes a jack-knifed chi-square statistic that is appropriate for the complex sample design. The four panels described earlier represent the effects of the presence or absence of a separate Hispanic question and the presence or absence of a separate multiracial category on respondents' choice of racial and ethnic categories. Log-linear analyses were conducted treating each of these effects as separate independent variables to isolate the impact of each one.

IV.3 Interpretation of Statistical Tests

A table based on the results of each jack-knifed chi-square test(s) is presented in Tables 1-18. Because of the complex nature of the analyses, only summary information for the analyses in section VII are shown in Tables 19 and 29. To test the panel effects, a hierarchical series of log-linear models were tested for their fit to the data. The first model tested was always whether there were any differences among the panels. If there were no significant differences among the panels, then 'No' appears in the first line, and no further statistical tests were conducted. If there were significant differences (p < .05) demonstrated across panels, the top line of this table indicates a 'Yes', and subsequent models were run to discover which best accounted for the data. In all cases, the simplest model that adequately fit the data was accepted, and this is indicated by a 'X' in the column next to that model. The models of panel effects that were tested included: (1) the effect of a separate Hispanic question (compares panels 1 and 2 with 3 and 4); (2) the effect of a multiracial category (compares panels 1 and 3 with 2 and 4); (3) the effect of both a separate Hispanic question and a multiracial category (i.e., both of the previous effects showed significant differences); and (4) the effect of both a separate Hispanic question and a multiracial category and their interaction (differences across the four panels showed a more complex pattern such that the effect of having a separate Hispanic question differed depending upon whether or not there was a multiracial category, and vice versa). The findings and interpretation of the best-fitting model for each analysis are discussed below. Further details about the log-linear model fitting, the jack-knifed chi-square and the exact significance levels of each statistical test reported in the following sections are given in Appendix B.

IV.4 Overview of Analyses of the Supplement

Section V contains the analyses of the Supplement questions by panel. These analyses include an examination by panel of the distributions of Hispanic origin, race, preferences for Hispanic origin being included as a racial category, preferences for racial terms, national origin of Hispanics, reasons multiracial respondents had for choosing the multiracial category, and the relation by panel between ancestry and racial identification. Section VI contains comparisons of the respondents' CPS racial and ethnic origin to their choices in the Supplement. Section VII presents analyses of the relations among demographic characteristics and racial and ethnic origin by panel. Additional analyses also focus on major racial and ethnic origin groups from the CPS in order to examine the demographic characteristics of people who changed their racial or ethnic identification from the CPS to the Supplement by panel. Finally, section VIII contains some analyses of the effect of the CPS data collection methods on the racial and ethnic distributions of the Supplement by panel. One particularly important preliminary analysis in section IV.5 is the examination of the distributions by panel of race and Hispanic origin from the CPS as a precautionary measure to ensure that there were no preexisting differences among panels in their racial or ethnic composition.

IV.5 Analysis of Race and Hispanic Origin from the CPS by Panel

The CPS data on race and Hispanic origin were collected during the first month's interview. As noted earlier, only 1/8 of the sample were asked the CPS race and origin questions in the same interview as the Supplement questions. The four different panels of the Supplement were randomized within clusters. This procedure should result in virtually identical racial and ethnic distributions (as measured by the CPS items) across the different panels. Because sampling differences in race and ethnicity by panel could impact the integrity of the statistical tests of the panel effects on the Supplement questions, tests were conducted to make sure that there were no significant differences in the CPS racial and ethnic distributions across the panels. The distribution of races from the CPS by the Supplement panel can be seen in Table 1, and the percentage of Hispanics and non-Hispanics from the CPS by panel can be seen in Table 2. Although there did appear to be some slight differences across panels, these differences were not statistically significant. On the other hand, some of these small differences could complicate later analyses. When this is likely, it will be indicated.

V. ANALYSIS OF SUPPLEMENT QUESTIONS BY PANEL

The purpose of the analyses in this section were to examine how the different questions about race and ethnicity that were asked in different panels of the Supplement affected: (1) the percentage of persons who identified as Hispanic; (2) the distribution of racial identification; (3) whether Hispanics prefer to indicate their ethnic origin through a separate question or in the race question; (4) what racial and ethnic terms are preferred by members of each group; (5) the national origin of Hispanics; (6) the reasons people have for identifying as multiracial; and (7) the relation between respondent's ancestry and racial identification. The analyses comparing different panels of the Supplement show differences in racial and ethnic distributions and reveal which experimentally manipulated factors explain the differences.

V.1 Hispanic Origin from the Supplement

For panels 1 and 2, the first question in the Supplement was the Hispanic origin question, and in panels 3 and 4 the first question was the race question with Hispanic included as one of the choices (see Appendix A). Table 3 shows the distribution of Hispanics and Non-Hispanics for each panel based on respondents' answers to this first question. In panels 3 and 4, respondents who did not identify themselves as Hispanic were asked an additional question about whether they also considered themselves to be Hispanic. Table 4 shows the percentage of Hispanics for each panel after the additional questions to identify as Hispanic in panels 3 and 4 were given to respondents. As can be seen in both Tables 3 and 4, a higher percentage of people identified as Hispanic when they were asked a separate Hispanic question than when there was no separate question. A complex model including the effects of a separate Hispanic question and the interaction with a multiracial category was needed to explain the pattern of results in Table 3; however, the largest effect was whether there was a separated Hispanic origin question or a combined race and ethnic origin question. For Table 4, a model with only a separate or combined Hispanic origin question was sufficient to explain the patter of results. Specifically, 10.60 percent of the respondents who received a separate Hispanic question (panels 1 and 2 combined) identified as Hispanic. In contrast only 8.06 percent of the respondents who were not asked a separate question (panels 3 and 4 combined) identified as Hispanic when given the first opportunity to do so (Table 3)

and only 9.12 percent identified as Hispanic after given all opportunities to do so (Table 4). In Table 3, the examination of the interaction revealed that panels 3 and 4 were significantly different from each other. It is important to note that the higher proportion of Hispanics in panel 4 compared to panel 3 may be due to the multiracial respondents who were also Hispanic or may be an artifact of the random variation in the panels (see Table 2). Recently, del Pinal (1992) predicted that the percentage of Hispanics in the total U.S. population would drop, although by less than half a percentage point, if Hispanic ethnicity were treated as a racial category. These results indicate that a larger potential drop could occur.

V.2 Alternative Racial Distributions from the Supplement

Each panel of the Supplement contained a number of questions to elicit respondent's racial identification. Depending on their answers to the initial race question in each panel, respondents were asked to give further specification of their racial identification or were given additional opportunities to identify as multiracial or to list additional races to better describe themselves (see Appendix A). We constructed and analyzed several alternative racial distributions based on answers to one or more of these questions in the following subsections.

