Evaluation of Respondent-Driven Sampling Prevalence Estimators Using Real-World Reported Network Degree.

Area of Science:

• Epidemiology
• Network Science
• Statistical Modeling

Background:

• Respondent-driven sampling (RDS) is a key method for estimating prevalence of traits or diseases in hidden or marginalized populations.
• Accurate prevalence estimation is crucial for public health interventions and resource allocation.

Purpose of the Study:

• To evaluate the performance of Respondent-driven sampling (RDS) estimators under diverse conditions.
• To assess the impact of network structure, trait prevalence, and homophily on RDS accuracy.
• To compare different RDS estimators in simulated and real-world network settings.

Main Methods:

• Utilized large simulated social networks (N=20,000) based on real-world RDS degree data.
• Employed an empirical Facebook network (N=22,470) for evaluating estimators.
• Assessed estimators for both binary and categorical trait prevalence.

Main Results:

• Prevalence estimate variability is higher with real-world network degrees compared to assumed Poisson distributions, leading to reduced coverage.
• Newer RDS estimators show good performance when sample size is a significant fraction of the population.
• Bias in prevalence estimates emerges when the total population size remains unknown.

Conclusions:

• The choice of the optimal Respondent-driven sampling (RDS) estimator is context-dependent.
• Study-specific considerations, including statistical properties and population knowledge, are vital for selecting the best RDS estimator.
• Understanding network characteristics is crucial for improving the reliability of RDS in prevalence studies.