Correction for Participation Bias in Nonprobability Samples Using Multiple Reference Surveys.

Area of Science:

• Survey methodology
• Statistical inference
• Health research

Background:

• Nonprobability sampling is increasingly used in health research.
• Participation mechanisms in these samples are often unknown, leading to potential bias in estimates and associations.
• Existing methods for statistical inference from nonprobability samples are limited to a single reference sample.

Purpose of the Study:

• To propose a general framework for addressing participation bias in nonprobability samples using multiple reference surveys.
• To extend current statistical inference capabilities beyond single-reference sample limitations.
• To focus on calibration estimators for practical implementation and flexibility.

Main Methods:

• Developed a general framework accommodating multiple reference surveys.
• Focused on calibration estimators, a flexible special case within the framework.
• Proposed two variance estimation methods: Taylor linearization and leave-one-out jackknife.

Main Results:

• The proposed framework successfully addresses participation bias.
• Raking ratio calibration estimators showed satisfactory performance, especially with dispersed participation probabilities.
• Variance estimates for continuous outcomes were markedly smaller using the proposed methods.

Conclusions:

• The new framework and calibration estimators effectively mitigate bias in nonprobability samples.
• The methods offer practical advantages, particularly with limited microdata access.
• Demonstrated utility in a real-world study of Canadian working adults.