Changing Numbers, Changing Needs: American Indian Demography and Public Health (1996)

Eugene P. Ericksen

Statisticians drawing samples of African Americans can create efficient plans using well-known strategies (Ericksen, 1976). They can take advantage of residential segregation to select blocks with many blacks at higher rates than blocks with fewer. They can instruct interviewers to code the race of "door answerers" by observation, and to subselect African American households within their sampled blocks. Because African Americans are thought to be culturally homogeneous, estimates for subgroups defined by cultural factors are not needed. Instead, estimates for the usual subclasses defined by variables such as age, sex, education, and income are sufficient. Since these subclasses tend not to be geographically concentrated, a national design providing precise estimates for all African Americans is likely to provide precise estimates for these subgroups as well.

These strategies are not likely to work as well for the Native American and Alaska Native populations. Native Americans and Alaska Natives are neither as segregated nor as concentrated as blacks. While the average black in a typical American city might live on a block where 75 percent of the population is African American, such concentrations of population occur only for Native Americans living on certain reservations or for Alaska Natives living in certain rural areas. Only small proportions of Native Americans and Alaska Natives live in such areas (Beals et al., 1994).

At the same time, there is great cultural diversity among Native Americans and Alaska Natives. Their distinctive cultures are more likely to be maintained on reservations and in other areas where Native populations are concentrated. These cultural subgroups tend to live in separate areas, e.g., Navajos in Arizona, Eskimo and Aleuts in Alaska, Sioux in the Northern Great Plains. When a national survey is taken, most respondents from any one of these subgroups are likely to live in just a few areas. The result is a highly clustered subsample producing imprecise subgroup estimates. If the cultural patterns associated with different tribal groups are related to the survey subject of interest, e.g., health practices, this clustering will reduce the precision of national estimates as well. For example, the designers of the Strong Heart Study (Lee et al., 1990) thought rates of heart disease varied greatly among different groups in different geographic areas. The importance of local cultures may decrease the efficiency of a clustered sample, necessitating a larger sample and creating tougher choices between designs maximizing precision for overall estimates and those more likely to improve the precision of subclass estimates.

Beals et al. (1994) found that over 300 tribal groups are recognized formally by the federal government, and there may be as many as 500 groups in the United States today. These groups have diverse histories, and to the extent that they have maintained their traditional identities, combining groups as separate as Seminole and Sioux into one category called ''Indian" seems little different than combining Polish Jews and Scottish Protestants into one category called "European." While political arrangements and economic opportunities may have lessened the need for separate cultural identities within the large categories of Indian and European, the apparent surge in Native American identity that has occurred over the past 30 or 40 years makes this less likely for them.

Finally, while blacks can usually be identified by visual inspection, this is more difficult for the Native American population. Generations of intermarriage and cultural mixing have blurred the distinctions between Native and other Americans, and it seems likely that many persons who would identify themselves as Native American in one situation might select a different racial identity in another situation. This creates the serious problem of how to estimate the total size of the population we are trying to study (see also Passel, in this volume).

This paper does not present an easy solution to the problem of how to sample the Native American and Alaska Native populations, and the correct statistical design will surely differ depending on the objectives of each study. Rather, the paper describes in some detail the three problems above, and attempts to explain decisions that need to be made with regard to each. The discussion begins with the most serious problem, defining

the population, then turns to the issue of cultural subgroups, and finally to strategies for sampling. First, however, we look briefly at two examples of surveys of Native Americans as context for the discussion that follows. The final section of the paper presents concluding remarks.

The Survey of American Indians and Alaska Natives (SAIAN) (Cunningham, 1995) and the Strong Heart Study (Lee et al., 1990) are two examples of surveys that include important elements of the Native American population, but do not provide complete coverage. The SAIAN surveyed those persons eligible for coverage by the Indian Health Service (IHS), i.e., Native Americans or Alaska Natives in federally recognized tribes living in IHS service areas. This population, while national, included only 906,000 persons, just under half the nearly 2 million Native Americans and Alaska Natives counted in the 1990 census. To increase the cost-efficiency of the sample, the SAIAN excluded counties with fewer than 400 American Indians or Alaska Natives, and 2.8 percent of the otherwise eligible population lived in these counties. The SAIAN also excluded "sampling segments," i.e., individual blocks or census enumeration districts, that were located in the eligible counties, but had less than 0.5 percent Native American or Alaska Native population.

