Figure 2.

Extended simulation study: performance measures under different missingness mechanisms for x; β ₀= ‐0.693;β _x = 0.405; the population distribution of x is assumed to be invariant (case 1) or estimated in an external data set of size 10 000 (case 2) or 1000 (case 3). M1, x is missing completely at random; M2, x is missing at random conditional on y; M3, x is missing not at random dependent on x; M4, x is missing not at random dependent on x and y; error bars, ±1.95× Monte Carlo standard errors; filled and hollow points, empirical and average model standard errors, respectively; ×, out‐of‐range values. CRA, complete record analysis; MI, multiple imputation [Colour figure can be viewed at wileyonlinelibrary.com]