Abstract
IN A PREVIOUS PAPER, A METHOD FOR THE ELIMINATION OF BIAS-BY-SELECTION FROM BEFORE-AND-AFTER COMPARISONS HAS BEEN SUGGESTED. WHILE THE SUGGESTED ESTIMATOR IS UNBIASED, THERE IS REASON TO BE CONCERNED ABOUT ITS PRECISION. IN THIS PAPER EXPRESSIONS ARE OBTAINED FOR THE VARIANCE OF THE UNBIASED ESTIMATOR AND THE MEAN SQUARE ERROR OF THE COMMONLY USED BIASED ESTIMATOR. THEIR COMPARISON YIELDS CONDITIONS UNDER WHICH THE UNBIASED HAS THE SMALLER ERROR. IT IS ALSO SHOWN THAT THE SUM OF UNBIASED ESTIMATORS FOR N SITES IS ASYMPTOTICALLY NORMALLY DISTRIBUTED. THE USE OF THIS INFORMATION FOR SIGNIFICANCE TESTS AND THE DETERMINATION OF CONFIDENCE LIMITS IS ILLUSTRATED.(Author/publisher).