Estimating Norms by Item-Sampling

If a 70-item test is to be normed for a norms population of 1,000 individuals, is it better to give the entire test to a sample of 100 individuals, or to give different samples of 7 items to each of the 1,000 individuals? When tried out on actual data, the latter, “item-sampling” method was found to provide an estimate of the “true” norms distribution slightly superior to that obtainable in the majority of cases by the former method.