A latent variable structural equation model (SEM) was used to test the association between family density of alcoholism (one manifest continuous variable) and child behavior (the two latent variables, BD and CP, described above). The model was estimated using Maximum Likelihood and full information maximum likelihood missing data estimation in MPlus (Muthén and Muthén, 1998–2007). These estimation methods were used for all remaining structural models unless otherwise noted. The fit of the structural model was good: χ2(33)=71.30, p<.01, CFI=.99, RMSEA=.05. Statistically significant associations were found between density and BD (β=.10, p=.04) and between density and CP (β=.18, p=.00), indicating small but significant positive associations between density of alcohol problems in the family and ratings of these behaviors for children (betas are standardized). Variances explained for BD (R2=.01) and for CP (R2=.03) were small. When tested separately by gender (using a multiple-group SEM model), the fit was marginally improved by freeing the path between density and CP [χ2difference(1)=2.71, p=.10] such that density predicted CP better for boys (β=.25, p=.00) than for girls (β=.12, p=.09).
