Date:         Thu, 17 May 2001 11:05:03 +0200
Reply-To:     "Beyersmann, Jan /BDF HAM" <BeyersJ@HAMBURG.BEIERSDORF.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "Beyersmann, Jan /BDF HAM" <BeyersJ@HAMBURG.BEIERSDORF.COM>
Subject:      Re: permutation test
Content-Type: text/plain

Hi,

proc multtest offers resampling methods. proc multtest was designed for
multiple comparisons, though, so you have to check if it suits your problem.

Jan

> -----Original Message-----
> From: Changxi Li [SMTP:lichang@GPU.SRV.UALBERTA.CA]
> Sent: Wednesday, May 16, 2001 9:41 PM
> To:   SAS-L@LISTSERV.UGA.EDU
> Subject:      permutation test
>
> Hi, there,
>
> I am doing a variance analysis using PROC GLM. The procedure gave me the F
> values and the corresponding p-values for testing the effect of a
> treatment. I understand that we can choose a p-value threshold (like 5%)
> to reject or accept the hypothesis. However, I am wondering if we can do
> a permutation to generate an empirical p-value threshold based on my
> data. The idea is to randomly shuffle the observation data while keep the
> independent variables unchanged and assign the shuffled data back to each
> individual and do the same variance analysis, and then output the F or P
> values. After repeating the process for about 1000 times, we would have a
> distribution of F or p-values. Then we can choose an empirical p-value
> threshold to decide if we should reject the hypothesis. Can we do that
> by writing a SAS program and how? Your help will be highly appreciated.
>
> Thanks,
>
> Chang