Hidden Population Estimation with Indirect Inference and Auxiliary Information

Area of Science:

• Social Network Analysis
• Statistical Modeling
• Public Health Research

Background:

• Conventional survey methods face challenges in sampling hidden or stigmatized populations.
• Respondent Driven Sampling (RDS) is a key method for reaching these groups, but existing imputation techniques introduce bias.
• Accurate estimation of hidden population sizes is crucial for public health interventions.

Purpose of the Study:

• To develop an improved statistical method for estimating hidden population sizes using Respondent Driven Sampling (RDS).
• To address and reduce estimation bias inherent in current RDS imputation techniques.
• To enhance the precision of key RDS-derived metrics, including arrival rates and subgraph characteristics.

Main Methods:

• Modeling RDS as a stochastic process on social network graphs.
• Leveraging auxiliary participant information and indirect inference for improved imputation.
• Developing novel statistical techniques to correct for biased edge imputation in RDS.

Main Results:

• The proposed method significantly reduces bias in estimating the study participant arrival rate, sample subgraph, and overall population size.
• Improved precision was observed in key estimation parameters compared to existing methods.
• Demonstrated successful application in estimating the size of the People Who Inject Drugs (PWID) population in Estonia.

Conclusions:

• The novel indirect inference approach offers a more accurate and precise way to analyze RDS data.
• This method enhances the reliability of estimates for hidden populations, crucial for targeted health programs.
• The findings have direct implications for improving the accuracy of public health surveys in hard-to-reach populations.