Fig. 4

Simulation study based on 50 training and test sets and random co-data (left) or informative co-data (right). Estimated prior variance for various co-data models; (i) linear for linear co-data model; (ii) gam for generalised addive co-data model; (iii) scam.p for positive shape constrained co-data model; (iv) scam.pmi for positive and monotone increasing shape constrained co-data model, and (v) AD for adaptive discretisation. The lines indicate the pointwise median and the inner and outer shaded bands indicate the 25–75% and 5–95% quantiles respectively. Points indicate the true effect sizes \((\beta _k^0)^2\)