On which database should we judge scholarly records… or how the comprehensive argument is slowly winning over the selective one.

There was a time, towards the end of the 20th century, when things were simple. If you just wanted to count the publications of an author, an institution, a country, you had to refer to the databases of the Institute for Scientific Information (ISI), created and directed by Eugene Garfield. The most famous of these, the Science Citation Index, was built on the idea of selecting the most relevant journals to capture the heart of science, in the already long tradition of bibliotheconomics. And these core journals wer sufficient to draw a relevant picture of the whole of scientific content. Taking the part for the whole raised many questions about the representativeness of the journals present, data and calculation errors, biases in favour of certain disciplines, languages and countries, but as Margaret Thatcher said about her economic world: ‘There is no alternative’

From monopoly to oligopoly: commercial databases getting benchmarked.

Le monopole acquis par l’ISI résultait d’un lent travail d’accumulation monétaire, logistique et commerciale. Garfield a promu et défendu de manière continue ses outils auprès de financeurs, de bibliothécaires, d’universités et de revues. Pourtant cette position dominante n’était pas nécessairement assumée par Garfield lui-même, qui appelait à l’utilisation de tout jeu de données à des fins de production de connaissances1. Il a cependant fallu attendre le début des années 2000 pour que deux entreprises disposant de très gros moyens apparaissent comme de véritables concurrents à l’ISI devenu Thomson IS,: l’éditeur scientifique néerlandais Elsevier, et le géant du web américain Google.

Elsevier a entrepris la construction d’un instrument en relation avec des universités et des chercheurs en bibliométrie. Le 3 novembre 2004, l’entreprise a présenté Scopus, la première véritable concurrence ce que s’appelle désormais ISI Web of Science (WoS), en raison de son caractère généraliste. Alors que les jeux de données de l’ISI avait été fondés sur un double argument – l’excellence de leur contenu et la réduction de la masse d’informations – Scopus est un dispositif à vocation d’exhaustivité. Loin de refuser des revues au contenu “exotique” comme l’ISI avait pu le faire dans les années 1980, Elsevier ne cessa au contraire de mettre en avant la taille de ses jeux de données, son extension vers des domaines peu couverts jusqu’alors (e.g. sciences humaines et sociales, brevets, supports dans d’autres langues que l’anglais). Il ne s’agit pas d’un enjeu simplement cognitif, mais également commercial puisqu’Elsevier démarche les universités ou les pouvoirs publics pour leur vendre ses jeux de données en affirmant que Scopus offrait une meilleure couverture de l’activité scientifique et de son rayonnement et, par là même, permettait une meilleure évaluation de cette activité.

En novembre 2004, l’entreprise Google a mis en place un jeu de données associé à un moteur de recherche spécialisé, Google Scholar. Il est accessible à tous via un site internet et permet de faire des requêtes sur les auteurs, les titres, les résumés. Une différence concerne cependant le critère “for scientific information only” qui n’était pas repris pour circonscrire la recherche de l’information pertinente. Google Scholar ne sépare pas a priori les sites “académiques” des ceux qui ne le sont pas, mais utilise l’intégralité des jeux de données sur lesquels s’appuie le moteur généraliste, y repérant automatiquement des motifs lexicaux de références bibliographiques.

Ce choix de l’extension maximale est confirmé dans les développements ultérieurs : le contenu des ouvrages numérisés dans Google Books était accessible dans le jeu de données Scholar fin 2004, et Google a entrepris un travail de numérisation de fonds de revues scientifiques en 2007, en complément d’accords avec des éditeurs scientifiques commerciaux. Google a également proposé sur son site des procédures permettant aux bibliothécaires et aux dépôts d’archives d’inclure leurs fonds dans les données Scholar. Dans la droite ligne du modèle économique de Google, combinant gratuité des usages et vente d’espaces publicitaires plus ou moins personnalisés, tout texte numérisé est potentiellement indexé dans le jeu de données Scholar, y compris de faux textes à des fins de bibliometrical gaming2. Les utilisateurs ont alors accès à des productions en texte intégral, archivées sur des sites personnels ou dans des dépôts institutionnels, quel que soit leur format informatique, leur langue ou leur genre (rapport, papier de travail, acte de colloque, chapitre d’ouvrage, tribune dans la presse…), que leur contenu soit ou non réservé à des abonnés. De plus, le jeu de données Scholar est indexé de telle sorte qu’il permet le suivi des productions citantes et citées (Giles 2005), ce qui en fait un concurrent de facto à l’ISI WoS et au tout nouveau Scopus, même en l’absence de commercialisation directe d’un “produit” Google Scholar.

