Talk Authoritarian to Me: Measuring Authoritarianism Using Speech

• In Event: The Causes and Consequences of the Rise of Populist Sentiments

Abstract

Developments in machine learning have provided researchers with the methodological tools needed to extrapolate linguistic indicators of psychological differences from a variety of digital sources. This "text as data" approach has been used to uncover linguistic cues of personality traits in samples of the mass public and generate validated personality measures for political elites. While such methods tend to cleanly delineate linguistic patterns of speech linked to traits directly observed in the population (e.g. Big 5), they may have more difficultly detecting speech patterns related to dispositions that are conditional upon one's social context. However, one such disposition – authoritarianism – has structured the recent rise of right-wing populist leadership across Western democracies, making an understanding of the rhetoric associated with this trait of utmost importance.

This paper represents an initial examination of the language employed by individuals across the child-rearing authoritarian scale, under a variety of model specifications. Using a dataset of over 3,000 respondents who completed a 20 minute free-writing task as well as the child-rearing measure of authoritarianism, we apply six machine learning algorithms to train text-based classifiers of authoritarianism. Building upon earlier work on the conditional effects of authoritarianism, we apply our algorithms to the following subsets of respondents: 1) high/low education, 2) high/low cognitive fatigue, 3) high/low political sophistication, and 4) Whites/non-Whites. We compare the performance of our authoritarianism classifiers across these sub-groups. Our goal is to illuminate the circumstances under which authoritarianism can influence speech patterns, and construct an authoritarianism dictionary.

Authors

• Julie Wronski, University of Mississippi

• Jonathan D. Klingler, University of Mississippi

• Adam Ramey, New York University Abu Dhabi

• Gary E. Hollibaugh, University of Pittsburgh