Two-Phase Sampling Designs for Data Validation in Settings with Covariate Measurement Error and Continuous Outcome.

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Measurement errors are common in data collection and can bias statistical analyses.
• Researchers often use validation subsamples to correct for measurement error.
• Simple random sampling (SRS) is a common but often inefficient method for selecting validation samples.

Purpose of the Study:

• To propose optimal and nearly-optimal sampling designs for validation samples in the presence of measurement error.
• To improve the efficiency of statistical estimators (model-based and design-based) by enhancing validation sample selection.
• To compare the efficiency of proposed designs against traditional SRS.

Main Methods:

• Applied and extended concepts from two-phase sampling literature.
• Developed optimal and nearly-optimal designs for validation sample selection.
• Evaluated designs using theoretical analysis, simulations, and a real-world HIV cohort study dataset.

Main Results:

• Sampling schemes that leverage information from error-prone variables are substantially more efficient than SRS.
• Both model-based and design-based estimators benefit from improved sampling designs.
• The most effective design is dependent on the specific analysis method employed.

Conclusions:

• Optimized validation sample selection can substantially increase the efficiency of statistical analyses affected by measurement error.
• The choice of optimal sampling design is analysis-specific.
• Proposed methods offer significant improvements over standard SRS for measurement error correction.