Table 2
Performance of alternative approaches to handling missing data on delta-V
| Observed relative risk for restraint use | |||||
|---|---|---|---|---|---|
| Missing data pattern | Delta-V | True RR | Restrict to cases with complete data | Impute ignoring crash outcome | Impute conditional on crash outcome |
| *See text. | |||||
| Missing completely at random | High | 0.40 | 0.40 | 0.32 | 0.40 |
| Low | 0.40 | 0.40 | 0.23 | 0.40 | |
| Missing at random | High | 0.40 | 0.59 | 0.32 | 0.40 |
| Low | 0.40 | 0.69 | 0.18 | 0.40 | |
| Missing not at random* | High | 0.40 | 0.57 | 0.30 | 0.38 |
| Low | 0.40 | 0.73 | 0.19 | 0.42 | |
