A field experiment randomizes 150 schools to receive a new literacy intervention (treatment, n=75) or standard curriculum (control, n=75). Reading comprehension is assessed after 1 year. Schools are clustered within 5 geographic districts; treatment and control schools are often neighbors.
Surprising result: Treatment schools show lower reading gains (−0.12 SD) compared to control schools, opposite the intended effect observed in the pilot program.
Post-hoc investigation reveals: Control school teachers, learning informally that neighboring treatment schools received a new curriculum, independently adopted similar practices to "keep up." Roughly 40% of control schools partially implemented the treatment.
Which fundamental assumption of randomized experiments is violated in this design?
A field experiment randomizes 150 schools to receive a new literacy intervention (treatment, n=75) or standard curriculum (control, n=75). Reading comprehension is assessed after 1 year. Schools are clustered within 5 geographic districts; treatment and control schools are often neighbors.
Surprising result: Treatment schools show lower reading gains (−0.12 SD) compared to control schools, opposite the intended effect observed in the pilot program.
Post-hoc investigation reveals: Control school teachers, learning informally that neighboring treatment schools received a new curriculum, independently adopted similar practices to "keep up." Roughly 40% of control schools partially implemented the treatment.
Which fundamental assumption of randomized experiments is violated in this design?