The total sample size was 6,500, permitting useful national comparisons based on age, perceived health status, income, and place of residence. However, the sampling design resulted in two problems. First, there were no comparisons by tribal group or geographic area. Thus cultural factors unique to individual tribal groups that are crucial to understanding healthcare usage could not be studied with the SAIAN design. More important, there could be no generalizations from the study population to the million or so persons identifying themselves as Native American on the census who were not eligible for IHS services.

The Strong Heart Study attempted to overcome the first of these problems by focusing on three large and important groups: (1) the Gila and Salt River Pima/Maricopa Indian communities of Arizona, who were thought to be a low-risk population for cardiovascular diseases (Stoddart et al., n.d.); (2) seven tribes living off reservation in southwestern Oklahoma, thought to be a moderate-risk population; and (3) three Sioux tribes living in South and North Dakota, thought to be a high-risk population.

Each of these groups was sampled from tribal rolls, with about 1,500 persons being selected in each group. This design made tribal comparisons possible. It did not, however, overcome the second problem with the SAIAN design since the Strong Heart sample, selected from tribal rolls,

permitted no generalization to those Indians not included in a federally recognized tribal group. The SAIAN and Strong Heart study populations were both more likely than other Native Americans to live on reservations or in other areas of concentrated Native American population. These groups may be more important to government policies than their more assimilated brethren living in cities, suburbs, and other "non-Native" areas. Moreover, their Native American or Alaska Native identity is likely to be more distinct.

Another possibility for studying the Native American population is to use samples drawn for very large government surveys and to add some questions pertinent to Native concerns. For example, Eschbach and Supple (1995) report that a 1993 edition of the Current Population Survey included 1700 American Indians living in 758 households. The problem with this design, in addition to an imprecise definition of the covered population, is that the distributions of the Native American and total Current Population Survey populations are quite different: where one is sparse, the other tends to be numerous. The Current Population Survey design is likely to produce large numbers of Native American respondents in certain clusters; thus a substantial share of this subsample will appear in just a few locations, and the clusters will be quite different from one another. To the extent that place-to-place and group-to-group variations are important, this design reflects the inefficiencies of using a sample designed to produce precise estimates for the total population rather than for uniquely distributed subpopulations like Native Americans and Alaska Natives.

Passel (1976) observed the rapid and discrepant growth of the Native American population from 1960 to 1970. Using "best estimates" of the numbers of Native American births and deaths between 1960 and 1970 and assuming that international migration was negligible, he found that Native American population growth could not be accounted for by demographic factors. Even with plausible assumptions about errors in the estimated numbers of births or deaths or changes in the net undercounting of the Native American populations in the 1960 and 1970 censuses, the estimated increase in numbers of Native Americans was still implausible. The only reasonable explanation was that people who had not identified themselves as Native American in 1960 had done so in 1970.

This trend accelerated (Table 6-1) after 1970. Whereas the counted population of Native Americans and Alaska Natives had remained at between 300,000 and 400,000 from 1930 through 1950 and grown to just

over 600,000 in 1960, it tripled in the next 30 years, reaching 828,000 in 1970, 1.42 million in 1980, and 2.06 million in 1990.

We can perhaps focus more easily on the problem by considering those persons identified as Native American or Alaska Native who were born between 1930 and 1970 and were aged 0 to 40 in 1970, 10 to 50 in 1980, and 20 to 60 in 1990. They numbered 630,000 in the 1970 census, but in spite of the deaths that surely occurred between 1970 and 1990, grew to 918,000 in 1980 and to 1,079,000 in 1990. The growth from 1970 to 1980 was 46 percent, from 1980 to 1990 18 percent, and for the full 20-year period 71 percent. Such extreme growth has not been limited to this particular group. In 1980, there were 296,000 persons identified as Native American or Alaska Native who were aged 0 to 9; by 1990 there were 388,000 aged 10 to 19, an increase of 31 percent.

