More on the Citations Indicator in the THE Rankings

I am copying the whole of a comment to the previous post since it might be the key to the strange results of the citations indicator.

Perhaps someone from Thomson Reuters can confirm that this is method they were using.

Pablo said...

My bet is that TR uses the Leiden "Crown indicator" since this is what is embodied in their product InCites.

To cut it short, each paper is linked to a subdiscipline, a type of publication (letter, review, ...) and a year of publication. With this data for the whole world, it is easy to calculate the expected number of citations for a paper of a given type, in a given discipline, in a given year.
For a set of papers (e.g. all the papers of Alexandria university), the indicator is calculated as Sum(received citations)/Sum(expected citations).

This number can become very high if you have a small number of paper or if you look only at recent papers (if, on average you expect 0.1 citations for a recent paper in math, a single citation will give you a score of 10 for this paper!)

Note that Leiden as recently decided to change its favorite indicator for a mean(citations received/citations expected) which gives less weight for a few highly cited papers in a set. But it seems that TR has not implemented yet this new indicator.

Note also that, in order to avoid the overweight given to few papers in a small set, Leiden publish its own ranking of universities with thresholds on the total number of papers published.