Repository logo
 

Application of Data Science Methodologies to Explore, Model and Predict Population-Level Subjective Wellbeing Outcomes Using the New Zealand Integrated Data Infrastructure (IDI).

Files

Thesis(2.17 MB)

Date

2024

Authors

Supervisor

Stewart, Tom
Duncan, Scott
Pacheco, Gail

Item type

Thesis

Degree name

Doctor of Philosophy

Journal Title

Journal ISSN

Volume Title

Publisher

Auckland University of Technology

Abstract

The increasing recognition of population wellbeing as a key indicator of societal prosperity has spurred governments worldwide to formulate policies aimed at enhancing their citizens’ wellbeing. In New Zealand, the General Social Survey (GSS) provides subjective wellbeing measures for a subset of the population (~10,000 individuals). Although the GSS sample is representative of the New Zealand population across various sociodemographic groups, including factors such as ethnicity and socio-economic status, the available wellbeing data only covers ~0.2% of the population. This limitation affects the granularity of insights it can offer, hindering in depth exploration into (1) the determinants of population wellbeing and (2) the effects of government policies on wellbeing, especially within smaller and marginalised subgroups of the population who may be of high policy interest (e.g., people living in deprived regions). To address this challenge, detailed population-level wellbeing data that is sensitive enough to reflect the effects of policy change is essential. The microdata available within New Zealand's Integrated Data Infrastructure (IDI) presents an opportunity to bridge this gap. By leveraging the IDI's extensive dataset, this thesis aims to apply advanced data science methodologies to systematically explore, model, and predict GSS-based subjective wellbeing outcomes for the broader New Zealand population. To begin, a systematic scoping review was conducted to provide a comprehensive overview of the current literature around modelling health and wellbeing outcomes using machine learning. This foundational review identified prevailing trends, methodologies, and notably, the scarcity of studies predicting population wellbeing outcomes. Next, the thesis delved into understanding New Zealand’s current wellbeing through detailed cross-sectional and trend analysis of GSS data. Key associations with subjective wellbeing were observed across diverse demographic categories, including age, gender, ethnicity, and socio-economic status. These insights set the stage for the next part of the thesis focused on modelling subjective wellbeing outcomes. The core of the thesis focussed on the development and validation of statistical models for predicting subjective wellbeing within the GSS population. Census-level administrative variables were utilised as predictors, and the Random Forest emerged as an effective model for showcasing how data science techniques can predict wellbeing outcomes. Despite its strengths, capturing the variability of subjective wellbeing proved challenging, prompting a critical discussion on the need for methodological refinement. Lastly, the research extended to applying and then validating these predictive models in the broader New Zealand census population. While predictions generally aligned with GSS estimates across different demographic groups, several disparities underscored the complexities of accurately modelling wellbeing at the population level, which were discussed in the final chapter of this thesis. The thesis demonstrated the feasibility of predicting subjective wellbeing outcomes in a population with existing routinely collected data and advanced analytical techniques. It acknowledged the challenges of modelling subjective outcomes, and suggested avenues to enhance the precision and applicability of this research methodology. This work contributes to the broader understanding and enhancement of population wellbeing, underscoring the importance of comprehensive, representative data for informing policy and societal progress.

Description

Keywords

Source

DOI

Publisher's version

Rights statement

Permanent link

http://hdl.handle.net/10292/17900

Collections

Doctoral Theses