To understand this growth and its application to our sampling problem, we must look at the way racial data are obtained on the census. The person filling out the form, intended to be the head or spouse of head of household, indicates the race of every person living in the household. In cases of persons with mixed racial identities, the respondent uses judgment, and these judgments have apparently changed over time. There seem to be many people with some Native ancestry who must choose an identity, usually between white and Native American/Alaska Native, but often between black and Native American/Alaska Native. These

choices can be influenced by factors such as what group is dominant in the area where the household is located, political and economic opportunities that may be limited to persons identifying as Native American or Alaska Native, or just a psychological desire to choose a particular identity. For example, a child with a Native American father who grew up living with a white or black mother and perhaps with some of the mother's relatives in a household where the father was absent may decide to identify as Native American upon becoming an adult.

In many ways, the problem of Native American identity is similar to that of Hispanic identity. In both cases, persons of mixed ancestry are asked to choose one racial identity in a way that takes their individual situation into account. Studies by the U.S. Bureau of the Census (1979) showed that about 10 percent of persons identifying themselves as Hispanic on various surveys identified themselves as non-Hispanic upon being reinterviewed. The likelihood of consistent responses was greater for those with a shorter generational gap between themselves and their immigrant ancestors: while 99 percent of persons born in a Hispanic country reported themselves to be of Spanish origin in the 1970 census, this percentage fell to 73 percent for the third generation and to 44 percent for the fourth generation. Similarly, where there were Spanish ancestors on both sides of the family (mother and father), 97 percent reported being of Spanish origin on the census; only 21 percent did so when there were Spanish ancestors on just one side of the family.

One of the studies on which the Census Bureau reported was a special 1974 census taken in Gallup, New Mexico, where large numbers of Hispanics and Native Americans live. A reinterview study showed that 91.7 percent of those who had listed themselves as "American Indian" on the census did so again on a follow-up reinterview. The comparable proportion for Hispanic persons in Gallup was 89.2 percent.

In a way, Hispanic or Native American identity can be thought of as an attitude (Yancey et al., 1976). It can be subject to the local context, i.e., whether the person is living among others who are Hispanic or Native American, or the perceived political or economic opportunities that might result from selecting a particular identity. As people move from one place to another or as the economic or political climate changes, self-identification may change as well. At the same time, it seems likely that those whose ancestry is not mixed or who live on or near a reservation are more likely to identify themselves consistently.

The importance of context is also evident in the way parents assign racial identities to children living in mixed-race families. Passel (1991) examined the racial identities of children included in the 1970 census in households where the father and mother were of different races. In cases where one parent was black or white, the race of the father was dominant

(Table 6-2), being assigned 66 percent of the time. The particular race of the mother and father also mattered: where the father was black and the mother white or other, children were more likely than average (77 percent) to be assigned the father's race; where the mother was black and the father white or other, the likelihood of being assigned the father's race was considerably less (51 percent). Of the three groups examined in this study—black, white, and other (predominantly Asian and Native American)—parents who were other were least likely to live with children assigned to the same race. Where the father was other and the mother white or black, only 48 percent of children were identified as other; where the mother was other and the father white or black, only 23 percent of children were identified as other.

Passel's study reflects a racial dynamic that existed in 1970. Were the study to be repeated for 1990 or 2000, the patterns might well be different. In 1970, having a black parent increased the chance of a mixed-race child being called black, and to a lesser degree having a white parent increased the chances of being called white. To the extent that Asian or Native American identities have become more powerful predictors of racial identity since 1970, this pattern may have changed. In addition, there may be more mixed-race families, or the racial identities of fathers and mothers may be changing. All of these factors could contribute to a rapid growth of Native American and Alaska Native populations that cannot be accounted for simply by counting births and deaths. This complexity reflects the social-psychological nature of racial identities, as well as the effects of specific situations in which people live. And it leads to a vexing problem for surveys.

In 1990, the census counted over 2 million Native Americans and Alaska Natives on the basis of the race question on the short form administered

to all households. It is not clear how many of these individuals would be counted as Native American or claimed as members by any tribal group. Nor is it clear how many of those who identified themselves as Native American or Alaska Native on the census would so identify themselves to a survey interviewer. It may be that many of these people consider themselves to be "Indian" or "Eskimo" only for the purpose of what they put down when filling out the census form. Their answer may have neither political implication nor cultural meaning for them and may be irrelevant to their economic prospects. They may not be included in tribal rolls and may be unimportant for purposes of developing policy for the Native American and Alaska Native populations. Whether the correct number for public policy is the total claimed by tribal groups, the 2 million counted on the census, or some number in between is not clear.

