On Sep 13, 3:58 pm, djh <halitsk...@att.net> wrote: > In your post of 9/12@6:42, you wrote: > > ?But that's a single CI. If you have n independent CIs, and you want > to be 95% confident that they all contain their respective parameters > -- i.e., that all n parameters are in the n-dimensional box whose > corners are the endpoints of the CIs -- then you must use .95^(1/n), > not .95, as the c for each individual CI.? > > And I certainly don?t feel I can make a correct decision on which > parameters of the design are independent and which aren?t. > > Let?s ignore the cS vs cA distinction for the moment, since it will > only figure in a secondary thrust of the initial paper (if in fact, > any empirically significant difference turns out to distinguish > them.) The important member of this pair is cA (because it?s both > more accurate and varied), so let?s agree to consider just cA cells. > > Next, let?s agree to ignore uA in this discussion, since as you?ve > pointed out several times, uA is independent of neither uL nor uH. > > So, before considering the regression coefficients themselves, that > leaves us with: > > a) the six folds a1,a3,b1,b2,c1,v47 > b) the uL vs uH dichotomy > c) the random vs non-random dichotomy > > for a total of 12 independent non-random combinations that can be > compared against 12 corresponding random combinations. > > Let?s stop there for a moment and consider your ?n parameters in the > n-dimensional box?. Do (a-c) imply an n of 24 or an n of 12? (Or am > I seeing this entirely wrong-headedly, and n is something else here.)
n = 12: 6 folds x 2 u-ranges. Each CI is for the difference between a random and nonrandom regression coefficient.
> > Now let?s go to the 7 regression coefficients: > > eS > eI > uS > uI > euS(e) > euS(u) > eu(I)
I thought you would want to do 7 separate analyses, one for each regression coefficient. Then you look for scientifically interesting patterns within each of the 7 plots.
> > Of the 7 x 7 possible pairs of these coefficients, you?ve already > pointed out a number that aren?t independent, e.g. eS and euS(e) or > uS and uS(u). Similarly, I assume that eS and eI are not independent, > nor uS and uI. > > But what is the complete inventory of independent parameters based on > the regression coefficients we?re considering? Whatever that number > is, I assume we multiply that by whatever inventory you derive from > (a-c). > > But again, I have no idea whether I?m even remotely thinking about > this in a correct way, so please advise what you think n should be > (and why, if you have the time.) > > Thanks again, as always.