-------- Original Message --------
Subject: RE: Low sample size problem
Date: Mon, 3 Nov 2008 07:52:38 -0800 (PST)
From: Fred Sundstrom <[EMAIL PROTECTED]>
To: <[email protected]>
References: <[EMAIL PROTECTED]>

Obviously a good point!

What I am looking for is a way to distinguish between my groups (ie find out
from which of the five groups a randomly chosen fish comes from) and so my
idea was to use discriminant analysis to see how well the morphmetric data
can do that for me.

I am also using "specialists colleagues" to do this by eye from the same
photographs, but they have not been very good at it (chimp level). So my
second idea was to use the morphometric data to try and visualize the shape
differences with TpsRelw using the first two RW and "teach" my colleagues
(attracted by the smell of coffee and cookies...) what to look for and then
have them look at my photos again.

My (possibly erroneous) analysis using the first 6 RW and body length
classify fish with an accuracy of 72-100 % with the group with the lowest
sample size being the 100 % ones.

Cheers,
Fred


-----Original Message-----
From: morphmet [mailto:[EMAIL PROTECTED]
Sent: den 3 november 2008 04:04
To: morphmet
Subject: RE: Low sample size problem



-------- Original Message --------
Subject: RE: Low sample size problem
Date: Sat, 1 Nov 2008 17:32:37 -0700 (PDT)
From: F. James Rohlf <[EMAIL PROTECTED]>
Reply-To: [EMAIL PROTECTED]
Organization: Stony Brook University
To: [email protected]
References: <[EMAIL PROTECTED]>

The obvious question is what do you plan to do with the relative warp
scores? If you wish to use MANOVA or CVA to test for differences between
the two means then there is a sample size problem. However, you could
simply compute the Procrustes distance between the two means and then do
a permutation test to see whether this distance is unusually large. That
test does not require a full rank covariance matrix. In many cases this
test does not have as much power as the standard test but it may be
sufficient in your case.

------------------------
F. James Rohlf, Distinguished Professor
Ecology & Evolution, Stony Brook University
www: http://life.bio.sunysb.edu/ee/rohlf


-----Original Message-----
From: morphmet [mailto:[EMAIL PROTECTED]
Sent: Friday, October 31, 2008 3:35 PM
To: morphmet
Subject: Low sample size problem



-------- Original Message --------
Subject:        Low sample size problem
Date:   Fri, 31 Oct 2008 11:34:02 -0700 (PDT)
From:   Fred Sundstrom <[EMAIL PROTECTED]>
To:     <[email protected]>



Dear List



I have a problem with low sample sizes and request your advice.



I have used TpsDig2 to take 13 landmarks on fish from five different
groups. However, in one group I only have 8 specimens and in another
only 12 and unfortunately I cannot get more of these samples.

I understand that I should not used all 22 partial warp scores to
analyze this. My idea the is then to use the first 6 or 7 relative
warps, but is this ok? Or is this is even advisable or is there a
better
(alternative) way of dealing with this problem of more landmarks than
specimens?



Any input is much appreciated!



Best fishes,

Fred Sundstrom












--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org




--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org



--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org

Reply via email to