The misuse of the nonlinear field normalization method: Nonlinear field normalization citation counts at the paper level cannot be added or averaged

TL;DR

This paper highlights the improper use of nonlinear field normalization methods for citation counts, demonstrating mathematically that such normalized data cannot be meaningfully added or averaged, and classifies existing methods.

Contribution

It provides a mathematical proof explaining why nonlinear normalized citation counts cannot be aggregated and classifies field normalization methods into linear and nonlinear categories.

Findings

Mathematical proof that nonlinear normalized citation counts cannot be added or averaged.

Systematic classification of field normalization methods into linear and nonlinear.

Provides a theoretical basis for correct application of normalization methods.

Abstract

There is a very important problem that has not attracted sufficient attention in academia, i.e., nonlinear field normalization citation counts at the paper level obtained using nonlinear field normalization methods cannot be added or averaged. Unfortunately, there are many cases adding or averaging the nonlinear normalized citation counts of individual papers that can be found in the academic literature, indicating that nonlinear field normalization methods have long been misused in academia. In this paper, we performed the following two research works. First, we analyzed why the nonlinear normalized citation counts of individual papers cannot be added or averaged from the perspective of theoretical analysis in mathematics: we provide mathematical proofs for the crucial steps of the analysis. Second, we systematically classified the existing main field normalization methods into linear and nonlinear field normalization methods. The above two research works provide a theoretical basis for the proper use of field normalization methods in the future, avoiding the continued misuse of nonlinear data. Furthermore, because our mathematical proof is applicable to all nonlinear data in the entire real number domain, our research works are also meaningful for the whole field of data and information science.