People with a limited attachment to a Native group may also be difficult for a survey interviewer to identify as such. Interviewers are typically uncomfortable asking respondents to state their race, and such information is usually obtained by inspection. This approach suffices when the goal is to identify blacks, whites, and perhaps Asians, and to distinguish them from each other. Under such a scheme, less-numerous groups are likely to be undercounted. Between 1972 and 1993, over 29,000 persons aged 18 and over were interviewed for the General Social Survey (Davis and Smith, 1993). Information on race was obtained by inspection unless the interviewer was uncertain, in which case (s)he was supposed to ask. During this period, only 131 Native Americans or Alaska Natives were identified, about 0.4 percent of all samples. This is below the population proportion obtained by the census, which increased from 0.4 percent in 1970 to 0.7 percent in 1990. This difference of (0.7 - 0.4 =) 0.3 percent sounds minimal; however, if a sampling rate were specified assuming that the 0.7 percent figure was correct, the shortfall in sample size would be 100% × (1 - 0.4/0.7) = 43 percent. While some of this gap is probably due to undercounting—i.e., the interviewers never found some Native Americans who should have been included in the sample—much of it probably results from persons who call themselves "Native American" or "Alaska Native" not being identified as such by the interviewers.

The issue is complicated by a separate result from the 1990 census. One of the questions included on the long form administered to a sample of households asked people to indicate their ancestry, and more than one answer was allowed. The U.S. Bureau of the Census (1993a: Table 56) reports that 8.7 million people indicated Native American ancestry on the form. This suggests that if people are asked directly whether they have Native American or Alaska Native ancestry, a larger number will reply "yes" than if they are asked to make this choice from a list of presented alternatives.

Considering Native American or Alaska Native identity as an attitude, as suggested above, is perhaps a new way of conceptualizing the demographic classification. Especially with increasing rates of intermarriage, racial and ethnic identity is ambiguous for more and more people in the United States. It thus becomes important to focus on how questions about a person's racial identity are asked, because different ways of asking can lead to different answers. For example, if a person were asked to choose a race among black, white, Asian, Native American, or other, the answer might be different than it would be on the Current Population Survey. On the latter, a person is asked to choose an ethnic identity from a long list that includes groups such as German, English, Italian, Mexican-American, Chicano, Puerto Rican, Afro-American (black, Negro), or "another group not listed," but omits white, Native American, and Alaska Native. Still other results might be obtained in answer to a direct question, such as "Do you consider yourself to be a Native American or Alaska Native?"

It is also likely that answers given will vary with particular circumstances. For example, a person trying to register for a government program may feel that self-identifying in a certain way will affect his/her chances of being eligible. Similarly, a person might self-identify in one way, whereas someone else, perhaps a relative filling out a death certificate, might identify that person differently. Moreover, we cannot assume that all persons who are included on tribal rolls or are eligible for the IHS will indicate themselves to be "American Indian" on the census, nor, as is now obvious, can we assume the opposite—that all Native Americans or Alaska Natives who are eligible for the IHS can be found on tribal rolls.

In conclusion, people who want to survey Native Americans and Alaska Natives need to decide on a definition. If the decision is to include only persons on tribal rolls or those with specified numbers of Native American or Alaska Native parents or grandparents, it is likely that the total population will be substantially less than the 2 million counted by the census. If the definition includes persons who would identify themselves as Native American or Alaska Native on a survey when offered the full range of racial identities used by the census, then 2 million is probably the accurate population size. In this latter case, skilled interviewing is needed to find people who might have reason to identify themselves as Native American or Alaska Native in some situations, but as members of different groups in other situations.

Native Americans and Alaska Natives are culturally diverse, a conclusion based on both general understanding and reports of people who

have worked with them (e.g., Beals et al., 1994). Thus, grouping them may create the same problem that occurs when the Census Bureau groups all persons of Hispanic origin to publish aggregated statistics. As a result of such aggregation, the extreme poverty of Puerto Ricans living in some Northeastern cities is disguised by the middle-class nature of other Hispanics (Ericksen, 1985); the affluence of Cubans is combined with the poverty of Mexican Americans and the even greater poverty of Puerto Ricans to create aggregate statistics that represent none of the subgroups.

