Table 3. Regression results for the association between high-quality delivery facility and neonatal mortality (n = 6,668) 1 .
| Model | F test of IV strength | β (95% CI) |
|---|---|---|
| Main models | ||
| Unadjusted ordinary least squares | NA | 0.006 (-0.001, 0.013) |
| Instrumental variable 2 | 273.6 | -0.023 (-0.046, <-0.001) |
| Robustness assessment of instrumental variable model | ||
| Modifying the threshold of high quality | ||
| 1. Absolute threshold (≥0.80 of 1.00) for classifying facilities as high quality 2 | 310.7 | -0.030 (-0.058, -0.003) |
| 2. Lower threshold (top 33%) for classifying facilities as high quality 2 | 288.6 | -0.033 (-0.058, -0.007) |
| 3. No threshold: continuous quality index, standardized 2 | 95.9 | -0.027 (-0.054, <0.001) |
| Modifying the calculation of the quality index | ||
| 4. Weighted summary of quality items using PCA 2 | 278.5 | -0.026 (-0.048, -0.003) |
| Modifying the quality index to include data from clinical observations | ||
| 5. Alternative quality metric: facility quality and clinical quality of observed deliveries (20 additional items) 2 (n = 4,171) | 328.1 | -0.016 (-0.038, 0.005) |
1 Eighteen observations with missing values on covariates (17 for infant birth weight, 1 for maternal education) excluded from all analyses.
2 Adjusted for the following: urban, logged number of facilities within 20 km, wealth index quintiles, maternal secondary education, maternal age <18, male infant, multiple infant, LBW infant, primiparous mother.
IV: Instrumental variable; PCA: Principal components analysis