Fig. 4: Adjustment factors (means ± SD) for the collider bias that are estimated using the Slope-Hunter (SH) and the ‘Hedges-Olkin’ estimator of the Dudbridge et al.
From: A robust method for collider bias correction in conditional genome-wide association studies
(DHO)7 methods under different simulated scenarios. The means and the error bars are estimated from n = 1000 independent simulations. The true collider bias induced under different scenarios are depicted by the horizontal dashed black lines. In these simulation studies, estimates of association coefficients for 10,000 independent SNPs with a quantitative trait I, and a quantitative outcome (P) conditional on I are simulated assuming an underlying four-component model for effect-size distribution, where SNPs could have direct effects on I only (GI.), both traits (GIP), P only or on neither traits, such that GI. explains larger (Scenario 1), equal (Scenario 2), or lower (Scenario 3) proportions of the variation in I compared with the GIP variants (Methods). The direct effects on I and P could be correlated (ρd ≠ 0) in the same direction as the correlation due to the collider-bias effect (ρc), uncorrelated (ρd = 0), or correlated in the opposite direction of the correlation due to the collider-bias effect. Source data are provided as a Source Data file.
