Event Details

Statistics seminar in the School of Mathematical Sciences.

Speaker: Dr Lanxin Li from the University of Edinburgh.

Title: Estimating hidden population sizes using multiple systems estimation with censored data

Abstract: Estimating the size of hidden or hard-to-reach populations, such as people who inject drugs, is an important problem in epidemiology and public health. Multiple systems estimation (MSE) is a common approach used in epidemiological surveillance studies to estimate the size of a target population. In MSE, multiple data lists observing individuals in the target population are linked and collated in the form of an incomplete contingency table, providing the total number of individuals observed by each distinct combination of lists. Log-linear models are typically fitted to the observed cell counts, allowing for interactions between different lists and the associated number of individuals not observed by any lists estimated from the fitted model. This estimated number is combined with the total number of observed individuals to provide an estimate of the total population size.

The observed cell counts, unfortunately, are often imperfect or only partially available. One of the common issues arising in epidemiological data is that the lists can include individuals who are not members of the target population, leading to censored cell counts that only provide the upper bounds for the true cell count. Using the censored counts for statistical modelling leads to biased population size estimators. We propose to extend the standard log-linear model in the classical framework, taking into account the censoring information. We demonstrate the substantial level of bias that can be introduced when ignoring the censoring for simulated data, before applying the approach to real data relating to drug users.