Anne Fleur Kortekaas-Rijlaarsdam

65 MPH EFFECTS ON MATH PERFORMANCE: INFLUENCE OF BEHAVIOR 3 Analyses Our power calculations indicated that a sample size of 63 completed participants was sufficient to detect significant differences between placebo and medication with an alpha of .05 and 80% power for medium sized effects (Twisk, 2013). The assumption that the washout phase was sufficient to rule out carryover effects was checked using independent samples t-test comparing sums of results from methylphenidate and placebo for participants in the two treatment sequences (methylphenidate first versus placebo first) (Wellek & Blettner, 2012). First we tested the effects of methylphenidate on academic accuracy and productivity as well as on ADHD symptoms (SWAN) using generalized linear mixed model (GLMM) analysis (SPSS version 21; IBM, 2012) with a random intercept to control for within subject dependency of measurements. To test underperformance of the ADHD group compared to the TD group we compared accuracy and productivity of the ADHD group during placebo with accuracy and productivity of the TD group using independent sample t-tests. In addition we used independent sample t-tests to compare academic accuracy and productivity of the ADHD group during methylphenidate with accuracy and productivity of the TD group to estimate whether performance normalized. Next, for academic measures that improved significantly with methylphenidate, we tested whether these improvements were related to reductions in ADHD symptoms: The possible mediating effect of symptom improvement was estimated by the method of Baron and Kenny (Baron & Kenny, 1986). Using GLMM with a random intercept, we calculated the regression coefficients of the effects of methylphenidate on the mediator (β a ) and of the mediator on math accuracy or productivity (β b ) when fixed effects of both methylphenidate and the mediator were entered in the model. Next, the product term β a β b of the regression coefficients with accompanying confidence interval was calculated using the asymptotic normal distribution method (Krull & MacKinnon, 2001; Tofighi & Mackinnon, 2011), thereby assuming a covariance of zero between β a and β b (Tofighi, Mackinnon, & Yoon, 2009). If the 95% CI did not include zero, the effect of the mediator was significant. Finally, for those dependent measures showing significant effects of methylphenidate compared to placebo, possible moderating effects of learning ability and ADHD symptom severity (DISC-P) were tested using generalized estimating equations (GEE). For learning ability, z-scores (z=0 equals the mean of the TD group) were entered in the GEE model. Only z-scores for the relevant academic subject (math, reading or spelling) were entered as possible moderators. The GEE analyses contained the fixed effect of medication and all other moderators and mediators, as well as the interaction effect of the relevant moderators with medication. Interaction terms derived from the z-scores for ability and methylphenidate efficacy factors indicated whether academic ability was a moderator. The nature of significant interactions was explored using post