ONLINE APPENDIX

Three concerns with this study are that estimated relationships between diagnosis and future school behaviors may be artifacts of (1) the cut-point used to differentiate children with less severe versus more severe pre-diagnosis behaviors, (2) modeling strategy, or (3) stable unobserved differences between the teachers/classrooms of diagnosed and undiagnosed children. One reason results may be idiosyncratic to use of a 25th percentile cut-point is if, for example, diagnosed children are clustered around the 25th percentile cut-point. Resulting estimates may be upwardly biased for children with less severe pre-diagnosis behaviors and downwardly biased for children with more severe pre-diagnosis behaviors. However, estimates using cut-points ranging from the 10th to the 60th percentiles of the aggregate measure of pre-diagnosis ADHD-related behaviors are roughly stable, especially between the 15th and 50th percentile severity cut-points for children with less severe pre-diagnosis behavior problems (see Appendix Figure A.1; note that externalizing problems are reverse-coded so scales are comparable across outcomes). This indicates that results are not driven by choice of idiosyncratic cut-point. To assess whether PSM estimates are driven by modeling strategy, OLS estimates from the 25th percentile cut-point can be compared to the PSM estimates in the main text. Although estimates from the controlled OLS models tend to be larger in absolute values due to lesser sample balance, the general pattern of differences by severity group remains similar.

To examine whether results are driven by stable unobserved differences between the teachers/classrooms of diagnosed versus undiagnosed children (e.g., differences in class sizes or instructional practices that lead to differing constraints/supports or tolerance of behavioral issues), estimates are generated for diagnosed and undiagnosed children with the same teachers in 5th grade (Appendix Tables A.3 and A.4). Estimated effects of diagnosis are roughly twice as large, likely reflecting poorer sample balance given that only 15 percent of sample students have at least one diagnosed sample student in class. (However, these fixed-effects estimates are smaller, on average, than comparable estimates from the clustered subsample without fixed effects, as would be expected.) Nonetheless, patterns of difference between children with less severe versus more severe pre-diagnosis ADHD-related behaviors are consistent: diagnosed and medicated children with less severe pre-diagnosis ADHD-related behaviors fare worse than children with more severe behaviors.

A final question pertains to the visibility of the diagnosis by teachers. Analysis to this point has assumed—but not tested—that the differing effects of an ADHD diagnosis by pre-diagnosis behavioral severity may result, in part, from teachers contributing to children’s psychological processing of their diagnosis (i.e., not only from the diagnosed child’s internalization of negative stereotypes or stigma). To explicitly examine this possibility, supplemental matching analyses examine special education/educational accommodations receipt as another moderator of the relationship between diagnosis and future school behaviors. It is reasonable to assume teachers are aware of a diagnosis if the child is receiving special education. Because the same matching techniques continue to be used to pair children who are diagnosed and receiving special education with undiagnosed matches with the same pre-diagnosis behaviors, cognitive skills, and other characteristics, matching minimizes the effects of differential selection into special education from driving the estimated effects of “diagnosis with special education/accommodations” on observed outcomes relative to undiagnosed matches. Estimated differences are instead attributed to teachers’ explicit knowledge of the diagnosis/service-use apart from the underlying behaviors or other observed characteristics, with all caveats about risk of omitted variables bias.

Because so few children in this sample are diagnosed and receive special education services or educational accommodations (“special education” for short) without medication, I separately estimated the effects of diagnosis within each severity group for three subgroups, each relative to their undiagnosed matches: (1) diagnosed and unmedicated, no special education; (2) diagnosed and medicated, no special education; and (3) diagnosed and medicated, receiving special education. This analysis reveals that the net negative marginal effects of diagnosis appear for all three groups. However, the magnitudes are descriptively (but not statistically significantly) larger for diagnosed children receiving special education. This suggests that teachers’ increased likelihood of knowledge of a child’s diagnosis/service-use (the “external” label) does increase the negative effects of diagnosis, but only descriptively.