Unweighted sample sizes and the error attributable