On 10 Apr 2004 19:41:22 -0700, [EMAIL PROTECTED] (Roger Levy)
wrote:

> Richard Ulrich <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...
> > On 10 Apr 2004 09:37:16 -0700, [EMAIL PROTECTED] (Roger Levy)
> > wrote:
[ snip, some]
> > 
> > "There is no precise answer to this question, but the more 
> > independents, the more likelihood of multicollinearity. 
> > Also, if you have 20 independents, at the .05 level of 
> > significance you would expect one to be found to be 
> > significant just by chance. A rule of thumb is that there 
> > should be no more than 1 independent for each 10 cases in 
> > the sample. In applying this rule of thumb, keep in mind 
> > that if there are categorical independents, such as 
> > dichotomies, the number of cases should be considered to 
> > be the lesser of the groups (ex., in a dichotomy with 
> > 500 0's and 10 1's, effective size would be 10). "
> > ---- end of extract from Garson.
> 
> Thanks for the reference, but unfortunately it doesn't seem to use the
> term "non-case" at all so this doesn't really answer my question.

Once again, I wonder at how this communication is missing.
It seems clear to me that I am talking about the dichotomy of
the criterion, the 0s and 1s, and the extract is, too.  We can
call them Cases/Non-cases, or something else.  The total N
is the sum; but the effective N, for looking at power and 
robustness, is the smaller of the two numbers.

> 
> > 
> > You might find the whole document interesting to scan.
> > [snip, more of mine]
> > > 
> > > By a "distinct covariate vector" I mean the following: with n
> > > covariates (i.e., predictors) X_1,...,X_n, a covariate vector is a
> > > value [x_1,...,x_n] for a given data point.  So, for example, if I
> > > have a half-dozen binary covariates, there are 2^6=64 logically
> > > possible covariate vectors.
> > 
> > Now I wonder what computer program you are using. 
> 
> I'm sorry, but I fail to see how that is relevant at all.  The idea of
> a covariate vector is high-level and theoretical and has nothing to do
> with implementation.  As a term, incidentally, "covariate vector" is
> quite widespread (try searching for it in "Books" on Amazon) and I
> don't see where the confusion is arising.

I know about the ordinary covariate vectors.  But I think ML Logistic 
packages of the last decade have not worried about them, 
the 2^6 set.  Maybe I need to re-read some documentation, 
but that is why I wonder if you are working with old documentation.


-- 
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/                    .
=================================================================

Reply via email to