On Sep 25, 3:45 am, djh <halitsk...@att.net> wrote: > Thanks for the additional review - much appreciated as always. > > A few questions about the custom heteroscedastic t-test (CHTT) itself. > > 1. Minimal number of rows > > The CHTT is currently being executed with 12 x rows and 12 y rows. > What is the minimum number of rows required? Is 12 skirting the > acceptable lower limit, or would 5 or 6 rows suffice?
There's no fixed minimum number of rows. Even if there's only 1 row, the test will still be valid. If there is a limitation, it's with the df, which uses Satterthwaite's approximation and works best with large df.
> > 2. "Horizontal" Equitability of N's > > For a given pair of rows, what criterion decides when the two N's are > equitable? That is, what criterion decides that in the comparison for > a given row i, Nx and Ny are sufficiently close to rule out difference > between them as an interfering factor in the analysis? Is this an > "absolute criterion" or one that scales up or down with the average > sizes of the two N's, or something like that?
There is no need for N's in the same row to be close.
> > 3. "Vertical" Equitability of N's. > Suppose you have twelve rows in which each (Nx,Ny) pair is > "horizontally equitable" according to whatever criterion you provide > in answer to (2). Do the rows ideally have to be "vertically > equitable" as well, e.g. does the average size of (Nx,Ny) for row i > have to differ less than a given amount from the average size of > (Nx,Ny) for row j.
There is no need for the average N's to be close.
> > 3. Definition of rows by length intervals. > > You have always been suspicious about the use of length intervals to > define subsets of observations. If I can find another "natural" > definition of "row", do you see any reason why that definition could > not be used instead of length interval?
No problem, provided the effect of Length was negligible within each "row".
> > Thanks as always for considering these questions and the extra time > you'll have to spend formulating your answers in a way a novice can > understand.