This article is rated Start-class on Wikipedia's content assessment scale. It is of interest to the following WikiProjects: | ||||||||||||||||||||||||
|
Untitled
editWhy rank products as opposed to rank sums?
It seems that the product approach weights things quite differently from sums.
Both sums and products have advantages, what are the tradeoffs?
Answer: In general, both rank sums and rank products work quite well. Their performance on simulated data is compared in Breitling and Herzyk (J Bioinform Comput Biol. 2005 Oct;3(5):1171-89). The major difference is that the rank sums give high ranks to those entities that very consistently score highly, while the rank products favor those that get very high ranks most of the time, but allows some exceptions. That means that negative "outliers" get less weight in rank products that in rank sums. The choice between the two approaches depends on the aim of the study and the expected pattern of noise. E.g., the rank product approach is recommended for most gene expression studies. Note that the rank sums as discussed here are very different from the rank sums used in the traditional Wilcoxon rank-sum test.
RainerBreitling (talk) 07:59, 9 October 2008 (UTC)