Self-citation corrections for the Hirsch index

I propose to sharpen the index h, proposed by Hirsch as a useful index to characterize the scientific output of a researcher, by excluding the self-citations. Performing a self-experiment and also discussing in detail two anonymous data sets, it is shown that self-citations can significantly reduce the h index in contrast to Hirsch’s expectations. This result is confirmed by an analysis of 13 further data sets. Introduction. – About one year ago the physicist Hirsch [1] proposed an easily computable index h as an estimate of the visibility, importance, significance, and broad impact of a scientist’s cumulative research contribution. This index h is defined as the highest number of papers of a scientist that received h or more citations. Of course, for many people ”it is distasteful to reduce a lifetime’s work to a number” [2]. For others h is an ”elegantly simple” measure [3], which allows an easy comparison of the scientific achievement of a scientist in an unbiased way by a single number. As it can be determined easily by ordering the publication list according to the number of citations which is for example possible using the Science Citation Index provided by Thomson ISI in the Web of Science (WoS) data base, it has received immediate attention in the public [4] and the physics community [5–7] and is already widely recognized as a convenient measure in evaluations. Already a significant amount of literature in informetrics [8–13] has been dealing with this measure of visibility of a scientist. Different data sets have been evaluated to identify the most highly cited scientists in various fields [1,3,7,14]. A comparative study on committee peer review [15] of post-doctoral researchers in biomedicine suggested that the Hirsch index is indeed a promising (rough) measurement of the quality. The statistical correlation of the Hirsch index with standard bibliometric indicators and peer judgement was shown to be quite high for 147 chemistry research groups in the Netherlands [16]. A critical analysis of the Hirsch index of 187 evolutionary biologists and ecologists from the editorial boards of seven journals illustrates the risk of indiscriminate use of the index [2]. A quantitative investigation of the statistical reliability [6] has cast doubts on the accuracy and precision of the Hirsch index. Nevertheless the interest in this measure continues to grow [7–12,16–21]. It was shown [10] that the Hirsch index notion can be extended to the general framework of information production processes and that any system has a unique Hirsch index. Banks [21] has extended it to an index for scientific topics and compounds in order to identify hot topics and interesting materials. The Hirsch-Banks index is defined in analogy to h as the highest numbers of papers in a particular field or on a specific compound that received h or more citations. This extension has also received a lot of attention even beyond the scientific community, identifying nanotubes, nanowires and quantum dots as the most interesting topics in recent years. Other generalizations concern the comparison of entire research groups by their Hirsch index [16] and the utility for assessing the impact of journals [12, 22]. When identifying hot topics, it is obvious that one will be dealing with a set of publications which are heavily cited within the field which means that they are probably most often cited by people working on the same topic, i.e. by the same set of people who have written these publications. However, when assessing the scientific achievement of an individual scientist, the analogous kind of citations within the data set, namely the self-citations should ideally not be included, because they are not reflecting the impact of a publication. Of course, self-citations increase the h index, but Hirsch has argued that the effect is relatively small and that the necessary corrections for h would involve only very few if any papers. An analysis of a group of scientists in ecology and evolution [2], however, showed an average decrease of 12.3%. In contrast, the Hirsch indices of 31 influential scientists in information science dropped only between zero and three, on average by 0.9, or 6.6%, when self-citations were excluded [3]. In the present investigation I demonstrate that the influence of self-citations on the Hirsch index can be drastic, in particular for younger scientists with a low Hirsch index. Three different ways to sharpen the Hirsch index will be proposed.