Dr. Don Hendricks -6- August 28, 1972 higher probability and larger samples will detect smaller errors. AS a working guide, we recommend that a power of .90 be the mininimal vatue used to select sample sizes. Thus to select the sample size necded to make com- parisons between two islands (two areas, two groups of islands) first determine what difference should be detected (e.g., 1.3 means roughly 30% differences will be picked up) and choose a sample size Jarge enough to give a power of .90 or better (about 130 in this case). Lacking any firm choice as to criterion as to difference to be detécted, one can simply browse around in the table and get a general impression of what difference a given sample size will serve to detect. R (Ratio of “high" to "Tow" mean) n 100 1.3 B 1.4 [nn 1.5; efn 1.8 Bin 2.0 e{n 2.5 Bf{n 3.0 get n Bg 30 .691 22 .86 114 .80 9 83 4 54 «.83 50 .75 120 .89 75 89 47 881 25 .89 116 .85 11 .89 6 .76 130 .91 85 92 55 91] 27 9] 20 .91 12 92 8 89 140 .93 |100 .96 -93} 30 -94 422 -95 13 94 9 .93 60 In the above table, & denotes theprobability that samples of size n will detect ratios of mean concentrations of size R. As an example, if we expect the concentrations in one area to be 1.3 times that in another and take 100 samples in each area, then the t-test will have a probability of .83 of finding this difference (and probability of .17 of not identifying this difference, even though one as great as this actually exists). An inspection of the above table will show that one cannot have a high probability of finding differences of less than about two-fold unless rather large samples are used. / Sampling patterns . In beginning this letter, we pointed out the necessity for using random sampling if one is to be protected against criticism. There is one rather widely used alternative scheme, which is to locate samples in some systematic pattern. In a problem like the present one, a two-dimensional grid may be the most likely pattern, so that the samples are located at, say, x-hundred yard spacing. One argument for using such a pattern is that it may provide somewhat better estimates of the amount of soil to be moved in a clean-up