Modelling bias in combining small area prevalence estimates from multiple surveys.

Area of Science:

• Statistics
• Public Health
• Survey Methodology

Background:

• Small area estimation quality can be improved by integrating data from multiple surveys.
• Traditional methods often require individual-level data, which is not always available from commercial surveys.
• Aggregate estimates from commercial surveys often lack methodological transparency, necessitating bias modeling.

Purpose of the Study:

• To develop and apply Bayesian hierarchical models for small area estimation using multiple aggregate survey data.
• To address biases present in commercial survey data.
• To improve the accuracy and precision of smoking prevalence estimates in local authorities.

Main Methods:

• Proposed a series of Bayesian hierarchical models to accommodate additive biases.
• Explored fitting some models within a classical mixed-effects framework.
• Applied the methods to smoking prevalence data from seven surveys in the East of England.

Main Results:

• The developed models successfully adjusted for biases in commercial surveys.
• Incorporated information from all seven diverse surveys.
• Achieved more accurate and precise smoking prevalence estimates at the local authority level.

Conclusions:

• Bayesian hierarchical modeling provides a robust framework for combining aggregate survey data.
• The approach effectively handles biases from non-transparent commercial surveys.
• This method enhances the reliability of small area estimates for public health surveillance.