ABSTRACT
This interval was proposed by Burdick, Maqsood, and Graybill (1986). Although (4.3.4) is simple to compute, it can be appreciably wider than the exact Wald interval when the design is extremely unbalanced and AA is small. Thus , the exact procedure is particularly warranted if extreme unbalancedness exists. Donner, Wells, and Eliasziw (1989) have recommended the exact procedure for the type of unbalanced data encountered in family studies. Other intervals on AA based on normal approximations have been studied by Donner and Wells (1986) and Mian, Shoukri, and Tracy (1989) in designs where I > 25. Groggel, Wackerly , and Rao (1988) present a rank based method for constructing a confidence interval on a scaled version of AA that does not require normality . Intervals on PA> Pe, and Ae are obtained from intervals on AA by using the relationships PA = X.A/(1 + AA), Pe = 1 - PA, and Ae = l/X.A.
