Survey data with sampling weights: is there a \"best\" approach?