Fig. 4From: A data-adaptive method for investigating effect heterogeneity with high-dimensional covariates in Mendelian randomizationResults of the simulation study showing mean squared error (MSE) of estimates with weak modification (strength of modification = 0.0) up to strong modification (strength of modification = 0.5). Top panel: Scenario A (all effect modifiers are non-colliders); middle panel: Scenario B (some effect modifiers are colliders); bottom panel: Scenario C (effect modifiers are colliders and influence the causal effect in a complex way). The black line represents results with no stratification. In each scenario, data are independently simulated 100 times, and the MSE represents the median value across simulationsBack to article page