It has been suggested in the above discussion that living arrangement indicators tend to vary across the states in systematic ways. This was particularly evident in the way that regional "patterns' seemed to be identifiable and persistent in these data. To examine the relationship between these living arrangements, and to draw some fundamental insight into their distribution, we performed a correlation analysis on statewide living arrangement indicators.(10)
This analysis uses three levels of indicators. Indicators 1 - 5 are population percentages of 1) children not living with two parents (family disruption), 2) own children living with mother, 3) own children living with father, 4) children relative (kinship) care, and 5), children living in unrelated care situations. Indicators 6 - 9 are percentages conditioned on less than two parents present for 6) own child living with mother, 7) own child living with father, 8) child living in relative (kinship) care and 9) child living in unrelated care situations. Indicator 10 is the percentage living in relative arrangements conditioned on no parents being present.
Correlation coefficients between each of these indicators are presented in Table 2.5, arranged in the format of a series of blocks (identified by letters) highlighting the three different levels of indicators being correlated--population percentages and two types of conditional percentages. The correlation is a measure of their mutual relationship between two variables, or the degree to which a change in the value of one variable can lead us to expect a change in the other. A correlation coefficient (r) can vary between 1.00 (a perfect positive relationship, where an increase in one implies an increase in the other) and -1.00 (a perfect negative relationship, where an increase in one implies a decrease in the other). When a correlation coefficient is 0.00, we say that the two variables are independent, that is, that information about value of one of the variables gives us no information about the expected value of the other. The square of the correlation coefficient (r2)is a statistical indicator of the actual amount of variation either of the variables can explain in the other variable.
Block A presents correlations of the "family disruption" indicator (percent of children not living with two parents) with the population-level percentages of the remaining four living arrangement types. The absolute percentages of children for both Mother-Only care (r=.97) and Relative Care (r=.89) are very strongly and positively correlated to the relative size of the available "pool" of children not living in two-parent families. Conversely, the percentage of children in Unrelated Care arrangements does not appear to co-vary significantly with the level of family disruption (r=.10).
Block B presents correlations between the population-level percentages for each of the alternative living categories. Strong positive relationships are evident in the correlation of Mother-Only and Relative care (.84), and in the correlation between Father-Only and Unrelated care (.63). Mother-Father and Relative-Father showed moderate positive relationships, while Mother-Unrelated and Relative-Unrelated showed no significant relationships.
The correlations in Block C represent relationships between the "family disruption" indicator and the conditional percentages of children in each living arrangement, given family disruption. This is an important set of relationships because, as we have seen, the final population percentages for these living arrangements are the product of the two percentages being correlated. The primary observed relationships here are a strong positive correlation (.71) between Relative Care and Family Disruption and a strong negative correlation (-.64) between Unrelated Care and Family Disruption. This can be interpreted as follows: as the presumed "pool" of children available for alternative care arrangements becomes larger (due to fewer children living in two-parent families), the likelihood that any child is in Relative Care becomes greater and the likelihood that any child is in Unrelated Care becomes smaller. This suggests that the increased "risk" of Relative Care caused by family disruption is further reinforced by an increasing likelihood of being in a relative arrangement. In the Unrelated Care case, increased "risk" is counteracted by a decreasing likelihood of being in an unrelated care arrangement. The conditional correlation between Mother-Only care and Family Disruption is positive but not statistically significant. Thus, while we saw in Block A that the percentage of children in mother-only care is clearly dependent on the level of family disruption, this finding suggests that the rate at which mother-only arrangements occur does not change significantly with the number of children at "risk." The correlation between Father-only and Family Disruption is mildly negative.
Skipping to Block E, we see correlations among the conditional percentages of moving to each arrangement given less than two parents. Here we see a very strong negative relationship between Mother-Only and Father-Only (-.93), strong negative relations between Mother-Only and Unrelated (-.73) and between Relative and Unrelated (-.59), and a strong positive relationship between Father-Only and Unrelated (.73). At this level of conditional likelihood, kinship care is independent of Mother-Only care, has a very weak negative relationship to Father-Only care, and a strong negative relation to Unrelated Care.
Block D, in the center, represents the correlations between the population percentages and the conditional percentages of each arrangement. All of the coefficients along the diagonal are positive, as a higher conditional percentage contributes to a higher population percentage. For Relative Care, this joint coefficient is very high (.94) because, as we have seen, these two percentages tend to increase together. The remaining coefficients in the table are mostly rather strong and follow a distinct pattern. Overall, the Mother-Only and Relative Care percentages vary together positively, the Father-Only and Unrelated Care percentages vary together positively, and the Mother-Only and Relative Care percents both vary negatively with the Father-Only and Unrelated percents(11).
Blocks F, G, and H introduce the second type of conditional percentage, that of being in a relative care arrangement given the condition that no parent is present. It is most readily interpreted as representing the Relative-Care versus Unrelated-Care dimension. This number is necessarily positively related to each of the other Relative Care percentages and negatively related to the Unrelated Care percentages. What is of interest is its strong positive correlation to the population percents of family disruption (.78) and Mother-Only (.80), and moderate positive correlation to the conditional percentage for Mother-Only care (.39).
The clear implication of this correlation analysis is that, at the state level, kinship care arrangements appear to be a response attached to what we have termed "family disruption," measured by the percentage of children not living with two parents. This "disruption" can be a product of either the nonformation or the breakup of families. Kinship care levels also are seen to co-vary closely with levels of Mother-Only care, while the relationships between these living arrangements and the percentage of children in Unrelated and Father-Only care arrangements tend to be weak or negative. It appears from these findings that the processes or conditions that lead children into the Unrelated and Father-Only care arrangements are different, and often in opposition to, the processes and conditions that lead children to Mother-Only and Relative care arrangements.
(8)" This is an important substantive distinction. Child living arrangements are often reported by relationship to the household head. The 1994 Current Population Survey estimated that over 5.4 million American children lived in households headed by grandparents or other nonparent relatives. Of these children, only 43 percent (or 2.1 million) did not have a parent present in the household. Therefore, what we here call "kinship care" represents significantly fewer than half of the population of children living in households headed by relative adults.
(9) The interpretation that the two-parent nuclear family is "primary" and other arrangements "residual" is not empirically justified, and these data cannot support such a causal inference. Rather, this is an inductively grounded organizing principle, which is subject to future empirical examination and revision. The working hypothesis is that children are most likely to remain in nuclear families unless those families are disrupted. An ancillary hypothesis would imply that children are more likely to remain with a single parent than either relatives or strangers if the parent-child living arrangement is not disrupted. A single-parent home is considered more likely to be "at-risk" of disruption than a two-parent home, other things being equal, so we would expect a significant amount of adaptive caretaking to occur around children living with one parent only. Clearly some children live outside of a parental unit for reasons other than disruption of the household (e.g. protective removal, institutionalization, school choice, etc.), but these factors should not have a disproportionate effect of the overall pattern of living arrangements.
(10) It is important to notice that this is an "ecological" analysis that examines the relation between properties of state distributions of child living arrangements. Conclusions cannot be casually assumed to apply at the individual level. Also, it should be noted that Washington, D.C. has been excluded from this correlation analysis to remove the extremely skewed influence of its population-level indicators.
(11) The sole exception to this pattern is the conditional percent in relative care having a weak positive relationship to the population percentage in Father-Only Care.