Year: 2025
Type: preprint
Abstract: Field normalization plays a crucial role in scientometrics to ensure fair comparisons across different disciplines. In this paper, we revisit the effectiveness of several widely used field normalization methods. Our findings indicate that source-side normalization (as employed in SNIP) does not fully eliminate citation bias across different fields and the imbalanced paper growth rates across fields are a key factor for this phenomenon. To address the issue of skewness, logarithmic transformation has been applied. Recently, a combination of logarithmic transformation and mean-based normalization, expressed as ln(c+1)/mu, has gained popularity. However, our analysis shows that this approach does not yield satisfactory results. Instead, we find that combining logarithmic transformation (ln(c+1)) with z-score normalization provides a better alternative. Furthermore, our study suggests that the better performance is achieved when combining both source-side and target-side field normalization methods.
Source: arXiv (Cornell University)
Authors: Lu, Xinyue, Li, Li, Shen, Zhesi
Language: English
Subfield: Statistics, Probability and Uncertainty
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W6979230325
Cites:
Cited by:
Related to: