Using Arguments to Persuade: Experimental Evidence

Abstract

Models of communication, deliberation, and persuasion highlight message quality, i.e., the credibility of a message, to be a major determinant of persuasiveness. The recent literature on persuasion and narratives argues that convincing messages are backed up by stories or models, i.e., by arguments. As yet, there is no common empirical measure of message persuasiveness. We propose counting arguments used to support claims as a simple, context-independent empirical measure of message persuasiveness. We show that this measure is easy to implement using simple Natural Language Processing and Machine Learning techniques, and that it has predictive value with regard to changes in beliefs and behavior. In a two-wave experiment, we collected voting intentions and text data from randomized chat interactions before a ballot and voting choices after a ballot. We find that the increased use of arguments induces more vote changes.

Keywords: Chat, Persuasion, Opinion Change, Survey Experiment, Textual Analysis, Voting

JEL Classification: D01, D04, D72, D83