Measuring Intergroup Trust with Social Media Data: A Case Study of How Americans Evaluate Chinese before and during the COVID-19 Pandemic

Intergroup trust, a crucial element of social cohesion, is the focus of this study. We introduce a novel computational method, guided by theory, to measure intergroup trust using social media data. This method, unlike traditional questionnaire approaches, is both simple and effective. Using Google News Word2Vec embeddings, we represent six dimensions derived from three classical theoretical models in social perception. This semantic space representation allows us to assess the possibility of synthesis using principal component analysis. The key finding is that the six dimensions can be combined into two fundamental dimensions: Warmth-Communion (WC) and Competence-Agency (CA). The WC dimension emphasizes subjective assessments such as trustworthiness and moral character. On the other hand, the CA dimension emphasizes objective abilities. Consequently, the WC dimension offers a new way to measure intergroup trust using textual data.

We used this new measure to analyze the perception of the Chinese by Americans before and during the COVID-19 pandemic. Our analysis was based on a dataset of 3 million tweets from Twitter originating from American IP addresses. These tweets underwent screening by an automated algorithm to identify those that were relevant to expressing opinions about the Chinese. Each tweet was scored based on the WC and CA dimensions. The results show that Americans' trust in the Chinese declined significantly after the pandemic outbreak. Specifically, (1) there was a statistically significant drop in Americans' evaluation of the Chinese in the WC dimension, but (2) no significant change in the CA dimension. The results confirm that there is a growing trend of discrimination against Chinese individuals by Americans. However, Americans' perception of the competence of Chinese individuals has remained consistent even after the onset of the COVID-19 pandemic. This study also confirms the validity of utilizing a theory-guided computational approach to assess intergroup trust through textual data.