Topic:
The same four proportional weighting factors work for each 00/01/10/11 when 0.25 is subtracted from each !!!
Replies:
506
Last Post:
Nov 20, 2012 9:21 PM
Thanks for the clarification of your last post, and also for the "editing" of my intuitive version of what you said re the x1 and x2 AAR t-test results for the S63 and S60 groups.
You asked:
"We don't usually talk about the significance of the difference between p-values. You seem to want to say something more than "something is going on with x1, but there's no evidence that it's also happening with x2". What do you want to say?"
In response, let me say first that my answer will be quite different than what it would have been BEFORE I obtained the t-test results (shown in the three tables in my last post) for the actual variables ln(c/e), ln(c/u), and ln(c/L) for segment data collected using the S63 study group and its energetic equivalent, the S60 study group.
Before these results for ln(c/e), ln(c/u), and ln(c/L) were in hand, i.e. when I originally asked you to look at the t-test results on the AAR's for the x1 and x2 drivers for the S63 and S60 groups, I was simply hoping that you could use these t-test results to find some substantive and acceptable way to distinguish the S63 group from the S60 group (which you succeeded in doing via the K-S test and also the observation about orderings of p-values and differences between the S63 and S60 means.)
But now that these results for ln(c/e), ln(c/u), and ln(L), are in hand, I need to answer the question "What do you want to say?" by asking you another question, which I?m of course hoping will have an affirmative answer.
The question is as follows.
On the one hand, we know from the last three tables I posted that the S63 and S60 groups differ significantly when ln(c/L) and ln(c/e) are t- tested, but not when ln(c/u) is t-tested.
On the other hand, we know from your K-S analysis of the two x1 and x2 AAR tables that S63 and S60 ttest differently when the residuals of the x1 driver are t-tested, but not when the residuals of the x2 driver are t-tested.
So to someone as naive and ignorant as I am, it seems intuitively reasonable that these two results should somehow be related (because it?s the x2 driver that?s ln(c/u) on ln(c/L, whike the x1 driver is ln(c/e) on ln(c/L), and it?s therefore the ln(c/u) variable that?s producing less of a difference between the S63 and S60 groups in bothe cases.)
But can you relate the two results thru some kind of formal deductive process, i.e. ?if this is so, then this must be so, and so this must also be so, etc.?
If you can, it would be wonderful for the development of Paper I.
Finally, I want to mention one other thing regarding the two AAR tables and three variable t-test result tables. Since the results on the variables themselves are so strong, you?re probably irritated that I?m not yet taking your advice to drop the effort to use the dichotomized AAR?s as logistic regression predictors and start using the underlying variables themselves, i.e. ln(c/u), ln(c/e), and ln(c/ L) rather than the dichotomized AARs for ln(c/e) on ln(c/L) and ln(c/ u) on ln(c/L).
My answer at present, subject to further elaboration and refinement in the future, is that we?re dealing with a situation in which several billion years of mutation have managed to destroy a lot of constancy that might otherwise be in the data, and for that reason, it may actually be better to try and deal with the data by finding subsets of the data that DEPART from norms in the same way. And this is, of course, is what the residual dichotmization identifies ? subsaets of the data that depart in the same ways from the norms established by the two driver correlations.
