1. It is unclear why the regression coefficients are given in Figure 1.
2. The results of empirical study were similar between MAR and NMAR approaches. I wonder only a few time points (e.g., 3) lead to the conclusion.
3. In stimulation study, the parameter recovery of true model was absent. In fact, it should be provided as baseline to compare other approaches.
4. Two ad hoc approaches performed poorly when dealing with data generated under NMAR. In literatures, it suggests that the mean imputation should be avoided because of its inefficiency. So it could be expected that other imputation methods may yield more acceptable outcomes.
1. It was assumed that the intercept have equal varaiance 1 across three timepoint. Therefore, in figure 1 growth model, there are three 1 under intercept. However, the authors did not explain why the residual variance will be different for linear slope and quadratic slope.
2. I don't think so. In simulation, it was found that when the percentage of missing and the dependence of missing on the intercept and slope are both high, it will lead to servious violation of MAR. For the empirical data, the mean of dropout was about 10%. Therefore, the results between MAR and NMAR were similar.
3. Yes. You are right.
4. Multiple imputation which use plausilbe value will estimate the standard error more properly than mean imputation.