We took a number of steps to test the interventions
under realistic circumstances like those that would be in
force when the interventions were scaled-up. We included
a heterogeneous sample of schools even though this
introduced more variance in outcomes, and we allowed
schools to control when and by whom intervention activities
would be administered. To retain schools that could
not offer extended time for the study, we also dispensed
with extensive survey batteries and psychological process
measures such as daily diaries (e.g., Walton & Cohen,
2011). By taking these steps and reducing the “superrealization
bias” (Cronbach et al., 1980) that can lead to
large effect sizes in trials with small, highly controlled
samples (Ioannidis, Cappelleri, & Lau, 1998; Slavin &
Smith, 2009), we expected to find more modest but more
realistic and broadly reliable effects.