Date:         Fri, 20 Jun 2008 15:16:59 -0500
Reply-To:     Robin R High <rhigh@UNMC.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Robin R High <rhigh@UNMC.EDU>
Subject:      Re: One-way Anova
Comments: To: Y Zhu <yzhu97@YAHOO.COM>
In-Reply-To:  <200806201552.m5KAl46e004686@malibu.cc.uga.edu>
Content-Type: text/plain; charset="US-ASCII"

Zhu,

You didn't say much about the data (sample sizes) or which variance
equality test you applied -- if it was Bartlett, I suggest you start over
and look at the results with the Levene test with absolute values:

 HOVTEST=levene( TYPE= abs )

Though, even if marginally "significant", ANOVA is likely to be robust
against small departures from equal variances.  If you want to take into
account the unequal variances, you can applied Weighted Least Squares with
a WEIGHT statement (which requires some external work ahead of the PROC
GLM code) or choose the welch option. With 8 groups you may want to
consider unequal variance ANOVA with PROC MIXED (syntax very similar to
GLM) that tests for unequal variances in all or a subset of the groups
with a REPEATED statement and GROUP= option.  MIXED allows you to do
multiple comparisons with the LSMEANS statement and adjust=simulate option
(a recommended choice for the unequal variance case).  The mixed
documentation should have details on how to implement it.

Robin High
UNMC







Y Zhu <yzhu97@YAHOO.COM>
Sent by: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
06/20/2008 10:53 AM
Please respond to
Y Zhu <yzhu97@YAHOO.COM>


To
SAS-L@LISTSERV.UGA.EDU
cc

Subject
One-way Anova






Hi,

I am using PROC GLM for one-way Anova. I have 8 groups in my data and the
test for equality of variances does not hold. Data is normally
distributed. What would be an alternative to perform ANOVA in this
situation and do pariwise comparison to get confidence intervals ?

Thanks,
Zhu