Quantitative evaluation of alternative field normalization procedures

The use of citation numbers for the assessment of research quality has become highly relevant in modern science. Although it is well known that scientific domains strongly differ in terms of citation rates, bibliometric indicators currently used in research assessment are often based on the sole use of raw citation numbers. This necessarily leads to unfair evaluation procedures in cross-disciplinary contexts. For this reason, there is an increasing trend towards the formulation of normalization procedures able to suppress disproportions in citation numbers among scientific domains, and thus to lead to more fair cross-disciplinary evaluation criteria. In this paper, we rigorously test the performance of several field normalization procedures devoted to this purpose. We find that four procedures discussed in the literature do worse than the usual normalization with field averages. The latter drastically reduces citation disproportions among scientific disciplines. Finally, we find that a recently introduced two-parameters normalization scheme reduces citation disproportions to a level very close to the best achievable level of reduction.