The (Exp(ß)-1) values for logistic regressions.

14 predictor variables (plus one interaction variable) were used to predict the target variable (citation counts): (1) article age (Age), (2) journal impact factor (JIF), (3) number of co-authors (Auth_N), (4) number of references cited (Ref_N) , (5) number of pages (Page_N), (6) Science vs. Social Science (Sci), (7) Review article vs. ordinary article (Review), (8) US co-author vs. no US co-author (USA), (9) open access vs not (OA), (10) non-additive interaction between OA and Age (Age*OA), (11) OA mandated vs. not (M), (12) mandating institution CERN (CERN), (13) mandating institution Southampton ECS (South), (14) mandating institution U. Minho (Minho), (15) mandating institution Queensland U. Technology (Queens). Four logistic regression models estimated the size of the independent contribution of each of the 15 predictor variables to predicting the citation counts using four different cut-off values and comparison ranges (selected on the basis of the overall citation count distribution in Figure 3): (M1) articles with 0 vs lo (1–4) citations; (M2) lo (1–4) vs. med-lo (5–10) citations; (M3) lo (1–4) vs. med-hi (10–19) citations; (M4) lo (1–4) vs. hi (20+) citations. Note that OA is a significant independent contributor to citations in all but the lowest of these four citation ranges. The effect is displayed as a bar graph in Figure 5. (Boldface values for Exp(ß) indicate differences significant at p<0.01 and italic values indicate differences significant at 0.01≤p<0.05.)