Fig 1.
The current bibliometrics incentives model.
Fig 2.
Standards of practice and publication patterns.
Font size represents relative numbers of publications in each category.
Fig 3.
Number of studies published and not published with positive or negative outcomes.
Fig 4.
Research activity levels by study outcome.
The number of papers published each year on the same topics as the studies included in the present analysis (means ± SEM). Publication years range from two years before (-2) until two years after the year the studies included in the present analysis were either completed or published (year 0). For example, even two years before the publication of the studies reporting positive results, on average, 544 ± 92 papers were already being published each year on those same topics. Two years before the publication of the studies reporting negative results, only 102 ± 56 papers were being published each year on those topics.
Table 1.
Studies divided into 5 groups based on research activity levels.
Fig 5.
Success rates increase and number of topics decrease as research activity levels increase.
(A) Percent successful, or positive, statistically significant results, at median activity levels for 5 equal groups (n’s = 22). (B) The number of interventions or topics of research is shown at each activity level. Symbol size indicates the number of studies being conducted on the same topic. The Y axes is inverted in this panel in order to convey the concept that, over time, only a small number of reliable findings rise to the top.
Fig 6.
Raw citation counts and RCRs are related to research activity levels.
(A) RCR values were calculated approximately 8 years after the papers were published but RCR values and 2-year raw citation numbers were very highly correlated, r(86) = 0.96, and both raw citation numbers and RCR values increased as activity levels increase. (B) Publications reporting positive, statistically significant results were more highly cited than studies reporting negative, non-statistically significant results, and those differences were still evident even with article-level normalized RCR values.