Does the Leiden Manifesto need a "Skewness Principle"?

[Andreas Strotmann]

Does the Leiden Manifesto need to add a "skewness" principle, i.e., an
admonition to always take into account that bibliometric data is likely to
suffer from skewed error distributions, including a reminder that for
typical skewed error distributions, your usual network statistics are
likely way off the truth unless the data used for analysis is of excellent
quality, both wrt. completeness and wrt. correctness?

Slides for a presentation in which we argue for this are now online on
Researchgate (researchgate.net/publication/279162430). Since this
presentation was mainly meant to ignite discussion, but the ISSI conference
program does not actually reserve time for discussion after ignite
presentations, I would like to offer this topic up for discussion on this
mailing list already.

It is true that most people in our field are aware of the problem and its
solutions to some extent, but fact is that I see publications that violate
the proposed principle all too frequently. Admittedly, in-depth research
asking just how good network data needs to be so you can trust results from
analyzing it has only very recently begun, but the references that I
collected in these slides seem to indicate that the problem (while still
manageable) has been severely underestimated.

Hoping for a lively discussion,

 -- Andreas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.asis.org/pipermail/sigmetrics/attachments/20150624/943ce104/attachment.html>