New empirical evidence for the design of group randomized trials in education