Opinion correlation across societies
A comparison of policy-opinion structure in surveys and social media across the US, Europe, and China.
Contents
Research question
How are opinions on different policy topics associated, and how does this structure differ across societies and data sources? The public repository compares nine topics using ANES, EVS, and WVS surveys alongside region-specific Twitter and Weibo samples.
Method
The analysis calculates pairwise opinion correlations, topic-semantic similarity, and the effective dimensionality of respondent-by-topic opinion matrices. It includes shuffled baselines for social-media correlations, topic-keyword embeddings, spectral measures, bootstrap intervals, and robustness analyses for sampling and missingness.
The public README documents survey recoding, LLM-assisted labeling, and BERT relevance and opinion models used to produce cleaned user-level opinions.
Status
Analysis code and topic reference material are public in opinion-structure-across-societies. Its import script explicitly identifies opinion_correlation as the earlier working directory. Raw social-media posts are not distributed; survey microdata are subject to their providers’ terms.
This page describes the research workflow. It does not report numerical findings, publication status, or an individual authorship breakdown.