12-16-2012 08:23 PM
I have never run the stepwise /backward regression before. I am wondering on what basis I can decide the values for slentry (sig level to allow the variable in the model) and slstay (sig level to keep the variable in the model).
Any guidance/help will be appreciated.
12-17-2012 08:05 AM
Guidance #1. Don't do backward/forward/stepwise regression.
Guidance #2. See guidance number 1.
There is an abundant number of posts and papers regarding the dangers of this method. The resulting models have poor predictive ability and the regression parameters range from moderately to severely biased. See the NESUG paper by Flom and Cassell: "Stopping stepwise: Why stepwise and similar selection methods are bad, and what you should use" http://www.nesug.org/proceedings/nesug07/sa/sa07.pdf
Still if you must deliberately use a poor tool, stick to the default values.
12-17-2012 05:47 PM
Thanks Steve for advising me on this! It's great to know about the latest guidelines on SAS/STAT procedures.