Mitesser, Heinz, Havemann, and Gl¨ aser 1 Measuring Diversity of Research by Extracting Latent Themes from Bipartite Networks of Papers and References Oliver Mitesser * Michael Heinz † Frank Havemann ‡ JochenGl¨aser § June 9, 2008 Abstract Inspired by the hypothesis that the diversity of research might decline as a result of new sci- ence policy measures we explore the potential of bibliometric measures for analysing the diver- sity of research at meso- and macro-levels of (na- tional sub-) fields, countries, and organisations. Our aim is to render changes in the diversity of research landscapes measurable and therefore comparable in time series as well as between dif- ferent countries. We discuss different method- ological approaches and some results based on a method that extracts latent themes from bipar- tite networks of research papers and their cited references by singular value decomposition of the citation matrix. 1 Introduction The diversity of science appears to be mov- ing to the centre stage of science policy discus- sions. Recent approaches to the governance of science by performance-based block funding for universities have the potential to affect diver- sity. These attempts to increase the selectiv- ity of research funding reduce the number of funded units and are thus likely to diminish di- versity (Adams and Smith 2003). At a more subtle level, diversity is threatened by the adap- * Technische Universit¨ at Darmstadt, Universit¨ ats- und Landesbibliothek Darmstadt, Germany, oliver dot mitesser at gmx dot de † Humboldt-Universit¨ at zu Berlin, Institute for Li- brary and Information Science (IBI), Germany, michael dot heinz at ibi dot hu-berlin dot de ‡ s. footnote before, frank dot havemann at ibi dot hu-berlin dot de § Freie Universit¨ at Berlin, Institut f¨ ur Soziologie, Ger- many, Jochen dot Glaser at FU-Berlin dot de tive behavior of scientists. Whenever science policy increases punishment for failure, e. g. via reduced funding, researchers are likely to choose projects that are safe in that they are approved of by the scientific community and have a high probability of success. Such safe projects follow the mainstream of a field and use approaches that are known to yield results. Research that deviates from the mainstream is increasingly un- likely to be pursued, which reduces the diversity of problem formulations and research strategies in a field (Harley and Lee 1997; Whitley 2007). These arguments, albeit persuasive, lack em- pirical foundation. While the micro-mechanisms that make researchers move flock to the main- stream could be identified (Gl¨ aser and Laudel 2007; Gl¨ aser et al. 2008), no convincing mea- surement of research diversity at higher levels of aggregation has so far been provided. Opinions of scientists on the subject cannot be considered as reliable evidence for two reasons. Firstly, per- ceptions of a changing diversity depend on scien- tists’ individual scientific perspectives and their opinions about science policy. They may there- fore be biased. Secondly, quality and marginal- ity of a scientific enterprise are often insepara- ble. Nonconformist approaches might be per- ceived as bad science by the majority. Con- versely, scientists might rationalise insufficient recognition of their work as being due to the specificity rather than quality of their work. Testing the above-described ‘homogenisation thesis’ requires measures of diversity that do not depend on scientists’ perceptions of that diver- sity. Bibliometric indicators can be used to con- struct these measures because they are unobtru- sive and objective, i. e. they neither affect the be- haviour they measure nor depend on scientists’ opinions about the attribute that is measured. H. Kretschmer & F. Havemann (Eds.): Proceedings of WIS 2008, Berlin Fourth International Conference on Webometrics, Informetrics and Scientometrics & Ninth COLLNET Meeting Humboldt-Universit¨ at zu Berlin, Institute for Library and Information Science (IBI) This is an Open Access document licensed under the Creative Commons License BY http://creativecommons.org/licenses/by/2.0/