The following three findings by Chawki Hajjem have been added
(text in French but graphs self-explanatory) to his research website:
(1) The OA advantage -- the higher citation counts for self-archived
articles compared to non-self-archived ones in the same journal/year --
is *not* due only or largely to a quality/impact self-selection bias
(with higher-impact articles tending to be self-archived more).
Quality bias is only one of the many factors underlying the OA advantage,
and it is not the largest of them.
OA advantage = EA + AA + QB + OA + UA
(2) The OA advantage is not just an artifact of the algorithm used to trawl the
web to look for the full-text version of articles: A signal detection analysis on
a random sample of 100 articles the algorithm classed as OA and 100 it classed as
non-OA revealed a detection-sensitivity (d') of 1.936 (which is very high)
and almost zero bias (between positive and negative errors: B = 0.988 which is
near a perfect 1).
(3) The proportion of articles that are OA varies from country to country
(10-20%), but not substantially; there appear to be no systematic national
differences. Nor are there systematic differences between disciplines in
the practise of OA provision, which also varies between 10-20%. There
do seem to be some discipline differences, however, in the size of the
OA advantage, but it is not clear how this may interact with discipline
differences in article or citation numbers.
These findings complement Chawki's findings on the OA advantage in Biology
and Social Sciences
Which confirm and extend Tim Brody's findings in Physics/Maths: