Date:         Wed, 25 Oct 2006 23:18:41 -0400
Reply-To:     Kitty Lee <lee.kitty@YAHOO.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Kitty Lee <lee.kitty@YAHOO.COM>
Subject:      Bootstrap OLS

I have a question about using bootstrapping to do hypothesis testing in OLS.
Say for example, I have

Y=a+b1X1+b2X2+e

I'm asked to test whether b1=b2. Normally I would run 2 regressions, one
without restriction, and another with restrict b*=b1+b2 and do F test. But
someone told me I could do the same test with bootstrapping.

The idea is that I can first calculate

ts=b1-b2 from the full sample. Use ts as the critical value for rejection
region.

Do bootstrapping, calculate bs=b1-b2 from bootstrap sample.

Then see how many bs-ts falls to the right (or left) of the critical value.

Does this make sense?

I wasn't 100% convinced.  From my full sample, ts=b1-b2=10. If I do the
above, what exactly am I testing? Am I testing b1=b2 or simply b1-b2=10?


K.