The first row of this table says, for example, that the probability of the dipeptide AN being encoded by a dicodon in S63 is 1/8, since AN can be encoded by eight dicodons and only one of these is in S63. (Note that if you subtract each probability in the table above from 1, you get the probabilities for the corresponding control group S711.)
What I discovered last night is that if you take the average of these probabilities for the control group C711, the result (about 0.85) is very close to the average co-efficient for the 12 ln(c/u) on ln(c/L) correlations for the segment data obtained from the a1 fold using the C711 group of dicodons, whereas if you do the same for the S63 study group, the result is only about half of the average coefficient of the S63 correlations. And the same result applies for the S60 and C537 groups, and the S119 and S1058 groups.
So if this result were to hold up consistently across all six folds, I THINK the claim could be made that the lower average probabilities for the study groups are pulling more than their weight to generate the coefficients of ln(c/u) on ln(c/L), whereas the higher average probabilities for the control groups are pulling about their weight to generate the coefficients of ln(c/u) on ln(c/L). This is what I meant when I posted last night that ln(c/u) on ln(c/L) is not as weak as it seems to be.
If you want to see the details of the above, I will send you PDF's - it's too much to put into post format here.