Dissertation

Methods to estimate mortality in small populations: review, evaluation, and application

Abstract

Background: The mortality of a population often serves as a proxy for its health. As such, knowledge about human mortality is not only of academic but also of public interest. In small populations, however, the study of mortality is challenged by erratic 
death counts. As a consequence, researchers have proposed different methods to estimate smooth mortality schedules. In practice, choosing among a number of methods is not a trivial task as in-depth knowledge about the advantages and disadvantages of a method is needed. To address this challenge, this thesis reviews, evaluates, and applies promising demographic methods to estimate full mortality schedules in small populations. 
Objectives: Chapter 1 sets out to provide the necessary demographic and statistical background as well as perspectives on methodological research. Chapter 2 aims to provide a state-of-the-art overview of three promising demographic methods; namely TOPALS (Gonzaga and Schmertmann 2016), D-splines (Schmertmann 2021b), and a Bayesian model (Alexander et al. 2017). Chapter 3 is dedicated to an evaluation of TOPALS and D-splines; two methods that incorporate knowledge about human mortality. Its focus is on providing evidence about their robustness in different settings. Chapter 4 aims to provide insights on temporal and spatial patterns at the subnational level in Austria, Czechia, and Slovakia. The thesis concludes with a discussion of the results and an outlook on future research in Chapter 5.
Methods and results: By leveraging the four phases of methodological research by Heinze et al. (2024), Chapter 2 provides an in-depth overview of current research. It shows that, among others, there is a lack of comparison studies, detailed case studies, and software. Through a systematic simulation study, Chapter 3 shows that both TOPALS and D-splines are sensitive to the population size and the incorporated demographic knowledge. At the same time, the chapter poses new questions about simulation designs to study these types of methods. Chapter 4 applies the Bayesian model by Dharamshi et al. (2025) to 243 regions of Austria, Czechia, and Slovakia from 2003–2019. With respect to life expectancy at birth, the results show that Austria is the most homogenous country and that Slovak men are the most disadvantaged. Furthermore, the study shows that Czech and Slovakian regions are consistently lagging behind Spain, a leading country in terms of life expectancy at birth. It also confirms that the Bayesian model by Dharamshi et al. (2025) is a viable option to study mortality patterns across time and space. Overall, the perspectives offered in this thesis, should prove useful in advancing rigorous methodological research that addresses the problems of practitioners.