The impact of extreme observations in citation distributions

This paper studies the role of extremely highly cited articles in two instances: the measurement of citation inequality, and mean citation rates. Using a dataset, acquired from Thomson Scientific, consisting of 4.4 million articles published in 1998-2003 in 22 broad fields with a five-year citation window, the main results are the following. Firstly, both within each of 22 broad fields and in the all-sciences case, citation inequality is strongly affected by the presence of a handful of extreme observations, particularly when it is measured by citation inequality indices that are very sensitive to citation differences in the upper tail of citation distributions. Secondly, the impact of extreme observations on citation averages is generally much smaller. The concluding Section includes some practical lessons for students of citation inequality and/or users of high-impact indicators