LATENT DEMOGRAPHIC PROFILE ESTIMATION IN HARD-TO-REACH GROUPS

Area of Science:

• Social Sciences
• Statistics
• Network Analysis

Background:

• Standard surveys often exclude hard-to-reach groups (e.g., homeless, HIV+ individuals).
• Demographic data for these populations, especially in developing nations, is frequently unavailable.
• Existing network-based methods for reaching these groups can be complex and require specialized sampling.

Purpose of the Study:

• To present statistical models for estimating demographic characteristics of hard-to-reach groups.
• To introduce Aggregated Relational Data (ARD) as a method for gathering information on these populations.
• To develop techniques for estimating latent demographic profiles using ARD.

Main Methods:

• Utilized social network structure to estimate demographic characteristics via Aggregated Relational Data (ARD).
• Proposed a Bayesian hierarchical model for estimating latent demographic profiles.
• Developed two estimation techniques: a Markov-chain Monte Carlo algorithm and a simple estimate based on a missing data approach for new data collection.

Main Results:

• Successfully estimated age and gender profiles for six hard-to-reach groups using ARD.
• Demonstrated the utility of ARD in standard surveys without special sampling requirements.
• Evaluated the accuracy of the proposed simple estimation technique through simulation studies.

Conclusions:

• ARD provides a feasible and effective method for collecting demographic data on hard-to-reach populations.
• The proposed Bayesian model and estimation techniques offer valuable tools for social science research.
• This approach enhances the inclusivity of surveys and improves understanding of marginalized communities.