Barron, 2004), those interactions that did not reach p < 0.10 were removed to improve interpretability of the main effects (Aiken, West, & G., 1991). Preliminary analyses (see Supplemental Tables S1-S2) showed interactions of neighborhood alcohol availability with sex in relation to high-risk drinking, as well as an interaction of FH with race/ethnicity in relation to dependence, so we used stratified models for subgroup analyses, even though three-way interactions (such as sex X FH X outlet density) were not statistically significant. All demographic covariates and an indicator for survey year were included in the stratified models, although only selected coefficients are reported (full models available upon request).
