Who is responsible for publication bias?

Studies with negative results could remain unpublished because authors fail to write manuscripts and submit them to journals, because such studies are peer reviewed less favourably, or because editors simply do not want to publish negative results. The peer review process is sometimes unreliable and susceptible to subjectivity, bias and conflict of interest (Peters 1982, Godlee 1999).  Experimental studies in which test manuscripts were submitted to peer reviewers or journals showed that peer reviewers are more likely to referee favourably if results were in accordance with their own views (Mahoney 1977, Epstein 1990, Ernst 1994). For example, when a selected group of authors was asked to peer review a fictitious paper on transcutaneous electrical nerve stimulation (TENS) they were influenced by their own findings and preconceptions .  Other studies have shown no association between publication of submitted manuscripts and study outcomes (Abbot 1998, Olson 2002), suggesting that although peer reviewers may hold strong beliefs which will influence their assessments, there is no general bias for or against positive findings.

A number of studies have directly asked authors why they had not published their findings. The most frequent answer was that they were not interesting enough to merit publication (e.g. journals would be unlikely to accept the manuscripts) (Easterbrook 1991, Dickersin 1992, Stern 1997, Weber 1998, Decullier 2005) or the investigators did not have enough time to prepare a manuscript (Weber 1998, Hartling 2004).  Rejection of a manuscript by a journal was rarely mentioned as a reason for not publishing. Selective submission of papers by authors rather than selective recommendation by peer reviewers and selective acceptance by editors thus appears to be the dominant contributor to publication bias.   In addition, Dickersin et al. examined the time from manuscript submission (to the journal JAMA) to full publication and found no association between this time and any study characteristics examined, including statistical significance of the study results (Dickersin 2002). Thus, time-lag bias may also be the result of delayed submission of manuscripts for publication by authors rather than by delayed publication by journals.