V.2.1 Use of the Multiracial and Hispanic Categories in the Race Question

Respondents to the Supplement answered at least one race question and were given at least one opportunity to identify as multiracial in all panels. In the race question, respondents were given "Hispanic" as a choice in panels 3 and 4 and were given a multiracial category as an option in panels 2 and 4. Respondents who selected the multiracial category were asked to specify their races in a follow-up question. Table 5 shows the racial distribution based on respondents' answers to the first race question they answered on the Supplement and includes a further breakdown of those who identified themselves as multiracial in panels 2 and 4. Overall, the percentages of persons who identified as White ranged from 74.66 percent to 79.88 percent, depending upon the panel. Between 10.27 percent and 10.66 percent of the people identified as Blacks. The percentage of people who identified as American Indian, Eskimo, or Aleuts ranged from 0.73 percent to 1.06 percent, and the percentage of people identifying as Asian or Pacific Islanders ranged from 3.25 percent to 3.83 percent. As can be seen in the lower portion of Table 5, of the respondents who said they were multiracial and also identified more than one race (in panels 2 and 4), the most common second race was American Indian. Other researchers have reported similar findings. For example, Snipp (1986) found that over 5 million persons reporting a race other than American Indian in the 1980 decennial census also reported an American Indian ancestry, and Passel and Berman (1986) found that in the November, 1979 CPS, about 10 million persons claimed some degree of American Indian ancestry.

A sizable percentage of the people who identified themselves as multiracial in panels 2 and 4 reported only 1 race when asked to identify what races they were, raising the possibility that a number of respondents who chose the multiracial category did not fit the definition of multiracial being used for this research, i.e., "belonging to two or more of the racial groups listed on the race question for that panel." Those who fit this definition are termed "multiracial;" those who did not are termed "indeterminate multiracial." Subsequent analysis revealed that a portion of the 1-race multiracial reporters did fit the definition of multiracial, while a portion of those who chose two racial categories on the list did not fit the definition.

The 1-race multiracial reporters included those who named a single race on the list, e.g., black, white, as well as those whose single race listed was "Something else." A portion of the latter group subsequently described themselves as members of racially-mixed groups, e.g., Creole, Eurasian. Such respondents do conform to the definition of multiracial being used for this research, and they are included in the multiracial category. The 2-race "indeterminate multiracials" are those who named one race on the list and "Something else," when the "Something else" entries consisted of ethnicities, e.g., Scottish, German.

If Hispanics who identified as multiracial but reported only 1 race, or 1 race and an ethnicity, are treated as "indeterminate multiracials" in panel 2, the "indeterminate multiracial" rate is 63 percent, with Hispanics comprising 15 percent of this group. Even without these Hispanics, almost 50 percent of those identifying as multiracial on panel 2 fell into the "indeterminate multiracial" group. When Hispanic is listed as a race in panel 4, the "indeterminate multiracial" rate drops to 20 percent.

V.2.2 Examination of Multiracial and Hispanic Categories

Table 6 shows the same racial distribution without the specific breakdowns of the multiracials and includes persons identifying as "Something else" and a few respondents who did not know or refused to answer the race question in the "All Other" category. Three

statistical comparisons were made, focusing on panels 2 and 4 only, to examine the effect of including Hispanic in the list of races on the percentage of people who identified as White, multiracial, and "All Other". There were significant differences between panels 2 and 4 in the percentage of persons identifying as White dropping from 79.74 percent with a separate Hispanic question (panel 2) to 74.66 percent with Hispanic included as a race (panel 4). There were no significant differences between panels 2 and 4 on the percentage of people identifying as multiracial (1.65 percent and 1.55 percent for panels 2 and 4, respectively), but there were significant differences between panels 2 and 4 in the percentage of people not choosing one of the racial categories (i.e., classified as "All Other"). When Hispanic is offered as a race (panel 4), only 1.23 percent of the respondents did not identify themselves with one of the racial categories offered, while 3.97 percent did not identify with one of the racial categories offered when Hispanic was not one of the choices (panel 2). These findings are less extreme than but similar to the results of the 1990 decennial census, in which Hispanics accounted for about 96 percent of the 10 million persons choosing "other race" on the race question (McKenney, Bennett, Harrison, and del Pinal, 1993).

An additional statistical comparison was made, focusing on panels 1 and 2 only, to examine the effect of including a multiracial category on the percentage of people who did not identify with one of the racial categories and compose the "All Other" category. There was a statistically significant effect for the presence of the multiracial category with only 3.97 percent of the respondents in panel 2 (multiracial category) not identifying themselves with one of the racial categories offered, compared to 5.03 percent of the respondents in panel 1 (no multiracial category) not selecting one of the racial categories.

V.2.3 Comparisons of Racial Distributions Across All Panels

Table 7 shows the racial distribution based on the first race question also including respondents who chose Hispanic (from panels 3 and 4), and multiracial (from panels 2 and 4), into the category of "All Other" in order to allow statistical comparisons across all four panels simultaneously. The percentage of people identifying with each racial category differed across the four panels, and these differences were statistically significant. The simplest model that provided an adequate fit to the data included both the effects of the separate Hispanic question and the inclusion of a multiracial category. In other words, both of the factors that were experimentally manipulated across the four panels affected the overall race distribution. Separate models for each race were run as follow-up tests to determine the effect of the panels on the percentage of people identifying with each racial group. There were no significant differences across panels in the percentage of people identifying as Black or Asian or Pacific Islander. The percentage of people identifying as White was influenced only by whether there was a separate Hispanic question or not, with 75.22 percent of the respondents identifying as White when Hispanic was included in the list of races (panels 3 and 4 combined) compared to 79.81 percent who identified as White when Hispanic origin was a separate question (panels 1 and 2 combined). The percentage of people identifying themselves as American Indians was influenced only by the presence of a multiracial category, with 0.76 percent of the respondents identifying as American Indian when there was a multiracial category offered in the list of races (panels 2 and 4 combined) compared to 1.02 percent identifying as American Indian when there was no multiracial category (panels 1 and 3 combined). These analyses show that including Hispanic as a category in the race question will likely lower the proportion of people currently identifying as White, and including a multiracial category will likely lower the percentage of people identifying as American Indian or Alaska native.

V.2.4 Racial Distributions based on Additional Opportunities to Identify as Multiracial

