A simulative comparison of respondent driven sampling with incentivized snowball sampling--the "strudel effect"

Area of Science:

• Epidemiology
• Network Analysis
• Public Health Research

Background:

• Respondent driven sampling (RDS) and incentivized snowball sampling (ISS) are key methods for sampling hard-to-reach populations, particularly people who inject drugs (PWID).
• Understanding the comparability of these sampling techniques is crucial for accurate public health surveillance.

Purpose of the Study:

• To assess the statistical differences between simulated Respondent driven sampling (RDS) estimates and an actual incentivized snowball sampling (ISS) sample of people who inject drugs (PWID).
• To evaluate the prevalence of infections (HIV, Hepatitis A, B, C, syphilis, Chlamydia) and behavioral risks using both sampling methods.

Main Methods:

• Simulated RDS samples were generated based on a sociometric ISS sample of PWID in Vilnius, Lithuania.
• Prevalence estimates from simulated RDS samples were compared to the original ISS sample for various health indicators and risk behaviors.

Main Results:

• Simulated RDS samples, when seeded from the largest network component, largely replicated the characteristics of the entire ISS sample.
• No statistically significant differences were found between the large component and the overall ISS sample for key characteristics.
• Over 99% of simulated RDS point estimates fell within the confidence intervals of the original ISS prevalence values.

Conclusions:

• RDS and ISS can produce statistically similar prevalence estimates when population characteristics are concentrated within large, dominant network components.
• The 'strudel effect' describes how isolated network components may be under-sampled, leading to potentially different prevalence values in those specific groups.
• This study highlights the importance of network structure in understanding the representativeness of RDS and ISS sampling methods.