Below are the coefficients, slopes, intercepts, and confidence intervals (for slopes and intercepts) for two cases of our ln(c/e) on ln(c/L) regression, using data from the a1 (helical hemoglobin) fold for the constant ratio length interval 24-27.
The first case (S63-R) was obtained using the completely random set of 63 dicodons which I derived by the method I showed you here at sci.stat.math a few days ago.
The second case (S63) was obtained using the team's S63 set of "dicodons of interest" with which you are now quite familiar.
N 166 227 Coeff 0.645914679 0.748241963 Int 1.602960254 2.259414374 Int CI Low (95%) 1.270975726 2.005368037 Int CI High (95% 1.934944783 2.513460710 Slope 1.105286834 1.472866628 Slope CI Low (95%) 0.903241787 1.301307846 Slope CI High (95%) 1.307331881 1.644425409
As you can see:
i) there is no overlap between the CI's of the intercepts (1.934955783 high vs 2.0053680367 low);
ii) there is only an overlap of 0.006024036 between the CI's of the slopes (1.307331881 high vs 1.301307846 low).
So, can the two regressions be considered to differ in a statistically meaningul way? Or not? Or, do you need more information to answer this question, e.g. residuals, input values, etc.
Regarding these questions, please note that we will have a total of 72 such cases to evaluate when all is said and done (6 folds * 12 constant-ratio intervals.)
But I am hoping you can tell us that this first one is promising enough to bother doing the remaining 71.
One final note: the "c" used in the calculation of ln(c/e) on ln(c/L) was the new "averaged c" I mentioned in my last post. When the old "simplified c" was used, the CI overlaps were somewhat greater. So, if you do think the above case is a promising start, then it was worth going the extra mile to refine our definition of "c".