Respondents in panels 1 and 3 were not given a multiracial category in the list of races, but were asked an additional question in the Supplement on whether they would have liked to have had a multiracial category to identify themselves. Respondents who wanted a multiracial category for themselves were then asked what races they considered themselves to be in addition to the first race they had indicated. Respondents in panels 3 and 4 who did not select the multiracial category were asked if they wanted to list any additional races they wanted to better describe themselves. Table 8 shows the racial distribution after these opportunities to identify as multiracial were given to respondents. While it is unlikely that these additional questions would be included in most surveys or censuses, there was a similar pattern of findings to the initial race variables described above, with the simplest model that provided an adequate fit to the data including both the effects of the separate Hispanic question and the inclusion of a multiracial category. Separate models for each race were run as follow-up tests to determine the effect of the panels on percentage of people identifying as a member of each racial group. Again, there were no significant differences across panels in the percentage of people identifying as Black or Asian or Pacific Islander. The percentage of people identifying themselves as American Indians was influenced equally by either the presence or absence of a multiracial category or the presence or absence of a separate Hispanic question, with either model explaining the data equally well. Specifically, 0.63 percent of the respondents identified as American Indian when there was a multiracial category offered in the list of races (panels 2 and 4 combined) and 0.78 percent of the respondents identifying as American Indian when no multiracial category was offered (panels 1 and 3 combined). Similarly, 0.63 percent of the respondents identified as American Indian when there was a separate Hispanic question (panels 1 and 2 combined) and 0.77 percent of the respondents identified as American Indian when Hispanic was included in the list of races (panels 3 and 4 combined). The percentage of people identifying as White was influenced by both the presence of a separate Hispanic question and a multiracial category. Specifically, 73.48 percent of the respondents identified themselves as White when Hispanic was included in the list of races (panels 3 and 4 combined) compared to 76.52 percent of the respondents identifying as White when Hispanic was a separate question. In addition, 74.41 percent of the respondents identified as White when there was a multiracial category (panels 2 and 4 combined) compared to 75.57 percent of the respondents identifying as White when there was no multiracial category (panels 1 and 3 combined).

V.3 Preference for Including Hispanic as a Racial Category

Hispanics were asked whether they preferred to identify themselves from a list of races that included Hispanic or through a separate question asking whether they are Hispanic. This question was asked differently for persons who had Hispanic included as a racial category (panels 3 and 4) than for those who were asked a separate question (panels 1 and 2). Respondents who were given Hispanic in the list of races were asked if they wanted a separate question on Hispanic origin, and respondents who were asked a separate question were asked if they preferred that Hispanic be included in the list of races. For comparison purposes, respondents' answers were coded to reflect the percentage of Hispanics from each panel who wanted to have Hispanic listed as a racial category. As can be seen in Table 9, well over half of Hispanics (68.56 percent overall) preferred that Hispanic be included in the list of races. This is consistent with the work of Kamasaki, 1994; and Kissam, et al, 1992, who also reported a preference for an Hispanic racial category among Hispanic respondents.

The percentage of people preferring that Hispanic be included as a racial category differed across the panels. Specifically, 73.71 percent of the respondents wanted Hispanic included with the list of races when they received the question that way (panels 3 and 4 combined) compared to 62.20 percent of the respondents preferring that Hispanic be included with the list of races when they received a separate Hispanic question (panels 1 and 2 combined).

V.4 Preferred Racial and Ethnic Terms

People who identified themselves as Hispanic, White, Black, American Indian, or multiracial were given a list of terms describing their respective racial or ethnic group and were asked to choose which term they preferred, or whether they preferred a term not stated, or had no preference. Although these questions were identical across the four panels of the CPS Supplement, statistical tests were conducted to examine differences in the distributions of responses across the different panels because it is possible that the earlier questions or categories that respondents were offered may have influenced their preferences. However, there were no statistically significant panel effects in any case. There appeared to be some variation in preferences for terms for American Indians and multiracials, but these differences were not statistically significant and were based on a very small number of cases. It should be noted that all of the choices of preferred terms could have been influenced by the terms given in the race questions that were asked earlier.

The percentage of persons of each race (collapsed across all four panels) preferring each term can be seen in Table 10. Persons identifying as Hispanic were asked which of the following terms they preferred to describe themselves: Hispanic, Latino, Of Spanish Origin, Some other term, or No preference. The majority chose the term Hispanic as the one they preferred. Persons identifying as White were asked which of the following terms they preferred to describe themselves: White, Caucasian, European-American, Anglo, Some other term, or No preference. A majority chose the term White, but a sizable number had no preference. Persons identifying as Black were asked which of the following terms they preferred to describe themselves: Black, African-American, Afro-American, Negro, Colored, Some other term, or No preference. A large plurality chose the term Black, but about as many chose African-American or Afro-American (combined). In a 1993 survey of Black respondents in the Chicago area, Lavrakas, Schejbal, and Smith (1994) found that Black was the preferred racial term for 59.1 percent, compared to 37.4 percent who preferred African American. They noted a rapid shift in preference for these terms from their 1991 Chicago survey results, when 78.0 percent preferred "Black," compared to 22.0 percent who preferred African American.

Persons identifying as American Indians were asked which of the following terms they preferred to describe themselves: American Indian, Alaska Native, Native American, Some other term, or No preference. Over 50 percent chose the term American Indian or Alaska Native as the one they preferred, but a sizable number preferred Native American. Persons identifying as "multiracial" were asked which of the following terms they preferred to describe themselves: Multiracial, More than one race, Biracial, Mixed-race, Mestizo/Mestiza, Some other term, or No preference. A plurality chose the term "multiracial" as the term they preferred, but about the same number had no preference. The large number of respondents selecting "no preference" may have been a result of the sizable number of "indeterminate multiracials" mentioned earlier.

V.5 Hispanic Nationality From the Supplement

All respondents who indicated they were Hispanic on the Supplement were asked for their national origin. The breakdown of Hispanics by national origin for each panel in the Supplement is shown in Table 11. Analyses were conducted to test whether there were differences across panels in the distributions of the national origins of the Hispanic respondents. There were significant overall effects for a separate Hispanic question. Follow-up analyses were also conducted for each nationality separately to determine which groups demonstrated significant differences across panels. As can be seen in Table 11, the proportions of Mexican, Cuban, and Other Hispanic national origins were affected by having a separate Hispanic question. Specifically, the respondents who identify as Hispanic from the list of races are composed of a greater percentage of people with Mexican national origin (66 percent in panels 3 and 4 combined) than the respondents who identify as Hispanic in a separate question (about 60 percent in panels 1 and 2 combined). In contrast, the respondents who identify as Hispanic in a separate question are composed of a greater percentage of people with Cuban and Other Hispanic national origins (about 4 percent Cuban and 13 percent Other Hispanic in panels 1 and 2 combined) than the respondents who identify as Hispanic from the list of races (about 2 percent Cuban and 9 percent Other Hispanic in panels 3 and 4 combined). In other words, Hispanics of different national origins differ in terms of how likely they are to identify themselves as Hispanic depending upon whether they are asked a separate Hispanic question or Hispanic is a choice in a list of races. Some groups, such as Cubans and "Other Hispanic" are less likely to be included as Hispanics when Hispanic is included in the list of races.

V.6 Reasons for Identifying as Multiracial

Respondents who chose the multiracial category when it was on the list of races in panels 2 and 4 were asked three follow-up questions concerning their reasons for choosing the multiracial category. They could select any or all of the following reasons for identifying as multiracial: (1) their parents were from different racial groups; (2) their ancestors before their parents were from different racial groups; and (3) the specific group they belong to is mixed. The percentage of the multiracial respondents selecting each of these reasons is shown in Table 12 for panels 2 and 4 separately. Having parents from different racial groups was the reason chosen by the most respondents in both panels, but each of the reasons was selected by more than half of the persons identifying as multiracial. Analyses were conducted to test whether there were any differences between these two panels in the reasons that multiracial respondents gave for identifying themselves as multiracial. Although there appear to be some differences, they are not statistically significant.

