Discussions of data resource issues have been interwoven throughout this report, as these have steered the research activity and discussion at least as much as have substantive concerns and questions. Certain of the approaches and analyses were chosen specifically to exploit unique data opportunities, while others had to be qualified carefully due to the specificity of the information that was available. As a result, this task took a patchwork format, with each type of available information being exploited to provide one vantage point on the complex set of questions involved in kinship care. It is clear that our ability to describe formal and informal kinship care is highly limited by the range of information that is available for consideration. Several methodological issues that are particularly salient for describing kinship care are discussed in some detail.
Definition of Child Relationships
The "American family " includes a diverse set of social institutions. While the current family landscape is dominated by traditional two-parent, two-generational nuclear families, other family forms are common -- particularly among specific subgroups of the population. Tracking the living arrangements and household situations of children is an important part of understanding the status of children in our society and an important first step in evaluating child well-being. In this work, oriented as it is to investigating alternative forms of caregiving for children, it is necessary to view the family environment from a perspective that centers on each child. In order to define informal kinship care arrangements, we need to identify the relationship of each child to their caregivers and to know whether or not a parent is also in the home.
Much of the attention given to family patterns and even to kinship caregiving does not presume such a fine definition. It is well documented that many children reside in complex family units headed by a grandparent or other relative. It is also known that, in many of these multi-generational homes, the grandparent is the effective caregiver, even in the case that a parent is present. But in our attempt to define alternative care structures, our interest is directed to those children who are being cared for by a relative in the absence of a parent. These are unambiguous arrangements where the custody and care of the child have clearly passed from the parent to another family member. This is not the type of information that is most commonly sought from household-based census and survey data. Typically, the organization of the household unit itself (including the presence or absence of children) is of greater interest, and the defined head of the household is the primary person described. After all, the household is the living unit that associates persons with residential space, it is a definable social and economic unit, and the notion of a household easily subsumes a broad range of possible combinations of people and arrangements.
In order to identify the specific situation of each child from data collected on a household level, it is necessary to refer to a matrix of information that defines each individual person and their relation to the others in the household-- information that often is collected as part of a census or household survey. In the portion of this project that was based on data from the Current Population Survey, this information was available and the complex task of defining a child's relationship in the household was performed as part of the data analysis effort. In the portion of this project based on Census data, though, the detailed individual-level relationship data was not publicly available, so the analysis had to rest on summary data as tabulated and provided by the Census Bureau.
From the 1990 census data, as distributed in the common Summary Tape File 3 format, the family relationships of all children within the household can be identified, reflecting real improvement in their sensitivity to the existence of subfamilies and complex living units. But the only other variable that can be directly associated with this information is a two-category age classification (0-4 and 6-17 years). More detailed information is available for other classes of children ("own" children, children classified by relation to head-of-household, etc.). Other variables, such as race/ethnicity and poverty status can only be evaluated on an aggregate geographical level.
Sampling of "rare" events
The main reason that sampled data sources present a particular problem for this subject area is that kinship caregiving, both formal and informal, is a relatively low-probability event in American family life. While a fairly large number of American children lived in homes with some non-parent adult relatives in 1990, only two to three percent did so in the absence of either parent. Thus a random sample of 1,000 children would only be expected to identify somewhere around twenty to thirty children in kinship care -- hardly enough to support detailed decomposition and analysis. As we commonly observe with more familiar findings based on sampling results (such as election polls), the confidence interval within which we can "trust" estimates drawn from sampled data are often expressed with a range of plus or minus several percentage points. When the true underlying number is around two to three percent, an estimate that is accurate within plus or minus two points does not provide a level of precision that is satisfactory for comparison, for trend analysis, or that encourages any further analysis.