Just as a survey of the Hispanic population will be more meaningful if separate statistics are published for Puerto Ricans, Mexican Americans, and Cubans, a survey of the Native American and Alaska Native populations will be more meaningful if separate statistics can be published for key subgroups. The problem is to decide what the subgroups should be. Two of them should no doubt be Native Americans and Alaska Natives. The problem is that the first of these groups is nearly 20 times larger than the second. If we want separate estimates of equal reliability for Native Americans and Alaska Natives, substantial oversampling of the latter population will be required, increasing total survey costs. Similarly, we must oversample further if we desire separate estimates for individual tribal groups (e.g., Navajo) or combined groups (e.g., southwestern Indians).

Alternatively, separate estimates may be desirable for those living on or off reservations. Sampling only those who live on reservations would be wrong for at least two reasons. One is that only a minority of the population, by any definition, lives on a reservation. The second is that even among those who would be certain to identify their race as Native American or Alaska Native, the reservation population is probably very different from the nonreservation population. Their jobs, living standards, and even laws are likely to differ. It also seems likely that health conditions among the reservation and nonreservation populations differ greatly.

Whether a study focuses on reservations or not, the cultural diversity among the various tribal groups needs to be considered. While Census Bureau reports and other commentaries frequently describe Native Americans as one group, there are important cultural and historical differences among Native groups. Whether the similarities outweigh the differences is a matter for substantive judgment. Moreover, once the differences have been recognized, even more expert judgment is required to determine how the various tribal groups might be combined to compute subclass estimates. Unless one has the resources for a very large sample, it is likely that only a small number of subgroups, perhaps four to six, can be recognized. How this determination is made—whether by Native Americans versus Alaska Natives, by reservation versus nonreservation

populations, by identification of certain very large groups (such as the Navajo), or by regional separation of populations—is an important issue with policy implications. While the statistician can point out the need to make these judgments, the judgments themselves must be made by subject matter experts. Because the number and types of subclasses of interest affect the overall sample size, we need to specify these subclasses as part of the survey planning process.

Once the population has been defined and important subclasses identified, we can turn our attention to how to construct the sample. Geographic dispersion creates a substantial sampling problem. Using the Census Bureau definition, the Native American and Alaska Native populations combined comprise 0.8 percent of the total U.S. population. Most of these people live in areas where their population is sparse, i.e., where Native Americans and Alaska Natives comprise a small share of the total population. Out of a population of 2 million, only 579,000, fewer than one-third, live in counties where they comprise at least 10 percent of the total population. The household screening necessary to find eligible respondents could vastly increase survey costs.

To understand this problem, it helps to recognize the statistical principle of optimal allocation. By this rule, if a population can be divided into strata, the rate at which we should sample among the various strata should be proportional to the reciprocal of the costs of obtaining the average interview in each stratum. For example, if the costs in one stratum were four times greater than the costs in a second stratum, we would sample the first stratum at half the rate of the second. Use of optimal allocation to determine sampling rates within strata maximizes the precision of sample estimates for a fixed total cost. In other words, if we have x dollars to spend on data collection, we know how to allocate the sample across the strata to minimize the standard errors of the sample estimates. Given that we have a fixed amount of money to spend on data collection, optimal allocation tells us what the sampling rate and therefore sample sizes should be in the various strata to minimize sampling error.

For a survey of Native Americans and Alaska Natives, costs would be determined largely by screening rates. Screening involves contacting each sample household and determining whether an eligible respondent lives there. For a limited screening task, such as whether an eligible Native person lives at a particular address, it is reasonable and consistent with past experience to assume that the cost of screening is about one-tenth the cost of obtaining a complete interview. The optimal sampling plan that would result if the ratio of screening to interviewing costs were as high as

one-fifth or as low as one-twentieth is not greatly different from the plan that would result with a ratio of one-tenth.

In an area where everyone was eligible, say, on a reservation, the cost per interview would be 1.1C, where C is the cost of actually going through the questionnaire with one respondent. In an area where 50 percent of the households included an eligible respondent, the cost would be 1.2C, since there would be two screenings per interview; where 10 percent of the households were eligible, the cost would be 2.0C; and where 1 percent of the households (close to the national average) were eligible, the cost would be 11.0C. For these four examples, if we set the sampling rate in the area where everyone was eligible at f, then the optimal sampling rates in the other areas would be .96f, .74f, and .32f, respectively. In other words, the sampling rates in areas where 50 or 100 percent of the population was Native American or Alaska Native would be almost the same (f and .96f); where only 10 percent of the population so qualified, the sampling rate would be a little bit less (.74f); but where only 1 percent of the population so qualified, it would be a great deal less (.32f). Table 6-3 shows a larger set of examples.

