Operant variability when reinforcement is delayed