Re: Model Size? Please Help

Dear ***,

Thanks a lot for your suggestion, I was also thinking about tobit,
however I am using SPSS and I think it is not possible to perform a
tobit with SPSS 13.0. In any case I greatly appreciate your help.

Nurhan Seyahi, MD

richardstartz@xxxxxxxxxxx yazdi:
On 3 Sep 2006 13:51:44 -0700, "nseyahi" <nseyahi@xxxxxxxxx> wrote:

Thanks a lot Bob,

Well I measured calcification score on subjects, Calcification can be
either absent (then you have a score of 0) or present (then you measure
the extend of the calcification). As you can guess the data is highly
skewed to left (because of all this zero values) I used ln+1
transformation but it is still skewed. As u suggested I was used
logistic regression, and I don't feel comfortable with the idea of
using linear regression on this data. But a referee (probably non
statistician) and the editorial committee asked me to perform a
logistic regression; that is the problem.

Nurhan Seyahi, MD



nseyahi wrote:
I examined 100 subjects, and measured a quantitative variable. However
the measured value was zero in 85 of the subjects, therefore I have 15
quantitative values. I am using log+1 transformation and I want to
perform a multivariate linear regression analysis. My question is that
given the sample size of 100 can I use up to 10 predictive variables in
my model or must I use up to 2 predictive variables, because in 85 of
the measurements the value was 0 (but we still have a measurement). Any
help will be greatly appreciated.

I'm not sure I would use regression on this data anyway. You could
consider using logistic regression, with the response being whether the
observation was zero or not.

There are probably other alternatives: it might help if you described
the measurements a bit more, so that we have more to go on.

Bob

--
Bob O'Hara
Department of Mathematics and Statistics
P.O. Box 68 (Gustaf Hällströmin katu 2b)
FIN-00014 University of Helsinki
Finland

Telephone: +358-9-191 51479
Mobile: +358 50 599 0540
Fax: +358-9-191 51400
WWW: http://www.RNI.Helsinki.FI/~boh/
Journal of Negative Results - EEB: www.jnr-eeb.org

You might want to try a tobit which, speaking loosely, estimates the
probability of being zero as one issue and estimates a linear
regression for the non-zero items.
-*** Startz

.

Quantcast