As noted earlier, a number of respondents who selected the multiracial category in both panels were classified as "indeterminate multiracials" because they did not explicitly identify more than one race. These indeterminate multiracials may have different reasons for choosing the multiracial category than the people classified as clearly multiracial. Analyses were conducted distinguishing between the multiracials and indeterminate multiracials in order to test for differences between them, including differences in panel effects. The results of these analyses can be seen in Table 13. Although some of the groups may appear to differ, none of the differences were statistically significant.

V.7 Ancestry and Racial Identification in the Supplement

All respondents were asked to give "ancestry or ethnic origin" in a free-response format. These responses were coded by the Census Bureau and were collapsed into 6 major categories: (1) European; (2) Spanish speaking; (3) Sub-Saharan or unspecified Black; (4) Asian and Pacific Islander; (5) American Indian and Alaska Native; and (6) All Other. As can be seen in Table 14, 97 percent of the respondents with European ancestry across all four panels identified as White. Across all four panels, over 95 percent of respondents with sub-Saharan or unspecified Black ancestry identified Black as their racial group, and 68-74 percent of respondents with Asian or Pacific Islander ancestries identified with the Asian or Pacific Islander racial group. In contrast, only 23 percent to 31 percent of respondents who gave an American Indian or Alaska Native ancestry identified themselves with that racial group, but the majority (54 percent to 60 percent) identified as White. This is consistent with Snipp's finding that 11 percent of persons reporting their race as American Indian in the 1980 decennial census indicated a European group as their first or only ethnic ancestry, while 6 percent reported Hispanic, African, or other non-American Indian ethnicity as their first ancestry (Snipp, 1986).

About 63 percent of respondents who gave ancestries of Spanish-speaking countries identified as White in panels 1 and 2, but only about 17 percent identified as White in panels 3 and 4. When "Hispanic" was an option in the list of races in panels 3 and 4, 78 percent of the respondents who had Spanish-speaking ancestries identified as Hispanic. The vast majority of respondents who gave ancestries that were classified as "All Other" identified as White.

An examination of the effect of the Supplement panel on respondents' racial identification and ancestry indicated significant overall panel effects for the relation between ancestry and racial identification, so these analyses were followed up by separate tests for each race. As can be seen in the results table following Table 14, there were significant panel effects involving Ancestry and racial identifications of White and Asian or Pacific Islander. Additional contrasts were conducted to examine the specific effects for each of these groups. There were no significant panel effects on the relation between ancestry and racial identification for Blacks or American Indians.

The panel effects for Whites were due to the effect of a separate Hispanic question on the how respondents with Spanish speaking ancestries identified. Specifically, 63 percent of respondents of Spanish ancestry identified as White when there was a separate Hispanic question (panels 1 and 2), and only 17 percent identified as White when Hispanic was included on the list of races (panels 3 and 4). In contrast, the percentage of respondents with other ancestries who identified as White was the same across panels.

The panel effects on Asian or Pacific Islander ancestries showed an extremely small effect for having a multiracial category that was statistically significant, but may not be meaningful. Respondents reporting an ancestry other than Asian or Pacific Islander were slightly more likely to identify as Asian or Pacific Islander when there was no multiracial category (panels 1 and 3) than when there was a multiracial category (panels 2 and 4).

VI. ANALYSIS OF SUPPLEMENT RACE AND HISPANIC ORIGIN BY CPS RACE AND HISPANIC ORIGIN BY PANEL.

The purpose of the analyses in this section was to examine how the different questions about race and ethnicity that were asked in different panels of the Supplement affected: (1) the agreement between identification of Hispanic origin in the CPS and in the Supplement; (2) the agreement between racial identification in the CPS and the Supplement; and (3) how respondents with different Hispanic nationalities (as determined from the CPS) identified their race in the Supplement. These analyses allow one to see the possible effects of changing the race and ethnicity questions from their present form on the CPS to the different versions on the four panels of the Race and Ethnicity Supplement.

VI.1 Hispanic Origin from the Supplement and the CPS

One might expect some small level of disagreement between the CPS classifications of Hispanic origin and those from the Supplement because the questions are different. Hispanic origin is determined on the CPS by a general question that asks a respondent about his or her origin or descent (see Appendix A). Respondents who indicate that their origin was Mexican American, Chicano, Mexican, Puerto Rican, Cuban, Central or South American, or Other Spanish are classified as Hispanic. In contrast, Hispanic origin in panels 1 and 2 is determined by asking respondents if they are Hispanic, Latino, or of Spanish origin in a separate question. Hispanic origin in panels 3 and 4 is determined by respondents selecting Hispanic from the list of races.

The agreement between persons identifying as Hispanic on the Supplement and persons classified as Hispanic on the CPS are shown in Table 15. Although a complex model that included both the effects of having a separate Hispanic question and a multiracial category and their interaction were required to adequately explain this pattern of results, the use of a separate Hispanic question was by far the most important factor. Specifically, 94.26 percent of the people who were classified as Hispanic on the CPS identified themselves as Hispanic on the Supplement in a separate Hispanic question (panels 1 and 2 combined). However, only 80.23 percent of the people who were classified as Hispanic on the CPS also identified themselves as Hispanic from the list of races (panels 3 and 4 combined). Clearly, measuring Hispanic origin using a category in the list of races misses a relatively large percentage of people who are currently classified as Hispanic on the CPS. There was much greater agreement between the CPS classification of Hispanic origin and the Supplement identification of Hispanic origin when a separate Hispanic question was included.

VI.2 Racial Distributions from the Supplement and the CPS

Some small level of disagreement between racial identification on the CPS and identification in the Supplement may also result from the fact the questions differ to varying degrees. The race question from the CPS simply asks a respondent what their race is and gives the following response options: White; Black; American Indian, Aleut, Eskimo; Asian or Pacific Islander; and Other-Specify (see Appendix A). The first race question from panel 1 of the Supplement most closely resembles the CPS race question and asks respondents "Which one of the following list are you: White; Black; American Indian, Eskimo, or Aleut; Asian or Pacific Islander; or Something else." The race questions in the other panels included additional response categories such as Hispanic (in panels 3 & 4) and a multiracial category (in panels 2 & 4) and can be found in Appendix A.

The agreement between the CPS race question and the Supplement race variable with the complete breakdown of multiracials in panels 2 and 4 is shown in Table 16. There is generally high agreement (over 85 percent) between CPS race and the Supplement race across all 4 panels for persons identifying as White, Black, and Asian or Pacific Islander. The percentage of respondents classified as American Indian on the CPS who also identified that way on the supplement ranged from 58.94 percent to 74.50 percent. Persons who identified as "Other" on the CPS are most likely to identify as "Something else" on the supplement in panels 1 and 2 (64.52 percent and 54.97 percent, respectively); however, they are most likely to identify as Hispanic in panels 3 and 4 (72.03 percent and 75.36 percent, respectively).

