Combining cluster sampling and link-tracing sampling to estimate the size of a hidden population: asymptotic properties of the estimators

TL;DR

This paper analyzes the asymptotic properties of estimators for hidden population size using combined cluster and link-tracing sampling, extending previous models to account for heterogeneity in link probabilities.

Contribution

It establishes conditions under which the maximum likelihood estimators are consistent and asymptotically normal, considering heterogeneity in link-probabilities.

Findings

Both unconditional and conditional MLEs are consistent.

The estimators have different asymptotic normal distributions.

Conditions for model validity are provided.

Abstract

F\'elix-Medina and Thompson (2004) proposed a variant of link-tracing sampling to estimate the size of a hidden population such as drug users, sexual workers or homeless people. In their variant a sampling frame of sites where the members of the population tend to gather is constructed. The frame is not assumed to cover the whole population, but only a portion of it. A simple random sample of sites is selected; the people in the sampled sites are identified and are asked to name other members of the population which are added to the sample. Those authors proposed maximum likelihood estimators of the population size which derived from a multinomial model for the numbers of people found in the sampled sites and a model that considers that the probability that a person is named by any element in a particular sampled site (link-probability) does not depend on the named person, that is, that the probabilities are homogeneous. Later, F\'elix-Medina et al. (2015) proposed unconditional and conditional maximum likelihood estimators of the population size which derived from a model that takes into account the heterogeneity of the link-probabilities. In this work we consider this sampling design and set conditions for a general model for the link-probabilities that guarantee the consistency and asymptotic normality of the estimators of the population size and of the estimators of the parameters of the model for the link-probabilities. In particular we showed that both the unconditional and conditional maximum likelihood estimators of the population size are consistent and have asymptotic normal distributions which are different from each other.