Comparing different implementations of similarity for disparity measures in studies on interdisciplinarity

The recent literature on interdisciplinarity in quantitative science studies focuses on the development and interpretation of measures of diversity. These measures consider the mutual similarity or distance between disciplines in the applied subject scheme. Most of the studies on this topic are well aware of the importance of the choice of the proper classification scheme as it may lead to large differences in the obtained diversity scores. The central objective of the present study is to investigate the underlying properties of distinct similarity matrices used as starting point in quantifications of disparity. The study screens ten combinations of three different classification schemes and five implementations of citation or reference-based similarities using a version of a cosine similarity. In addition, each of the ten combinations are calculated for nine sliding time windows. The ten combinations are scored on different evaluative criteria both of quantitative and qualitative nature: stability, discriminative power, density, skewness and deviation and ease of calculation. The study provides the required tools for an informed choice on the appropriate similarity measures in future research on and application of diversity measures. Based on the investigated criteria, the study favors the use of bibliographic coupling on a medium-resolution granularity subject classification.