Respondent2Vec: A Vector-Based Framework for Belief Analysis with Survey Data

• In Event: 20420 - Section on Sociology of Culture Roundtables
  In Refereed Roundtable (90 minute): Table 1: Culture, Cognition, and Beliefs

Abstract

Alongside the broader X2vec framework, this study introduces Respondent2vec, a novel method for ideological analysis that vectorizes survey respondents to explore belief systems. Using this approach, we investigate the causes of belief gaps and discuss the potential for advancing computational cultural analysis with the vector-based framework. By embedding respondents in a multi-dimensional belief space, belief gaps are quantified through Euclidean and cosine distances, capturing variations in belief strength and ideological structure both within and across social groups. To illustrate its application, we analyze belief gaps across gender, education, and income groups using World Values Survey (WVS) data from China (1990–2018). The results reveal that women exhibit lower ideological variance than men, while belief gaps across education and income levels differ depending on the measurement method. These findings demonstrate the potential of Respondent2vec as a robust tool for quantifying ideological patterns, offering fresh insights for social science research. Future applications could extend to cross-national comparisons and network-based analyses.

Authors

• Yangyu Wang, University of Chicago

• Jinfeng Wu, Hong Kong University of Science and Technology

• Xuewei Xu