Date: Tue, 10 Jul 2007 12:01:14 -0700 Paige Miller "SAS(r) Discussion" Paige Miller http://groups.google.com Re: repeated t-test and record the p-values. To: sas-l@uga.edu <1184093411.301206.198930@57g2000hsv.googlegroups.com> text/plain; charset="us-ascii"

On Jul 10, 2:50 pm, gsantu_h...@yahoo.co.in wrote: > Hi guys, > > I need to perform several (6000) t-tests for 6000 genes and record the > p-vaues fo reach of these test in a dataset. Such that from looking at > the p-values I can say thatwhich genes are significantly different. > > Plz help me out from this problem

Whenever you do such a thing, you must realize that if you do the t- tests such that you have a five percent chance of concluding there is a difference when there is NO difference, then lets see, 6000*0.05 = 300 tests will give you the wrong answer (wrong specifically in the direction of saying there is a difference when there is no difference). And that's if the tests are uncorrelated.

In order to find a "better" procedure, you need to define what risks you want to take. In other words, what is the risk you wish to take in maiking a Type I error, and what is the risk you wish to take for making a Type II error? If you want to allow a relatively high number of Type I errors, and a relatively low number of Type II errors, then one procedure might be chosen. If you want the risks to be the other way around, a different procedure might be chosen. So I can't really help you further at this time ... you need to answer these questions about risks.

And if you don't know what a Type I and Type II error are, then that is the place to start ... look in almost any basic statistics text book.

-- Paige Miller paige\dot\miller \at\ kodak\dot\com

Back to: Top of message | Previous page | Main SAS-L page