Date:         Thu, 29 Mar 2012 20:13:37 -0400
Reply-To:     Rick Wicklin <Rick.Wicklin@SAS.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Rick Wicklin <Rick.Wicklin@SAS.COM>
Subject:      Re: Improving bootstrap performance of sas code

Good tips to remember. Thanks, Roland.

I've previously mentioned David Cassell's SGF 2007 paper on bootstrapping
(http://www2.sas.com/proceedings/forum2007/183-2007.pdf), but the first
published article that I've read that mentions these techniques in a
general context is
Novikov, I., 2003, "A Remark on Efficient Simulations in SAS,"
Journal of the Royal Statistical Society. Series D (The Statistician),
52(1), 83--86
http://www.jstor.org/stable/4128171
(To be fair, Novikov references a 2000 paper, and I think it was the late
90's when Phil Gibbs at SAS started encouraging SAS customers to use this
approach.)

In his paper, Novikov points out that for his simulation examples the BY
processing approach might be 80 times more efficient than the macro loop
approach. And, as I point out in my course on Simulation of Data in SAS,
it has the added bonus of enabling you to run the simulation in an
interactive session without filling up your log. Plus, as Roland points
out, it unifies the concepts of bootstrapping and simulation (and
imputation, too, if you think about it!).

By the way, there are only a few seats left for my SGF tutorial this year,
so if you want to take a course on Simulation in SAS, you should sign up
soon:
http://support.sas.com/events/sasglobalforum/2012/sunday.html#rw

Best wishes,
Rick Wicklin
Statistical programming and SAS/IML blog: http://blogs.sas.com/content/iml