The error range of an estimate is mostly a function of the size of the sample, though it also depends on the underlying prevalence of the characteristic being estimated. Therefore, to discuss national patterns and trends about kinship caregiving, we must turn to very large samples or to population-based information if we hope to obtain reliable information. The Current Population Survey, which is based on annual national samples of about 11,000 [households], is the best source of this type of information. In rough terms, the national one-year estimates for 1994 produce the child living arrangement estimates at the following levels of accuracy:
||90% confid interval +/-
||+/- as % of estimate
|N children w/ 2 parents
|N children in kinship care
|% children in kinship
|% white ch in kinship
|% afr-amer ch in kinship
|% hisp ch in kinship
As we can see from these numbers, the CPS provides a rather close estimate of the national kinship care population, here as 3.1% of all children plus or minus one-quarter of one percent. As we move from a national to any sub-national estimates (region, state, ethnic group, etc) the precision of the estimates decays as the N decreases. Therefore, the CPS provides poor state estimates of kinship care levels except in the few largest states. Looking at race/ethnicity in the numbers above, it can be seen that the relative size of the error margin for white children in kinship care is about one-fourth of that for Hispanic children, and less than one-fifth of the error margin for African-American children.
The Problem of Different Sub-Populations
When we attempt to interpret any simple univariate frequency or a bivariate relation describing the association between the level of children in kinship care and another variable, say region or age, we are implicitly assuming (at least for the moment) that this relationship is not somehow influenced by the impact of other factors. When we have reason to believe that other factors do intervene in the relationship, good research procedure suggests that variables be explicitly introduced into the analysis to represent the influence of these factors. The racial/ethnic subpopulation estimates for children in kinship care above point to the type of factor that a well-designed analysis would always evaluate. On the national level, the 1994 CPS data estimate the kinship care levels at 1.8% of the total for white children, 3.4% for Hispanic children, and 8.0% for African-American children. This is an interesting, and not unexpected, finding in its own right. Kinship and extended family relations have been more prevalent in African-American families for a long time. However, these differences require that we ask a certain empirical questions before we continue. Are four times as many African-American children as white children in kinship care arrangements because they are differentially exposed to the same causal conditions? Are different causal factors operating independently on these different racial/ethnic subgroups of children?
Arguments can be developed in support of both interpretations. To support the "same causes" position, African-Americans are more likely than white Americans to be exposed to forces that we already know to be associated with increased family breakdown, such as poverty, unemployment, program receipt, and teenage childbearing. In contrast, our observation of ethnic cultural patterns also suggests that the practice of sharing child-rearing responsibilities across members of the extended family is accepted and considered more normative across a much broader segment of the African-American community than among white Americans, thus supporting the "different causes" position. This argument will not be unravelled here(26), but it highlights the significance of the issue. Because there are such wide differences between racial/ethnic subgroups in the formation of family groups and in the prevalence of kinship caregiving, we should not expect all relations or explanations to be the same across the entire population. Race and ethnicity should be paramount in the study of family structure and living arrangements, and any study that cannot explicitly control for racial/ethnic effects will necessarily be limited in its ability to define and describe these phenomena.
In the Current Population Survey portion of this report, resource limitations resulted in race/ethnicity only being evaluated as a univariate category. Because race/ethnicity is such an important discriminator variable in describing relative caregiving, far more descriptive value could be gained by controlling some of the other relationships (i.e. for age, region, metro/nonmetro, poverty) by race/ethnicity. For example, the South showed significantly higher levels of relative caregiving than the other three regions of the nation. Is this due to the higher proportion of African-Americans in the South? Do whites in the South also have higher kinship care levels than whites in other regions? Is there a difference between urban and rural southerners of any race? These are the type of questions where racial controls would be useful. They are necessary to help see if racial/ethnic factors alone explain other relationships, and to help look for variability along other dimensions within each race/ethnic group. Direct racial/ethnic classification of children and families was available in the preceding analysis only for the four-state data on children in formal kinship foster care and for the records of Illinois kinship AFDC cases. None of the analyses of informal kinship were able to address racial subclassifications, due to the fact that the Census-based data for kinship care cases were not available in a format classified by race.