Accounting for selection bias in association studies with complex survey data

Area of Science:

• Epidemiology
• Biostatistics
• Survey Methodology

Background:

• Representative data from hidden populations are crucial for understanding sexually transmitted diseases like HIV.
• Simple random sampling is often infeasible for these groups, necessitating complex survey designs.
• Complex sampling can lead to biased estimates if unequal selection probabilities are not addressed.

Purpose of the Study:

• To explore how survey sampling designs can induce selection bias in epidemiological studies.
• To present a novel statistical approach for analyzing complex survey data efficiently and validly.

Main Methods:

• Utilized directed acyclic graphs to illustrate selection bias mechanisms.
• Developed and applied a novel maximum likelihood method for complex survey data analysis.
• Employed simulated data to validate the new method and compare it with existing techniques.

Main Results:

• Demonstrated how specific sampling designs and subject-matter factors can introduce selection bias.
• The proposed maximum likelihood approach optimizes statistical efficiency without compromising validity.
• Simulations confirmed the effectiveness of the novel method compared to other analytic strategies.

Conclusions:

• Effective analysis of complex survey data from hidden populations requires careful consideration of sampling design.
• The novel maximum likelihood method offers a valid and statistically efficient solution for analyzing such data.
• This approach can improve the accuracy of epidemiological estimates for hard-to-reach groups.