On Aug 31, 5:31 pm, djh <halitsk...@att.net> wrote: > 1. I have a particular and a general question regarding this part of > your last post: > > ?In analysis of variance terms, all the designs mentioned above > include an additional 2-level factor: random vs nonrandom. You want > to create a dependent variable (calculated separately for the random > and nonrandom data but possibly crossing the levels of the other > factors) that will be meaningful (in the sense mentioned earlier), > interpretable (the aspect of meaning that escaped me earlier), > and that interacts (in the anova sense) with the random-nonrandom > factor. I know of no way to automate the search for such a d.v., > and I am in no position to suggest scientifically interpretable > functions. I see my role in this as checking suggested d.v.s for > possible technical meaninglessness and attempting to provide a test > of the interaction.? > > 1.1 Particular question: > > What precisely do you mean by ?crossing the levels of the other > factors??
"possibly crossing the levels of the other factors" meant only that the formula for the d.v. might involve data from more than one cell. I don't know why you would want to do that, but it wouldn't be illegal (although it might make the analysis more difficult).
> > If the idea here is simple, could you simply reword? (Because I seem > to be blocking on what you mean ...) > > If it?s not simple, could you take a moment to explain? > > 1.2 General Question: > > I don?t think I?ll have any trouble finding at LEAST one DEPENDENT > variable that satisfies your conditions. > > But my question is the following. > > Can my INDEPENDENT variable(s) be any one or more of the various > regression coefficients, SE?s, variances, and covariances reported in > the pair of files that I?ve sent you?
Probably not. I say "probably" because I don't understand the question. Your i.v. is the R vs NR dichotomy. If you want to construct a variable using a formula that involves some sort of i.v. -> d.v. interpretation the data values, then fine, but what you end up with is still going to be the d.v. for your analysis.
> > I assume the answer to this question is ?yes?, because you wrote this > in your post of 824@8:23: > > "Since you want to talk about regression-related relations, the > statistic you construct should be something for which the result of > the random vs nonrandom comparison would not be implied by, or an > artifact of, differences in the univariate or bivariate > distributions of the predictors. That probably eliminates anything > that is purely correlation-related, leaving only functions of the so- > called "structural parameters" -- the regression coefficients and the > error variances. (In this regard, note that intercepts are regression > coefficients)." > > But I just want to check to be absolutely certain that I correctly > understood what you wrote. > > 1. You wrote: > > ?The overlap of the random and nonrandom dicodon sets can probably be > ignored, but if you can identify the posts in which the randomization > was discussed then I'll have another look at the matter. My guess is > that my earlier approval was (partially) a function of the amount of > overlap being not too far from what would be expected by chance, but > now the potential problem is the actual amount of overlap.? > > The relevant posts are: > > May 28, 1:37 (#300) mine > May 28 1:35 (#301) yours > May 28, 2:46 (#302) mine > > (The post numbers are the ordinals when you scroll through the thread > ?flat?, not in tree view.) > > If you need to know the degree of overlap between: > > the non-random set 63 and the random set R63 > the non-random set 119 and the random set R120 > the non-random set 60 and the random set R58 > > please let me know and I will compute it. > > Also, note that although R63 is a set that was constructed as per my > post of May 28@1:37, R58 was constructed from R63 in exactly the same > way as 60 was constructed from 63, and R120 was constructed fom > (R63,R58) in exactly the same way as 119 was constructed from (63,60). > > In particular: > > a) 60 is the set of dicodons obtained by ?reverse complementation? of > the dicodons in 63, and R58 is the set of dicodons obtained by > ?reverse complementation? of the dicodons in R63. (Reverse > complementation simply means to: 1) write a dicodon in reverse order; > ii) translate this reversed string by the rules t->a, c->g, a->t, g->c.) > > b) 119 is the union of 63 and 60, and R120 is the union of R63 and > R58. > > c) there are only 60 dicodons in R60, not 63, because the dicodon D? > resulting from reverse complementation of a dicodon D in 63 was > dropped if it contained a stop codon or if D? happened to be in 63 > > d) there are only 58 dicodons in R58, not 63, for the same reasons as > stated in (c) for the sets 63 and R60.