A closer examination of Table 16 also reveals how respondents changed their identification from the CPS to the Supplement race question(s) depending on the panel they were in. Persons who identified as White on the CPS but do NOT identify as White on the Supplement are most likely to identify as 'Something else' in panels 1 and 2, and are most likely to identify as Hispanic in panels 3 and 4. Persons who identified as Black on the CPS but do NOT identify as Black on the Supplement are most likely to identify as "Something else" in panels 1 and 3, and are most likely to identify as multiracial in panels 2 and 4. Respondents who said they were American Indian on the CPS but who did NOT identify as American Indian in the Supplement were most likely to identify as White across all 4 panels. Respondents who identified as Asian or Pacific Islander on the CPS but who did not identify themselves that way in the Supplement were most likely to identify as "Something else" across all 4 panels.

Table 17 also shows the agreement between racial identification on the CPS and on the Supplement with Hispanic and multiracial categories by panel, but without the further breakdown of multiracials. An overall analysis was done to test if there were significant differences by panel for agreement between the CPS race and the Supplement race variable. A complex model with the effects of the separate Hispanic question and the multiracial category interacting was required to explain adequately the pattern of results shown in Table 17.

A series of follow-up statistical tests were conducted to contrast the percentage of persons choosing each race on both the CPS and the Supplement across panels. There were no significant differences across panels for Blacks, American Indians, and Asian or Pacific Islanders, but there were significant differences across panels for Whites. In this later case the agreement on the response to the race question on the CPS and the Supplement was influenced by having a separate Hispanic question. Over 95 percent (panels 1 and 2 combined) of the people who identified as White on the CPS also identified themselves as White in the Supplement when there was a separate Hispanic question. This agreement drops to about 91 percent (panels 3 and 4 combined) when there was no separate Hispanic question. In other words, the presence of an Hispanic category in the list of races is drawing respondents who were identifying as White on the CPS.

An additional analysis was conducted across panels looking at respondents who identified as "Other" on the CPS and who did not identify with one of the racial categories (i.e., they identified as "Something else" or did not know or refused to answer the race question) and were classified as "All Other" on the Supplement. There were significant differences across panels in agreement between CPS and the Supplement for the "Other" and "All Other" categories that depended on the presence of a separate Hispanic origin question. Specifically, when there was a separate Hispanic origin question, about 60 percent of the respondents (panels 1 and 2 combined) who chose "Other" on the CPS were also classified as "All Other" on the Supplement; however, when Hispanic was included in the list of races only about 11 percent (panels 3 and 4 combined) were classified as "All Other" on the supplement. Thus, changing the race question from the current CPS version by adding an Hispanic category appears likely to lower the proportion of Whites and persons who are currently classified as "Other."

VI.3 Racial Identification in the Supplement and Hispanic Nationality from the CPS

The racial identification of Hispanics with different national origins was also examined by utilizing the measure of Hispanic national origin from the CPS and racial identification in the Supplement. There were no panel differences on the CPS measure of Hispanic national origin itself. However, as can be seen in Table 18, Hispanics of different national origins identify themselves differently in terms of race depending upon the panel of the Supplement they were in. Analyses were focused on the two most common racial identifications made by Hispanics: "White" and "All Other", and follow-up analyses were conducted to examine whether there were panel effects on how respondents with each different Hispanic national origin identified. The results of these analyses can be seen at the bottom of Table 18. Hispanics with a Mexican national origin were much more likely to identify as White when there was a separate Hispanic question (63 percent for panels 1 and 2 combined) than when Hispanic was available as a racial category (11 percent in panels 3 and 4 combined). Similarly, Hispanics with other Spanish national origins were more likely to identify as White when there was a separate Hispanic question (54 percent for panels 1 and 2 combined) than when Hispanic was available as a racial category (about 23 percent in panels 3 and 4 combined). A more complex model was required to explain the pattern of findings for Hispanics with Central or South American national origins identifying as White, but the effect of the separate Hispanic question was the largest effect with about 52 percent of Hispanics with Central or South American Origin identifying as White when there was a separate Hispanic question (panels 1 and 2 combined), and only about 16 percent identifying as White when Hispanic was available as a racial category (panels 3 and 4 combined). Although there were no statistically significant panel effects for Hispanics of Cuban origin, they exhibited a pattern of identification that appeared somewhat different from other Hispanics in panels 3 and 4. A majority of Hispanics of Cuban origin in panel 3 and nearly 50 percent in panel 4 identified as White even though the Hispanic category was offered. When no Hispanic category was offered in panels 1 and 2, over 90 percent of Hispanics with Cuban Origin identified as White. This pattern of racial identification for Mexican origin and Cuban origin Hispanic respondents is consistent with the findings of the 1990 Panel Study of Income Dynamics conducted by the Institute for Survey Research at the University of Michigan. For Hispanics reporting a single race when given a list of racial categories that included "Latino," 88 percent of Cubans reported as White and 9 percent as Latino, compared to Mexicans, 56 percent of whom reported as White and 35 percent of whom reported as Latino (Duncan et. al, 1992). Bates et al. (in press) found that Cubans, compared to other Hispanic groups, were most likely to report their race as White when the race question followed a question on Hispanic origin.

The only significant panel effect for Hispanics who were classified in the "All Other" category was found for those with Central or South American origins. A complex model was required to explain the pattern of findings, but the effect of the separate Hispanic question was again the largest effect with about 38 percent of Hispanics with Central or South American Origin falling in the "all other" category when there was a separate Hispanic question (panels 1 and 2 combined), and only about 3 percent being "All Other" when Hispanic was available as a racial category (panels 3 and 4 combined). These analyses provide further evidence that certain subgroups of Hispanics are more or less likely to change their racial identification depending on the way the race and ethnic origin questions are asked.

VII. ANALYSIS OF SUPPLEMENT RACE AND HISPANIC ORIGIN BY CPS DEMOGRAPHIC CHARACTERISTICS BY PANEL.

The purpose of the analyses in this section is to examine the effects of demographic characteristics of respondents on the differences found across panels in the racial and Hispanic origin distributions and also on the changes in racial and Hispanic origin identification from the CPS to the Supplement panels. Thus, the focus of these analyses is not on the association among race or Hispanic origin and a set of demographic characteristics (these associations are well documented elsewhere), but more specifically on how the different demographic groups responded to the questions in the panels of the Supplement. In other words, how do the people who identified as Hispanic or White compare in one panel to the people who identified the same in another panel.

We chose a variety of demographic characteristics available from the CPS that reflected the characteristics of the area in which the household was located, of the household as a whole, and of particular household members. Specifically, the area characteristics included region of the country, population of the area (Census place size), whether the area was urban or rural, whether the household was located within a Metropolitan Statistical Area (MSA) or not, and the degree of poverty in the area. The characteristics of the household itself included family income, household size, and whether the household members owned or rented the housing unit. The characteristics of the household members included age, education, labor force status (employed/unemployed/not in the labor force), U. S. citizenship, length of time in the United States, nativity of the respondent, nativity of the respondent's mother and father, and the respondent's relationship to the reference person. Because of the complex nature of these analyses and space limitations, these findings will be discussed only briefly.

