-----Original Message-----

From: jsm_44092 [mailto:

tpr42345@...]

Sent: Friday, January 02, 2004 10:16 PM

>When you state "multivariate regression" I assume you mean one

>dependent variable with multiple explanatory variables. Correct?

>That is "multiple regression". Multivariate regression, which is

>performed by running the univariate regressions, is when there is

>more than one dependent variable. The two terms are often used

>interchangeably in certain literature, but they are not the same.

The terminology may be discipline-dependent. In econometrics,

I usually see the term "multivariate regression". See, e.g. MIT's

calendar for the syllabus for its undergrad econometrics class, 14.32:

http://ocw.mit.edu/OcwWeb/Economics/14-32EconometricsSpring2003/Calendar/index.htm
After the usual review of probability and statistical inference, the

real meat of the course starts with lectures 6 - 11, which start

with the simple "bivariate regression" model (although I usually use

the term "univariate regression", and evidently you do also) and then

move to "Introduction to multivariate regression".

What you seem to be calling "multivariate regression" -- models with more

than one dependent variable, in which several regressions are run -- are

called "simultaneous equation models" in econometrics (the regressions

however need not be, and usually are not, univariate ones). This same

progression of "univariate regression", "multivariate regression", and

"simultaneous equation models" is e.g. used by Wiley to describe the

contents of one of its econometrics books:

http://www.wiley.com/WileyCDA/WileyTitle/productCd-0471802603.html
On the other hand, the econometrics class the Univ. of Virginia

evidently uses the terminology that you're advocating: "univariate

regression" being the second item on its schedule and "multiple

regression" the fifth item:

http://www.people.virginia.edu/~sns5r/classes/undergrad/econ372.html
In practice, econometricians are almost never running univariate

regressions, and usually just call something "a regression" with

the understanding that it will have multiple explanatory variables.

If there are multiple dependent variables, then it's "simultaneous

equations".

--MKT