Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Bad advice in regression tutorial: Using stepwise to fix collinearity problems

169 views
Skip to first unread message

Bruce Weaver

unread,
Apr 14, 2010, 5:17:44 PM4/14/10
to
The tutorial on multiple linear regression (available via Help) starts
with a model containing 10 explanatory variables (main effects only)
in which there are some indications of multicollinearity. It then
says:

"Now try to fix the collinearity problems by rerunning the regression
using z scores of the dependent variables and the stepwise method of
model selection. This is in order to include only the most useful
variables in the model."

Anyone who is contemplating following that advice should take a look
at *this* advice first:

http://www.stata.com/support/faqs/stat/stepwise.html

Frank Harrell's 6th point is particularly interesting in light of the
advice given in the SPSS tutorial:

6. It [stepwise] has severe problems in the presence of
collinearity.

Another good resource is Mike Babyak's article on overfitting
regression models:

http://www.psychosomaticmedicine.org/cgi/content/abstract/66/3/411

--
Bruce Weaver
bwe...@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/Home
"When all else fails, RTFM."

0 new messages