On Oct 8, 11:35 am, djh <halitsk...@att.net> wrote: > The two tables below were computed from data obtained by running the > analysis with "C" subsets of the same cardinalities as their > corresponding "S" subsets, as per your previous OK of this design > change. > > Components of t-numerator > [(1S-1C) - (R1S-R1C)] > for mean ln(e)?s for 1:R1 > > uL uH > 1S-1C R1S-R1C 1S-1C R1S-R1C > > a1 -0.045 -0.060 -0.287 -0.041 > a3 0.170 0.074 -0.275 -0.025 > b1 -0.122 -0.030 -0.162 -0.074 > b47 0.020 0.014 -0.124 -0.061 > c1 -0.006 -0.007 -0.171 -0.101 > c2 -0.118 -0.033 -0.163 0.029
I'm glad you decided to look inside the numerator, at its components. Looking at just the numerator or (worse) its absolute value might mislead you. The best way to gain understanding is to plot the four means (nS,nC,RnS,RnC) so that the difference between the differences are apparent. Don't let the fact that there will be many such plots deter you. The more important the result is, the greater the need to look at the plots.
> > 2-tailed p's for > paired t-tests of > uL 1S-1C vs R1S-R1C ln(e) means > and > uH 1S-1C vs R1S-R1C ln(e) means > > uL .751 > uH .008 > > (Note that these results were those obtained by leveraging across all > length intervals as specified by your ?point 4? in your post of 10/2 @ > 3:8.)
I doubt it. I got the same p-values by doing two ordinary paired t's with 5 df, which is wrong. ALL tests must use the procedure from my Oct 2 post (although the df's may be so big that nothing substantial would change if you got p-values from the normal distribution instead of the t distribution).
In anova terms, you're trying to say that there is a three-way interaction, N-R x S-C x L-H; that there is a large simple two-way interaction, N-R x S-C at uH; and that the same simple interaction at uL is negligible. (The convention is that factors not mentioned in the interaction are averaged over; in this case, that means Length and Fold.)
> > So, if we had 20 rows for 20 folds in the first table above, and if we > still obtained 2-tailed p?s similar to the .751 (uL) and .008 (uH) > obtained for six folds, could we legitimately argue from this result? > (And if the answer is ?no?, could you take a moment to explain why > not ?)
??? In anova terms, you have been treating Fold as a fixed effect: the particular folds in the study were not chosen randomly, and the conclusions apply to those folds and no others. Do you want to switch now and treat Fold as a random effect, so that you can generalize to folds not in the study? If so then all the previous analyses must be done differently.
> > Thanks as always for considering these questions.