I've always though that using the Box-Cox method (using the likelihood-profile, e.g., as done in MASS) is a good way to find out about how to stabilize the variance (and thus implicitly get a grip on what the variance function looks like).
Somebody recently queried me that if the Likelihhood profile assumes uses the Normal distribution, then there may be a problem in finding the transformation that acheives variance-stabilization - since, e.g.., with gamma distributed data the log-transform transform stabilizes the variance while the cube-root attains approximate normality.
So my question is:
1) Is the box-cox likelihood-profile method (e.g., as in boxcox()) useless in finding a transformation to stabilize variances?
2) If so, what is a good method to find a power-transformation to stabilize the variance?
Ottar Bjørnstad
--------------------------------------------
Ottar N. Bjornstad http://www.uio.no/~ottarnb/
Until 2nd of April: NCEAS, 735 State St., Suite 300, Santa Barbara, CA 93101-3351
Tel. + 1 (805) 892 2511, Fax. + 1 (805) 892 2510
After that: Div. Zool., Dept. Biol., Univ. of Oslo, Box 1050 Blindern, N-0316 Oslo, Norway, Tel. + 47 22858370, Fax. + 47 22854605, mailto:ottarnb@bio.uio.no
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news