Administering the NSAP as an add-on to the NSCH not only allows for a representative sample of adopted children, but it also allows for comparisons of adopted children with children in the general population on the health and well-being measures collected by the NSCH. For our analyses, we merged variables from the NSCH onto NSAP records (i.e., we created a linked NSAP-NSCH file). This enabled us to calculate estimates pertaining to NSCH variables for adopted children. Estimates pertaining to adopted children in this Chartbook are always based on the NSAP sample.45 Estimates per-taining to all children are based on the NSCH sample and do not exclude adopted children, in order to represent the general population of U.S. children.
To yield representative samples of children in each state, the NSCH used complex sampling methods involving the clustering of children within households and stratification of households within states. The complex sampling methodology means that, in order to generate estimates of variance that are not biased downward, analyses must take advantage of stratum and primary sampling unit (PSU) identifiers, as well as weights that have been adjusted for unequal sampling probability.46 The weights, which have also been adjusted for non-response and which were further adjusted to match pre-existing population control totals, are also necessary in order to generate parameter estimates (e.g., percentages) that can be extrapolated to the overall populations of children that the NSCH—and the NSAP—are intended to represent.
We used the statistical software package Stata in order to account for the complex sampling methodology in both the NSCH and NSAP. Additionally, all analyses were weighted. We generally avoided reporting estimates for which the relative standard error exceeded 0.3, and - at a minimum—flagged such estimates in the appendix tables to denote their imprecision. We also omitted value labels for percentages with relative standard errors exceeding 0.3 from the figures in the Chartbook. Additionally, we tested whether variables were associated by examining chi-square statistics. When chi-square statistics indicated that variables were associated, we tested differences between pairs of groups of sampled children (such as those adopted from foster care versus those adopted internationally or privately and domestically, as well as those adopted by relatives versus non-relatives) by calculating t-statistics for each difference. To test whether differences between adopted children and all children were statistically significant, we calculated t-statistics that accounted for the fact that the NSAP was a subsample of the NSCH. All comparisons between groups that are highlighted in text are statistically significant at the .05 level of significance; notable differences or associations that are statistically significant at the .10 level were also in some cases mentioned and footnoted as “marginally significant” at the .10 level.