Some properties of an adjusting-magnitude schedule of reinforcement: implications for models of choice