Date: Nov 26, 2012 12:40 AM
Author: Ray Koopman
Subject: Re: As per your suggestion in the other thread, scaled e on scaled u,<br> c, L
On Nov 25, 8:30 pm, djh <halitsk...@att.net> wrote:

> I'm sending you off-line a PDF with the usual statistics plus line fit

> plots and residual plots for a regression of the sort you suggested at

> the end of the other thread:

>

> e on (u,c,L)

>

> where e,u,c, and L have all been scaled into [0,1] using the following

> formula and intervals:

>

> x' = (x - x_min) / (x_max - x_min)

Any such linear rescaling has no effect on ordinary multiple

regression. I suggested that rescaling only as a possible way of

normalizing the data prior to the rotation/projection analyses

described in your post of 11/15 @ 9:16pm.

>

> e: [221.735, 308.65]

> c: [1,252]

> u: [.0079,36]

> L: [2,253]

>

> I've also sent off-line the raw and scaled values for u,e,c,L

> (these are for all observations in a1_1_N_S, across all lengths).

>

> I don't know if the residual plots for scaled u and c are too

> heteroscedastic to make the regression meaningful.

>

> And even if these two residual plots are sufficiently homoscedastic,

> I don't know if you'll find the regression itself meaningful or

> interesting.

>

> But I figured I would send it along for what it's worth, inasmuch as

> scaled e varies directly with scaled u and c, which is what should be

> the case.