Fig. 1From: Instrumental variable analysis in the context of dichotomous outcome and exposure with a numerical experiment in pharmacoepidemiologyRelative bias (rB) of the methods. a: True model b: conventional model; c : 2SRI with instrument pr; d: GMM with instrument pr. Low, Medium and High indicate the corresponding level of confounding and the instrument strength grows from a, b, c, d sequence to the next (from left to right)Back to article page