Randomized interventions and “real” treatment effects: A cautionary tale and an example