Model-based inference for small area estimation with sampling weights.

Area of Science:

• Statistics
• Public Health
• Epidemiology

Background:

• Small area estimation (SAE) is crucial for health outcome assessment in data-sparse regions.
• Health surveys often involve complex designs with unequal sampling weights.
• Existing SAE methods, like hierarchical Bayesian models, may not adequately incorporate these sampling weights.

Purpose of the Study:

• To develop a predictive model-based approach for small area estimation (SAE) that incorporates sampling weights from complex survey designs.
• To estimate the prevalence of binary health outcomes for both sampled and non-sampled individuals accurately.
• To apply the developed method for estimating asthma prevalence across districts.

Main Methods:

• Utilized hierarchical Bayesian models within a predictive modeling framework.
• Explicitly incorporated survey sampling weights into the Bayesian models.
• Conducted a simulation study to compare the proposed method against standard and elaborate SAE approaches.
• Applied the method to real-world data for estimating asthma prevalence.

Main Results:

• The proposed weighted hierarchical Bayesian method significantly reduced mean squared error compared to standard SAE approaches.
• The method demonstrated comparable or superior performance to more complex methods when response variables correlated with sampling weights.
• Accurate estimation of asthma prevalence was achieved across districts.

Conclusions:

• The developed predictive model-based SAE approach effectively integrates sampling weights, improving estimation accuracy.
• This method offers a robust solution for health outcome estimation in complex survey settings.
• The approach provides reliable health statistics for policy and planning, particularly for conditions like asthma.