Commonly, the hypergeometric test and Fisher's exact test are two main statistical approaches used when computing a gene set enrichment for a functional term (e.g., among the top 10 differentially expressed genes in an assay, 5/10 genes map to a functional term that contains a total of 8 genes).
The Fisher's exact test, from what I understand, is a test of independence where individual observations are assumed to be independent. Is this also the case for the hypergeometric test? It seems that one of the biggest arguments for using one test over the other (in this application) is that the Fisher's exact test can be used for small datasets; however, there seldom seems to be debate over data independence.
