2004年-世界发展银行全球_Environmental_Determinants_of_Child_Mortality_in_Rural_China___A_Competing_Risks_Approach_26页_300kb

Environmental Determinants of Child Mortality in Rural China: A Competing Risks Approach

Core Content

This paper investigates the environmental determinants of child mortality in rural China using a competing risks model, with the aim of evaluating whether considering cause of death provides more accurate or meaningful insights into the effectiveness of policy interventions compared to a simpler all-cause mortality model. The study is based on data from the 1992 China National Health Survey (CHS), which provides detailed cause-of-death information for children under five.

Main Points

• Data Source: The 1992 CHS was conducted by the China Statistical Bureau and UNICEF, covering 29 provinces (excluding Tibet) and collecting information on the cause of death for children under five.
• Cause of Death Information: While the CHS data on cause of death may be noisy, it is relatively reliable compared to other methods, such as verbal autopsies, which are more accurate but also more expensive.
• Competing Risks Model: The paper uses a competing risks model to analyze the effect of environmental variables (safe water, sanitation, and clean cooking fuel) on child mortality, as well as the impact of female education.
• Hazard Models Compared: Three models are used: Weibull (W), Piece-wise Weibull (PWW), and Competing Risks (CR). The results show that the competing risks model does not significantly differ from the simpler hazard models in terms of overall policy conclusions, but it supports the causal interpretation of the impact of safe water access on child mortality.
• Policy Simulations: The study simulates four policy interventions:
  • Universal access to safe water
  • Universal access to basic sanitation
  • Use of clean cooking fuels
  • Universal female primary education attainment
• Key Findings:
  • Access to safe water has the most significant impact on reducing child mortality, with a substantial effect even when considering different models.
  • Policies targeting poor households or poor localities yield greater health benefits than untargeted policies.
  • The elasticity of U5 mortality reduction with respect to safe water access is higher in poor areas than in all rural areas.
  • The CR model helps in validating causal relationships by showing that certain causes (like birth-related deaths and neonatal tetanus) are not influenced by access to safe water, reinforcing the causal interpretation of the model's findings.
  • Environmental variables such as sanitation and clean cooking fuels do not show statistically significant effects on child mortality in the models tested.

Key Environmental Variables

• Access to Safe Drinking Water: Shows the strongest and most consistent impact on reducing child mortality across all models.
• Basic Sanitation Facilities: Has a less significant effect, with the impact not being statistically significant in some models.
• Clean Cooking Fuels: Also has a limited effect, though the study suggests that more research is needed to determine its true impact.
• Female Primary Education Attainment: Has a significant impact on child survival, but only in the PWW model.

Methodological Considerations

• The study uses cluster means of environmental variables to address potential endogeneity, assuming that at the community level, environmental conditions are more exogenous than at the household level.
• The cause of death data is imputed using a multinomial logit model, assigning unknown causes to the most probable category based on socio-economic and demographic characteristics.
• The paper acknowledges that unobserved heterogeneity is a challenge in hazard models but argues that it is unlikely to significantly affect the results given the Weibull specifications used.
• The PWW model is used to capture age-specific differences in mortality causes, with three age intervals: first month, 12 months, and 60 months.

Policy Implications

• The results suggest that targeted interventions in poor areas or households can have a more substantial impact on child survival than untargeted ones.
• The study supports the use of cause of death information in future demographic and health surveys (DHS), particularly for high-mortality countries.
• It also recommends that future DHS instruments be modified to include cause of death data in a cost-effective manner, as it can improve the analytical strength of the surveys and help validate the causal effects of environmental interventions.

Conclusion

The paper concludes that while the competing risks model does not significantly alter overall policy conclusions about environmental impacts on child mortality, it provides important validation for causal relationships, especially regarding the effect of safe water access. The findings support the idea that targeted policies can yield better results, and that cause of death data is a valuable tool for improving the accuracy and interpretation of child mortality analyses in low-income settings.