When the InSIDE assumption for multivariable MR‐Egger was satisfied, αj′ and βX1j were drawn from independent distributions, and when it was violated, they were drawn from a multivariate normal distribution with cor(α′,βX1)=0.3. The above 4 scenarios were applied to the simulated data when βXk were generated independently for all k, with the parameters in the covariance matrix set to: σ12=0.03; σ22=0.02; σ32=0.04; and ρ 12=ρ 13=ρ 23=0. The 4 scenarios were repeated when βXk were correlated for all k(ρ 12=0.2, ρ 13=−0.3, ρ 23=0.1). The mean F‐statistics were greater than 200 and I 2 statistics greater than 99% in each scenario; values are provided in Web Tables A1 and A2. In total, data were simulated for 32 different choices of parameters.
