MODELING THE VISIBILITY DISTRIBUTION FOR RESPONDENT-DRIVEN SAMPLING WITH APPLICATION TO POPULATION SIZE ESTIMATION

Area of Science:

• Statistics
• Epidemiology
• Social Network Analysis

Background:

• Respondent-driven sampling (RDS) is crucial for studying hidden populations but relies on self-reported network sizes, which are prone to bias.
• Current RDS estimators approximate inclusion probabilities using self-reported network size (degree), leading to potential inaccuracies.

Purpose of the Study:

• To enhance the successive sampling population size estimation (SS-PSE) framework for RDS data.
• To introduce a "visibility" measurement error model to replace unreliable self-reported network sizes.
• To improve the accuracy of population size and prevalence estimations from RDS.

Main Methods:

• Developed an enhanced SS-PSE framework incorporating a measurement error model for participant "visibility."
• Modeled the number of individuals a participant can recruit.
• Applied the visibility SS-PSE framework to RDS data from three populations in Kosovo.

Main Results:

• The visibility model effectively smooths degree distributions and handles missing/invalid network size data.
• Demonstrated the performance of the enhanced SS-PSE framework on real-world RDS data.
• Inferred visibilities provide a more robust measure than self-reported network sizes.

Conclusions:

• The proposed visibility modeling framework offers a significant improvement over traditional SS-PSE methods for RDS.
• This approach can mitigate biases associated with self-reported network sizes in hidden population studies.
• The framework shows potential for extension to prevalence estimation in future research.