Comment on: Lemire, Daniel (2014) Though unrefereed, arXiv has a better h-index than most journals?
Arxiv includes both unrefereed and refereed versions of papers.
Distinguish citation from access-date (early access) and access-locus.
Peer-reviewed publication is not the same thing as (or not only) access-provision:
Journals provide both peer review and access (to subscribers only, if journal is subscription-based).
Repositories provide access (to peer-reviewed journal articles and sometimes to earlier unrefereed drafts).
Hence repositories do not have citation counts or h-indexes: just access-locus statistics; their citation counts are parasitic on journal citation counts (and especially journal peer review).
Users access whatever version they can access, but they cite the journal article (the canonical, archival “version of record”).
The only exception is unrefereed drafts — but even there, it is the author’s draft that is being cited; the repository is just the access-locus:
Unrefereed drafts used to be cited as “name, title, unpublished (or ‘in prep’)” and refereed, accepted drafts used to be cited as “name, title, journal, in press).”
Adding an OA access-locus to the journal citation is becoming an increasingly common (and desirable) scholarly practice, but it does not change the fact that what is being cited is the work, and the canonical version of the work is the refereed, published version-of-record, regardless of access-locus.
Hence repositories do not have citation counts; they just have access-locus (download) counts.
(Some interesting statistics can, however, be done on the citation of unrefereed vs refereed versions, i.e., early access.)