Xu, Puti ORCID: https://orcid.org/0009-0001-4066-8537 (2025) A Data-Driven Analysis of Pandemic Educational Impacts and the Integration of Large Language Models in Global Learning Environments. PhD thesis, Victoria University.

Abstract

Both COVID-19 and the rise of large language models (LLMs) have significantly influenced global education systems. The COVID-19 pandemic, along with its associated policies, profoundly shaped educational practices. The rapid surge in COVID-19 cases and deaths forced numerous countries to implement lockdown measures, resulting in the transformation of traditional face-to-face learning modes into remote or online learning systems. This shift has substantially impacted various dimensions of education, including academic performance and attendance rates across multiple demographic groups worldwide. In parallel with these policy-driven changes, LLMs have been rapidly developed and adopted across multiple disciplines. These models are now widely utilized as examiners and tutors, influencing both students and educators, and presenting both positive opportunities and notable challenges. Although previous studies have examined the pandemic’s impact on academic performance and attendance rates, comprehensive comparative analyses that address variations across multiple countries, different phases of the pandemic, and diverse ethnic groups remain scarce. Furthermore, most research on LLMs has been limited to a single disciplinary context, primarily within the medical field. Addressing these research gaps, this thesis investigates the educational impacts of the COVID-19 pandemic—including its effects on global academic performance and attendance rates among various ethnic groups—while also examining the multifaceted roles of LLMs in education from the perspectives of both students and educators. The first part of the thesis compares academic performance across different phases of the COVID-19 pandemic—pre-pandemic, during the pandemic, and post-pandemic—using standardized tests such as TOEFL and GMAT. The subsequent chapters analyze the effects of COVID-19 case numbers and vaccination rates on childcare and school attendance among multiple ethnic groups in New Zealand, including M¯aori, European/P¯akeh¯a, Asian, and Pacific populations. The third research focus explores the diverse roles LLMs play in educational development across various disciplines, such as medicine, programming, engineering, and language studies, where LLMs function as both examiners and tutors. This section also systematically assesses the pandemic’s influence on education from both student and educator perspectives. The thesis contributes to existing knowledge in several aspects. First, it highlights both the differences and similarities in standardized test performance across countries, demonstrating that the most significant changes occurred at the onset of the pandemic, and that the patterns of change varied depending on test difficulty, home-edition test requirements, and participant characteristics on a global scale. Additionally, through Spearman correlation and Mean Absolute Percentage Error(MAPE) results derived from multiple machine learning regression analyses, the findings illustrate that while the pandemic greatly influenced attendance rates, the years with the most significant impacts varied. Moreover, the ethnic groups most affected differed between childcare and school attendance rates. Regarding large language models, existing studies reveal that the latest versions of GPT have achieved substantial breakthroughs in examination accuracy across various disciplines; however, these accuracy rates are inconsistent across fields. Both students and educators also report concerns regarding overreliance on these models, as well as ethical issues such as cheating and plagiarism.

Search Google Scholar

Repository staff login