dc.description.abstract | One method for assessing quality of research outputs across
different technical disciplines is comparing citations
received by the research output documents. However, crossdiscipline
citation comparison studies require discipline
normalization, in order to eliminate discipline differences in
cultural citation practices and discipline differences in the
number of active researchers available to cite. The 'definition'
of, and number of documents used to represent, a
discipline become critical. This study attempted to determinewhether
the citation characteristics (average, median)
ofa discipline's domain stabilized as the domain's size was
decreased. A sample of papers (classified as research
articles only, not review articles, by the Institute for ScientificInformation)
published in the journal Oncogene in 1999
wasclustered hierarchically, and the citation averages and
medians were computed for each cluster at different cluster
hierarchical levels. The citation characteristics became
increasingly stratified as the clusters were reduced in size,
raising serious questions about the credibility of a selected
denominator for normalization studies. An interesting side
result occurred when all the retrieved articles were sorted
by number of citations. Thirteen ofthe fifty most highly cited
research articles had 100 or more references, whereas zero
of the fifty least cited research articles had 100 or more
references.
Keywords: citation analysis; citation normalization;
document clustering; research evaluation | en_US |