Table 3.
Estimates of 3 Target Parameters Using 2 Approaches to Handle Missing Data in the Example Study of Maternal Mental Illness and Child Behavior, Longitudinal Study of Australian Children (Waves 1–3), 2004–2008
| Parameter | Estimate (SE) | 95% CI | Is Estimate Reliablea if We Adopt: | |
|---|---|---|---|---|
| m-DAG E? | m-DAG J? | |||
| Proportion of mentally ill mothers at wave 1 | ||||
| Available-case analysis | 0.21 (0.01) | 0.20, 0.22 | No | No |
| MICE | 0.21 (0.01) | 0.20, 0.23 | No | No |
| Mean SDQ scoreb of children at wave 3 | ||||
| Available-case analysis | 7.48 (0.09) | 7.31, 7.65 | No | No |
| MICE | 7.74 (0.09) | 7.57, 7.90 | No | No |
| Regression-adjusted difference in mean SDQ scorec | ||||
| Available-case analysis | 0.59 (0.20) | 0.20, 0.98 | Yes | No |
| MICE | 0.64 (0.21) | 0.23, 1.06 | Yes | No |
Abbreviations: CI, confidence interval; m-DAG, missingness directed acyclic graph; MICE, multiple imputation by chained equations; SDQ, Strengths and Difficulties Questionnaire; SE, standard error.
a This indicates whether the estimate can be considered reliable according to which m-DAG from Figure 2 is adopted, based on the recoverability of the parameter in that m-DAG.
b Range, 0–40. A higher score indicates increased behavioral difficulties.
c Comparing mentally ill mothers with non–mentally ill mothers.