To investigate the impact of ADHD and class context on children's on-task behavior, generalized estimation equation (GEE)
models with an exchangeable working correlation matrix were used. GEE is an appropriate technique to take account of the
correlations among repeated observations of the same participant without the covariance structure being of central interest
(Zeger & Liang, 1986). The most popular form of inference on GEE regression parameters is the Wald Chi-square test. As the dependent variables, we used two continuous scores: (a) the proportion of time on-task and (b) the duration of each on-task interval (i.e., on-task span). In a first set of analyses, predictors were class structure (whole group vs. small group vs. individual work), group (ADHD vs. control), and the group x class context condition interaction term. The standardized mean difference between groups (i.e., Cohen's d = (M1 − M2) divided by the SDpooled) was additionally calculated and defined as small (0.2), moderate (0.5), and large (0.8) effect sizes. Although effect size is not strictly speaking the same thing as clinical significance, we assume that they will be strongly correlated. In contrast to statistical power, the effect size is independent of sample size and varies with precision of measurement (Swanson et al., 2001). Precision in this study was increased by averaging observation data across four time blocks. In a second set of analyses, the analyses above were repeated with academic content (highly academic vs. academic vs. nonacademic vs. instructional transitions) as the context condition. In these two sets of analyses, we additionally controlled for ODD problems and averaged academic performance as these variables may also contribute to variations in classroom on-task behavior.