Methodological Differences in Understandings of Civic & Ethnic Myths in America

• In Event: History, Memory, and Identity in Qualitative and Multi-Method Research (Virtual Session)

Abstract

Bringing the Humans Back In: Methodological Differences in Historically-Grounded Understandings of Civic and Ethnic Myths in America

Over the last several years, the opportunities to deepen our understanding of the dynamics of political discourse have greatly expanded. Increased computing power and the advent of machine learning, coupled with changes to the way public engagement takes place through social media, have been central factors in advancing our understanding of political debates in America and beyond. In this paper, we reflect on these advancements and explore their applicability to unpack how deeply embed historical ideas manifest themselves through social media. To do so, we report our findings from a study of all tweets sent by Presidents Trump and Biden during the 2020 presidential campaign (n=4321). The paper examines differences between, one the one hand, human conducted coding using a qualitative framework used by researchers interested in historical discourses and, on the other hand, unsupervised topic modelling – now commonplace through machine learning applications. Our analysis focuses on the applicability of machine learning applications to uncover complex, historically situated ideas related to race, ethnicity and nationalism – specifically the long-standing divide in American political culture between civic and ethnic nationalism. Through this comparison, we show the promise and perils of the two methods when conducting discourse analysis in these areas. A particular issue for machine learning is the reality that computers are not aware of historical circumstances that may affect perceptions of what political actors say or allude to. Extended rhetorical devices may make allusions, through what has become known as ‘dog whistle politics,’ that cannot be detected without human intervention. This paper therefore compares the findings of both methods, highlights differences between them, and suggests some studies may be better suited to one method or the other.

Authors

• Marcus Closen, University of Toronto

• Alexandre Fortier-Chouinard, University of Toronto

• Catherine Ouellet, University of Toronto

• Robert Schertzer, University of Toronto

• Eric Taylor Woods, University of Plymouth