Thanks for clarifying "i.v" vs "d.v" - I see what you mean now. (The problem is that I'm learning statistics by learning the techniques first, not the overall theoretical framework. This is great in one sense, like learning the multiplication tables before Peano, but it also prevents me sometimes from understanding the framework in which you're making a particular remark.)
The following should make clear why I said in the post title that you're either a genius, or very very experienced, or both.
1. Nomenclature in the Table Below (Section 3)
Re = ln(c/L) on ln(c/e) Ru = ln(c/L) on ln(c/u) Reu = ln(c/L) on (ln(c/u),(ln(c/e)) Re:e = ln(c/e) predictor in Re Ru:u = ln(c/u) predictor in Ru Reu:e = ln(c/e) predictor in Reu Reu:u = ln(/cu) predictor in Reu Slv = Slope Variance IntV = Intercept Variance uL = 0 < u <=1 (uLow) uH = u > 1 (uHigh) cS = c "Simple" cA = c "Average"
2. Construction of Table Below (Section 3)
The value in every cell of the table below is a two-tailed P resulting from a Paired Two-sample t-Test of 12 cells from the a1_63_S (non- random data) spreadsheet which I've sent you against the corresponding 12 cells from the a1_R63 S (random data) spreadsheet which I've sent you.
For example, in the SlV section of the table below, the P of .00005 in the (Reu:e,uH,Ca) cell results from t-testing the values in:
a) rows 38-49 of the C_2_2 column of the a1_63_S spreadsheet
Results of t-Testing (Paired Two-Sample) the Slope and Intercept Variances Resulting from Execution of Regressions Re, Ru, and Reu on Length Intervals 1-12 in Non-Random (a1 63 S) Data and Random (a1 R63 S) Data
Since the results in the above table are readily interpretable scientifically, I hope that you can adjudge the results above as technically legitimate and meaningful.
If so, I will program generation of the above table in Perl, using the PERL module Statistics::TTEst to do the t-testing and the PERL module Statistics::Distribution to get the "P" from the "t". (When doing so, I will also include t-tests of the SlopeIntercept Covariance for the Re and Ru regressions, and the off-diagonal SlopeIntercept Covariances for the Reu regression.)
Assuming again the legitimacy of the above results, please note that in addition to generating the above table for non-random (a1 63 C) data vs random (a1 R63 C) data, the above table also has to be generated for the other five folds and the other ten dicodon sets/ subsets. It may be that the "P-pattern" changes as we move through these other data frames.
5. Comments (II)
Assuming again the legitimacy of the above results, they indicate that we cannot legitimately derive ANY logistic regression predictors from the Re and Ru regressions, at least not when assessing the structural alignability of subsequences in the (a1 63 S) data frame. As we run the other folds and dicodon set/subsets, it may be that this restriction continues to be "in force", or it may be that other data frames allow us to derive logistic regression predictors from the Re and Ru regressions. (I suspect that this will turn out to be the case, based on the systematic results we obtained for different folds in our original investigations of structural alignability results via different logistic regression predictors.