Date: Tue, 6 Mar 2001 14:46:53 -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: multivariate normality test ?
Content-Type: text/plain
Nothing is bothering me. I have to tell you that my time is
valuable (perhaps not to you, but to me and my employer). So,
if you want assistance, I have to find out if it is worth
expending my effort to do so. It is worth expending my effort
if I can learn something while doing so. Otherwise, it is not
worth my while. Sometimes I reply to posters questions without
expecting to learn anything new because I can do so relatively
quickly. In the case of MVN testing, you are not satisfied
with the Mardia test that you have available to you. I asked
(twice, and politely I believe) if you would share why you felt
it necessary to perform a test for MVN. You were the one
soliciting help, not I. I don't see too many people responding
to your question, but I did take some time to ask more about
your need. I asked you to think about your data, and what you
would do with it if the test for MVN failed. For this I am
asked what is bothering me? I would turn the question around:
what is bothering you? Or, I could get really personal and ask
you to sign your name to your postings. Something to think
about!
Dale
>Date: Tue, 6 Mar 2001 15:56:05 -0600
>Reply-To: yogab <yogab@HOTMAIL.COM>
>From: yogab <yogab@HOTMAIL.COM>
>Subject: Re: multivariate normality test ?
>To: SAS-L@LISTSERV.UGA.EDU
>
>Ha.... I dont know what is bothering you. All we are asking is
>for some help or light on this MVN testing. If you could or some one
>could share some expericence of MVN it would be great. If you think
>otherwise its fine too.
>
>I can understand from your reply that its difficult proportion
>to test for MVN, because of low tollerance.
>
>thanks for sharing you experience.
>
>yoga
>
>
>
>
>Dale McLerran wrote:
>
>> My own experience is that testing normality assumptions for a
>> single variable is problematic. Given sufficient sample size,
>> it is almost impossible to accept the null hypothesis that the
>> single variable is normally distributed. That is, you will almost
>> always find a "significant" departure from normality even for a
>> single variable that you know is normally distributed. Testing
>> for multivariate normal is a more difficult proposition than
>> testing for univariate normal.
>>
>> Multivariate normal is a very strict assumption that in all
>> likelihood will not be strictly met. Quite often, tests which
>> "require" multivariate normality assumptions for obtaining
>> the distribution of the test statistic will tolerate some
>> departure from MVN. Just saying "In our experiment it is critical
>> to establish MVN" doesn't inform us of anything that we didn't
>> already know. I am not trying to be rude, but I am trying to
>> press you into telling us more about your situation. Why is it
>> critical in your experiment to establish MVN? And, if you can't
>> establish MVN, what will happen to your experiment? Will you
>> have to just throw your hands up and say "There is no way to
>> properly analyze these data because we could not establish MVN"?
>>
>> Dale
>>
>> >Date: Sat, 3 Mar 2001 12:33:33 -0600
>> >Reply-To: yogab <yogab@HOTMAIL.COM>
>> >From: yogab <yogab@HOTMAIL.COM>
>> >Subject: Re: multivariate normality test ?
>> >To: SAS-L@LISTSERV.UGA.EDU
>> >
>> >Thanks for replying my message. In our experiemnt its crictical to establish
>> >MVN. Why do you think its difficult to estabilish. ? Looking at the
>> >literature
>> >i could see tests with higher convergence, so iam not really clear of these
>> >test.
>> >
>> >If you could throw some light on any reference,comments of your
>> >experience or links would be grateful.
>> >
>> >thanks in advance
>> >
>> >yoga@tamu.edu
>> >
>> >
>>
>> ---------------------------------------
>> 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/
---------------------------------------
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/
| 
