A multiple-record systems estimation method that takes observed and unobserved heterogeneity into account

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Background:

• Estimating population size from multiple lists is crucial in epidemiology.
• Traditional models often assume homogeneity of capture probabilities and list independence, which are frequently violated in real-world data.
• Heterogeneity of individuals and lack of control over data sources are common in epidemiological studies.

Purpose of the Study:

• To present a novel statistical model for estimating unknown population size from multiple lists.
• To address violations of homogeneity and marginal independence assumptions common in epidemiological data.
• To extend existing methods for estimating population undercount, particularly within strata defined by covariates.

Main Methods:

• Development of a statistical model for population size estimation from multiple lists.
• Incorporation of methods to handle heterogeneity in capture probabilities and list dependencies.
• Application of profile log-likelihood techniques for confidence interval estimation.

Main Results:

• The proposed model provides a robust method for population size estimation when standard assumptions are unmet.
• The model allows for the estimation of undercount within specific strata defined by categorical covariates.
• Confidence intervals for stratum-specific undercounts are derived using profile log-likelihood methods.

Conclusions:

• The developed model offers a significant advancement for population size estimation in complex epidemiological settings.
• It provides a flexible framework for analyzing data from multiple imperfect sources.
• The methods enable more accurate assessment of undercounting, crucial for public health surveillance and resource allocation.