VII.1 Hispanic Origin and Demographic Characteristics

A series of analyses were conducted to see whether the seventeen demographic characteristics mentioned above influenced the differences found in Hispanic origin by panel. In these analyses, the associations among demographic characteristics and responses to the different ethnic origin questions are not necessarily causal. The racial and ethnic groups affected by the treatments have certain demographic characteristics. When some of these respondents change their racial or ethnic identification in response to the treatment conditions, it follows that the demographic makeup of the resulting racial and ethnic classes change. A summary of the results of these analyses are shown in Table 19. The only demographic variables that showed a relation with Hispanic origin that differed across panels were respondent's nativity, mother's nativity and father's nativity.

The results for respondent's nativity can be seen in Table 20. The percentage of respondents who were born in a Spanish speaking country and who identify as Non-Hispanic is slightly smaller when there is a separate Hispanic question (panels 1 and 2) than when a Hispanic category is included as part of the race question (panels 3 and 4). The pattern of results for mother's nativity and father's nativity are quite similar as can be seen in Tables 21 and 22. Hispanics in panels 1 and 2 were more likely to have parents born in non-Spanish-speaking countries (i.e., within the U.S. and outside the U.S.) than Hispanics in panels 3 and 4. Non-Hispanics in panels 1 and 2 were less likely to have parents born in Spanish-speaking countries than Non-Hispanics in panels 3 and 4. Thus, the combined race and ethnicity question results in a higher percentage of persons born in Spanish-speaking countries and a higher percentage of persons with parents born in Spanish-speaking countries not identifying as Hispanic.

VII.2 Race and Demographic Characteristics

A similar series of analyses to those described above were conducted to see whether the association among the seventeen demographic characteristics mentioned above were related to the panel differences for racial identification. A summary of the results of these analyses are also shown in Table 19. Six demographic characteristics appeared to have differing relations with racial identification across panels. Specifically, the association among racial identification and the extent of poverty in the area, U. S. citizenship, years in the United States, respondent's nativity, mother's nativity, and father's nativity, showed differences by panel. All of these panel differences were due to the effect of including Hispanic as a race or not and appear to be largely due to the Hispanics who identified as White in panels 1 and 2, but identified as Hispanic in panels 3 and 4. Again, the associations among demographic characteristics and response to the race questions are not necessarily causal. The demographic changes in the make-up of the 'White' racial category (which was most affected by the treatment conditions or panels) are discussed here.

The distribution of respondents living in poverty areas for each race by panel can be seen in Table 23. There were significantly more Whites living in poverty areas in panels 1 and 2, where many Hispanics identified as White, than in panels 3 and 4, where Hispanic was included in the list of races. The distribution of U. S. citizenship for each race by panel is shown in Table 24. There were significantly more Whites who were naturalized or not citizens of the U. S. in panels 1 and 2 than in panels 3 and 4. For persons who came to the U. S., the distribution of the year they came to the U. S. for each race by panel is displayed in Table 25. The percentage of Whites who came to the U. S. is significantly higher during both time periods in panels 1 and 2 compared to panels 3 and 4. The distribution of respondent's nativity for each race by panel is shown in Table 26. There were significantly more Whites with Spanish nativity in panels 1 and 2, where Hispanics were identified with a separate question, than in panels 3 and 4, where Hispanic was included in the list of races. The pattern of results was very similar for respondent's mother's nativity and father's nativity as can be seen in Tables 27 and 28.

VII.3 Hispanic Origin and Demographic Characteristics for Hispanics from the CPS

A series of analyses were conducted separately for respondents who identified as Hispanic in the first CPS interview to examine whether their demographic characteristics affected the relation between the CPS and the Supplement ethnicity measure across panels. These analyses allow one to see which demographic characteristics are associated with the respondent's changing their Hispanic origin from the CPS to the Supplement by panel. A summary of the results of these analyses are shown in Table 29. Six demographic characteristics appeared to have some effects. They are the region of the country, the population of the place, household size, labor force status, mother's nativity and father's nativity. Three of these panel differences had to do with the separate Hispanic origin question, and the others involved the multiracial category also and a more complex combination of these two factors. In the case of these of analyses, the relations of demographic characteristics with different responses to the different ethnic origin questions seem more likely to be causal. Thus, actual changes in the consistency of response by demographic groups are presented and discussed. Again, because of the complex nature of these analyses and space limitations, only the largest and most relevant findings will be discussed in this and following sections.

The distribution of region for Hispanics by panel can be seen in Table 30. Hispanics in the Midwest and West were more likely to identify the same on the CPS and the Supplement than Hispanics in the Northeast and South in panels 3 and 4; however, there were no regional differences in ethnic identification for panels 1 and 2. The distribution of the Census place size for Hispanics by panel can be seen in Table 31. In panel 3, Hispanics who were from more populated areas were more likely to identify the same on the CPS and the Supplement than Hispanics from the smallest areas (less than 5,000). In panel 1, the opposite pattern occurred; Hispanics from the least populated areas were more likely to identify the same in the CPS and the Supplement than Hispanics from the most populated areas. The distribution of household size for Hispanics by panel can be seen in Table 32. In panels 3 and 4, persons in larger households were more likely to identify consistently as Hispanic on both the CPS and the Supplement than persons in smaller households. There were uniformly high levels of agreement across household sizes in panels 1 and 2. The distribution of labor force status for Hispanics can be seen in Table 33, and the pattern is complex and difficult to interpret meaningfully. The pattern for the respondent's mother's and father's nativity were similar as can be seen in Tables 34 and 35. Across all panels, Hispanics who had parents born in non-Spanish speaking countries outside of the U. S. had lower agreement in their identification between the CPS and the Supplement than Hispanics whose parents were born in the U.S. or a Spanish speaking country. However, the differences in agreement were much greater in panels 3 and 4 than they were in panels 1 and 2.

VII.4 Race and Demographic Characteristics for Whites from the CPS

A similar series of analyses were conducted separately for respondents who identified as White in the first CPS interview to examine whether the relation among their racial identification in the Supplement and their demographic characteristics differed by panel. A summary of the results of these analyses are also shown in Table 29. Seven demographic characteristics appeared to have differing relations with racial identification across panels. Specifically, the association among racial identification and poverty area, owner/renter status, education , U. S. citizenship, respondent's nativity, their mother's nativity, and their father's nativity showed differences by panel. Five of these panel differences involved the effect of a separate Hispanic origin question, and the others had to do with both the effect of a separate Hispanic question and the effect of a multiracial category or a more complex combination of these two factors. Most of these effects are explained by the shifting of Hispanics who identified as White in panels 1 and 2, but identified as Hispanic in panels 3 and 4.

