Measuring anxiety in textual data
What is it about?
This paper proposes an approach to create a dictionary of words indicative of anxiety, which can then be used to detect the presence of anxiety in language. The paper illustrates the creation of a dictionary designed to study political texts. The findings show that some topics debated in parliaments are more likely to generate anxiety (for instance, foreign affairs and immigration). There are a number of dictionaries (or lexicons) available to measure emotions such as anxiety, but these may not be suitable for all purposes. For instance, some general-purpose dictionaries include words such as "war" or "terrorism" as indicators of a speaker's anxiety. When studying political texts, it would not make sense to use these words, since the job of politicians is to speak about issues such as war and terrorism. When using word occurrences to measure psychological traits, it matters to understand which of the words in the dictionary drive the variations in the indicator, and assess whether these words are substantively relevant. The approach proposed in the paper overcomes these limitations, by using word embeddings (a technique in natural language processing) to identify automatically the relevant words related to anxiety in a given domain. The validity of the method is tested by evaluating its accuracy when incorporated in machine learning models. The lexicon created to study political texts is released openly on GitHub and can be used to applications in the field of political science: https://github.com/lrheault/anxiety Moreover, the entire raw corpus of Canadian parliamentary debates used in this paper is available to the research community and can be downloaded here: https://www.lipad.ca/data/ The paper was published in the peer-reviewed proceedings of a conference in the field of computer science (the Conference on Empirical Methods in Natural Language Processing, EMNLP) and can be cited as: Rheault, Ludovic. 2016. "Expressions of Anxiety in Political Texts." Proceedings of the 2016 EMNLP Workshop on Natural Language Processing and Computational Social Science. Austin, Texas: 92-101.
Why is it important?
Anxiety is a key emotion affecting human behavior. There is a large literature on the impact of anxiety in politics, but more generally the emotion is fundamental when studying decision-making in situations of uncertainty. The methodology proposed in this paper may be of interest to psychologists, sociologists, economists, computer scientists, and others interested in measuring anxiety in texts. Additionally, the findings represent in this paper represent the first attempt to measure the anxiety expressed in the language of politicians in parliament. They provide useful insights for scholars interested in politics.
The following have contributed to this page: Ludovic Rheault