Designs to guard against outliers in the présence or absence of model bias