Hi.

There has been all sorts of advice on the wisdom of keeping or removing outliers. Some of that advice has included discussion of what actually constitutes an outlier. It's a fuzzy definition, but whatever one ends up doing, part of exploratory data analysis should include detection of outliers. Paige Miller and Ron Fehd in particular mentioned SAS procedures to use in outlier detection. I'd like to add a bit to this. I am of the school that believes outlier detection is best practiced as a "holistic" discipline, based on all the data to analyze, rather than looking at each variable separately. The ROBUSTREG and PLS PROCs mentioned by Paige allow one to do this (I didn't check the program link Ron provided). There are also some functions in PROC IML that take a holistic approach to outlier detection, MCD and MVE. It's cool stuff, but has a bit of a steep learning curve.

-- TMK -- "The Macro Klutz"

On Wed, 14 May 2008 08:29:43 -0700, Eversmann <rifazrazeek@GMAIL.COM> wrote:

>hi all, > >this is more of a stats question (than SAS only..).. > >i was wondering whats the best way to remove outliers (extreme values >in your data)... > >at the moment i am using percentiles (p25, p75 etc.. in proc >summary)... problem is what happens when you have only 2 or 3 >values... ? i am reading some sas meterial as to how this works.. > >but in general .. what are the best ways of removing outliers... may >be you higher level decicion makers can post a few comments... > >many thanks