Table 3.
Results for regressions of log total cost on service volume and other potential predictors
| Variablea | Model specification | ||||
|---|---|---|---|---|---|
| 1 | 2 | 3 | 4 | 5 | |
| Intercept | 9.44 (0.45) | 9.44 (0.25) | 9.43 (0.16) | 9.43 (0.16) | 9.48 (0.13) |
| Service volume | |||||
| log(doses) | – | 1.10 (0.03) | 1.10 (0.03) | 1.13 (0.04) | 1.04 (0.20) |
| log(doses) squared | – | –0.04 (0.02) | –0.01 (0.02) | –0.01 (0.02) | 0.10 (0.03) |
| Other predictors | |||||
| log(GDP) | – | – | 0.41 (0.16) | 0.31 (0.17) | 0.39 (0.14) |
| Government owned | – | – | –0.13 (0.09) | –0.14 (0.09) | –0.15 (0.08) |
| Hospital | – | – | 0.34 (0.09) | 0.31 (0.09) | 0.27 (0.08) |
| Percent outreach | – | – | 0.11 (0.04) | 0.11 (0.04) | 0.08 (0.03) |
| Percent management | – | – | 0.13 (0.03) | 0.13 (0.03) | 0.13 (0.03) |
| DTP3 per dose | – | – | 0.15 (0.02) | 0.15 (0.02) | 0.13 (0.02) |
| Rural | – | – | – | –0.02 (0.06) | –0.04 (0.06) |
| ANC4 | – | – | – | 0.14 (0.08) | 0.12 (0.07) |
| Wealth ratio | – | – | – | –0.06 (0.04) | –0.07 (0.03) |
| Random effects included | |||||
| Country r.e.s for intercept | + | + | + | + | + |
| Province r.e.s for intercept | + | + | + | + | + |
| Country r.e.s for log(doses) | – | – | – | – | + |
| Variance parameters | |||||
| Error term | 0.86 (0.04) | 0.39 (0.02) | 0.35 (0.02) | 0.35 (0.02) | 0.32 (0.01) |
| SD of country r.e.s, intercept | 0.93 (0.48) | 0.55 (0.27) | 0.34 (0.24) | 0.32 (0.23) | 0.24 (0.21) |
| SD of province r.e.s, intercept | 0.44 (0.10) | 0.19 (0.04) | 0.19 (0.04) | 0.20 (0.04) | 0.14 (0.03) |
| SD of country r.e.s, log(doses) | – | – | – | – | 0.46 (0.23) |
| WAICb | 828.2 | 335.2 | 273.2 | 270.9 | 212.4 |
| Sample size | 316 | 316 | 316 | 316 | 316 |
aCountry and province random effects not shown. Predictors are standardized; thus, fitted coefficients for continuous variables (e.g., log(doses)) represent the increase in log total costs observed for a 1.0 standard deviation increase in the variable. Values in parentheses represent standard errors
bWatanabe-Akaike information criterion (WAIC) describes out-of-sample prediction accuracy for the fitted model, with lower values suggesting better model fit
SD standard deviation, r.e. random effect