Statement of Mary C. Waters
Department of Sociology, Harvard University
Before the Subcommittee on Government Management, Information and Technology
of the House Committee on Government Reform and Oversight
Hearing on Multiracial Identification
25 July 1997
Thank you Mr. Chairman and Members of the Committee for inviting me to speak to you today about the very important issue of how the federal government should measure race and ethnicity. My name is Mary Waters and I am a sociologist and demographer who specializes in racial and ethnic identity. I have written extensively about how Americans think about their ancestry and backgrounds and how the answer censuses and surveys that ask about those backgrounds. I was a participant in the National Academy of Sciences workshop on the federal standards for racial and ethnic classification and I have consulted to the Census Bureau and Bureau of Labor Statistics on how to test various ways of measuring a multiracial category and on the design and reporting of the census ancestry question. I currently have a grant from the NICHD to examine patterns of intermarriage and the identity choices parents make for their children when filling out the census form, and how interracial children should be counted in long run population projections.
I will be commenting today on the Recommendations of the Interagency Committee for teh Review of the Racial and Ethnic Standards, as published in Part II of the Federal Register for July 9, 1997. I think the report does a very good job of summarizing the pros and cons of each of the actions the committee considered taking. I also think the report does a very good job summarizing the four years of research on this topic that the government has undertaken. I will concentrate my remarks today on the recommendation of the committee regarding the issue of allowing people to report more than one race. I will try to outline some of the issues the report raises but does not answer completely, and some questions and issues I think the recommendations raise that need to be thought about before adopting the committee's recommendations. My approach is to think about the social science issues and the demographic and statistical questions this new approach raises, I leave it to others to debate the political pros and cons of the recommendation.
My reading of the recommendation of the committee to allow people to "mark one or more" or "select one or more" race is that this recommendation tries to meet concerns on both sides of this issue. Allowing people to check more than one race does allow people the recognition of their ancestries they have been requesting. It no longer forces a person with parents of different races to identify only with one parent's race. Allowing people to "mark one or more" also allows a greater degree of historical continuit with previous federal categories by not including a separate "multiracial" category and by preserving information about which races an individual chooses.
Of course like most middle roads, it does not necessarily satisfy either side. It does change the way we measure race, thus upsetting those who want to preserve the status quo, and it does not provide a separate multiracial category, thus upsetting those people who have requested such recognition.
I see two important unresolved issues in this recommendation. The first and most important has to do with the methods and principles for tabulating and reporting the counts of people that will result from this new method. The second has to do with implementation of this new method in the field and resulting problems.
TABULATING RESULTS
The recommendation to allow people to choose more than one race effectively decouples response from reporting. The committee does not specify how federal agencies will process the data to meet legislative needs, or how the general counts of the population from the census will be presented. Allowing more than one response means that there needs to be consistent rules on how those who choose more than one race will be tabulated. These rules will be necessary in order to have comparable data across agencies and to meet requirements of the law to have mutually exclusive and exhaustive data. The interagency committee report summarizes three possible ways of tabulating the data that were outlined in the Race and Ethnic Targeted Test (RAETT) report. The report also briefly suggests two other possible methods. There are two other methods the report does not mention that I will also introduce here. Thus I will briefly describe seven possible methods:
The three possible tabulation methods illustrated in the RAETT test are:
1-Single race approach. This approach would count all people who marked more then one race in a "multiple race" category, similar to the "other" race category that the census uses now. The information then that a person identified as Asian and White, for example, would be lost. The benefit of this there is no ambiguity in counting and no chance of double counting.
2-All inclusive approach. This approach would count a person in each race that they marked. This would involve double or triple counting. If a person said they were Asian and White they would be counted in both of those categories. This would mean that the counts would add to more than 100% and you would end up with a count of races, not of people.
3-Historical series approach. This approach would reclassify those who chose more than one race back into a single race in a set of mutally exclusive categories that add up to 100%. This approach, it was stressed in the report (page 37), produced counts that were statistically the same as those produced by the status quo race question, for all groups except for the Alaska Native targeted sample.
Two other possibilities that were mentioned in the interagency report, but not used in illustrative samples in the RAETT report are:
4-An algorithm that distributes responses from a multiracial category in proportion to the distributions of the current single race categories (page 61 of the interagency report). The report currently notes that this method could mispresent the multiracial respondents, even though it would not change the relative sizes of the single race categories.
5-An algorithm similar to the one used now by the National Center for Health Statistics to impute multiracials from states such as Georgia, Indiana and Michigan into the standard race categories. These individuals are first classified as "other" and then using other characteristics are given an imputed or "estimated" race.