OJS

https://direct.mit.edu/qss/article/doi/10.1162/qss_a_00228/114119/Recalibrating-the-Scope-of-Scholarly-Publishing-A

  1. Garfield 1979 []
  2. Labbe []

The short history of the h-index or… being one click away to determine whether you are a succesful scientist

Sometimes, newness really happens in the academic world. Take bibliometrics indicators: for at least a century they have been usally proposed and discussed by specialized scientists from a scientometrics background in their field journals (currently JASIST, Scientometrics, …) and nobody else cared, at least for some time. But in 2005, something very peculiar happened: the h-index was coined by a total stranger to that field and had an instant success, which endures until now. How did that happen and why such a success?1

Physicist J.E. Hirsch proposed the h-index in a working paper posted on August 3rd, 2005 on arXiv, the famous open archive developed by physicists in Los Alamos. In his manuscript, he discussed the issue of comprehensively evaluating a researcher and wished, in a radical way, to subsume their whole career into a simple and practical measurement: an integer number. To do so, he considered that the production and its uses had to be taken into account through a citation measurement. So, number h is the greatest number for which h articles by an author have at least h citations. For example, for 5 articles cited at least 5 times, h is 5; likewise, for 50 articles cited at least 50 times each, h is 50.

Yet, the use of algorithms concerning authors was not new. Eugene Garfield, the founder of the Institut for ScIentific Information (ISI), claimed to regularly predict Nobel prizes using the Science Citation Index and had then developed the “ISI highly cited”, presenting results for a tiny fraction of “top”researchers ; similarly, a few disciplines like economics and management had a long tradition of ranking “top authors”. But, to our knowledge, no algorithm had to that date been specifically designed to evaluate authors . This novelty also comes from the lack of consideration Hirsch displayed for the existing litterature: there were only four references in his manuscript and only one in the field.

This departing from the scientometric tradition enabled Hirsch to make several shifts. Firstly, he did not take into account journals in which articles are published, probably because in high energy physicists, journals are used to archive knowledge more than to make discoveries public. Secondly, he excluded the pitfalls of the number and order of coauthors, which are of little relevance in physics but crucial in biomedical research for individual evaluation. Thirdly, his index combined two elements considered as heterogeneous in the scientometric tradition: production on the one hand, and use on the other. And finally fourthly, whereas scientometricians are always very cautious about individual analysis and save it for “outliers”, Hirsch proposed a measurement which applies to all researchers, and as a cherry on the cake, argued that it would be of some use for the allocation of research funds.

How did he make such a bold move? Based on numbers crunched in the case of high-energy physicists, HIrsch forged a model of the “successful scientist”. As the result of his algorithm very much depends on the duration of a researcher’s career, the h divided by the number of years of the career was considered as a good indicator by Hirsch.


“An h index of 20 after 20 years of scientific activity, characterizes a successful scientist. […] an h index of 40 after 20 years of scientific activity, characterizes outstanding scientists, likely to be found only at the top universities or major research laboratories. […] an h index of 60 after 20 years, or 90 after 30 years, characterizes truly unique individuals”.


At that point, Hirsch woud have probably been considered as a bibliometrics crackpot, a talented physicist that happens to crunch citations numbers in his pastime and posting his “personal views” on a website.

An instant success,
an impressive series of implementations

To the surprise (and probably horror) of the scientometrics community, the h-index was taken up at a staggering rate. As soon as his manuscript had been available on ArXiv, Hirsch received extensive feedback from his physicist colleagues and, in view of the shared enthusiasm, an open archive specialized in high energy physics, SPIRES (Stanford Physics Information REtrieval System), implemented the algorithm on its dataset only two weeks later. The same day, August 17th, 2005, Nature presented Hirsch’s proposition and highlighted his colleagues’ enthusiasm, while an editorial entitled “Rating Games” discussed the respective role of metrics and peer review. Shortly afterwards, in November 2005, two of Hirsch’s colleagues published the manuscript as an article2 in the Proceedings of the National Academy of Sciences, thus confirming physicists’ keen interest in this new measurement.

The popularization of the h-index took a new turn a yer later with a bibliometric tool developed by Ann-Will Harzing, Publish or Perish (PoP). In October 2006, this management professor at the University of Melbourne put online a small software, operationalizing the h-index calculation. That way, for any author whose name is entered by the user, irrespective of the discipline, PoP calculates their h-index in a single click, based on the nascent Google Scholar dataset. This tool could be downloaded for free and has thus allowed the magic algorithm to reach users far beyond audiences specialized in scientometrics. Researchers and institutions adopted this bibliometric tool so fast that the British Medical Journal published a spoof article describing the different pathologies it generates. Meanwhile, in May 2007 Elsevier had included the h-index in Scopus; Thomson-Reuters likewise had changed its “ISI Highly Cited” and integrated this index into the WoS in 2008. Thus, in just three years, individual measurement became a practical operation drawing on bibliometric tools easily accessible to each researcher.

