Diversity and interdisciplinarity: Should variety, balance and disparity be combined as a product or better as a sum? An information-theoretical and statistical estimation approach

Abstract

Diversity is a central concept not only in ecology, but also in the social sciences and in bibliometrics. The discussion about an adequate measure of diversity is strongly driven by the work of Rao (Sankhyā Indian J Stat Series A 44:1-22, 1982) and Stirling (J R Soc Interface 4:707-719, 2007). It is to the credit of Leydesdorff (Scientometr 116:2113-2121, 2018) to have proposed a decisive improvement with regard to an inconsistency in the Rao-Sterling-diversity indicator that Rousseau (Scientometr 116:645-653, 2018) had pointed out. With recourse to Shannon's probabilistically based entropy concept, in this contribution the three components of diversity “variety”, “balance”, and “disparity” are to be reconceptualized as entropy masses that add up to an overall diversity indicator dive. Diversity can thus be interpreted as the degree of uncertainty or unpredictability. For "disparity", for example, the concept of mutual information is used. However, probabilities must be estimated statistically. A basic estimation strategy (cross tables) and a more sophisticated one (parametric statistical model) are presented. This overall probability-theoretical based concept is applied exemplarily to data on research output types of funded research projects in UK that were the subject of the Metric Tide Report (REF 2014) and ex-ante evaluation data of a research funding organization. As expected, research output types depend on the research area, with journal articles having the strongest individual balance among the output types, i.e., being represented in almost all research areas. For the ex-ante evaluation data of 1,221 funded projects the diversity components were statistically estimated. The overall diversity of the projects in terms of entropy is 55.5% of the maximal possible entropy.