A Labour-saving Method of Sampling

Abstract

IF a fraction p of a population have the attribute A, then it is well known that if m members out of a sample of N have this attribute, the best estimate of p is and its standard error is or Supposing, therefore, that we want our estimate of p to be correct within a standard error of 10 per cent of its value, we must count a sample containing 100(1 – p) members with the attribute A. If we do not know p roughly before-hand we do not know how large to take our sample. For example, if we wish to estimate the frequency of a type of blood corpuscle, and count 1,000 blood corpuscles in all, we should get such values as 20 ± 1·3 per cent, or 1 ± 0·31 per cent. The former value would be needlessly precise for many purposes. The latter would not differ significantly from an estimate of 2 per cent.