Family Data, Analysis of

Family Roles

In textbooks about families, one of the first issues considered is “What is a family?” This discussion often leads to controversy regarding which roles should be included in the definition of the family. Mother, father, and child are the conventional roles associated with family, but gay and lesbian [Page 619]couples also raise children together; some families include adoptive or stepparents, natural siblings, and stepchildren; and many households consist of single parents. For the purposes of this discussion, it is not important to define the family by specific roles, but the fact that there are distinguishable roles remains important. Most family research is organized around family roles (e.g., mother, father, parent, child, and sibling).

Levels of Analysis

Statistically, family group data are typically studied as a two-level system. The family as a group is considered level 2, and the individuals within the family are considered level 1. Put differently, individuals are nested within the family, just as students may be said to be nested within a classroom. This is important statistically because the sample size differs at the two levels. Fifty couples may be observed in a study of marital relationships, a relatively small sample, but there would be 100 individuals in the study, a relatively large sample. The ability to detect a significant relation between two variables (i.e., the statistical power of the analysis) depends heavily on the sample size. The level at which a problem is analyzed, for example group versus individual, is called the level of analysis.

In family research, the level of analysis is usually the family or the dyad (a two-person group) even though the unit of analysis (i.e., the level at which measurements are made) may be the individual (e.g., a personality measure) or a particular relationship (e.g., how much one family member trusts another). This is so because, as mentioned earlier, the outcome scores of individuals within the group are usually correlated or nonindepen-dent. To the extent that the scores are correlated, they cannot be counted as two separate individuals statistically (i.e., the N, or sample size for the analysis, must be adjusted). Some statistical procedures (e.g., hierarchical linear modeling) make these adjustments automatically and are capable of simultaneously testing predictions at both the group and individual level. It is important to include enough families or dyads in a sample so that effects at the group level can be reliably estimated.

...