Richard,
I am far from an expert on bootstrap methods, and to make matters
worse, you did not give us (or at least me) enough information to know
exactly what you colleague did. Never the less, at the risk of making a
fool of my self, I think your colleague may be barking up the wrong
tree. Rather than using bootstrap methods to generate standard
deviations of the dissimilarity indices, perhaps he (or she) should be
using bootstrap methods to calculate a large number of estimates of the
dissimilarity indices. The confidence limits would then be the range of
each index that includes 95% of the values, i.e. the range that excludes
the lower and upper tails of the distribution. Your colleague could then
explore each interval to see if they include zero.
I hope this helps.
John
John Sorkin M.D., Ph.D.
Chief, Biostatistics and Informatics
Baltimore VA Medical Center GRECC,
University of Maryland School of Medicine Claude D. Pepper OAIC,
University of Maryland Clinical Nutrition Research Unit, and
Baltimore VA Center Stroke of Excellence
University of Maryland School of Medicine
Division of Gerontology
Baltimore VA Medical Center
10 North Greene Street
GRECC (BT/18/GR)
Baltimore, MD 21201-1524
(Phone) 410-605-7119
(Fax) 410-605-7913 (Please call phone number above prior to faxing)
[log in to unmask]
>>> Richard Bailey <[log in to unmask]> 1/15/2007 12:06 PM >>>
I have a question regarding an analysis a colleague h
Dear allstaters,
I have a question regarding an analysis a colleague has carried out. As
far as I understand it, he has calculated indexes of dissimilarity
between three species of tree (three individual trees of each species)
based on the species identities and abundances of a large number of
insects captured in each tree. He has used bootstrapping, producing 1000
bootstrap samples, to produce standard deviations for the dissimilarity
index for each of two comparisons (tree species A versus B; A versus C;
there is no reason to test B vs C). He has then carried out a t test
using these standard deviations and assuming 1000 degrees of freedom
(for the 1000 bootstrap samples). The result is highly significant,
despite in one case the standard deviations strongly overlapping.
I'm sure there's something wrong with this, but I can't put my finger
on what. I can't see how there can be 1000 degrees of freedom - these
bootstrap samples are not 1000 independent data points, are they?
I would have used (nonparametric) bootstrapping to produce a curve for
the dissimilarity index for each comparison (A vs B; A vs C), and
examined whether there is any overlap between the least similar 95% of
bootstrapped values for A vs C (expected to be more dissimilar) and the
most similar 95% of values for A vs B (expected to be more similar).
Any advice gratefully received!
Richard.
___________________________________________________________
New Yahoo! Mail is the ultimate force in competitive emailing. Find out
more at the Yahoo! Mail Championships. Plus: play games and win prizes.
http://uk.rd.yahoo.com/evt=44106/*http://mail.yahoo.net/uk
Confidentiality Statement:
This email message, including any attachments, is for the sole use of
the intended recipient(s) and may contain confidential and privileged
information. Any unauthorized use, disclosure or distribution is
prohibited. If you are not the intended recipient, please contact the
sender by reply email and destroy all copies of the original message.
|