Multilevel Regression and Poststratification: A Modeling Approach to Estimating Population Quantities From Highly...

Area of Science:

• Epidemiology
• Biostatistics
• Population Health

Background:

• Recruiting representative samples for large-scale population health studies is challenging.
• Nonparticipation and attrition can lead to biased samples, even in well-designed studies.
• Addressing participation bias is crucial for accurate population health estimation.

Purpose of the Study:

• To evaluate the utility of multilevel regression and poststratification (MRP) for correcting participation bias in large cohort studies.
• To assess MRP's effectiveness in estimating population descriptive quantities.
• To compare MRP with traditional survey weighting methods.

Main Methods:

• An extensive case study using baseline data from the Australian "Ten to Men" longitudinal study of male health (2013-2014).
• Application of multilevel regression and poststratification, a method previously used for election forecasting.
• Analyses conducted using the RStan Bayesian computational package.

Main Results:

• MRP demonstrated greater consistency and precision across diverse population subsets compared to conventional survey weights.
• Estimates for smaller population subgroups showed significant shrinkage towards the national average.
• The method proved effective in addressing potential participation bias.

Conclusions:

• Multilevel regression and poststratification is a promising analytical technique for improving the accuracy of population health estimates from large-scale surveys.
• MRP offers a robust approach to mitigate biases arising from nonresponse and attrition in cohort studies.
• This method enhances the reliability of descriptive quantities derived from health research data.