Explanatory statistical modelling of influences of demographic experience on political identity

Abstract

This thesis seeks to create, and relate, holistic depictions of demographic and political identity, using data from the Australian National University’s 2016 Australian Election Study. We thus create new spectra of identity, and produce a multivariate model to explore demographic influences on political ideology. From a statistical perspective, we build a foundation for, and expand upon, techniques for model selection. From a political science perspective, we create a modern, data-driven Australian political spectrum, produce key findings on how this spectrum is influenced by demography, and build a stronger understanding of the ways political thought can diverge from expectations. We first note that the 2016 Australian Election Study is too complicated to model directly. Instead, we seek simple representations of demographic and political identities. After exploring the literature on mathematical ways of reducing the dimensions of variables, we produce a new spectrum of political ideology for Australia, as well as a new spectrum of demographic identity. The spectra are mathematically designed to be comprised of axes representing the issues that most unite, and most divide, Australians. Our new political spectrum is interpreted in the context of current conceptions of political thought. We explore models to connect our demographic and political spectra, with the goal of explaining relationships between them in a clear and concise manner. We do not attempt to make predictions about individuals, but rather, explain relationships that exist in the population at large. We seek to build a multivariate model, to describe all dimensions of our new political spectrum simultaneously. We introduce, construct, and prove results relating to four candidate models, each of which elucidates key relationships in subtly different ways. We explain that a traditionally-used model to predict multiple outcomes simultaneously, the multivariate regression model, makes assumptions about the data that cannot be justified. We develop new tools for comparing models. In our model selection process, we put emphasis on the models’ errors; it is dangerous to ignore the ways in which our expectations might be wrong. Our tools for comparing the four candidate models thus place emphasis on selecting a model that is most accurate when it comes to its error distribution’s assumptions. Our tools are placed in the context of the historical literature on the issue, as well as in conjunction with other criteria by which a model could be selected. Our model’s results are interpreted in a political science context. We explore, with reference to other research in the area, emergent associations between demographic and political identity. These include the positive relationships between socio-economic status and education, and social inclusivity, and between stage of life and trust in authority, as well as the negative relationships between stage of life and social inclusivity, and socio-economic status and high social spending. We discuss the relationships between different political views that persist after accounting for demographic influences. These associations are of particular relevance in light of trends towards populist movements around the world, with evidence of low trust in ‘elites’ being common especially among a small cluster of socially non-inclusive people. With a process established for relating underlying constructs in large and complex surveys, our methodologies have the capacity to be implemented to surveys of varying geographical and temporal origin. This leads to two paths for future analysis: questioning how the associations between demographic and political identity have changed over time in Australia; and questioning how these associations differ internationally.