observations, 79 (25.57%) had ^zig

observations, 79 (25.57%) had ^zig values in that range. For the 12P4 vs. 12P6 comparison, 42 of 71 (51.15%) misclassified observations had ^zig 2 1⁄20:5; 0:9Þ. In all, 87 of the other 349 (24.93%) observations had ^zig 2 1⁄20:5; 0:9Þ. Therefore, for both the 12 P3 and 12 P6 and 12 P4 and 12P6 designs, observations classified into different components are more than twice as likely to be near component boundaries. This, taken together with the result of the 12P3 and 12P6 compar- ison, provides a nice validation of our clustering results based on the 12P6 SID. Furthermore, they suggest that we could have obtained very similar results had a 12P3 or 12P4 SID been used. That three of 12 breads might well have sufficed here is, in itself, an interesting result.