The changing face of journal metrics
As communication evolves, new methods help evaluate the impact of scholarly journals
By Mike Taylor and Judith Kamalski Posted on 28 November 2012
For decades, a principal measure of an article's impact on the scholarly world has been the number of citations it has received.1An increasing focus on using these citation counts as a proxy for scientific quality provided the catalyst for the development of journal metrics, including Garfield’s invention of the Impact Factor in the 1950s2.
The migration from paper to electronic delivery (particularly online) has enabled better understanding and analysis of citation count-based impact measurements and created a new supply of user-activity measurements: page views and downloads. Over the past few years, the growing importance of social networking — combined with a rising number of platforms making their activity data publicly available — has resulted in new ways of measuring scholarly communication activities: one encapsulated by the term altmetrics5. Although we have added these new metrics to this chart, we are not suggesting that superseding generations necessarily replace the earlier ones. In fact, the Relative Impact Measure is still used substantially, even though network analysis exists.
A third mature product is ImpactStory (formerly total-impact.org), an always-free and open application. Altmetrics applications acquire the broadest possible set of data about content consumption. This includes HTML page views and PDF downloads, social usage, (e.g. tweets and Facebook comments), as well as more specialized researcher activities, such as bookmarking and reference sharing via tools like Mendeley, Zotero and CiteULike. A list of the data sources used by ImpactStory appears below. As well as counting activities surrounding the full article, there are also figure and data re-use totals. Altmetric.com also takes into account mass media links to scholarly articles.
An example of the Altmetric.com donut, which can be found on many Scopus articles. This one, from the paper "How to Choose a Good Scientific Problem" in Molecular Cell, shows that (at time of writing) the article has been mentioned 89 times on a variety of platforms and saved as a bookmark by more than 4,000 people.
What do all these numbers mean?
Although there is some evidence to link social network activity, such as tweets, with ultimate citation count (Priem & Piwowar et al, 20126, Eysenback, 20117), this field is still in its early stages, and a considerable number of areas still require research. Further investigation aims to uncover patterns and relationships between usage data and ultimate citation, allowing users to discover papers of interest and influence they might previously have failed to notice. Planned areas of research include:
- Scholarly consumption versus lay consumption. With so much benefit to be gained from encouraging public engagement in science, we need new ways of tracking this. After all, while members of the public are unlikely to cite articles in a formal setting, we may well see increased social sharing. Analysis of usage data might reveal striking differences between scholarly and lay usage patterns. For example, references to research amongst the general public may be primarily driven by mass media references – just as the mass media might be influenced by academic work going viral on Twitter and Facebook: whereas one might hypothesize that activity measured in specialized scholarly tools, such as Mendeley, would be less subject to this influence. This information could be critical in allowing publishers and platform owners to tweak their systems so as to best support use and report on wider usage to funding agencies.
- When does social networking become marketing and when does it become gaming or cheating? There has been criticism8 that the JIF can be increased by excluding or including reference counts from certain types of articles, and by journals' self-citation policies. Social data is just as prone to influence. For example, while authors' tweets about their papers are perfectly legitimate social marketing of the type previously done through email groups, and while it's reasonable to assume that some mentions of this type will go 'viral' and thus be propelled towards mass media mentions and possibly drive citations, there will inevitably be concerted efforts to build momentum that goes beyond natural self/network marketing. A sophisticated analysis of social networking mapped against author networks might be able to detect and downplay this type of activity.
- What other factors influence ultimate impact? As we expand our ability to understand what drives scholarly impact and how usage patterns should be interpreted, the scope should increase to include other non-social facets. For example, do cross-discipline papers get a wider readership than simply the disciplines targeted? Do papers with a lay abstract attract a wider lay audience? To what extent does the inclusion of a high-ranking contributor boost citation above what might be predicted?
- Do any particular consumption activities predicate others? Is there a computable conversion rate for moving from one activity to another? How do these vary over time and by discipline? What activities lead to citation? Are there papers that are less well cited - or not cited at all - that nevertheless appear to have impact in other ways?
Elsevier is fully engaged with the altmetrics movement. For example, in 2013 the Elsevier Labs team aims to co-publish large scale research that will begin to explore the relationship between the different facets and to establish a framework for understanding the meaning of this activity. It aims to build on the current work to found an empirically-based discipline that analyses the relationship between social activity, other factors and both scholarly and lay consumption and usage. By working together to combine knowledge at Elsevier, we intend to show that no single measurement can provide the whole picture and that a panel of metrics informed by empirical research and expert opinion is typically the best way to analyze the performance of a journal, an author or an article.