The nested structure of these data (individuals within families) presents a potential analytic challenge because related individuals share common family influences. As a result, there is potential for interdependence among observations. Researchers traditionally measure the degree of interdependence by intraclass correlations (ICCs) among the observed variables. However, statisticians have argued that the “design effect,” which takes into account the average cluster size, is more important in determining the extent of interdependence in the data (Muthén and Satorra, 1995). A design effect of ∃2.0 is generally considered to be a meaningful threshold, and the average cluster size (1.22) and small ICC (0.07) of this sample yield a small design effect of 1.015 due to the fact that 46 of 54 families included in the analyses have only one individual. Thus, only 7% of the variance in future drinking quantity in this sample is at the between-family level, suggesting that clustering does not pose a problem for single-level analyses. An exploratory multilevel model found that residual variance at the between-family level was not significant, supporting the decision not to utilize a hierarchical
