Table 4.
Performance of different imputation models in terms of genetic effect estimation: IM0 refers to imputation carried out using the observed frequencies of the α3.7-globin deletions, IM1 includes four SNPs as imputation covariates (rs1800629, rs3211938, rs334, and rs542998), IM2 includes eight phenotypes and socio-environmental factors (Hb, mild anemia, malaria parasite positivity, transect, altitude, and ethnicity), and IM3 includes all variables in IM1 and IM2
| IM0 | IM1 | IM2 | IM3 | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Estimation bias | Estimation bias | Estimation bias | Estimation bias | |||||||||
| Log- | (CI coverage, %) | Log- | (CI coverage, %) | Log- | (CI coverage, %) | Log- | (CI coverage, %) | |||||
| likelihood | likelihood | likelihood | likelihood | |||||||||
| Missing completely at randoma | bias (%) | λ1 | λ2 | bias (%) | λ1 | λ2 | bias (%) | λ1 | λ2 | bias (%) | λ1 | λ2 |
| Pmiss = 10% | −8.80 (1.00) | 0.04 (100) | −0.06 (100) | −8.80 (1.00) | 0.03 (100) | −0.06 (100) | −1.16 (0.13) | 0.10 (100) | −0.07 (100) | −1.14 (0.13) | 0.10 (100) | −0.08 (100) |
| Pmiss = 25% | −8.90 (1.02) | 0.07 (100) | −0.04 (100) | −8.95 (1.02) | 0.07 (100) | −0.06 (100) | −1.02 (0.12) | 0.11 (100) | −0.09 (100) | −1.00 (0.11) | 0.10 (100) | −0.10 (100) |
| Pmiss = 50% | −9.12 (1.04) | 0.12 (100) | −0.09 (100) | −9.16 (1.05) | 0.12 (100) | 0.13 (100) | −0.93 (0.11) | 0.13 (100) | −0.15 (100) | −0.93 (0.11) | 0.13 (100) | −0.16 (100) |
| Missing data from one villageb | ||||||||||||
| Kilimanjaro | ||||||||||||
| Mokala | −0.19 (0.02) | 0.02 (100) | 0.01 (100) | −1.47 (0.17) | 0.24 (80) | 0.13 (100) | 7.83 (0.89) | −0.31 (8) | 0.41 (100) | 2.63 (0.30) | 0.19 (100) | 0.64 (0) |
| Machame | −0.06 (0.01) | 0.01 (100) | <−0.01 (100) | −0.90 (0.11) | 0.06 (100) | −0.08 (100) | 4.39 (0.50) | −0.16 (100) | 0.25 (100) | 1.62 (0.19) | −0.02 (100) | 0.20 (100) |
| Ikuini | 0.21 (0.02) | −0.02 (100) | −0.01 (100) | −0.80 (0.09) | 0.07 (100) | −0.01 (100) | 1.20 (0.13) | −0.09 (100) | −0.05 (100) | 0.51 (0.06) | −0.02 (100) | 0.07 (100) |
| Kileo | 0.46 (−0.05) | −0.05 (100) | −0.07 (100) | −0.19 (0.02) | 0.00 (100) | −0.05 (100) | 2.71 (0.31) | −0.15 (100) | 0.05 (100) | 1.98 (0.23) | −0.07 (100) | 0.14 (100) |
| South Pare | ||||||||||||
| Bwambo | −0.30 (0.03) | 0.01 (100) | −0.07 (100) | −0.30 (0.03) | 0.01 (100) | −0.07 (100) | −0.26 (0.03) | 0.01 (100) | −0.06 (100) | −0.25 (0.02) | 0.01 (100) | −0.06 (100) |
| Mpinji | 0.21 (−0.02) | −0.02 (100) | −0.01 (100) | −1.50 (0.17) | 0.22 (90) | 0.10 (100) | 8.66 (0.99) | −0.34 (0) | 0.42 (100) | 2.80 (0.32) | 0.16 (100) | 0.64 (0) |
| Goha | −0.15 (0.03) | 0.03 (100) | 0.04 (100) | −0.46 (0.05) | 0.03 (100) | −0.05 (100) | 1.07 (0.12) | −0.08 (100) | −0.03 (100) | 0.26 (0.26) | 0.01 (100) | 0.10 (100) |
| Kadando | −0.33 (0.04) | 0.06 (100) | 0.07 (100) | −0.94 (0.11) | 0.11 (100) | 0.03 (100) | 1.27 (0.15) | −0.06 (100) | 0.06 (100) | 1.14 (0.13) | 0.01 (100) | 0.22 (100) |
| West Usambara | ||||||||||||
| Kwadoe | 0.02 (<0.01) | <−0.01 (100) | −0.02 (100) | −0.20 (0.02) | −0.00 (100) | −0.08 (100) | 4.05 (0.46) | −0.10 (100) | 0.35 (99) | 2.73 (0.31) | −0.08 (100) | 0.20 (100) |
| Funta | 0.04 (<0.01) | 0.02 (100) | 0.05 (100) | −0.30 (0.03) | 0.02 (100) | −0.04 (100) | 0.00 (0.00) | 0.00 (100) | −0.02 (100) | 0.12 (0.01) | −0.01 (100) | −0.03 (100) |
| Tamota | −0.66 (0.07) | 0.05 (100) | −0.05 (100) | −1.60 (0.18) | 0.25 (62) | −0.02 (100) | 9.63 (1.10) | −0.38 (4) | 0.38 (96) | 2.74 (0.31) | 0.23 (83) | 0.65 (13) |
| Mgila | 0.24 (0.03) | 0.07 (100) | 0.21 (100) | −0.65 (0.07) | 0.13 (100) | 0.16 (100) | 8.56 (0.98) | −0.21 (87) | 0.47 (58) | 4.35 (0.50) | −0.02 (100) | 0.43 (70) |
| Tanga coast | ||||||||||||
| Mgome | −0.75 (0.09) | 0.04 (100) | −0.32 (100) | −1.32 (0.15) | 0.18 (88) | 0.05 (100) | 8.15 (0.93) | −0.33 (23) | 0.41 (94) | 3.38 (0.39) | 0.10 (100) | 0.59 (8) |
Results based on 100 MCAR data sets in which each data set was analyzed by MICE using 25 imputed data sets generated from chains of 25 iterations and random initial conditions.
Results based on 100 imputed data sets generated by MICE using chains of 25 iterations and random initial conditions.