Skip to main content

Table 4 Final model to predict crude death rate, South Sudan (2013–2018)

From: A method for small-area estimation of population mortality in settings affected by crises

Fixed effect Relative rate 95% CI p-value
Intercept 0.00014 0.00008 to 0.00022  < 0.001
Region
Northeast [Ref.]   
Northwest 0.54 0.41 to 0.72  < 0.001
Southern 0.80 0.51 to 1.25 0.326
Main livelihood type
Agriculturalists [Ref.]   
Agro-pastoralists 0.82 0.55 to 1.22 0.329
Pastoralists 1.24 0.69 to 2.23 0.478
Displaced to Protection of Civilians camps 0.52 0.34 to 0.81 0.004
Rate of insecurity events (per 100,000 people per month, lag = 4 months)
0 [Ref.]   
0.01 to 0.99 1.16 1.02 to 1.32 0.021
 ≥ 1.00 1.32 1.08 to 1.62 0.008
Uptake of measles vaccine (doses administered per 100,000 people per month)
0 [Ref.]   
0.1 to 199.9 0.83 0.69 to 0.99 0.042
200.0 to 399.9 0.76 0.60 to 0.97 0.025
 ≥ 400.0 0.56 0.43 to 0.74  < 0.001
Terms of trade purchasing power index (Kg of white wheat flour that an average goat can be exchanged for; 3 months running average, lag = 3 months) 0.992 0.987 to 0.996  < 0.001
Rate of violent incidents affecting humanitarian staff (per 100,000 per month, lag = 4 months)
0 [Ref.]   
 ≥ 0 1.19 1.04 to 1.36 0.010
Incidence rate of confirmed or probable measles cases (per 100,000 per month)
0 [Ref.]   
 ≥ 0 1.30 1.15 to 1.47  < 0.001
Model performance metric Value Notes
Dawid–Sebastiani score (internal prediction) 26.9 \(\frac{{\left( {{\text{observed }} - {\text{predicted}}} \right)^{2} }}{{{\text{variance}}}} + 2 \times \log \left( {{\text{variance}}} \right)\)
Dawid–Sebastiani score (out-of-sample prediction) 29.2 Based on tenfold cross-validation (CV)
Relative bias (on CV)  − 0.064 \(\frac{{{\text{predicted}} - {\text{observed}}}}{{{\text{observed}}}}\)
Relative 95% precision (mean across strata on CV) 1.011 \(\frac{{0.5 \times \left( {{\text{upper}} 95\% {\text{CI}} - {\text{lower}} 95\% CI} \right)}}{{{\text{predicted}}}}\)
Coverage of 80% confidence intervals (on CV) 0.754 Proportion of stratum observations falling within the confidence interval of the prediction
Coverage of 95% confidence intervals (on CV) 0.901  
  1. Note that the predictors and values below differ from the original model presented in the study report, as they arise from an improved fitting procedure. Random effects are omitted