Looking at it another way, costs are reasonably consistent in areas where the proportion eligible varies from 10 to 100 percent, but they increase sharply when the proportion eligible falls below 10 percent and especially when it falls below 5 percent. Because there are clear advantages

to area sampling where this strategy is feasible, its use can be recommended in areas where the proportion eligible is at least 10 percent. Where the proportion is between 5 and 10 percent, we might consider use of area sampling, but in areas where the proportion eligible is less than 5 percent, we should probably adopt other strategies.

Table 6-4 shows the distribution of counties by state where Native Americans and Alaska Natives are concentrated. For example, in Colorado there is one county, Montezuma, where at least 10 percent of the population is Native American or Alaska Native; the eligible population totals 2,141. The calculations in Table 6-4 indicate that over the total United States, about 29 percent of Native Americans and Alaska Natives live in counties where they are at least 10 percent of the population. In these counties they are 28.3 percent of the total, so the costs of screening would be moderate. If we dropped the limit to 5 percent of the county's population to try to capture some of the remaining 71 percent, it is doubtful that we would increase the proportion of eligible counties greatly or capture much of the remaining population. A better strategy would be to substitute smaller geographic areas for counties. In other words, we would use area sampling in all towns, townships, and census enumeration

districts where at least 10 percent of the population is Native American or Alaska Native. This set of areas might well include half of the Native population, but a substantial number would still be omitted.

Supplementing the area sample would be difficult. One strategy would be to use tribal rolls, adding all listed persons who lived outside the designated set of places where area sampling was used. This process would be costly and error prone. Listed addresses, especially in rural areas, are often inexact, and some may be out of date as well. Moreover, it would be expensive to get even a sample of tribal rolls from the different tribes, and their quality is likely to be uneven.

A second alternative strategy is ''multiplicity sampling." This strategy can take many forms, but one version would work as follows. We would take a sample of persons living on reservations or in concentrated areas. In each case, we would obtain a list of designated relatives, perhaps parents, grown children, and siblings. Addresses would be obtained for each of these designated persons, and a subset of those living in places other than where area sampling was used would be added to the sample. We would need to be careful to account for differential probabilities of sampling. If we included all relatives in the sample, a person with six designated relatives living in "Native areas" would have a higher chance of selection than a person with one such relative. Judicious subsampling, fastidious record keeping, and high-quality interviews providing the correct list of relatives and their addresses would be needed for multiplicity sampling to work well.

The proposed strategy has several disadvantages: (1) careful interviewing is sometimes difficult to maintain; (2) it would be difficult to obtain an equal probability sample given the variation in the numbers of living relatives; (3) even if we kept careful records, the costs of traveling to and interviewing those living away from concentrations of eligible persons would still be high; (4) Native Americans and Alaska Natives with no relatives living in the designated areas would be left out of the survey; and (5) interviewing so many relatives of persons already interviewed would increase sampling errors because the clusters of selected relatives would be more similar to each other than two randomly selected respondents would be, increasing the "design effects" due to cluster sampling and thus decreasing the efficiency of the sample.

Both area sampling supplemented by lists and multiplicity sampling are complex and risky procedures. They have the potential to limit the biases due to omitting Native persons living away from areas of population concentration, but increase cost as well as risk. Since the bias is not

eliminated entirely and since the question of how to define the population of concern remains ambiguous, it is not obvious that the advantages are worth the added cost and risk.

Given the diversity of the Native American and Alaska Native cultures, the sparse distribution of their populations, problems in identifying the population of concern, and the varying objectives of different surveys, no one sampling plan will suffice for all surveys. Rather than trying to devise such a plan, it may be more useful to indicate some of the key decisions the survey taker must make.

One is to decide whether to use list or area sampling. List sampling simplifies many aspects of the survey as the population is readily defined to be list members; addresses are given; and, as was the case in the Strong Heart Study, we can select large enough samples from each list to permit explicit comparisons of cultural groups. The major disadvantage of the list sampling approach is that unlisted persons are omitted, and most persons identifying as Native American or Alaska Native on the 1990 census are not included on any tribal roll.