The distribution of poverty by race for each panel for persons identifying as White in the first CPS interview is shown in Table 36. In panels 3 and 4, Whites who lived in poverty areas were more likely to change their identification from the CPS to the Supplement than Whites in non-poverty areas, and they were more likely to change their identification than all Whites in panels 1 and 2, regardless of where they lived. The distribution of household owners/renters by race for each panel is shown in Table 37. In panels 3 and 4, Whites who rented their homes were more likely to identify differently on the CPS and Supplement than Whites who were home owners, and they were more likely to identify differently on the CPS and Supplement than all Whites in panels 1 and 2, regardless of their owner/renter status. The distribution of education by race for each panel is shown in Table 38. Whites with less education showed less agreement in their identification between the CPS and the Supplement across all panels, but there were much larger differences on agreement between Whites of different educational levels in panels 3 and 4 than in panels 1 and 2. The distribution of U. S. citizenship by race for each panel is shown in Table 39, and the largest differences appear for the effect of a separate Hispanic question. Across all panels, Whites who were native citizens showed higher agreement in their identification between the CPS and Supplement than Whites who were naturalized citizens, who showed higher agreement in their identification than Whites who were not citizens. However, the differences between these groups for panels 3 and 4 were much greater than the differences exhibited between these groups in panels 1 and 2. The distribution of respondent's nativity by race for each panel is shown in Table 40. Whites born in a Spanish-speaking country were much more likely to identify the same on the CPS and the Supplement in panels 1 and 2 than in panels 3 and 4. Whites born in the U. S. were also more likely to identify the same on the CPS and the Supplement in panels 1 and 2 than in panels 3 and 4. The distribution of mother's and father's nativity by race for each panel are shown in Tables 41 and 42. The largest differences appear for the effect of a separate Hispanic question on Spanish nativity, with similar patterns to that shown for respondent's nativity. Whites with parents born in Spanish-speaking countries showed much lower levels of agreement in their identification on the CPS and the Supplement in panels 3 and 4 than in panels 1 and 2. Again, Whites born in the U. S. showed a similar pattern, and were more likely to identify the same in the CPS and the Supplement in panels 1 and 2 than in panels 3 and 4.

VII.5 Race and Demographic Characteristics for Other Races from the CPS

A series of analyses were also conducted separately for respondents who identified as Black in the first CPS interview to examine whether their demographic characteristics affected the relation between the CPS and Supplement racial measures across panels. There were no demographic characteristics that appeared to have differing relations with racial identification across panels. These analyses were not conducted on the American Indian or Asian or Pacific Islander racial groups due to their small sample sizes.

VIII. ANALYSIS OF METHODOLOGICAL EFFECTS IN THE SUPPLEMENT

Several factors related to the data collection procedures of the CPS could potentially influence responses to the Supplement questions. The CPS is an interviewer-administered survey, and the results from this Supplement may differ from a survey which asked the same questions in a self-administered questionnaire. Interviews with households that are selected into the sample are attempted for a total of eight months. Households are interviewed for four consecutive months one year, are not interviewed for eight months, and then are interviewed for the same four consecutive months the following year. Each month's CPS sample is composed of eight rotation groups which reflect the number of CPS interviews that the household has given. One-eighth of the sample is being interviewed for the first time, 1/8 the second time, etc. The first interview and the fifth interview (after eight months of no interviews) are done in person by an interviewer from the Census Bureau. The second to fourth and sixth to eighth interviews are typically done by telephone. The majority of telephone interviews are done by the same interviewers who originally visited the household, but telephone interviews are also completed by interviewers working in one of three Census Bureau centralized telephone facilities. Finally, although interviewers were encouraged to ask the Supplement questions of all of the adults (over 15 years old) in the household, information could be obtained by proxy reports of any adult living in the household. Typically, about 50 percent of the labor force information in the CPS is obtained from proxy reports. For the Race and Ethnicity Supplement, about 55 percent of the data came from proxy reports.

The purpose of the analyses in this section was to examine the effects on racial and Hispanic origin distributions by panel for four methodological factors associated with CPS data collection. The factors examined were: (1) the length of time the respondent has been in the sample, (2) whether the interview was a personal visit or telephone interview, (3) whether the interview was done by a field representative or an interviewer calling from a centralized telephone facility (for telephone interviews only), and (4) whether the supplement data was obtained from the person directly or from a proxy.

Analyses were conducted to determine if any of these methodological factors influenced the racial or Hispanic origin distributions differently depending upon the panel. There were no significant effects on either the Hispanic origin or racial distribution of the Supplement by panel for the length of time the respondent had been in the sample and whether the telephone interview had been done at a centralized telephone facility or by a field interviewer. However, it should be noted that the cases interviewed by telephone at the centralized facility were not a random subset of cases. There were significant effects for the mode of the interview (personal visit versus telephone) on the racial distribution that can be seen in Table 43. There was a tendency for fewer respondents to identify as American Indian in telephone interviews when there was a multiracial category (panels 2 and 4) than when there was no multiracial category (panels 1 and 3). There were no significant effects on the Hispanic origin distribution for the mode of the interview. Significant effects on the Hispanic origin distribution was shown for whether the data was obtained from a self-report or a proxy. As can be seen in Table 44, the pattern of results for Hispanic origin reported by self and proxy is difficult to interpret meaningfully. Fewer Hispanics were identified by proxy reporters in panels 3 and 4 than in panels 1 and 2; however even fewer Hispanics in panel 3 were identified by proxy reports than in panel 4. It is important to note that the conclusions that can be drawn from the self-proxy differences are minimal because self and proxy reporting status was confounded by self-selection, and was not randomly determined. There were no significant differences in the Supplement racial distribution by self and proxy reporting.

IX. REFERENCES

Bates, Nancy A.; de la Puente, Manuel, DeMaio, Theresa J., and Martin, Elizabeth A., "Research on Race and Ethnicity: Results from Questionnaire Design Tests," Proceedings of the 1994 Annual Research Conference, Bureau of the Census, 1994, Rosslyn, pp. 435-445.

Bates, Nancy; Martin, Elizabeth A., DeMaio, Theresa J., and de la Puente, Manuel. Questionnaire Effects on Measurements of Race and Spanish Origin. In press, Journal of Official Statistics.

del Pinal, Jorge H. and Lapham, Susan J., "Impact of Ethnic Data Needs in the United States, Challenges of Measuring an Ethnic World: Science, Politics, and Reality," Proceedings of the Joint Canada-United States Conference on Measurement of Ethnicity, April 1-3, 1992, Ottawa, Canada. Washington, DC: U. S. Government Printing Office, pp. 447-475.

Duncan, Greg J., Hill, Martha S., Lepkowski, James, Garza, Rodolfo de la Garza, Falcon, Angelo, Garcia, Chris, and Garcia, John. Panel Study of Income Dynamics, 1968-1990 [Latino Sample, 1990]. [Computer file]. Ann Arbor, MI: Survey Research Center [producer], 1992. Ann Arbor, MI: Consortium for Political and Social Research [distributor], 1992.

Elias-Olivares, Lucia and Farr, Marcia, Sociolinguistic Analysis of Mexican-American Patterns of Non-Response to Census Questionnaires, Report submitted to the Census Bureau under Joint Statistical Agreement 88-25, Ethnographic Exploratory Research Report #16, 1991.

