New Empirical Evidence for the Design of Group Randomized Trials in Education