Each regression equation included (a) SNP, (b) digit span forward, (c) digit span backward, (d) SNP × digit span forward, and (e) SNP × digit span backward entered hierarchically in this specified order as predictors of change in CBCL-AP scores. This sequential approach allowed us to assess whether higher-order working memory manipulation interacted with genetic background above and beyond any lower-order working memory maintenance effects and interactions. Change scores for digit span forward, backward and attention problems were extracted as standardized residuals, obtained by predicting the follow-up score from the baseline score. Using the residual to quantify change negates the correlation between change score and baseline performance, facilitating the measurement of variability in change over time after what is predicted from baseline status (Glass & Hopkins, 1996). We applied a conservative Bonferroni correction threshold of p < .0063 for this analysis, which corrects for testing eight non-independent SNPs. Post-hoc probing of significant moderator effects required plotting and testing the ‘simple slopes’ of the relation between ADHD symptoms and working memory by individual genotypes for each SNP (Holmbeck, 2002). Models were
