Robustness test for list combination in multiple systems estimation

Lanxin Li, University of Edinburgh

Co-authors: Ruth King, University of Edinburgh; Serveh Sharifi Far, University of Edinburgh

Abstract: The multiple systems estimation (MSE) method has been increasingly used in epidemiology to estimate the incidence and prevalence of a disease, using linked data lists. In these applications, log-linear models that consider interactions between pairs of lists are typically applied. As a trade-off between model complexity and interpretability, lists presumed to be dependent or conceptually similar may be proposed to be combined into a single source for model fitting. However, the impact of list combinations on the estimation process is generally ignored. This paper explores the robustness of MSE in terms of using combined lists through simulation studies motivated by a real-world epidemiological study under various scenarios with different dependence structures of lists. Checking list correlations before combining them is encouraged for better insight into modeling.