Evinger, Suzanne, "How Shall We Measure Our Nation's Diversity?" Chance, vol.8, no. 1, 1995, pp. 7-18.

Fay, Robert E. CPLX Contingency Table Analysis for Complex Sample Designs Program Documentation. U. S. Bureau of the Census, Washington, DC, 1989.

Hahn, Robert A. and Stroup, Donald F. "Race and Ethnicity in Public Health Surveillance: Criteria for the Scientific Use of Social Categories," Public Health Reports, vol.109, no. 1, 1994, pp. 7-15.

Harrison, Roderick J. and Bennett, Claudette, "Racial and Ethnic Diversity," in Farley, Reynolds (ed.) The1990 Census Research Project, Changes and Challenges: America 1990, Volume II, Russell Sage Foundation, New York, forthcoming.

Kamasaki, Charles K., "An Hispanic Assessment of the Federal Standards for Race and Ethnicity," presentation to the Workshop on Race and Ethnicity Classification, Committee on National Statistics, National Academy of Sciences, February 17, 1994, Washington.

Kissam, Edward, Herrera, Enrique, and Nakamoto, Jorge M., Hispanic Response to Census Enumeration Forms and Procedures, Report submitted to the Census Bureau under Contract No. 50-YABC-2-66027, Task Order No. 46-YABC-2-0001, March 1993.

Lavrakas, Paul J., Schejbal, Judith A., and Smith, Tom W., "The Use And Perception of Ethno-Racial Labels: African American" And/Or "Black," Proceedings of the 1994 Annual Research Conference, Bureau of the Census, 1994, Rosslyn, pp. 505-515.

Marks, Jonathan, "Black, White, Other," Natural History, December, 1994, pp.32-35.

Martin, Elizabeth; DeMaio, Theresa A.; and Campanelli, Pamela A., "Context Effects of Race and Hispanic Origin," Public Opinion Quarterly, vol. 54, no. 4, pp.551-566, 1990

McKay, Ruth B. and de la Puente, Manuel, "Cognitive Research in Designing the CPS Supplement on Race and Ethnicity," Proceedings of the 1995 Annual Research Conference, Bureau of the Census, 1995, Rosslyn, pp.435-445.

McKenney, Nampeo; Bennett, Claudette; Harrison, Roderick; and del Pinal, Jorge, "Evaluating Racial And Ethnic Reporting In The 1990 Census," Proceedings of the American Statistical Association, Section on Survey Methods Research, 1993, pp. 66-74.

Passel, Jeffrey S. and Berman, Patricia A., "Quality of 1980 Census Data for American Indians," Social Biology, 33, 1986, pp. 163-182.

Rodriguez, Clara E., "Challenges and Emerging Issues: Race and Ethnic Identity Among Latinos, " Committee on National Statistics, National Research Council, February 17, 1994, Washington.

Royston, Patricia, Bercini, Deborah, Sirken, Monroe, and Mingay, David, "Questionnaire design research laboratory." Proceedings of the Section on Survey Research Methods. Washington, DC: American Statistical Association, 1986, pp. 707-713.

Smith, Tom W., "Changing Racial Labels: From 'Colored' to 'Negro' to 'Black' to 'African American,'" Public Opinion Quarterly, vol. 56, no. 4, pp.496-514, 1992.

Snipp, Matthew C., "Who Are American Indians? Some Observations About the Perils and Pitfalls of Data for Race and Ethnicity." Population Research and Policy Review, 1986, 5:237-252.

TABLES

APPENDIX A. RACE AND ETHNICITY SUPPLEMENT QUESTIONS AND RACE AND ETHNICITY QUESTIONS FROM THE CPS

A.1 Supplement Questions

PANEL 1

SA1a First, are you one of the following: Hispanic,
Latino, or of Spanish origin?

<1> Yes
<2> No

SA3a Which one of the following list are you?
(READ ENTIRE LIST)

<1> White
<2> Black
<3> American Indian, Eskimo, or Aleut
<4> Asian or Pacific Islander
<5> Something else

PANEL 2

SB1a First, are you one of the following: Hispanic,
Latino, or of Spanish origin?

<1> Yes
<2> No

SB3a Which one of the following list are you?
(READ ENTIRE LIST)

<1> White
<2> Black
<3> American Indian, Eskimo, or Aleut
<4> Asian or Pacific Islander
<5> Multiracial
<6> Something else

(If Multiracial)

SB4a Which of the following list do you consider
yourself to be? (READ ENTIRE LIST. MARK ALL THAT APPLY.)

<1> White
<2> Black
<3> American Indian, Eskimo, or Aleut
<4> Asian or Pacific Islander
<5> Something else

PANEL 3

SC1a First, which one of the following list are you?
(READ ENTIRE LIST.)

<1> White
<2> Black
<3> Hispanic, Latino, of Spanish origin
<4> American Indian, Eskimo, or Aleut
<5> Asian or Pacific Islander
<6> Something else

PANEL 4

SD1a First, which one of the following list are you?
(READ ENTIRE LIST.)

<1> White
<2> Black
<3> Hispanic, Latino, of Spanish origin
<4> American Indian, Eskimo, or Aleut
<5> Asian or Pacific Islander
<6> Multiracial
<7> Something else

(If Multiracial)

SD1c I will read the list again and ask you to tell me
which ones you consider yourself to be:
(READ ENTIRE LIST. MARK ALL THAT APPLY.)

<1> White
<2> Black
<3> Hispanic, Latino, of Spanish origin
<4> American Indian, Eskimo, or Aleut
<5> Asian or Pacific Islander
<6> Something else

Hispanic national origin question

The following question on Hispanic national origin appeared on all four panels:

(If Hispanic)

Which one of the following are you? (READ EACH ITEM TO THE RESPONDENT)

Mexican or Mexican-American or Chicano
1. Puerto Rican
2. Cuban
3. Central American or South American
4. Other Hispanic, Latino, or Spanish Origin group

Questions on Hispanic as a racial category

The following question on Hispanic as a racial category / separate question appeared on Panels 1 and 2:

(If Hispanic)

Earlier, when I read you the list of White; Black; American Indian, Eskimo, or Aleut; Asian or Pacific Islander; (Multiracial on Panel 2); and Something else, you told me that you are [Fill from response on racial category]. To best describe yourself, would you like to have had "Hispanic, Latino, or of Spanish origin" added to this list?

Yes
5. No

The following question on Hispanic as a racial category / separate question appeared on Panels 3 and 4:

(If Hispanic)

To best describe yourself, would you have liked to have had two separate questions at the beginning, one where you could identify as Hispanic, Latino, or of Spanish origin, and another where you could also identify as White; Black; American Indian, Eskimo, or Aleut; Asian or Pacific Islander; or Something else?

Questions on preferred racial and Hispanic-origin terms

The following questions on preferred racial and Hispanic origin terms were asked across all four panels:

(If White on race question)

Earlier you told me that you are White. Which one of the following terms do you prefer? (READ ALL TERMS)

White
6. Caucasian
7. European American
8. Anglo