Fig. 3
Relative bias in the associations between predictors and health outcome at 50% response-rate. Notes: × 1 to the left, and × 2 to the right. Solid bars represent situations where extreme scores are present even if they are under-represented in the samples. Dotted bars represent situations where the 10% most extreme scores on the health outcome are totally missing from the samples. bnon 1 = dependency between liability of non-response and each of the three predictors (× 1, × 2, × 3). b non 2 = dependency between liability of non-response and the health outcome. Different degrees of dependency between study variables and liability of non-response are modeled. Relative bias is calculated by dividing the difference between the estimated and the true value on the true value of bpred = 0.20 for × 1 and 0.30 for × 2