White noise assumptions revisited : Regression metamodels and experimental designs for simulation practice