Re: Richard Ulrich EXPOSED his own Ignorance in LINEAR Regression Models and other matters in Statistics!
- From: jweedon@xxxxxxxxxxxxx
- Date: Thu, 22 Dec 2005 07:57:03 -0500
On Wed, 21 Dec 2005 22:32:35 -0500, Richard Ulrich
<Rich.Ulrich@xxxxxxxxxxx> wrote:
>Who decides what to *call* a "linear regression model"?
Well, in the context of a statistics discussion group, statisticians.
And there's a particular reason for the (admittedly somewhat
confusing) standard nomenclature, which comes down to the issue of
tractability of the mathematics:
Any polynomial regression model that is linear *in the parameters* and
has errors that are normally and independently distributed, is an
instance of a single, fairly simple and well understood mathematical
model, whose parameters (along with their respective asymptotic
covariances & p-values) can be easily estimated using non-iterative
OLS methods. This is known as the general linear model.
Further generalizations, e.g., to non-normal distributions and
correlated error terms, are less tractable and have smaller and more
recent literatures. These are generalized linear models.
On the other hand, models that cannot be specified linearly in the
parameters can't be counted on to have by any of these relatively
simple & well-known properties, and are therefore generally much more
of a headache. That's why the definition.
JW
.
- Prev by Date: Post-Stratification Variance Help
- Next by Date: Re: Etymology of mean
- Previous by thread: Post-Stratification Variance Help
- Next by thread: Need help with evaluation/testing (probably simple)
- Index(es):
Relevant Pages
|
