need help with model building

Main Page | Report this Page

Science Forum Index » Statistics - Education Forum » need help with model building

Page 1 of 1

Author

Message

Jean K

Posted: Fri Feb 08, 2008 2:07 pm

Guest

Hi,

I have searched the discussions in this group and still need help
building a predictive model. If anyone can respond I would really
appreciate it!

I want to find the significant predictors of Depression scores in
caregivers of patients.

Possible predictors:
sex
age
relationship to care recipient
tumor type (in care recipient)
income
emotional stability
social support
economic burden
neuropsych functions (5 subscales)

I have heard from another statistician to run univariate regressions
first on each predictor and then choose the significant ones to
include in the model. Then use backwards elimination regression to get
the final model. Given that these predictors are selected by my the
researchers to be included in the model, how can I go about building a
model? I am still quite a novice at model building, especially after
reading about the horrors of stepwise regression, I am unsure how best
to proceed.

Thanks!

Herman Rubin

Posted: Fri Feb 08, 2008 10:25 pm

Guest

In article <f13eda1e-c345-4dec-aa18-df288c35faf4@u10g2000prn.googlegroups.com>,
Jean K <cjkuo584@gmail.com> wrote:

Quote:

Statistics cannot do your thinking for you; you have to
use your judgment about the model.

Anyhow, a predictor can be quite important and also
insignificant by itself. A good example is the effect
of temperature and rainfall on early growth; temperature
seems irrelevant by itself, but the negative correlation
between temperature and rainfall makes temperature
important in the combined model.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@stat.purdue.edu Phone: (765)494-6054 FAX: (765)494-0558

Richard Ulrich

Posted: Fri Feb 08, 2008 11:59 pm

Guest

On Fri, 8 Feb 2008 16:07:44 -0800 (PST), Jean K <cjkuo584@gmail.com>
wrote:

Quote:

Since you don't have a randomized study, you are looking
at a sample of convenience, and the only conclusions that
you can draw in an interesting way are the ones that you
are willing to investigate and wonder about and philosophize
about, as some sort of advocate. I'm glad you have seen
comments on stepwise -- but the main alternative is
"knowing what you are after."

So, you have to consider what is primary, versus what may be
secondary or have a deeper use (but is not interesting alone).

That is, if I was looking for something publishable here,
I would figure that the interesting outcomes - from that
whole list - might be sex, age, and income/economic burden.
The rest are potential "confounds", and might be considered
for whatever they *add* to the main findings.

As Herman is wont to say, the PI should define the problem;
so, consider my personal-take as what *I* would do as a
would-be PI, given my own knowledge, etc. But you do
need to construct and define the problem before you
lay on with statistical tools.

--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html

Page 1 of 1 All times are GMT - 5 Hours The time now is Fri Aug 29, 2008 4:16 pm Science Forum Index