The second decision is how to identify the population of concern. One choice is including persons on tribal lists, as was done by the SAIAN and the Strong Heart Study. This has the advantage of providing a clear definition, but excludes many people who may be of concern to meet the study goals. The alternative to this approach is self-identification, including either those who select Native American or Alaska Native from a list of proffered racial alternatives or those who say "yes" when asked if they have Native ancestry. On the other hand, self-identification involves problems of unreliability; for example, people who identify with a particular race on one survey may not do so again when offered the same choice at a later date. The approach will lead to an enlarged Native population, but this will be an advantage only if Native persons not included on tribal lists really matter to the study.

A third decision is how to define subgroups of interest. Explicit tribal comparisons must be limited in number for any survey with a small enough sample to be accomplished within a budget that would be realistic for most studies. As discussed before, this is a decision for substantive experts, and unless individual tribes are to be compared, the grouping of similar tribes will presumably be based on geographic, economic, or cultural similarities.

All of these decisions are conditioned in part by the need to define the population of interest. The population of persons included on tribal rolls and served by an agency such as the IHS is much smaller than the population of persons identifying as American Indian on a survey. It seems reasonable to advise that before good decisions can be made on the three issues just identified, study directors must decide whether to focus on

persons included on a tribal list or those who identify as Native American or Alaska Native.

Beals, J., E.M. Keane, and S.M. Manson, 1994 Population Studies of Older American Indians and Alaska Natives. Unpublished paper. National Center for American Indian and Alaska Native Mental Health Research, Denver, CO.

Cunningham, P.J. 1995 Health Care Access, Utilization and Expenditures for American Indians and Alaska Natives Eligible for the Indian Health Service. Unpublished paper presented at the National Academy of Sciences workshop on the Demography of American Indians and Alaska Natives, May 22-23.

Davis, J.A., and J.W. Smith 1993 General Social Surveys (1972-1993): Cumulative Codebook. Chicago, IL: National Opinion Research Center.

Ericksen, E.P. 1976 Sampling a rare population: A case study. Journal of the American Statistical Association 71:816-822.

Ericksen, E.P., ed. 1985 The State of Puerto Rican Philadelphia. Philadelphia, PA: Institute for Public Policy Studies, Temple University.

Eschbach, K., and K. Supple 1995 Employment, Household Structure, and the Health Insurance Coverage of American Indians, whites and blacks. Unpublished paper presented at the National Academy of Sciences workshop on the Demography of American Indians and Alaska Natives, May 22-23.

Lee, E.T., T.K. Welty, R.R. Fabsitz, L.D. Cowan, A.L. Ngoc, A.J. Oopik, A.J. Cucciara, P.J. Savage, and B.V. Howard 1990 The strong heart study—A study of cardiovascular disease in American Indians: Design and methods. American Journal of Epidemiology 132:6, 1141-1155.

Passel, J. 1976 Provisional evaluation of the 1970 census count of American Indians. Demography 13:398-409.

1991 Demographic Analysis: A Report on Its Utility for Adjusting the 1990 Census. Unpublished paper submitted to U.S. Secretary of Commerce Robert Mosbacher. (June).

Stoddart, M.K., B. Jarvis, B. Blake, R.R. Fabsitz, B.V. Howard, E.T. Lee, and T.K. Welty. no Recruitment of American Indians in Epidemiologic Research: The Strong Heart date Study. Unpublished paper, Center for Epidemiologic Research, Oklahoma City, OK.

U.S. Bureau of the Census 1970 1970 Census of Population: Characteristics of the Population. Volume 1. Washington, D.C.: U.S. Bureau of the Census.

1979 Coverage of the Hispanic population in the 1970 census. Current Population Reports, Series P-23, No. 82. Washington, D.C.: U.S. Department of Commerce.

1990 1990 census of population, supplementary reports, detailed ancestry groups for states. (1990 CP-S-1-2) Series 1-2. Washington, D.C.: U.S. Department of Commerce.

1993a Statistical Abstract of the United States: 1993 (113th edition). Washington, D.C.: U.S. Department of Commerce.

1993b Current Population Reports, P25: U.S. Population Estimates, by Age, Race, & Hispanic Origin: 1980 to 1991. Washington, D.C.: U.S. Department of Commerce.

Yancey, W.L., E.P. Ericksen, and R.L. Juliani 1976 Emergent ethnicity: A review and reformulation. American Sociological Review 41:391-403.