Finally, there are two other possibilities that could be used to process these data:
6-Assign a weight to a person reporting more than one race, and then count them as a fraction of a person in each race that they report. For instance if you were calculating the income of Asians and Whites in a particular census tract and a person said they were both Asian and White, you would count their income in each race's count, but with a weight of .5. Likewise if a person reported three races, they would be counted in all three races but with a weight of .33. This would be similar to a sampling weight and is done often in work using the census ancestry question which allows multiple responses. However this method has very negative historical connotations as we shamefully began as a country, counting slaves as 3/5 of a person for congressional apportionment. It also would yield fractional counts of people for geographic areas and I am not sure how counts which include fractions of people would be viewed legally in discrimination cases and the like.
7-Assign multiracial people randomly and equally to one of their races. Thus of all people who say they are Asian and American Indian, 50% would be randomly chosen to be Asian and 50% would randomly be chosen to be American Indian.
My impression from reading the interagency report is that the method described as the "historical series approach" is the most likely approach because it affords the greatest continuity with previous data. As I understand it, the historical series approach assigns a person who reports their race to be both white and any one of the other OMB categories, (Black, Hispanic, Asian, or American Indian) to be in the latter category, not white. Since the majority of current intermarriage does involve a white spouse this method would classify a large number of multiple race people into a single race. In effect however the "historical series approach" uses the "one dro rule". Those people who are part white are classified as the nonwhite race. This may be the current"best guess" about how these people are "socially indentified" in their communities, but it does recreate the classification dilemma that the self identified multiracial community is trying to move beyond.
Under this method if a person reports two of the nonwhite groups — say a person who identifies as Asian and Black, they would be put into a "multiple race" category. I assume people who listed three races would also be classified into the multiple race response. This method leaves open the question of how those in the multiple race category would be classified if the intent of the classification was to exactly replicate current OMB categories, with no room for the category of multiple race.
As the interagency report makes clear we do have some experience with overlapping categories now because of the separate race and Hispanic categories. When data on race and Hispanic origin ore presented based on the two question format, we in effect double count, because Hispanics can be of any race. Directive 15 states that when a single question is used to report data, in effect Hispanic trumps race. If a person is both Hispanic and White, they are counted as Hispanic, and the category white does not include Hispanics. The illustration of the historical series in the RAETT report seems to operate on a similar principle. I think it will be very important for OMB to very clearly lay out all of the combinations of responses possible to the question and how each one will be classified if this approach is the one suggested. It also should be specified if a "multiple response" category will be allowed for all purposes or if further imputation will be used for this category for some legislative or reporting needs.
IMPLEMENTATION
Federal data on race and ethnicity are gathered in many different agencies using a variety of different forms and techniques. I think the biggest possible impact of this change of allowing more than one race to be reported will be between data sets that are based on self identification and those based on observer identification. Self identification will allow people to report more than one race. Some data however is collected through observation. It is often difficult for observers to accurately estimate one race; the error rate will increase dramatically if they estimate more than one race. This is especially true if we consider that while a very small proportion of Americans report more than one race in any of the tests conducted thus far — the CPS, NCS and the RAETT — a large proportion of Americans have some mixed ancestry in their backgrounds. This is particularly true of the American black population, many of whomhave distant white ancestors, but who would never identify as anything other than black. If an observerwer "guessing" about whether any one individual African American was multiracial or not, this could wildly inflate the counts of multiple race people. I would therefore suggest that attention be paidto the federal agencies that now collect data based on observer idenification, and to the kinds of instructions about this issue that would be given to such observers. one possible solution to this problem would be to only allow reporting more than one race if the data are collected through self identification. The implications of two different modes of data collection would have to be discussed further however. This issue of the disjuncture between self identification and observer identification might call for further research or at least increased vigilance in reconciling these two methods.
Finally, if people are instructed to mark one or more races on the census and other self identification forms, the instructions to respondents will be extremely important. Will the instructions to respondents ask them to choose races they "identify" with? Will the instructions specify how many generations to go back in determining race? It will be important not only to come up with understandable instructions which help respondents, but also to standardize those instructions across all methods of gathering data. If the census for instance asks for races you identify with, but schools ask for the races of parents and then assign them to children, there could be a disjuncture between the different methods of collecting the data.
CONCLUSION
In conclusion, I would like to stress a point I made in my last testimony before this committee and I think it is worth repeating again today. The political debate about this issue has tended to concentrate on the counts and identities of African Americans. This is very understandable given the history of race relations in this country. All of the statistical and demographic research however points to this change having the biggest effect on American Indians, Alaskan Natives, and Asian and Pacific Islanders. These groups could be very much changed by this new method because of their relatively small size and their relatively high intermarriage rates. Law makers and government statisticians should pay particular attention to howthese groups will be affected and should probably make a special effort to make sure they are consulted.