On Nov 22, 6:24 am, djh <halitsk...@att.net> wrote: > Thanks for the correction re the squatre term .... duh! > > Am sending off-line a zip file with one summ file and 12 detl files > for the regression Ruq = lnc on (lnu,(lnu)^2). > > The summary file has 12 rows (one per length interval) with the > following col hdrs: > > Fold a1 > Frame N > DiSet 1 > SetTyp S > NmbrDics 63 > uLevl uA (all) > cTyp cA (avg) > LenID 1-12 > obsN Observation Count > uMean mean ln(u) > qMean mean q, where q = (ln(u))^2 > RuqResVar Residual Variance > RuqInt Intercept > RuqSl_u Slope for ln(u) > RuqSl_q Slope for q > RuqIntSE Intercept SE > RuqSl_uSE ln(u) Slope SE > RuqSl_qSE q Slope SE > CRuq_1_1 CoVar Cell (1,1) > CRuq_1_2 etc. > CRuq_1_3 > CRuq_2_1 > CRuq_2_2 > CRuq_2_3 > CRuq_3_1 > CRuq_3_2 > CRuq_3_3 > > The detail files are for lengths 1-12 and have the col hdrs: > > ChnID Observation ID > u Raw u > c Raw c > L Raw L > lnc ln(c) > lnu ln(u) > (lnu)^2 ((ln(u))^2 > RuqRes Associated Residual > > (You can cross-check the residual here using the coefficients in the > summary file for the corresponding length interval.) > > If I were to develop 2-ways and 3-ways as you previously instructed me > to, I would operate on files like the summary file to develop averages > per fold across length intervals, and then roll-up these intermediate > results across folds. > > But you?ve implied that something different has to be done here, so I > will await further instruction/clarification. In the meantime, I will > develop the summary files for the remaining cells of the model.
I'm looking at the detail data. A tentative preliminary conclusion is that you should probably be regressing c on u/(1+u).