On Mon, 26 Apr 2004 20:27:57 +0100, "Rod" <[EMAIL PROTECTED]> wrote:
> > "Richard Ulrich" <[EMAIL PROTECTED]> wrote in message > news:[EMAIL PROTECTED] > > On Sun, 25 Apr 2004 09:48:28 +0100, "Rod" <[EMAIL PROTECTED]> > > wrote: > > > > > Strictly speaking it is still non-linear because the errors are centred > on x > > > not e^-x. Probably won't make a big difference though > > > > > > > Aren't the errors centered in Y? > > > > Isn't X assumed to be measured without error? > > (so, e^-x is equally without error.) > > Fair point. > Would still want to convince myself that these errors could all come from > the same normal distribution. > If Y is getting big (say instead, we were looking at e^(x)) this may not be > true as error is often proportional to Y. > Sure, it is true that the errors might not be normal. Or independent, or homogeneous. OLS regression might not be the efficient way to estimate it, but it is linear in the coefficients (how most statisticians use the term 'linear'). - which is not universal, by the way. I think we noted that engineers use 'nonlinear' fairly liberally. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
