[Note: this message will be posted on Evoldir, Ecolog-L, and the popbiol
listervs, and I apologize in advance if you get multiple copies. I will post
summaries of replies to all three lists.]
I am try to get a sense of whether a consensus is developing among evolutionary
and population biologists on the use of the "false discovery rate" (Q-value) in
multiple comparisons instead of the various forms of the Bonferroni correction.
Since Rice's 1989 paper (Evolution 43:223), the Bonferroni correction has been
widely used for adjusting the significance level of multiple comparisons to
minimize type I errors. In fact, in some fields, as I have learned the hard
way, it has become a virtual requirement for publication. But with these
corrections, the power of the analyses are lowered, perhaps artificially and
unnecessarily, and some biologists have argued that they probably should not be
used (e.g. Moran,2003; Oikos 100:443). My own work often involves large numbers
of comparisons of gene frequencies by chi-square or related statistics, to
evaluate the significance of allelic frequency differences among populations or
the potential reality of linkage disequilibrium detected by programs like
LinkDos. I have used the Bonferroni correction in the past, but often with the
feeling that I was throwing many babies out with the bathwater, especially with
preliminary surveys, which are often "fishing expeditions."
It seems to me that use of the Q value instead of the Bonferroni correction
could restore a great deal of power to multiple comparisons. [see Verhoeven et
al., 2005; Oikos 108:643 for a discussion and review of literature]. Yet I see
relatively few papers, particularly in population genetics, that use the Q
value, and many that still use the Bonferroni correction. Of course, this
apparent delay could simply reflect the fact that most papers that are in the
curent issues of journals were submitted around a year ago or more. Or it could
simply stem from the conservatism of research communities (including referees of
journal articles) that don't spend a lot of time worrying about type I vs type
II errors.
I cordially invite comment from anyone who has a viewpoint or experience with
this issue.
Bruce J. Turner
Dept. Biol. Sci.
VPISU
Blacksburg, VA 24061
(540)-231-7444 (V)
(540)-231-9307 (F)
_______________________________________________
Pop-bio mailing list
Pop-bio@...
http://www.bio.net/biomail/listinfo/pop-bio
