Figure 3.
Mean average treatment effect estimates for 50% missing data in a binary outcome under the MAR X mechanism (odds of Y missing 1.5 (weak MAR) or 2.5 (strong MAR) times higher per standard deviation increase in X), incorrectly specified analysis model. Horizontal reference lines illustrate the least false average treatment effect in the absence of missing data. Error bars correspond to empirical standard errors (±1 standard error) across 2000 simulated datasets.