[BioC] Multtest Question!

Sandrine Dudoit sandrine@stat.berkeley.edu
Tue, 11 Feb 2003 09:55:43 -0800 (PST)

We used Gaussian p-values simply as an example.
You can use the function mt.maxT to obtained permutation unadjusted
p-values as well as adjusted p-values.

In general, the Gaussian p-values should be adequate when you have
fairly large sample sizes in both groups, so that AVERAGES in both
groups are approximately normally distributed.

The assumptions behind t-distribution p-values are more restrictive:
it is assumed that INDIVIDUAL expression measures have a Gaussian
distribution. This is not very realistic.

You can find a discussion of t and z-tests in pretty much any
introductory statistics textbook. I like "Statistics" by Freedman,
Pisani and Purves.

Best regards,

Sandrine Dudoit, Ph.D.               E-mail: sandrine@stat.berkeley.edu
Assistant Professor                  Tel: (510) 643-1108
Division of Biostatistics            Fax: (510) 643-5163
School of Public Health		     http://www.stat.berkeley.edu/~sandrine
University of California, Berkeley
140 Earl Warren Hall, #7360
Berkeley, CA 94720-7360

On Tue, 11 Feb 2003, Puhong Gao wrote:

> Hi, all,
> In the documentation of the Multtest R package, the raw p values were generated based on the standard Gaussian distribution. Can someone explain to me why it should be like that? What situations can it be applicable to? The p values based on the standard Gaussian distribution are much smaller than those based on the t-distributions. The consequence of this is more genes will be picked up as significant. How can this be explained? Your inputs will be very appreciated.
> Have a great day,
> Puhong Gao, PhD
> Dept. of Developmental Psychobiology
> NYSPI/Columbia University
> 1051 Riverside Drive
> New York, NY 10032
> (212) 543-5710
> 	[[alternate HTML version deleted]]
> _______________________________________________
> Bioconductor mailing list
> Bioconductor@stat.math.ethz.ch
> http://www.stat.math.ethz.ch/mailman/listinfo/bioconductor