This cycle of implementation was provisionally finalized by the opening of Google Scholar Citations (GSC) in the summer of 2011. With this new service, every academic could create and have control on her/his profile page on Google Scholar, and could decide to make it public or not. Whatever the choice, GSC would then automatically compute three metrics:
the widely used h-index, the i-10 index, which is the number of articles with at least ten citations, and the total number of citations to your articles.
At this point, the definition of the h-index wasn’t anymore needed and thousands of academics quickly made theirs available online. As Paul Wouters and Rodrigo Costas soon noted in their 2012 manuscript, this was a typical example of what they named “technologies of narcissism”, a mirror through which you and others would look in order to measure your influence, evaluate your importance, worry about your deficiencies.

From journals to articles to authors:
new policies for evaluation

Then, despite professional scientometricians harshly criticizing the h-index for its crudeness, pointing out its variations and limits or taming it to make a g-index or a v-index, the utopic/dystopic vision of Hirsch had come into reality. Conversely, it is because of its very crudeness that made it easier to implement in databases & more readable to lay researchers. But its availability is not sufficient to explain the duration and intensity of uses – the orignial paper will probably pass the 10.000 citations mark in 2020. There are two main and very different reasons for its popularity, which we must analytically distinguish. The first one is the implementation of the algorithm to new objects, such as research groups or even journals. Confronted with its unexpected success and eager to use the new data avaialable, scientometricians got into an h-index frenzy3. While the dominant popular bibliometrics index, both within the community and outside of it, had been for 30 years the Journal Impact Factor (JIF) then prouced and owned by Thomson ISI, the growing success of the h-index made it a challenger in the bibliometrical index “market”. Many papers compared the pros and cons of each algorithm, based on different datasets.

Nevertheless, this “good index” competition shouldn’t hide a second reason for which the h-index became so popular and discussed. Its use was sustained by a political agenda in assesment and evaluation, best represented by the San Francisco Declaration on Research Assessment (DORA) published in 2013. This complex text is often subsumed as an anti-bibliometrics statement, in which signing institutions promise they won”t use the JIF as a way to evaluate research, would it be for hiring, promoting, giving grants, etc. Beyond this simple vision, there are more nuanced recommandations that oppose journal-based metrics, but don’t refuse bibliometrics as a whole. JIF is seen as a bad way to perform quantified assesment, where article-level metrics, whatever they are (citations, downloads, views, social networks mentions…) are more realistic of the “impact” of a given research.

The problem with these new metrics is that nobody really knows what they are used for and what they really mesure4. Consequently, the most popular article-level metric remains the number of citations in a given database (Web of Science, Scopus, Crossref, Google Scholar). Rather than summing these numbers for a given journal, the aggregation is made on a given author. It is so simple to perform on the same databases, that what was absurd a few years ago has become natural. The “h revolution” therefore went beyond Hirsch’s own vision on two points. Firstly, his h-index was to be used for senior scientists, it is now also being applied to/by early and mid-carrer researchers as a more “ethical way” to judge their impact. Secondly, its extension goes hand in hand with a potential transformation of the model of scientific communication, to a post-journal world, in which any kind of text could be cited and counted. Rather than highlighting the fact that you actually passed the test of supposed prestigious journals, you just now give your h number and academic age, so everybody checks whether you really are the successful scientist you pretend to be . ((Please cite selected papers of the author of this post so he may finally become one)).

  1. This post is partially adapted from Pontille David, Torny Didier, « La manufacture de l’évaluation scientifique. Algorithmes, jeux de données et outils bibliométriques », Réseaux, 2013/1 (n° 177), p. 23-61. DOI : 10.3917/res.177.0023 []
  2. There are almost no differences between the ArXiv manuscript from mid-August (V3) and the published PNAS paper []
  3. See for an early litterature review, Bornmann, Lutz, and Hans‐Dieter Daniel. “The state of h index research.” EMBO reports 10.1 (2009): 2-6. []
  4. See Haustein, Stefanie, Timothy D. Bowman, and Rodrigo Costas. “Interpreting” altmetrics”: Viewing acts on social media through the lens of citation and social theories.” on ArXiv []
Search OpenEdition Search

You will be redirected to OpenEdition Search