Date: Wed, 14 Feb 2001 14:16:10 -0800
Reply-To: Dale McLerran <dmclerra@MY-DEJA.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Dale McLerran <dmclerra@MY-DEJA.COM>
Subject: Re: Bartlett's test
Content-Type: text/plain
David,
In some experiments, you may run into small numbers of observations
for at least some combinations of factors. I think you could get
into trouble performing a test for homogeneity of variance when
some groups have few observations. Would you have any qualms
with fitting the full model, generating the residuals, and then
performing an HOVTEST on the residuals for each factor separately?
This may not be an optimal test, but it seems to me that it
should suffice for most problems.
Dale
>Date: Wed, 14 Feb 2001 11:12:19 -0800
>Reply-To: Cassell.David@EPAMAIL.EPA.GOV
>From: "David L. Cassell" <Cassell.David@EPAMAIL.EPA.GOV>
>Subject: Re: Bartlett's test
>To: SAS-L@LISTSERV.UGA.EDU
>
>Christopher Tong wrote:
>> Many thanks to Dale McLerran and David Cassell, particularly
>> for the latter's extensive and thought-provoking comments
>> on my question.
>
>You're welcome.
>
>> According to my version of SAS, HOVTEST is only available for 1-way
>ANOVA,
>> however. Perhaps there is no analog for higher-way ANOVA?
>
>The analog is the same, but you're running this through a single variable!
>Bear in mind where the HOVTEST option is. If you want this for a more
>complex
>structure, then you will want to build a dummy variable which has a
>different
>level for every combination of all your independent variables.
>
>> Is there a better way than just plotting the residuals vs. factor levels
>> for a given factor, and repeating for each factor? (i.e., eyeballing.)
>
>This is workable.. assuming you have enough experience looking at residual
>plots
>to be able to tell the difference between actual features and artifacts
>caused
>by the way the human mind groups data. Residual plots can give you immense
>amounts of useful info when done properly, and I strongly recommend them
>for
>all such analyses. But you can also get the test as I described above.
>Still,
>look over those residual plots. There is a lot more that can go wrong than
>heteroscedasticity, and you can see just about all the problems if you look
>at
>the right residual plots. Back when I was teaching intro stats, I always
>had
>my students do regressions and ANOVAs which had problems that residual
>plots would
>flag.
>
>David
>--
>David Cassell, OAO Corp. Cassell.David@epa.gov
>Senior computing specialist
>mathematical statistician
---------------------------------------
Dale McLerran
Fred Hutchinson Cancer Research Center
mailto: dmclerra@fhcrc.org
Ph: (206) 667-2926
Fax: (206) 667-5977
---------------------------------------
------------------------------------------------------------
--== Sent via Deja.com ==--
http://www.deja.com/
| 
