After the publication of this article and after further work on this problem, we realized that several formulas derived in the article only held for special cases (MVN with identical ps(y1)). The formulas for general cases are provided here.
In Section 4, should be
In Section 5, equation (3) should be . The definition of ϖj,k should be . By assuming nonfuture dependence, we obtain for the future data
In Section 6, for the MNAR analysis of the Growth Hormone study, the first bullet should be for MVN model only. For OMVN model, does not appear in the posterior distribution and we restrict .
With the correction of the formulas, we observed changes in the illustrative analysis results of the Growth Hormone study. Specifically, in Table 2, the posterior mean(SD) from MVN under MNAR are 75(7.5) and 71(4.9) for EG and EP at Month 12, respectively. From OMVN under MAR, it is 81(8.0) and 78(7.9) for EG at Month 6 and 12 and 81(9.0) and 73(6.5) for EP at Month 6 and 12, respectively. For OMVN under MNAR, it is 78(8.3) and 75(8.2) for EG at Month 6 and 12 and 79(9.1) and 71(6.6) for EP at Month 6 and 12, respectively. The differences at Month 12 are 5.5(8.7), 4.0(8.9), 5.3(10.3), and 4.0(10.5) for MVN under MAR and MNAR and OMVN under MAR and MNAR, respectively. In Table 3, it is 3.6(10.6), 4.6(10.3), 4.2(10.4), and 4.0(10.5) for the OMVN column. We also found small changes in the simulation results in Table 5. We present the updated Tables 2, 3, and 5 in the web-based supplement.
There were no changes to the article’s conclusions or recommendations.