[BioC] ChIP on Chip normalization | Shorth

[David Lee Duewer]

>> effectively one-sided.
> "midpoint of the shorth" as a centrality parameter ...
> it  is often less susceptible to asymetric tails than mean or median

The shorth (shortest half) IS a less susceptible location estimate than the median to OUTLIERS that asymmetrically contaminate a symmetric base distribution.  This does NOT imply that the midpoint of the shorth is an adequate centrality estimate for asymmetric base distributions.  Transformations (log is most obvious) that move the base distribution towards symmetry can aid figuring out a useful, stable location estimate.  (In non-genechip applications, I've found that "median of the shorth" is "better" than midpoint, but that doesn't say that it's "good"...)

The shorth is, however, a wizard "best guesstimate" of the "if it wasn't contaminated by outliers" interquartile range (IQR) for use with robust estimation of DISPERSION.

D

David Lee Duewer
National Institute of Standards and Technology
100 Bureau Drive Stop 8390
Gaithersburg, MD  20899-8390
USA
	[[alternative HTML version deleted]]